]> git.proxmox.com Git - mirror_zfs-debian.git/blob - module/zfs/dmu_send.c
projects / mirror_zfs-debian.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Imported Upstream version 0.6.5.3
[mirror_zfs-debian.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 arc_flags_t aflags = ARC_FLAG_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 arc_flags_t aflags = ARC_FLAG_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 uint64_t offset;
525 arc_flags_t aflags = ARC_FLAG_WAIT;
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 static int
840 dmu_adjust_send_estimate_for_indirects(dsl_dataset_t *ds, uint64_t size,
841 uint64_t *sizep)
842 {
843 int err;
844 /*
845 * Assume that space (both on-disk and in-stream) is dominated by
846 * data. We will adjust for indirect blocks and the copies property,
847 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
848 */
849
850 /*
851 * Subtract out approximate space used by indirect blocks.
852 * Assume most space is used by data blocks (non-indirect, non-dnode).
853 * Assume all blocks are recordsize. Assume ditto blocks and
854 * internal fragmentation counter out compression.
855 *
856 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
857 * block, which we observe in practice.
858 */
859 uint64_t recordsize;
860 err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
861 if (err != 0)
862 return (err);
863 size -= size / recordsize * sizeof (blkptr_t);
864
865 /* Add in the space for the record associated with each block. */
866 size += size / recordsize * sizeof (dmu_replay_record_t);
867
868 *sizep = size;
869
870 return (0);
871 }
872
873 int
874 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
875 {
876 int err;
877 uint64_t size;
878
879 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
880
881 /* tosnap must be a snapshot */
882 if (!ds->ds_is_snapshot)
883 return (SET_ERROR(EINVAL));
884
885 /* fromsnap, if provided, must be a snapshot */
886 if (fromds != NULL && !fromds->ds_is_snapshot)
887 return (SET_ERROR(EINVAL));
888
889 /*
890 * fromsnap must be an earlier snapshot from the same fs as tosnap,
891 * or the origin's fs.
892 */
893 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
894 return (SET_ERROR(EXDEV));
895
896 /* Get uncompressed size estimate of changed data. */
897 if (fromds == NULL) {
898 size = dsl_dataset_phys(ds)->ds_uncompressed_bytes;
899 } else {
900 uint64_t used, comp;
901 err = dsl_dataset_space_written(fromds, ds,
902 &used, &comp, &size);
903 if (err != 0)
904 return (err);
905 }
906
907 err = dmu_adjust_send_estimate_for_indirects(ds, size, sizep);
908 return (err);
909 }
910
911 /*
912 * Simple callback used to traverse the blocks of a snapshot and sum their
913 * uncompressed size
914 */
915 /* ARGSUSED */
916 static int
917 dmu_calculate_send_traversal(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
918 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
919 {
920 uint64_t *spaceptr = arg;
921 if (bp != NULL && !BP_IS_HOLE(bp)) {
922 *spaceptr += BP_GET_UCSIZE(bp);
923 }
924 return (0);
925 }
926
927 /*
928 * Given a desination snapshot and a TXG, calculate the approximate size of a
929 * send stream sent from that TXG. from_txg may be zero, indicating that the
930 * whole snapshot will be sent.
931 */
932 int
933 dmu_send_estimate_from_txg(dsl_dataset_t *ds, uint64_t from_txg,
934 uint64_t *sizep)
935 {
936 int err;
937 uint64_t size = 0;
938
939 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
940
941 /* tosnap must be a snapshot */
942 if (!dsl_dataset_is_snapshot(ds))
943 return (SET_ERROR(EINVAL));
944
945 /* verify that from_txg is before the provided snapshot was taken */
946 if (from_txg >= dsl_dataset_phys(ds)->ds_creation_txg) {
947 return (SET_ERROR(EXDEV));
948 }
949 /*
950 * traverse the blocks of the snapshot with birth times after
951 * from_txg, summing their uncompressed size
952 */
953 err = traverse_dataset(ds, from_txg, TRAVERSE_POST,
954 dmu_calculate_send_traversal, &size);
955 if (err)
956 return (err);
957
958 err = dmu_adjust_send_estimate_for_indirects(ds, size, sizep);
959 return (err);
960 }
961
962 typedef struct dmu_recv_begin_arg {
963 const char *drba_origin;
964 dmu_recv_cookie_t *drba_cookie;
965 cred_t *drba_cred;
966 uint64_t drba_snapobj;
967 } dmu_recv_begin_arg_t;
968
969 static int
970 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
971 uint64_t fromguid)
972 {
973 uint64_t val;
974 int error;
975 dsl_pool_t *dp = ds->ds_dir->dd_pool;
976
977 /* temporary clone name must not exist */
978 error = zap_lookup(dp->dp_meta_objset,
979 dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name,
980 8, 1, &val);
981 if (error != ENOENT)
982 return (error == 0 ? EBUSY : error);
983
984 /* new snapshot name must not exist */
985 error = zap_lookup(dp->dp_meta_objset,
986 dsl_dataset_phys(ds)->ds_snapnames_zapobj,
987 drba->drba_cookie->drc_tosnap, 8, 1, &val);
988 if (error != ENOENT)
989 return (error == 0 ? EEXIST : error);
990
991 /*
992 * Check snapshot limit before receiving. We'll recheck again at the
993 * end, but might as well abort before receiving if we're already over
994 * the limit.
995 *
996 * Note that we do not check the file system limit with
997 * dsl_dir_fscount_check because the temporary %clones don't count
998 * against that limit.
