1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2012, Martin Matuska <mm@FreeBSD.org>. All rights reserved.
27 * Copyright 2014 HybridCluster. 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/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 #ifdef __FreeBSD__
58 #undef dump_write
59 #define dump_write dmu_dump_write
60 #endif
61
62 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
63 int zfs_send_corrupt_data = B_FALSE;
64
65 static char *dmu_recv_tag = "dmu_recv_tag";
66 static const char *recv_clone_name = "%recv";
67
68 static int
dump_bytes(dmu_sendarg_t * dsp,void * buf,int len)69 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
70 {
71 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
72 struct uio auio;
73 struct iovec aiov;
74 ASSERT0(len % 8);
75
76 fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
77 aiov.iov_base = buf;
78 aiov.iov_len = len;
79 auio.uio_iov = &aiov;
80 auio.uio_iovcnt = 1;
81 auio.uio_resid = len;
82 auio.uio_segflg = UIO_SYSSPACE;
83 auio.uio_rw = UIO_WRITE;
84 auio.uio_offset = (off_t)-1;
85 auio.uio_td = dsp->dsa_td;
86 #ifdef _KERNEL
87 if (dsp->dsa_fp->f_type == DTYPE_VNODE)
88 bwillwrite();
89 dsp->dsa_err = fo_write(dsp->dsa_fp, &auio, dsp->dsa_td->td_ucred, 0,
90 dsp->dsa_td);
91 #else
92 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
93 dsp->dsa_err = EOPNOTSUPP;
94 #endif
95 mutex_enter(&ds->ds_sendstream_lock);
96 *dsp->dsa_off += len;
97 mutex_exit(&ds->ds_sendstream_lock);
98
99 return (dsp->dsa_err);
100 }
101
102 static int
dump_free(dmu_sendarg_t * dsp,uint64_t object,uint64_t offset,uint64_t length)103 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
104 uint64_t length)
105 {
106 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
107
108 /*
109 * When we receive a free record, dbuf_free_range() assumes
110 * that the receiving system doesn't have any dbufs in the range
111 * being freed. This is always true because there is a one-record
112 * constraint: we only send one WRITE record for any given
113 * object+offset. We know that the one-record constraint is
114 * true because we always send data in increasing order by
115 * object,offset.
116 *
117 * If the increasing-order constraint ever changes, we should find
118 * another way to assert that the one-record constraint is still
119 * satisfied.
120 */
121 ASSERT(object > dsp->dsa_last_data_object ||
122 (object == dsp->dsa_last_data_object &&
123 offset > dsp->dsa_last_data_offset));
124
125 /*
126 * If we are doing a non-incremental send, then there can't
127 * be any data in the dataset we're receiving into. Therefore
128 * a free record would simply be a no-op. Save space by not
129 * sending it to begin with.
130 */
131 if (!dsp->dsa_incremental)
132 return (0);
133
134 if (length != -1ULL && offset + length < offset)
135 length = -1ULL;
136
137 /*
138 * If there is a pending op, but it's not PENDING_FREE, push it out,
139 * since free block aggregation can only be done for blocks of the
140 * same type (i.e., DRR_FREE records can only be aggregated with
141 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
142 * aggregated with other DRR_FREEOBJECTS records.
143 */
144 if (dsp->dsa_pending_op != PENDING_NONE &&
145 dsp->dsa_pending_op != PENDING_FREE) {
146 if (dump_bytes(dsp, dsp->dsa_drr,
147 sizeof (dmu_replay_record_t)) != 0)
148 return (SET_ERROR(EINTR));
149 dsp->dsa_pending_op = PENDING_NONE;
150 }
151
152 if (dsp->dsa_pending_op == PENDING_FREE) {
153 /*
154 * There should never be a PENDING_FREE if length is -1
155 * (because dump_dnode is the only place where this
156 * function is called with a -1, and only after flushing
157 * any pending record).
158 */
159 ASSERT(length != -1ULL);
160 /*
161 * Check to see whether this free block can be aggregated
162 * with pending one.
163 */
164 if (drrf->drr_object == object && drrf->drr_offset +
165 drrf->drr_length == offset) {
166 drrf->drr_length += length;
167 return (0);
168 } else {
169 /* not a continuation. Push out pending record */
170 if (dump_bytes(dsp, dsp->dsa_drr,
171 sizeof (dmu_replay_record_t)) != 0)
172 return (SET_ERROR(EINTR));
173 dsp->dsa_pending_op = PENDING_NONE;
174 }
175 }
176 /* create a FREE record and make it pending */
177 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
178 dsp->dsa_drr->drr_type = DRR_FREE;
179 drrf->drr_object = object;
180 drrf->drr_offset = offset;
181 drrf->drr_length = length;
182 drrf->drr_toguid = dsp->dsa_toguid;
183 if (length == -1ULL) {
184 if (dump_bytes(dsp, dsp->dsa_drr,
185 sizeof (dmu_replay_record_t)) != 0)
186 return (SET_ERROR(EINTR));
187 } else {
188 dsp->dsa_pending_op = PENDING_FREE;
189 }
190
191 return (0);
192 }
193
194 static int
dump_write(dmu_sendarg_t * dsp,dmu_object_type_t type,uint64_t object,uint64_t offset,int blksz,const blkptr_t * bp,void * data)195 dump_write(dmu_sendarg_t *dsp, dmu_object_type_t type,
196 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
197 {
198 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
199
200 /*
201 * We send data in increasing object, offset order.