999 */
1000 error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT,
1001 NULL, drba->drba_cred);
1002 if (error != 0)
1003 return (error);
1004
1005 if (fromguid != 0) {
1006 dsl_dataset_t *snap;
1007 uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1008
1009 /* Find snapshot in this dir that matches fromguid. */
1010 while (obj != 0) {
1011 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1012 &snap);
1013 if (error != 0)
1014 return (SET_ERROR(ENODEV));
1015 if (snap->ds_dir != ds->ds_dir) {
1016 dsl_dataset_rele(snap, FTAG);
1017 return (SET_ERROR(ENODEV));
1018 }
1019 if (dsl_dataset_phys(snap)->ds_guid == fromguid)
1020 break;
1021 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
1022 dsl_dataset_rele(snap, FTAG);
1023 }
1024 if (obj == 0)
1025 return (SET_ERROR(ENODEV));
1026
1027 if (drba->drba_cookie->drc_force) {
1028 drba->drba_snapobj = obj;
1029 } else {
1030 /*
1031 * If we are not forcing, there must be no
1032 * changes since fromsnap.
1033 */
1034 if (dsl_dataset_modified_since_snap(ds, snap)) {
1035 dsl_dataset_rele(snap, FTAG);
1036 return (SET_ERROR(ETXTBSY));
1037 }
1038 drba->drba_snapobj = ds->ds_prev->ds_object;
1039 }
1040
1041 dsl_dataset_rele(snap, FTAG);
1042 } else {
1043 /* if full, then must be forced */
1044 if (!drba->drba_cookie->drc_force)
1045 return (SET_ERROR(EEXIST));
1046 /* start from $ORIGIN@$ORIGIN, if supported */
1047 drba->drba_snapobj = dp->dp_origin_snap != NULL ?
1048 dp->dp_origin_snap->ds_object : 0;
1049 }
1050
1051 return (0);
1052
1053 }
1054
1055 static int
1056 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
1057 {
1058 dmu_recv_begin_arg_t *drba = arg;
1059 dsl_pool_t *dp = dmu_tx_pool(tx);
1060 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1061 uint64_t fromguid = drrb->drr_fromguid;
1062 int flags = drrb->drr_flags;
1063 int error;
1064 uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
1065 dsl_dataset_t *ds;
1066 const char *tofs = drba->drba_cookie->drc_tofs;
1067
1068 /* already checked */
1069 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
1070
1071 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
1072 DMU_COMPOUNDSTREAM ||
1073 drrb->drr_type >= DMU_OST_NUMTYPES ||
1074 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
1075 return (SET_ERROR(EINVAL));
1076
1077 /* Verify pool version supports SA if SA_SPILL feature set */
1078 if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
1079 spa_version(dp->dp_spa) < SPA_VERSION_SA)
1080 return (SET_ERROR(ENOTSUP));
1081
1082 /*
1083 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1084 * record to a plan WRITE record, so the pool must have the
1085 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1086 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1087 */
1088 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) &&
1089 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA))
1090 return (SET_ERROR(ENOTSUP));
1091 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) &&
1092 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
1093 return (SET_ERROR(ENOTSUP));
1094
1095 /*
1096 * The receiving code doesn't know how to translate large blocks
1097 * to smaller ones, so the pool must have the LARGE_BLOCKS
1098 * feature enabled if the stream has LARGE_BLOCKS.
1099 */
1100 if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
1101 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
1102 return (SET_ERROR(ENOTSUP));
1103
1104 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1105 if (error == 0) {
1106 /* target fs already exists; recv into temp clone */
1107
1108 /* Can't recv a clone into an existing fs */
1109 if (flags & DRR_FLAG_CLONE) {
1110 dsl_dataset_rele(ds, FTAG);
1111 return (SET_ERROR(EINVAL));
1112 }
1113
1114 error = recv_begin_check_existing_impl(drba, ds, fromguid);
1115 dsl_dataset_rele(ds, FTAG);
1116 } else if (error == ENOENT) {
1117 /* target fs does not exist; must be a full backup or clone */
1118 char buf[MAXNAMELEN];
1119
1120 /*
1121 * If it's a non-clone incremental, we are missing the
1122 * target fs, so fail the recv.
1123 */
1124 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
1125 return (SET_ERROR(ENOENT));
1126
1127 /* Open the parent of tofs */
1128 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
1129 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
1130 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
1131 if (error != 0)
1132 return (error);
1133
1134 /*
1135 * Check filesystem and snapshot limits before receiving. We'll
1136 * recheck snapshot limits again at the end (we create the
1137 * filesystems and increment those counts during begin_sync).