202 * See comment in dump_free() for details.
203 */
204 ASSERT(object > dsp->dsa_last_data_object ||
205 (object == dsp->dsa_last_data_object &&
206 offset > dsp->dsa_last_data_offset));
207 dsp->dsa_last_data_object = object;
208 dsp->dsa_last_data_offset = offset + blksz - 1;
209
210 /*
211 * If there is any kind of pending aggregation (currently either
212 * a grouping of free objects or free blocks), push it out to
213 * the stream, since aggregation can't be done across operations
214 * of different types.
215 */
216 if (dsp->dsa_pending_op != PENDING_NONE) {
217 if (dump_bytes(dsp, dsp->dsa_drr,
218 sizeof (dmu_replay_record_t)) != 0)
219 return (SET_ERROR(EINTR));
220 dsp->dsa_pending_op = PENDING_NONE;
221 }
222 /* write a DATA record */
223 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
224 dsp->dsa_drr->drr_type = DRR_WRITE;
225 drrw->drr_object = object;
226 drrw->drr_type = type;
227 drrw->drr_offset = offset;
228 drrw->drr_length = blksz;
229 drrw->drr_toguid = dsp->dsa_toguid;
230 if (bp == NULL || BP_IS_EMBEDDED(bp)) {
231 /*
232 * There's no pre-computed checksum for partial-block
233 * writes or embedded BP's, so (like
234 * fletcher4-checkummed blocks) userland will have to
235 * compute a dedup-capable checksum itself.
236 */
237 drrw->drr_checksumtype = ZIO_CHECKSUM_OFF;
238 } else {
239 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
240 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
241 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
242 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
243 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
244 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
245 drrw->drr_key.ddk_cksum = bp->blk_cksum;
246 }
247
248 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
249 return (SET_ERROR(EINTR));
250 if (dump_bytes(dsp, data, blksz) != 0)
251 return (SET_ERROR(EINTR));
252 return (0);
253 }
254
255 static int
dump_write_embedded(dmu_sendarg_t * dsp,uint64_t object,uint64_t offset,int blksz,const blkptr_t * bp)256 dump_write_embedded(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
257 int blksz, const blkptr_t *bp)
258 {
259 char buf[BPE_PAYLOAD_SIZE];
260 struct drr_write_embedded *drrw =
261 &(dsp->dsa_drr->drr_u.drr_write_embedded);
262
263 if (dsp->dsa_pending_op != PENDING_NONE) {
264 if (dump_bytes(dsp, dsp->dsa_drr,
265 sizeof (dmu_replay_record_t)) != 0)
266 return (EINTR);
267 dsp->dsa_pending_op = PENDING_NONE;
268 }
269
270 ASSERT(BP_IS_EMBEDDED(bp));
271
272 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
273 dsp->dsa_drr->drr_type = DRR_WRITE_EMBEDDED;
274 drrw->drr_object = object;
275 drrw->drr_offset = offset;
276 drrw->drr_length = blksz;
277 drrw->drr_toguid = dsp->dsa_toguid;
278 drrw->drr_compression = BP_GET_COMPRESS(bp);
279 drrw->drr_etype = BPE_GET_ETYPE(bp);
280 drrw->drr_lsize = BPE_GET_LSIZE(bp);
281 drrw->drr_psize = BPE_GET_PSIZE(bp);
282
283 decode_embedded_bp_compressed(bp, buf);
284
285 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
286 return (EINTR);
287 if (dump_bytes(dsp, buf, P2ROUNDUP(drrw->drr_psize, 8)) != 0)
288 return (EINTR);
289 return (0);
290 }
291
292 static int
dump_spill(dmu_sendarg_t * dsp,uint64_t object,int blksz,void * data)293 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
294 {
295 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
296
297 if (dsp->dsa_pending_op != PENDING_NONE) {
298 if (dump_bytes(dsp, dsp->dsa_drr,
299 sizeof (dmu_replay_record_t)) != 0)
300 return (SET_ERROR(EINTR));
301 dsp->dsa_pending_op = PENDING_NONE;
302 }
303
304 /* write a SPILL record */
305 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
306 dsp->dsa_drr->drr_type = DRR_SPILL;
307 drrs->drr_object = object;
308 drrs->drr_length = blksz;
309 drrs->drr_toguid = dsp->dsa_toguid;
310
311 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
312 return (SET_ERROR(EINTR));
313 if (dump_bytes(dsp, data, blksz))
314 return (SET_ERROR(EINTR));
315 return (0);
316 }
317
318 static int
dump_freeobjects(dmu_sendarg_t * dsp,uint64_t firstobj,uint64_t numobjs)319 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
320 {
321 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
322
323 /* See comment in dump_free(). */
324 if (!dsp->dsa_incremental)
325 return (0);
326
327 /*
328 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
329 * push it out, since free block aggregation can only be done for
330 * blocks of the same type (i.e., DRR_FREE records can only be
331 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
332 * can only be aggregated with other DRR_FREEOBJECTS records.