1138 */
1139 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1140 ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred);
1141 if (error != 0) {
1142 dsl_dataset_rele(ds, FTAG);
1143 return (error);
1144 }
1145
1146 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1147 ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred);
1148 if (error != 0) {
1149 dsl_dataset_rele(ds, FTAG);
1150 return (error);
1151 }
1152
1153 if (drba->drba_origin != NULL) {
1154 dsl_dataset_t *origin;
1155 error = dsl_dataset_hold(dp, drba->drba_origin,
1156 FTAG, &origin);
1157 if (error != 0) {
1158 dsl_dataset_rele(ds, FTAG);
1159 return (error);
1160 }
1161 if (!origin->ds_is_snapshot) {
1162 dsl_dataset_rele(origin, FTAG);
1163 dsl_dataset_rele(ds, FTAG);
1164 return (SET_ERROR(EINVAL));
1165 }
1166 if (dsl_dataset_phys(origin)->ds_guid != fromguid) {
1167 dsl_dataset_rele(origin, FTAG);
1168 dsl_dataset_rele(ds, FTAG);
1169 return (SET_ERROR(ENODEV));
1170 }
1171 dsl_dataset_rele(origin, FTAG);
1172 }
1173 dsl_dataset_rele(ds, FTAG);
1174 error = 0;
1175 }
1176 return (error);
1177 }
1178
1179 static void
1180 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
1181 {
1182 dmu_recv_begin_arg_t *drba = arg;
1183 dsl_pool_t *dp = dmu_tx_pool(tx);
1184 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1185 const char *tofs = drba->drba_cookie->drc_tofs;
1186 dsl_dataset_t *ds, *newds;
1187 uint64_t dsobj;
1188 int error;
1189 uint64_t crflags;
1190
1191 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
1192 DS_FLAG_CI_DATASET : 0;
1193
1194 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1195 if (error == 0) {
1196 /* create temporary clone */
1197 dsl_dataset_t *snap = NULL;
1198 if (drba->drba_snapobj != 0) {
1199 VERIFY0(dsl_dataset_hold_obj(dp,
1200 drba->drba_snapobj, FTAG, &snap));
1201 }
1202 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
1203 snap, crflags, drba->drba_cred, tx);
1204 dsl_dataset_rele(snap, FTAG);
1205 dsl_dataset_rele(ds, FTAG);
1206 } else {
1207 dsl_dir_t *dd;
1208 const char *tail;
1209 dsl_dataset_t *origin = NULL;
1210
1211 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
1212
1213 if (drba->drba_origin != NULL) {
1214 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
1215 FTAG, &origin));
1216 }
1217
1218 /* Create new dataset. */
1219 dsobj = dsl_dataset_create_sync(dd,
1220 strrchr(tofs, '/') + 1,
1221 origin, crflags, drba->drba_cred, tx);
1222 if (origin != NULL)
1223 dsl_dataset_rele(origin, FTAG);
1224 dsl_dir_rele(dd, FTAG);
1225 drba->drba_cookie->drc_newfs = B_TRUE;
1226 }
1227 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
1228
1229 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
1230 DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
1231 !newds->ds_large_blocks) {
1232 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1233 newds->ds_large_blocks = B_TRUE;
1234 }
1235
1236 dmu_buf_will_dirty(newds->ds_dbuf, tx);
1237 dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT;
1238
1239 /*
1240 * If we actually created a non-clone, we need to create the
1241 * objset in our new dataset.
1242 */
1243 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
1244 (void) dmu_objset_create_impl(dp->dp_spa,
1245 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1246 }
1247
1248 drba->drba_cookie->drc_ds = newds;
1249
1250 spa_history_log_internal_ds(newds, "receive", tx, "");
1251 }
1252
1253 /*
1254 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1255 * succeeds; otherwise we will leak the holds on the datasets.
1256 */
1257 int
1258 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1259 boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1260 {
1261 dmu_recv_begin_arg_t drba = { 0 };
1262 dmu_replay_record_t *drr;
1263
1264 bzero(drc, sizeof (dmu_recv_cookie_t));
1265 drc->drc_drrb = drrb;
1266 drc->drc_tosnap = tosnap;
1267 drc->drc_tofs = tofs;
1268 drc->drc_force = force;
1269 drc->drc_cred = CRED();
1270
1271 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1272 drc->drc_byteswap = B_TRUE;
1273 else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1274 return (SET_ERROR(EINVAL));
1275
1276 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1277 drr->drr_type = DRR_BEGIN;
1278 drr->drr_u.drr_begin = *drc->drc_drrb;
1279 if (drc->drc_byteswap) {
1280 fletcher_4_incremental_byteswap(drr,
1281 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1282 } else {
1283 fletcher_4_incremental_native(drr,
1284 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1285 }
1286 kmem_free(drr, sizeof (dmu_replay_record_t));
1287
1288 if (drc->drc_byteswap) {
1289 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1290 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1291 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1292 drrb->drr_type = BSWAP_32(drrb->drr_type);
1293 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1294 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1295 }
1296
1297 drba.drba_origin = origin;
1298 drba.drba_cookie = drc;
1299 drba.drba_cred = CRED();
1300
1301 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1302 &drba, 5, ZFS_SPACE_CHECK_NORMAL));
1303 }
1304
1305 struct restorearg {
1306 int err;
1307 boolean_t byteswap;
1308 vnode_t *vp;
1309 char *buf;
1310 uint64_t voff;
1311 int bufsize; /* amount of memory allocated for buf */
1312 zio_cksum_t cksum;
1313 avl_tree_t *guid_to_ds_map;
1314 };
1315
1316 typedef struct guid_map_entry {
1317 uint64_t guid;
1318 dsl_dataset_t *gme_ds;
1319 avl_node_t avlnode;
1320 } guid_map_entry_t;
1321
1322 static int
1323 guid_compare(const void *arg1, const void *arg2)
1324 {
1325 const guid_map_entry_t *gmep1 = arg1;
1326 const guid_map_entry_t *gmep2 = arg2;
1327
1328 if (gmep1->guid < gmep2->guid)
1329 return (-1);
1330 else if (gmep1->guid > gmep2->guid)
1331 return (1);
1332 return (0);
1333 }
1334