333 */
334 if (dsp->dsa_pending_op != PENDING_NONE &&
335 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
336 if (dump_bytes(dsp, dsp->dsa_drr,
337 sizeof (dmu_replay_record_t)) != 0)
338 return (SET_ERROR(EINTR));
339 dsp->dsa_pending_op = PENDING_NONE;
340 }
341 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
342 /*
343 * See whether this free object array can be aggregated
344 * with pending one
345 */
346 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
347 drrfo->drr_numobjs += numobjs;
348 return (0);
349 } else {
350 /* can't be aggregated. Push out pending record */
351 if (dump_bytes(dsp, dsp->dsa_drr,
352 sizeof (dmu_replay_record_t)) != 0)
353 return (SET_ERROR(EINTR));
354 dsp->dsa_pending_op = PENDING_NONE;
355 }
356 }
357
358 /* write a FREEOBJECTS record */
359 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
360 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
361 drrfo->drr_firstobj = firstobj;
362 drrfo->drr_numobjs = numobjs;
363 drrfo->drr_toguid = dsp->dsa_toguid;
364
365 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
366
367 return (0);
368 }
369
370 static int
dump_dnode(dmu_sendarg_t * dsp,uint64_t object,dnode_phys_t * dnp)371 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
372 {
373 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
374
375 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
376 return (dump_freeobjects(dsp, object, 1));
377
378 if (dsp->dsa_pending_op != PENDING_NONE) {
379 if (dump_bytes(dsp, dsp->dsa_drr,
380 sizeof (dmu_replay_record_t)) != 0)
381 return (SET_ERROR(EINTR));
382 dsp->dsa_pending_op = PENDING_NONE;
383 }
384
385 /* write an OBJECT record */
386 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
387 dsp->dsa_drr->drr_type = DRR_OBJECT;
388 drro->drr_object = object;
389 drro->drr_type = dnp->dn_type;
390 drro->drr_bonustype = dnp->dn_bonustype;
391 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
392 drro->drr_bonuslen = dnp->dn_bonuslen;
393 drro->drr_checksumtype = dnp->dn_checksum;
394 drro->drr_compress = dnp->dn_compress;
395 drro->drr_toguid = dsp->dsa_toguid;
396
397 if (!(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
398 drro->drr_blksz > SPA_OLD_MAXBLOCKSIZE)
399 drro->drr_blksz = SPA_OLD_MAXBLOCKSIZE;
400
401 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
402 return (SET_ERROR(EINTR));
403
404 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
405 return (SET_ERROR(EINTR));
406
407 /* Free anything past the end of the file. */
408 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
409 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL) != 0)
410 return (SET_ERROR(EINTR));
411 if (dsp->dsa_err != 0)
412 return (SET_ERROR(EINTR));
413 return (0);
414 }
415
416 static boolean_t
backup_do_embed(dmu_sendarg_t * dsp,const blkptr_t * bp)417 backup_do_embed(dmu_sendarg_t *dsp, const blkptr_t *bp)
418 {
419 if (!BP_IS_EMBEDDED(bp))
420 return (B_FALSE);
421
422 /*
423 * Compression function must be legacy, or explicitly enabled.
424 */
425 if ((BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_LEGACY_FUNCTIONS &&
426 !(dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4)))
427 return (B_FALSE);
428
429 /*
430 * Embed type must be explicitly enabled.
431 */
432 switch (BPE_GET_ETYPE(bp)) {
433 case BP_EMBEDDED_TYPE_DATA:
434 if (dsp->dsa_featureflags & DMU_BACKUP_FEATURE_EMBED_DATA)
435 return (B_TRUE);
436 break;
437 default:
438 return (B_FALSE);
439 }
440 return (B_FALSE);
441 }
442
443 #define BP_SPAN(dnp, level) \
444 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
445 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
446
447 /* ARGSUSED */
448 static int
backup_cb(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)449 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
450 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
451 {
452 dmu_sendarg_t *dsp = arg;
453 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
454 int err = 0;
455
456 if (issig(JUSTLOOKING) && issig(FORREAL))
457 return (SET_ERROR(EINTR));
458
459 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
460 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
461 return (0);
462 } else if (zb->zb_level == ZB_ZIL_LEVEL) {
463 /*
464 * If we are sending a non-snapshot (which is allowed on
465 * read-only pools), it may have a ZIL, which must be ignored.