1335 static void
1336 free_guid_map_onexit(void *arg)
1337 {
1338 avl_tree_t *ca = arg;
1339 void *cookie = NULL;
1340 guid_map_entry_t *gmep;
1341
1342 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1343 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1344 dsl_dataset_rele(gmep->gme_ds, gmep);
1345 kmem_free(gmep, sizeof (guid_map_entry_t));
1346 }
1347 avl_destroy(ca);
1348 kmem_free(ca, sizeof (avl_tree_t));
1349 }
1350
1351 static void *
1352 restore_read(struct restorearg *ra, int len, char *buf)
1353 {
1354 int done = 0;
1355
1356 if (buf == NULL)
1357 buf = ra->buf;
1358
1359 /* some things will require 8-byte alignment, so everything must */
1360 ASSERT0(len % 8);
1361 ASSERT3U(len, <=, ra->bufsize);
1362
1363 while (done < len) {
1364 ssize_t resid;
1365
1366 ra->err = vn_rdwr(UIO_READ, ra->vp,
1367 buf + done, len - done,
1368 ra->voff, UIO_SYSSPACE, FAPPEND,
1369 RLIM64_INFINITY, CRED(), &resid);
1370
1371 if (resid == len - done)
1372 ra->err = SET_ERROR(EINVAL);
1373 ra->voff += len - done - resid;
1374 done = len - resid;
1375 if (ra->err != 0)
1376 return (NULL);
1377 }
1378
1379 ASSERT3U(done, ==, len);
1380 if (ra->byteswap)
1381 fletcher_4_incremental_byteswap(buf, len, &ra->cksum);
1382 else
1383 fletcher_4_incremental_native(buf, len, &ra->cksum);
1384 return (buf);
1385 }
1386
1387 noinline static void
1388 backup_byteswap(dmu_replay_record_t *drr)
1389 {
1390 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1391 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1392 drr->drr_type = BSWAP_32(drr->drr_type);
1393 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1394 switch (drr->drr_type) {
1395 case DRR_BEGIN:
1396 DO64(drr_begin.drr_magic);
1397 DO64(drr_begin.drr_versioninfo);
1398 DO64(drr_begin.drr_creation_time);
1399 DO32(drr_begin.drr_type);
1400 DO32(drr_begin.drr_flags);
1401 DO64(drr_begin.drr_toguid);
1402 DO64(drr_begin.drr_fromguid);
1403 break;
1404 case DRR_OBJECT:
1405 DO64(drr_object.drr_object);
1406 DO32(drr_object.drr_type);
1407 DO32(drr_object.drr_bonustype);
1408 DO32(drr_object.drr_blksz);
1409 DO32(drr_object.drr_bonuslen);
1410 DO64(drr_object.drr_toguid);
1411 break;
1412 case DRR_FREEOBJECTS:
1413 DO64(drr_freeobjects.drr_firstobj);
1414 DO64(drr_freeobjects.drr_numobjs);
1415 DO64(drr_freeobjects.drr_toguid);
1416 break;
1417 case DRR_WRITE:
1418 DO64(drr_write.drr_object);
1419 DO32(drr_write.drr_type);
1420 DO64(drr_write.drr_offset);
1421 DO64(drr_write.drr_length);
1422 DO64(drr_write.drr_toguid);
1423 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1424 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1425 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1426 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1427 DO64(drr_write.drr_key.ddk_prop);
1428 break;
1429 case DRR_WRITE_BYREF:
1430 DO64(drr_write_byref.drr_object);
1431 DO64(drr_write_byref.drr_offset);
1432 DO64(drr_write_byref.drr_length);
1433 DO64(drr_write_byref.drr_toguid);
1434 DO64(drr_write_byref.drr_refguid);
1435 DO64(drr_write_byref.drr_refobject);
1436 DO64(drr_write_byref.drr_refoffset);
1437 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1438 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1439 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1440 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1441 DO64(drr_write_byref.drr_key.ddk_prop);
1442 break;
1443 case DRR_WRITE_EMBEDDED:
1444 DO64(drr_write_embedded.drr_object);
1445 DO64(drr_write_embedded.drr_offset);
1446 DO64(drr_write_embedded.drr_length);
1447 DO64(drr_write_embedded.drr_toguid);
1448 DO32(drr_write_embedded.drr_lsize);
1449 DO32(drr_write_embedded.drr_psize);
1450 break;
1451 case DRR_FREE:
1452 DO64(drr_free.drr_object);
1453 DO64(drr_free.drr_offset);
1454 DO64(drr_free.drr_length);
1455 DO64(drr_free.drr_toguid);
1456 break;
1457 case DRR_SPILL:
1458 DO64(drr_spill.drr_object);
1459 DO64(drr_spill.drr_length);
1460 DO64(drr_spill.drr_toguid);
1461 break;
1462 case DRR_END:
1463 DO64(drr_end.drr_checksum.zc_word[0]);
1464 DO64(drr_end.drr_checksum.zc_word[1]);
1465 DO64(drr_end.drr_checksum.zc_word[2]);
1466 DO64(drr_end.drr_checksum.zc_word[3]);
1467 DO64(drr_end.drr_toguid);
1468 break;
1469 default:
1470 break;
1471 }
1472 #undef DO64
1473 #undef DO32
1474 }
1475
1476 static inline uint8_t
1477 deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size)
1478 {
1479 if (bonus_type == DMU_OT_SA) {
1480 return (1);
1481 } else {
1482 return (1 +
1483 ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT));
1484 }
1485 }
1486
1487 noinline static int
1488 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1489 {
1490 dmu_object_info_t doi;
1491 dmu_tx_t *tx;
1492 void *data = NULL;
1493 uint64_t object;
1494 int err;
1495
1496 if (drro->drr_type == DMU_OT_NONE ||
1497 !DMU_OT_IS_VALID(drro->drr_type) ||
1498 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1499 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1500 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1501 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1502 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1503 drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(os)) ||
1504 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1505 return (SET_ERROR(EINVAL));
1506 }
1507
1508 err = dmu_object_info(os, drro->drr_object, &doi);
1509
1510 if (err != 0 && err != ENOENT)
1511 return (SET_ERROR(EINVAL));
1512 object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT;
1513
1514 if (drro->drr_bonuslen) {
1515 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8), NULL);
1516 if (ra->err != 0)
1517 return (ra->err);
1518 }
1519
1520 /*
1521 * If we are losing blkptrs or changing the block size this must
1522 * be a new file instance. We must clear out the previous file
1523 * contents before we can change this type of metadata in the dnode.