466 */
467 return (0);
468 } else if (BP_IS_HOLE(bp) &&
469 zb->zb_object == DMU_META_DNODE_OBJECT) {
470 uint64_t span = BP_SPAN(dnp, zb->zb_level);
471 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
472 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
473 } else if (BP_IS_HOLE(bp)) {
474 uint64_t span = BP_SPAN(dnp, zb->zb_level);
475 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
476 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
477 return (0);
478 } else if (type == DMU_OT_DNODE) {
479 dnode_phys_t *blk;
480 int i;
481 int blksz = BP_GET_LSIZE(bp);
482 arc_flags_t aflags = ARC_FLAG_WAIT;
483 arc_buf_t *abuf;
484
485 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
486 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
487 &aflags, zb) != 0)
488 return (SET_ERROR(EIO));
489
490 blk = abuf->b_data;
491 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
492 uint64_t dnobj = (zb->zb_blkid <<
493 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
494 err = dump_dnode(dsp, dnobj, blk+i);
495 if (err != 0)
496 break;
497 }
498 (void) arc_buf_remove_ref(abuf, &abuf);
499 } else if (type == DMU_OT_SA) {
500 arc_flags_t aflags = ARC_FLAG_WAIT;
501 arc_buf_t *abuf;
502 int blksz = BP_GET_LSIZE(bp);
503
504 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
505 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
506 &aflags, zb) != 0)
507 return (SET_ERROR(EIO));
508
509 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
510 (void) arc_buf_remove_ref(abuf, &abuf);
511 } else if (backup_do_embed(dsp, bp)) {
512 /* it's an embedded level-0 block of a regular object */
513 int blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
514 err = dump_write_embedded(dsp, zb->zb_object,
515 zb->zb_blkid * blksz, blksz, bp);
516 } else { /* it's a level-0 block of a regular object */
517 arc_flags_t aflags = ARC_FLAG_WAIT;
518 arc_buf_t *abuf;
519 int blksz = BP_GET_LSIZE(bp);
520 uint64_t offset;
521
522 ASSERT3U(blksz, ==, dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
523 ASSERT0(zb->zb_level);
524 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
525 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
526 &aflags, zb) != 0) {
527 if (zfs_send_corrupt_data) {
528 /* Send a block filled with 0x"zfs badd bloc" */
529 abuf = arc_buf_alloc(spa, blksz, &abuf,
530 ARC_BUFC_DATA);
531 uint64_t *ptr;
532 for (ptr = abuf->b_data;
533 (char *)ptr < (char *)abuf->b_data + blksz;
534 ptr++)
535 *ptr = 0x2f5baddb10c;
536 } else {
537 return (SET_ERROR(EIO));
538 }
539 }
540
541 offset = zb->zb_blkid * blksz;
542
543 if (!(dsp->dsa_featureflags &
544 DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
545 blksz > SPA_OLD_MAXBLOCKSIZE) {
546 char *buf = abuf->b_data;
547 while (blksz > 0 && err == 0) {
548 int n = MIN(blksz, SPA_OLD_MAXBLOCKSIZE);
549 err = dump_write(dsp, type, zb->zb_object,
550 offset, n, NULL, buf);
551 offset += n;
552 buf += n;
553 blksz -= n;
554 }
555 } else {
556 err = dump_write(dsp, type, zb->zb_object,
557 offset, blksz, bp, abuf->b_data);
558 }
559 (void) arc_buf_remove_ref(abuf, &abuf);
560 }
561
562 ASSERT(err == 0 || err == EINTR);
563 return (err);
564 }
565
566 /*
567 * Releases dp using the specified tag.
568 */
569 static int
dmu_send_impl(void * tag,dsl_pool_t * dp,dsl_dataset_t * ds,zfs_bookmark_phys_t * fromzb,boolean_t is_clone,boolean_t embedok,boolean_t large_block_ok,int outfd,vnode_t * vp,offset_t * off)570 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
571 zfs_bookmark_phys_t *fromzb, boolean_t is_clone, boolean_t embedok,
572 #ifdef illumos
573 boolean_t large_block_ok, int outfd, vnode_t *vp, offset_t *off)
574 #else
575 boolean_t large_block_ok, int outfd, struct file *fp, offset_t *off)
576 #endif
577 {
578 objset_t *os;
579 dmu_replay_record_t *drr;
580 dmu_sendarg_t *dsp;
581 int err;
582 uint64_t fromtxg = 0;
583 uint64_t featureflags = 0;
584
585 err = dmu_objset_from_ds(ds, &os);
586 if (err != 0) {
587 dsl_pool_rele(dp, tag);
588 return (err);
589 }
590
591 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
592 drr->drr_type = DRR_BEGIN;
593 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
594 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
595 DMU_SUBSTREAM);
596
597 #ifdef _KERNEL
598 if (dmu_objset_type(os) == DMU_OST_ZFS) {
599 uint64_t version;
600 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
601 kmem_free(drr, sizeof (dmu_replay_record_t));
602 dsl_pool_rele(dp, tag);
603 return (SET_ERROR(EINVAL));
604 }
605 if (version >= ZPL_VERSION_SA) {
606 featureflags |= DMU_BACKUP_FEATURE_SA_SPILL;
607 }
608 }
609 #endif
610
611 if (large_block_ok && ds->ds_large_blocks)
612 featureflags |= DMU_BACKUP_FEATURE_LARGE_BLOCKS;
613 if (embedok &&
614 spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA)) {
615 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA;
616 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
617 featureflags |= DMU_BACKUP_FEATURE_EMBED_DATA_LZ4;
618 } else {
619 embedok = B_FALSE;
620 }
621
622 DMU_SET_FEATUREFLAGS(drr->drr_u.drr_begin.drr_versioninfo,
623 featureflags);
624
625 drr->drr_u.drr_begin.drr_creation_time =
626 dsl_dataset_phys(ds)->ds_creation_time;
627 drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
628 if (is_clone)
629 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
630 drr->drr_u.drr_begin.drr_toguid = dsl_dataset_phys(ds)->ds_guid;
631 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
632 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
633
634 if (fromzb != NULL) {
635 drr->drr_u.drr_begin.drr_fromguid = fromzb->zbm_guid;