1524 */
1525 if (err == 0) {
1526 int nblkptr;
1527
1528 nblkptr = deduce_nblkptr(drro->drr_bonustype,
1529 drro->drr_bonuslen);
1530
1531 if (drro->drr_blksz != doi.doi_data_block_size ||
1532 nblkptr < doi.doi_nblkptr) {
1533 err = dmu_free_long_range(os, drro->drr_object,
1534 0, DMU_OBJECT_END);
1535 if (err != 0)
1536 return (SET_ERROR(EINVAL));
1537 }
1538 }
1539
1540 tx = dmu_tx_create(os);
1541 dmu_tx_hold_bonus(tx, object);
1542 err = dmu_tx_assign(tx, TXG_WAIT);
1543 if (err != 0) {
1544 dmu_tx_abort(tx);
1545 return (err);
1546 }
1547
1548 if (object == DMU_NEW_OBJECT) {
1549 /* currently free, want to be allocated */
1550 err = dmu_object_claim(os, drro->drr_object,
1551 drro->drr_type, drro->drr_blksz,
1552 drro->drr_bonustype, drro->drr_bonuslen, tx);
1553 } else if (drro->drr_type != doi.doi_type ||
1554 drro->drr_blksz != doi.doi_data_block_size ||
1555 drro->drr_bonustype != doi.doi_bonus_type ||
1556 drro->drr_bonuslen != doi.doi_bonus_size) {
1557 /* currently allocated, but with different properties */
1558 err = dmu_object_reclaim(os, drro->drr_object,
1559 drro->drr_type, drro->drr_blksz,
1560 drro->drr_bonustype, drro->drr_bonuslen, tx);
1561 }
1562 if (err != 0) {
1563 dmu_tx_commit(tx);
1564 return (SET_ERROR(EINVAL));
1565 }
1566
1567 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1568 tx);
1569 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1570
1571 if (data != NULL) {
1572 dmu_buf_t *db;
1573
1574 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1575 dmu_buf_will_dirty(db, tx);
1576
1577 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1578 bcopy(data, db->db_data, drro->drr_bonuslen);
1579 if (ra->byteswap) {
1580 dmu_object_byteswap_t byteswap =
1581 DMU_OT_BYTESWAP(drro->drr_bonustype);
1582 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1583 drro->drr_bonuslen);
1584 }
1585 dmu_buf_rele(db, FTAG);
1586 }
1587 dmu_tx_commit(tx);
1588 return (0);
1589 }
1590
1591 /* ARGSUSED */
1592 noinline static int
1593 restore_freeobjects(struct restorearg *ra, objset_t *os,
1594 struct drr_freeobjects *drrfo)
1595 {
1596 uint64_t obj;
1597
1598 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1599 return (SET_ERROR(EINVAL));
1600
1601 for (obj = drrfo->drr_firstobj;
1602 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1603 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1604 int err;
1605
1606 if (dmu_object_info(os, obj, NULL) != 0)
1607 continue;
1608
1609 err = dmu_free_long_object(os, obj);
1610 if (err != 0)
1611 return (err);
1612 }
1613 return (0);
1614 }
1615
1616 noinline static int
1617 restore_write(struct restorearg *ra, objset_t *os,
1618 struct drr_write *drrw)
1619 {
1620 dmu_tx_t *tx;
1621 dmu_buf_t *bonus;
1622 arc_buf_t *abuf;
1623 void *data;
1624 int err;
1625
1626 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1627 !DMU_OT_IS_VALID(drrw->drr_type))
1628 return (SET_ERROR(EINVAL));
1629
1630 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1631 return (SET_ERROR(EINVAL));
1632
1633 if (dmu_bonus_hold(os, drrw->drr_object, FTAG, &bonus) != 0)
1634 return (SET_ERROR(EINVAL));
1635
1636 abuf = dmu_request_arcbuf(bonus, drrw->drr_length);
1637
1638 data = restore_read(ra, drrw->drr_length, abuf->b_data);
1639 if (data == NULL) {
1640 dmu_return_arcbuf(abuf);
1641 dmu_buf_rele(bonus, FTAG);
1642 return (ra->err);
1643 }
1644
1645 tx = dmu_tx_create(os);
1646
1647 dmu_tx_hold_write(tx, drrw->drr_object,
1648 drrw->drr_offset, drrw->drr_length);
1649 err = dmu_tx_assign(tx, TXG_WAIT);
1650 if (err != 0) {
1651 dmu_return_arcbuf(abuf);
1652 dmu_buf_rele(bonus, FTAG);
1653 dmu_tx_abort(tx);
1654 return (err);
1655 }
1656 if (ra->byteswap) {
1657 dmu_object_byteswap_t byteswap =
1658 DMU_OT_BYTESWAP(drrw->drr_type);
1659 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1660 }
1661 dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx);
1662 dmu_tx_commit(tx);
1663 dmu_buf_rele(bonus, FTAG);
1664 return (0);
1665 }
1666
1667 /*
1668 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1669 * streams to refer to a copy of the data that is already on the
1670 * system because it came in earlier in the stream. This function
1671 * finds the earlier copy of the data, and uses that copy instead of
1672 * data from the stream to fulfill this write.
1673 */
1674 static int
1675 restore_write_byref(struct restorearg *ra, objset_t *os,
1676 struct drr_write_byref *drrwbr)
1677 {
1678 dmu_tx_t *tx;
1679 int err;
1680 guid_map_entry_t gmesrch;
1681 guid_map_entry_t *gmep;
1682 avl_index_t where;
1683 objset_t *ref_os = NULL;
1684 dmu_buf_t *dbp;
1685
1686 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1687 return (SET_ERROR(EINVAL));
1688
1689 /*
1690 * If the GUID of the referenced dataset is different from the
1691 * GUID of the target dataset, find the referenced dataset.