636 fromtxg = fromzb->zbm_creation_txg;
637 }
638 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
639 if (!dsl_dataset_is_snapshot(ds)) {
640 (void) strlcat(drr->drr_u.drr_begin.drr_toname, "@--head--",
641 sizeof (drr->drr_u.drr_begin.drr_toname));
642 }
643
644 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
645
646 dsp->dsa_drr = drr;
647 dsp->dsa_outfd = outfd;
648 dsp->dsa_proc = curproc;
649 dsp->dsa_td = curthread;
650 dsp->dsa_fp = fp;
651 dsp->dsa_os = os;
652 dsp->dsa_off = off;
653 dsp->dsa_toguid = dsl_dataset_phys(ds)->ds_guid;
654 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
655 dsp->dsa_pending_op = PENDING_NONE;
656 dsp->dsa_incremental = (fromzb != NULL);
657 dsp->dsa_featureflags = featureflags;
658
659 mutex_enter(&ds->ds_sendstream_lock);
660 list_insert_head(&ds->ds_sendstreams, dsp);
661 mutex_exit(&ds->ds_sendstream_lock);
662
663 dsl_dataset_long_hold(ds, FTAG);
664 dsl_pool_rele(dp, tag);
665
666 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
667 err = dsp->dsa_err;
668 goto out;
669 }
670
671 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
672 backup_cb, dsp);
673
674 if (dsp->dsa_pending_op != PENDING_NONE)
675 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
676 err = SET_ERROR(EINTR);
677
678 if (err != 0) {
679 if (err == EINTR && dsp->dsa_err != 0)
680 err = dsp->dsa_err;
681 goto out;
682 }
683
684 bzero(drr, sizeof (dmu_replay_record_t));
685 drr->drr_type = DRR_END;
686 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
687 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
688
689 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
690 err = dsp->dsa_err;
691 goto out;
692 }
693
694 out:
695 mutex_enter(&ds->ds_sendstream_lock);
696 list_remove(&ds->ds_sendstreams, dsp);
697 mutex_exit(&ds->ds_sendstream_lock);
698
699 kmem_free(drr, sizeof (dmu_replay_record_t));
700 kmem_free(dsp, sizeof (dmu_sendarg_t));
701
702 dsl_dataset_long_rele(ds, FTAG);
703
704 return (err);
705 }
706
707 int
dmu_send_obj(const char * pool,uint64_t tosnap,uint64_t fromsnap,boolean_t embedok,boolean_t large_block_ok,int outfd,vnode_t * vp,offset_t * off)708 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
709 boolean_t embedok, boolean_t large_block_ok,
710 #ifdef illumos
711 int outfd, vnode_t *vp, offset_t *off)
712 #else
713 int outfd, struct file *fp, offset_t *off)
714 #endif
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, fp, off);
751 } else {
752 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
753 embedok, large_block_ok, outfd, fp, off);
754 }
755 dsl_dataset_rele(ds, FTAG);
756 return (err);
757 }
758
759 int
dmu_send(const char * tosnap,const char * fromsnap,boolean_t embedok,boolean_t large_block_ok,int outfd,vnode_t * vp,offset_t * off)760 dmu_send(const char *tosnap, const char *fromsnap,
761 boolean_t embedok, boolean_t large_block_ok,
762 #ifdef illumos
763 int outfd, vnode_t *vp, offset_t *off)
764 #else
765 int outfd, struct file *fp, offset_t *off)
766 #endif
767 {
768 dsl_pool_t *dp;
769 dsl_dataset_t *ds;
770 int err;
771 boolean_t owned = B_FALSE;
772
773 if (fromsnap != NULL && strpbrk(fromsnap, "@#") == NULL)
774 return (SET_ERROR(EINVAL));
775
776 err = dsl_pool_hold(tosnap, FTAG, &dp);
777 if (err != 0)
778 return (err);
779
780 if (strchr(tosnap, '@') == NULL && spa_writeable(dp->dp_spa)) {
781 /*
782 * We are sending a filesystem or volume. Ensure
783 * that it doesn't change by owning the dataset.
784 */
785 err = dsl_dataset_own(dp, tosnap, FTAG, &ds);
786 owned = B_TRUE;
787 } else {
788 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
789 }
790 if (err != 0) {
791 dsl_pool_rele(dp, FTAG);
792 return (err);
793 }
794
795 if (fromsnap != NULL) {
796 zfs_bookmark_phys_t zb;
797 boolean_t is_clone = B_FALSE;
798 int fsnamelen = strchr(tosnap, '@') - tosnap;
799
800 /*
801 * If the fromsnap is in a different filesystem, then
802 * mark the send stream as a clone.
803 */
804 if (strncmp(tosnap, fromsnap, fsnamelen) != 0 ||
805 (fromsnap[fsnamelen] != '@' &&
806 fromsnap[fsnamelen] != '#')) {
807 is_clone = B_TRUE;
808 }
809
810 if (strchr(fromsnap, '@')) {
811 dsl_dataset_t *fromds;
812 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
813 if (err == 0) {
814 if (!dsl_dataset_is_before(ds, fromds, 0))
815 err = SET_ERROR(EXDEV);
816 zb.zbm_creation_time =
817 dsl_dataset_phys(fromds)->ds_creation_time;
818 zb.zbm_creation_txg =
819 dsl_dataset_phys(fromds)->ds_creation_txg;
820 zb.zbm_guid = dsl_dataset_phys(fromds)->ds_guid;
821 is_clone = (ds->ds_dir != fromds->ds_dir);
822 dsl_dataset_rele(fromds, FTAG);
823 }
824 } else {
825 err = dsl_bookmark_lookup(dp, fromsnap, ds, &zb);
826 }
827 if (err != 0) {
828 dsl_dataset_rele(ds, FTAG);
829 dsl_pool_rele(dp, FTAG);
830 return (err);
831 }
832 err = dmu_send_impl(FTAG, dp, ds, &zb, is_clone,
833 embedok, large_block_ok, outfd, fp, off);
834 } else {
835 err = dmu_send_impl(FTAG, dp, ds, NULL, B_FALSE,
836 embedok, large_block_ok, outfd, fp, off);
837 }
838 if (owned)
839 dsl_dataset_disown(ds, FTAG);
840 else
841 dsl_dataset_rele(ds, FTAG);
842 return (err);
843 }
844
845 int
dmu_send_estimate(dsl_dataset_t * ds,dsl_dataset_t * fromds,uint64_t * sizep)846 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
847 {
848 dsl_pool_t *dp = ds->ds_dir->dd_pool;
849 int err;
850 uint64_t size;
851
852 ASSERT(dsl_pool_config_held(dp));
853
854 /* tosnap must be a snapshot */
855 if (!dsl_dataset_is_snapshot(ds))
856 return (SET_ERROR(EINVAL));
857
858 /*
859 * fromsnap must be an earlier snapshot from the same fs as tosnap,
860 * or the origin's fs.
861 */
862 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds, 0))
863 return (SET_ERROR(EXDEV));
864
865 /* Get uncompressed size estimate of changed data. */
866 if (fromds == NULL) {
867 size = dsl_dataset_phys(ds)->ds_uncompressed_bytes;
868 } else {
869 uint64_t used, comp;
870 err = dsl_dataset_space_written(fromds, ds,
871 &used, &comp, &size);
872 if (err != 0)
873 return (err);
874 }
875
876 /*
877 * Assume that space (both on-disk and in-stream) is dominated by
878 * data. We will adjust for indirect blocks and the copies property,
879 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
880 */
881
882 /*
883 * Subtract out approximate space used by indirect blocks.
884 * Assume most space is used by data blocks (non-indirect, non-dnode).
885 * Assume all blocks are recordsize. Assume ditto blocks and
886 * internal fragmentation counter out compression.
887 *
888 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
889 * block, which we observe in practice.
890 */
891 uint64_t recordsize;
892 err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
893 if (err != 0)
894 return (err);
895 size -= size / recordsize * sizeof (blkptr_t);
896
897 /* Add in the space for the record associated with each block. */
898 size += size / recordsize * sizeof (dmu_replay_record_t);
899
900 *sizep = size;
901
902 return (0);
903 }
904
905 typedef struct dmu_recv_begin_arg {
906 const char *drba_origin;
907 dmu_recv_cookie_t *drba_cookie;
908 cred_t *drba_cred;
909 uint64_t drba_snapobj;
910 } dmu_recv_begin_arg_t;
911
912 static int
recv_begin_check_existing_impl(dmu_recv_begin_arg_t * drba,dsl_dataset_t * ds,uint64_t fromguid)913 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
914 uint64_t fromguid)
915 {
916 uint64_t val;
917 int error;
918 dsl_pool_t *dp = ds->ds_dir->dd_pool;
919
920 /* temporary clone name must not exist */
921 error = zap_lookup(dp->dp_meta_objset,
922 dsl_dir_phys(ds->ds_dir)->dd_child_dir_zapobj, recv_clone_name,
923 8, 1, &val);
924 if (error != ENOENT)
925 return (error == 0 ? EBUSY : error);
926
927 /* new snapshot name must not exist */
928 error = zap_lookup(dp->dp_meta_objset,
929 dsl_dataset_phys(ds)->ds_snapnames_zapobj,
930 drba->drba_cookie->drc_tosnap, 8, 1, &val);
931 if (error != ENOENT)
932 return (error == 0 ? EEXIST : error);
933
934 /*
935 * Check snapshot limit before receiving. We'll recheck again at the
936 * end, but might as well abort before receiving if we're already over
937 * the limit.
938 *
939 * Note that we do not check the file system limit with
940 * dsl_dir_fscount_check because the temporary %clones don't count
941 * against that limit.
942 */
943 error = dsl_fs_ss_limit_check(ds->ds_dir, 1, ZFS_PROP_SNAPSHOT_LIMIT,
944 NULL, drba->drba_cred);
945 if (error != 0)
946 return (error);
947
948 if (fromguid != 0) {
949 dsl_dataset_t *snap;
950 uint64_t obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
951
952 /* Find snapshot in this dir that matches fromguid. */
953 while (obj != 0) {
954 error = dsl_dataset_hold_obj(dp, obj, FTAG,
955 &snap);
956 if (error != 0)
957 return (SET_ERROR(ENODEV));
958 if (snap->ds_dir != ds->ds_dir) {
959 dsl_dataset_rele(snap, FTAG);
960 return (SET_ERROR(ENODEV));
961 }
962 if (dsl_dataset_phys(snap)->ds_guid == fromguid)
963 break;
964 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
965 dsl_dataset_rele(snap, FTAG);
966 }
967 if (obj == 0)
968 return (SET_ERROR(ENODEV));
969
970 if (drba->drba_cookie->drc_force) {
971 drba->drba_snapobj = obj;
972 } else {
973 /*
974 * If we are not forcing, there must be no
975 * changes since fromsnap.