1692 */
1693 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1694 gmesrch.guid = drrwbr->drr_refguid;
1695 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1696 &where)) == NULL) {
1697 return (SET_ERROR(EINVAL));
1698 }
1699 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1700 return (SET_ERROR(EINVAL));
1701 } else {
1702 ref_os = os;
1703 }
1704
1705 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1706 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1707 if (err != 0)
1708 return (err);
1709
1710 tx = dmu_tx_create(os);
1711
1712 dmu_tx_hold_write(tx, drrwbr->drr_object,
1713 drrwbr->drr_offset, drrwbr->drr_length);
1714 err = dmu_tx_assign(tx, TXG_WAIT);
1715 if (err != 0) {
1716 dmu_tx_abort(tx);
1717 return (err);
1718 }
1719 dmu_write(os, drrwbr->drr_object,
1720 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1721 dmu_buf_rele(dbp, FTAG);
1722 dmu_tx_commit(tx);
1723 return (0);
1724 }
1725
1726 static int
1727 restore_write_embedded(struct restorearg *ra, objset_t *os,
1728 struct drr_write_embedded *drrwnp)
1729 {
1730 dmu_tx_t *tx;
1731 int err;
1732 void *data;
1733
1734 if (drrwnp->drr_offset + drrwnp->drr_length < drrwnp->drr_offset)
1735 return (EINVAL);
1736
1737 if (drrwnp->drr_psize > BPE_PAYLOAD_SIZE)
1738 return (EINVAL);
1739
1740 if (drrwnp->drr_etype >= NUM_BP_EMBEDDED_TYPES)
1741 return (EINVAL);
1742 if (drrwnp->drr_compression >= ZIO_COMPRESS_FUNCTIONS)
1743 return (EINVAL);
1744
1745 data = restore_read(ra, P2ROUNDUP(drrwnp->drr_psize, 8), NULL);
1746 if (data == NULL)
1747 return (ra->err);
1748
1749 tx = dmu_tx_create(os);
1750
1751 dmu_tx_hold_write(tx, drrwnp->drr_object,
1752 drrwnp->drr_offset, drrwnp->drr_length);
1753 err = dmu_tx_assign(tx, TXG_WAIT);
1754 if (err != 0) {
1755 dmu_tx_abort(tx);
1756 return (err);
1757 }
1758
1759 dmu_write_embedded(os, drrwnp->drr_object,
1760 drrwnp->drr_offset, data, drrwnp->drr_etype,
1761 drrwnp->drr_compression, drrwnp->drr_lsize, drrwnp->drr_psize,
1762 ra->byteswap ^ ZFS_HOST_BYTEORDER, tx);
1763
1764 dmu_tx_commit(tx);
1765 return (0);
1766 }
1767
1768 static int
1769 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1770 {
1771 dmu_tx_t *tx;
1772 void *data;
1773 dmu_buf_t *db, *db_spill;
1774 int err;
1775
1776 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1777 drrs->drr_length > spa_maxblocksize(dmu_objset_spa(os)))
1778 return (SET_ERROR(EINVAL));
1779
1780 data = restore_read(ra, drrs->drr_length, NULL);
1781 if (data == NULL)
1782 return (ra->err);
1783
1784 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1785 return (SET_ERROR(EINVAL));
1786
1787 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1788 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1789 dmu_buf_rele(db, FTAG);
1790 return (err);
1791 }
1792
1793 tx = dmu_tx_create(os);
1794
1795 dmu_tx_hold_spill(tx, db->db_object);
1796
1797 err = dmu_tx_assign(tx, TXG_WAIT);
1798 if (err != 0) {
1799 dmu_buf_rele(db, FTAG);
1800 dmu_buf_rele(db_spill, FTAG);
1801 dmu_tx_abort(tx);
1802 return (err);
1803 }
1804 dmu_buf_will_dirty(db_spill, tx);
1805
1806 if (db_spill->db_size < drrs->drr_length)
1807 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1808 drrs->drr_length, tx));
1809 bcopy(data, db_spill->db_data, drrs->drr_length);
1810
1811 dmu_buf_rele(db, FTAG);
1812 dmu_buf_rele(db_spill, FTAG);
1813
1814 dmu_tx_commit(tx);
1815 return (0);
1816 }
1817
1818 /* ARGSUSED */
1819 noinline static int
1820 restore_free(struct restorearg *ra, objset_t *os,
1821 struct drr_free *drrf)
1822 {
1823 int err;
1824
1825 if (drrf->drr_length != -1ULL &&
1826 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1827 return (SET_ERROR(EINVAL));
1828
1829 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1830 return (SET_ERROR(EINVAL));
1831
1832 err = dmu_free_long_range(os, drrf->drr_object,
1833 drrf->drr_offset, drrf->drr_length);
1834 return (err);
1835 }
1836
1837 /* used to destroy the drc_ds on error */
1838 static void
1839 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1840 {
1841 char name[MAXNAMELEN];
1842 dsl_dataset_name(drc->drc_ds, name);
1843 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1844 (void) dsl_destroy_head(name);
1845 }
1846
1847 /*
1848 * NB: callers *must* call dmu_recv_end() if this succeeds.
1849 */
1850 int
1851 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1852 int cleanup_fd, uint64_t *action_handlep)
1853 {
1854 struct restorearg ra = { 0 };
1855 dmu_replay_record_t *drr;
1856 objset_t *os;
1857 zio_cksum_t pcksum;
1858 int featureflags;
1859
1860 ra.byteswap = drc->drc_byteswap;
1861 ra.cksum = drc->drc_cksum;
1862 ra.vp = vp;
1863 ra.voff = *voffp;
1864 ra.bufsize = SPA_MAXBLOCKSIZE;
1865 ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
1866
1867 /* these were verified in dmu_recv_begin */
1868 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1869 DMU_SUBSTREAM);
1870 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1871
1872 /*
1873 * Open the objset we are modifying.
1874 */
1875 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1876
1877 ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT);
1878
1879 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1880
1881 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1882 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1883 minor_t minor;
1884
1885 if (cleanup_fd == -1) {
1886 ra.err = SET_ERROR(EBADF);
1887 goto out;
1888 }
1889 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1890 if (ra.err != 0) {
1891 cleanup_fd = -1;
1892 goto out;
1893 }
1894
1895 if (*action_handlep == 0) {
1896 ra.guid_to_ds_map =
1897 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1898 avl_create(ra.guid_to_ds_map, guid_compare,
1899 sizeof (guid_map_entry_t),
1900 offsetof(guid_map_entry_t, avlnode));
1901 ra.err = zfs_onexit_add_cb(minor,
1902 free_guid_map_onexit, ra.guid_to_ds_map,
1903 action_handlep);
1904 if (ra.err != 0)
1905 goto out;
1906 } else {
1907 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1908 (void **)&ra.guid_to_ds_map);
1909 if (ra.err != 0)
1910 goto out;
1911 }
1912
1913 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1914 }
1915
1916 /*
1917 * Read records and process them.