976 */
977 if (dsl_dataset_modified_since_snap(ds, snap)) {
978 dsl_dataset_rele(snap, FTAG);
979 return (SET_ERROR(ETXTBSY));
980 }
981 drba->drba_snapobj = ds->ds_prev->ds_object;
982 }
983
984 dsl_dataset_rele(snap, FTAG);
985 } else {
986 /* if full, then must be forced */
987 if (!drba->drba_cookie->drc_force)
988 return (SET_ERROR(EEXIST));
989 /* start from $ORIGIN@$ORIGIN, if supported */
990 drba->drba_snapobj = dp->dp_origin_snap != NULL ?
991 dp->dp_origin_snap->ds_object : 0;
992 }
993
994 return (0);
995
996 }
997
998 static int
dmu_recv_begin_check(void * arg,dmu_tx_t * tx)999 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
1000 {
1001 dmu_recv_begin_arg_t *drba = arg;
1002 dsl_pool_t *dp = dmu_tx_pool(tx);
1003 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1004 uint64_t fromguid = drrb->drr_fromguid;
1005 int flags = drrb->drr_flags;
1006 int error;
1007 uint64_t featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
1008 dsl_dataset_t *ds;
1009 const char *tofs = drba->drba_cookie->drc_tofs;
1010
1011 /* already checked */
1012 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
1013
1014 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
1015 DMU_COMPOUNDSTREAM ||
1016 drrb->drr_type >= DMU_OST_NUMTYPES ||
1017 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
1018 return (SET_ERROR(EINVAL));
1019
1020 /* Verify pool version supports SA if SA_SPILL feature set */
1021 if ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
1022 spa_version(dp->dp_spa) < SPA_VERSION_SA)
1023 return (SET_ERROR(ENOTSUP));
1024
1025 /*
1026 * The receiving code doesn't know how to translate a WRITE_EMBEDDED
1027 * record to a plan WRITE record, so the pool must have the
1028 * EMBEDDED_DATA feature enabled if the stream has WRITE_EMBEDDED
1029 * records. Same with WRITE_EMBEDDED records that use LZ4 compression.
1030 */
1031 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA) &&
1032 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_EMBEDDED_DATA))
1033 return (SET_ERROR(ENOTSUP));
1034 if ((featureflags & DMU_BACKUP_FEATURE_EMBED_DATA_LZ4) &&
1035 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LZ4_COMPRESS))
1036 return (SET_ERROR(ENOTSUP));
1037
1038 /*
1039 * The receiving code doesn't know how to translate large blocks
1040 * to smaller ones, so the pool must have the LARGE_BLOCKS
1041 * feature enabled if the stream has LARGE_BLOCKS.
1042 */
1043 if ((featureflags & DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
1044 !spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
1045 return (SET_ERROR(ENOTSUP));
1046
1047 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1048 if (error == 0) {
1049 /* target fs already exists; recv into temp clone */
1050
1051 /* Can't recv a clone into an existing fs */
1052 if (flags & DRR_FLAG_CLONE) {
1053 dsl_dataset_rele(ds, FTAG);
1054 return (SET_ERROR(EINVAL));
1055 }
1056
1057 error = recv_begin_check_existing_impl(drba, ds, fromguid);
1058 dsl_dataset_rele(ds, FTAG);
1059 } else if (error == ENOENT) {
1060 /* target fs does not exist; must be a full backup or clone */
1061 char buf[MAXNAMELEN];
1062
1063 /*
1064 * If it's a non-clone incremental, we are missing the
1065 * target fs, so fail the recv.
1066 */
1067 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
1068 return (SET_ERROR(ENOENT));
1069
1070 /* Open the parent of tofs */
1071 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
1072 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
1073 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
1074 if (error != 0)
1075 return (error);
1076
1077 /*
1078 * Check filesystem and snapshot limits before receiving. We'll
1079 * recheck snapshot limits again at the end (we create the
1080 * filesystems and increment those counts during begin_sync).
1081 */
1082 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1083 ZFS_PROP_FILESYSTEM_LIMIT, NULL, drba->drba_cred);
1084 if (error != 0) {
1085 dsl_dataset_rele(ds, FTAG);
1086 return (error);
1087 }
1088
1089 error = dsl_fs_ss_limit_check(ds->ds_dir, 1,
1090 ZFS_PROP_SNAPSHOT_LIMIT, NULL, drba->drba_cred);
1091 if (error != 0) {
1092 dsl_dataset_rele(ds, FTAG);
1093 return (error);
1094 }
1095
1096 if (drba->drba_origin != NULL) {
1097 dsl_dataset_t *origin;
1098 error = dsl_dataset_hold(dp, drba->drba_origin,
1099 FTAG, &origin);
1100 if (error != 0) {
1101 dsl_dataset_rele(ds, FTAG);
1102 return (error);
1103 }
1104 if (!dsl_dataset_is_snapshot(origin)) {
1105 dsl_dataset_rele(origin, FTAG);
1106 dsl_dataset_rele(ds, FTAG);
1107 return (SET_ERROR(EINVAL));
1108 }
1109 if (dsl_dataset_phys(origin)->ds_guid != fromguid) {
1110 dsl_dataset_rele(origin, FTAG);
1111 dsl_dataset_rele(ds, FTAG);
1112 return (SET_ERROR(ENODEV));
1113 }
1114 dsl_dataset_rele(origin, FTAG);
1115 }
1116 dsl_dataset_rele(ds, FTAG);
1117 error = 0;
1118 }
1119 return (error);
1120 }
1121
1122 static void