1918 */
1919 pcksum = ra.cksum;
1920 while (ra.err == 0 &&
1921 NULL != (drr = restore_read(&ra, sizeof (*drr), NULL))) {
1922 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1923 ra.err = SET_ERROR(EINTR);
1924 goto out;
1925 }
1926
1927 if (ra.byteswap)
1928 backup_byteswap(drr);
1929
1930 switch (drr->drr_type) {
1931 case DRR_OBJECT:
1932 {
1933 /*
1934 * We need to make a copy of the record header,
1935 * because restore_{object,write} may need to
1936 * restore_read(), which will invalidate drr.
1937 */
1938 struct drr_object drro = drr->drr_u.drr_object;
1939 ra.err = restore_object(&ra, os, &drro);
1940 break;
1941 }
1942 case DRR_FREEOBJECTS:
1943 {
1944 struct drr_freeobjects drrfo =
1945 drr->drr_u.drr_freeobjects;
1946 ra.err = restore_freeobjects(&ra, os, &drrfo);
1947 break;
1948 }
1949 case DRR_WRITE:
1950 {
1951 struct drr_write drrw = drr->drr_u.drr_write;
1952 ra.err = restore_write(&ra, os, &drrw);
1953 break;
1954 }
1955 case DRR_WRITE_BYREF:
1956 {
1957 struct drr_write_byref drrwbr =
1958 drr->drr_u.drr_write_byref;
1959 ra.err = restore_write_byref(&ra, os, &drrwbr);
1960 break;
1961 }
1962 case DRR_WRITE_EMBEDDED:
1963 {
1964 struct drr_write_embedded drrwe =
1965 drr->drr_u.drr_write_embedded;
1966 ra.err = restore_write_embedded(&ra, os, &drrwe);
1967 break;
1968 }
1969 case DRR_FREE:
1970 {
1971 struct drr_free drrf = drr->drr_u.drr_free;
1972 ra.err = restore_free(&ra, os, &drrf);
1973 break;
1974 }
1975 case DRR_END:
1976 {
1977 struct drr_end drre = drr->drr_u.drr_end;
1978 /*
1979 * We compare against the *previous* checksum
1980 * value, because the stored checksum is of
1981 * everything before the DRR_END record.
1982 */
1983 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1984 ra.err = SET_ERROR(ECKSUM);
1985 goto out;
1986 }
1987 case DRR_SPILL:
1988 {
1989 struct drr_spill drrs = drr->drr_u.drr_spill;
1990 ra.err = restore_spill(&ra, os, &drrs);
1991 break;
1992 }
1993 default:
1994 ra.err = SET_ERROR(EINVAL);
1995 goto out;
1996 }
1997 pcksum = ra.cksum;
1998 }
1999 ASSERT(ra.err != 0);
2000
2001 out:
2002 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
2003 zfs_onexit_fd_rele(cleanup_fd);
2004
2005 if (ra.err != 0) {
2006 /*
2007 * destroy what we created, so we don't leave it in the
2008 * inconsistent restoring state.
2009 */
2010 dmu_recv_cleanup_ds(drc);
2011 }
2012
2013 vmem_free(ra.buf, ra.bufsize);
2014 *voffp = ra.voff;
2015 return (ra.err);
2016 }
2017
2018 static int
2019 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
2020 {
2021 dmu_recv_cookie_t *drc = arg;
2022 dsl_pool_t *dp = dmu_tx_pool(tx);
2023 int error;
2024
2025 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
2026
2027 if (!drc->drc_newfs) {
2028 dsl_dataset_t *origin_head;
2029
2030 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
2031 if (error != 0)
2032 return (error);
2033 if (drc->drc_force) {
2034 /*
2035 * We will destroy any snapshots in tofs (i.e. before
2036 * origin_head) that are after the origin (which is
2037 * the snap before drc_ds, because drc_ds can not
2038 * have any snaps of its own).
2039 */
2040 uint64_t obj;
2041
2042 obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj;
2043 while (obj !=
2044 dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) {
2045 dsl_dataset_t *snap;
2046 error = dsl_dataset_hold_obj(dp, obj, FTAG,
2047 &snap);
2048 if (error != 0)
2049 break;
2050 if (snap->ds_dir != origin_head->ds_dir)
2051 error = SET_ERROR(EINVAL);
2052 if (error == 0) {
2053 error = dsl_destroy_snapshot_check_impl(
2054 snap, B_FALSE);
2055 }
2056 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
2057 dsl_dataset_rele(snap, FTAG);
2058 if (error != 0)
2059 break;
2060 }
2061 if (error != 0) {
2062 dsl_dataset_rele(origin_head, FTAG);
2063 return (error);
2064 }
2065 }
2066 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
2067 origin_head, drc->drc_force, drc->drc_owner, tx);
2068 if (error != 0) {
2069 dsl_dataset_rele(origin_head, FTAG);
2070 return (error);
2071 }
2072 error = dsl_dataset_snapshot_check_impl(origin_head,
2073 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
2074 dsl_dataset_rele(origin_head, FTAG);
2075 if (error != 0)
2076 return (error);
2077
2078 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
2079 } else {
2080 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
2081 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
2082 }
2083 return (error);
2084 }
2085
2086 static void
2087 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
2088 {
2089 dmu_recv_cookie_t *drc = arg;
2090 dsl_pool_t *dp = dmu_tx_pool(tx);
2091
2092 spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
2093 tx, "snap=%s", drc->drc_tosnap);
2094
2095 if (!drc->drc_newfs) {
2096 dsl_dataset_t *origin_head;
2097
2098 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
2099 &origin_head));
2100
2101 if (drc->drc_force) {
2102 /*
2103 * Destroy any snapshots of drc_tofs (origin_head)
2104 * after the origin (the snap before drc_ds).
2105 */
2106 uint64_t obj;
2107
2108 obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj;
2109 while (obj !=
2110 dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) {
2111 dsl_dataset_t *snap;
2112 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
2113 &snap));
2114 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
2115 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
2116 dsl_destroy_snapshot_sync_impl(snap,
2117 B_FALSE, tx);
2118 dsl_dataset_rele(snap, FTAG);
2119 }
2120 }
2121 VERIFY3P(drc->drc_ds->ds_prev, ==,
2122 origin_head->ds_prev);
2123
2124 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
2125 origin_head, tx);
2126 dsl_dataset_snapshot_sync_impl(origin_head,
2127 drc->drc_tosnap, tx);
2128
2129 /* set snapshot's creation time and guid */
2130 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
2131 dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time =
2132 drc->drc_drrb->drr_creation_time;
2133 dsl_dataset_phys(origin_head->ds_prev)->ds_guid =
2134 drc->drc_drrb->drr_toguid;
2135 dsl_dataset_phys(origin_head->ds_prev)->ds_flags &=
2136 ~DS_FLAG_INCONSISTENT;
2137
2138 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2139 dsl_dataset_phys(origin_head)->ds_flags &=
2140 ~DS_FLAG_INCONSISTENT;
2141
2142 dsl_dataset_rele(origin_head, FTAG);
2143 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
2144
2145 if (drc->drc_owner != NULL)
2146 VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
2147 } else {
2148 dsl_dataset_t *ds = drc->drc_ds;
2149
2150 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
2151
2152 /* set snapshot's creation time and guid */
2153 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2154 dsl_dataset_phys(ds->ds_prev)->ds_creation_time =
2155 drc->drc_drrb->drr_creation_time;
2156 dsl_dataset_phys(ds->ds_prev)->ds_guid =
2157 drc->drc_drrb->drr_toguid;
2158 dsl_dataset_phys(ds->ds_prev)->ds_flags &=
2159 ~DS_FLAG_INCONSISTENT;
2160
2161 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2162 dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT;
2163 }
2164 drc->drc_newsnapobj = dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj;
2165 /*
2166 * Release the hold from dmu_recv_begin. This must be done before
2167 * we return to open context, so that when we free the dataset's dnode,
2168 * we can evict its bonus buffer.
2169 */
2170 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
2171 drc->drc_ds = NULL;
2172 }
2173
2174 static int
2175 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
2176 {
2177 dsl_pool_t *dp;
2178 dsl_dataset_t *snapds;
2179 guid_map_entry_t *gmep;
2180 int err;
2181
2182 ASSERT(guid_map != NULL);
2183
2184 err = dsl_pool_hold(name, FTAG, &dp);
2185 if (err != 0)
2186 return (err);
2187 gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
2188 err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
2189 if (err == 0) {
2190 gmep->guid = dsl_dataset_phys(snapds)->ds_guid;
2191 gmep->gme_ds = snapds;
2192 avl_add(guid_map, gmep);
2193 dsl_dataset_long_hold(snapds, gmep);
2194 } else {
2195 kmem_free(gmep, sizeof (*gmep));
2196 }
2197
2198 dsl_pool_rele(dp, FTAG);
2199 return (err);
2200 }
2201
2202 static int dmu_recv_end_modified_blocks = 3;
2203
2204 static int
2205 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
2206 {
2207 int error;
2208
2209 #ifdef _KERNEL
2210 char *name;
2211
2212 /*
2213 * We will be destroying the ds; make sure its origin is unmounted if
2214 * necessary.
2215 */
2216 name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2217 dsl_dataset_name(drc->drc_ds, name);
2218 zfs_destroy_unmount_origin(name);
2219 kmem_free(name, MAXNAMELEN);
2220 #endif
2221
2222 error = dsl_sync_task(drc->drc_tofs,
2223 dmu_recv_end_check, dmu_recv_end_sync, drc,
2224 dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2225
2226 if (error != 0)
2227 dmu_recv_cleanup_ds(drc);
2228 return (error);
2229 }
2230
2231 static int
2232 dmu_recv_new_end(dmu_recv_cookie_t *drc)
2233 {
2234 int error;
2235
2236 error = dsl_sync_task(drc->drc_tofs,
2237 dmu_recv_end_check, dmu_recv_end_sync, drc,
2238 dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2239
2240 if (error != 0) {
2241 dmu_recv_cleanup_ds(drc);
2242 } else if (drc->drc_guid_to_ds_map != NULL) {
2243 (void) add_ds_to_guidmap(drc->drc_tofs,
2244 drc->drc_guid_to_ds_map,
2245 drc->drc_newsnapobj);
2246 }
2247 return (error);
2248 }
2249
2250 int
2251 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
2252 {
2253 drc->drc_owner = owner;
2254
2255 if (drc->drc_newfs)
2256 return (dmu_recv_new_end(drc));
2257 else
2258 return (dmu_recv_existing_end(drc));
2259 }
2260
2261 /*
2262 * Return TRUE if this objset is currently being received into.
2263 */
2264 boolean_t
2265 dmu_objset_is_receiving(objset_t *os)
2266 {
2267 return (os->os_dsl_dataset != NULL &&
2268 os->os_dsl_dataset->ds_owner == dmu_recv_tag);
2269 }
2270
2271 #if defined(_KERNEL)
2272 module_param(zfs_send_corrupt_data, int, 0644);
2273 MODULE_PARM_DESC(zfs_send_corrupt_data, "Allow sending corrupt data");
2274 #endif
Mirror of ZFSOnLinux for Debian