dmu_recv_begin_sync(void * arg,dmu_tx_t * tx)1123 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
1124 {
1125 dmu_recv_begin_arg_t *drba = arg;
1126 dsl_pool_t *dp = dmu_tx_pool(tx);
1127 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
1128 const char *tofs = drba->drba_cookie->drc_tofs;
1129 dsl_dataset_t *ds, *newds;
1130 uint64_t dsobj;
1131 int error;
1132 uint64_t crflags;
1133
1134 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
1135 DS_FLAG_CI_DATASET : 0;
1136
1137 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
1138 if (error == 0) {
1139 /* create temporary clone */
1140 dsl_dataset_t *snap = NULL;
1141 if (drba->drba_snapobj != 0) {
1142 VERIFY0(dsl_dataset_hold_obj(dp,
1143 drba->drba_snapobj, FTAG, &snap));
1144 }
1145 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
1146 snap, crflags, drba->drba_cred, tx);
1147 if (drba->drba_snapobj != 0)
1148 dsl_dataset_rele(snap, FTAG);
1149 dsl_dataset_rele(ds, FTAG);
1150 } else {
1151 dsl_dir_t *dd;
1152 const char *tail;
1153 dsl_dataset_t *origin = NULL;
1154
1155 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
1156
1157 if (drba->drba_origin != NULL) {
1158 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
1159 FTAG, &origin));
1160 }
1161
1162 /* Create new dataset. */
1163 dsobj = dsl_dataset_create_sync(dd,
1164 strrchr(tofs, '/') + 1,
1165 origin, crflags, drba->drba_cred, tx);
1166 if (origin != NULL)
1167 dsl_dataset_rele(origin, FTAG);
1168 dsl_dir_rele(dd, FTAG);
1169 drba->drba_cookie->drc_newfs = B_TRUE;
1170 }
1171 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
1172
1173 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
1174 DMU_BACKUP_FEATURE_LARGE_BLOCKS) &&
1175 !newds->ds_large_blocks) {
1176 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1177 newds->ds_large_blocks = B_TRUE;
1178 }
1179
1180 dmu_buf_will_dirty(newds->ds_dbuf, tx);
1181 dsl_dataset_phys(newds)->ds_flags |= DS_FLAG_INCONSISTENT;
1182
1183 /*
1184 * If we actually created a non-clone, we need to create the
1185 * objset in our new dataset.
1186 */
1187 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
1188 (void) dmu_objset_create_impl(dp->dp_spa,
1189 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
1190 }
1191
1192 drba->drba_cookie->drc_ds = newds;
1193
1194 spa_history_log_internal_ds(newds, "receive", tx, "");
1195 }
1196
1197 /*
1198 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
1199 * succeeds; otherwise we will leak the holds on the datasets.
1200 */
1201 int
dmu_recv_begin(char * tofs,char * tosnap,struct drr_begin * drrb,boolean_t force,char * origin,dmu_recv_cookie_t * drc)1202 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
1203 boolean_t force, char *origin, dmu_recv_cookie_t *drc)
1204 {
1205 dmu_recv_begin_arg_t drba = { 0 };
1206 dmu_replay_record_t *drr;
1207
1208 bzero(drc, sizeof (dmu_recv_cookie_t));
1209 drc->drc_drrb = drrb;
1210 drc->drc_tosnap = tosnap;
1211 drc->drc_tofs = tofs;
1212 drc->drc_force = force;
1213 drc->drc_cred = CRED();
1214
1215 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1216 drc->drc_byteswap = B_TRUE;
1217 else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
1218 return (SET_ERROR(EINVAL));
1219
1220 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1221 drr->drr_type = DRR_BEGIN;
1222 drr->drr_u.drr_begin = *drc->drc_drrb;
1223 if (drc->drc_byteswap) {
1224 fletcher_4_incremental_byteswap(drr,
1225 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1226 } else {
1227 fletcher_4_incremental_native(drr,
1228 sizeof (dmu_replay_record_t), &drc->drc_cksum);
1229 }
1230 kmem_free(drr, sizeof (dmu_replay_record_t));
1231
1232 if (drc->drc_byteswap) {
1233 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1234 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1235 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1236 drrb->drr_type = BSWAP_32(drrb->drr_type);
1237 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1238 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1239 }
1240
1241 drba.drba_origin = origin;
1242 drba.drba_cookie = drc;
1243 drba.drba_cred = CRED();
1244
1245 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
1246 &drba, 5, ZFS_SPACE_CHECK_NORMAL));
1247 }
1248
1249 struct restorearg {
1250 int err;
1251 boolean_t byteswap;
1252 kthread_t *td;
1253 struct file *fp;
1254 char *buf;
1255 uint64_t voff;
1256 int bufsize; /* amount of memory allocated for buf */
1257 zio_cksum_t cksum;
1258 avl_tree_t *guid_to_ds_map;
1259 };
1260
1261 typedef struct guid_map_entry {
1262 uint64_t guid;
1263 dsl_dataset_t *gme_ds;
1264 avl_node_t avlnode;
1265 } guid_map_entry_t;
1266
1267 static int
guid_compare(const void * arg1,const void * arg2)1268 guid_compare(const void *arg1, const void *arg2)
1269 {
1270 const guid_map_entry_t *gmep1 = arg1;
1271 const guid_map_entry_t *gmep2 = arg2;
1272
1273 if (gmep1->guid < gmep2->guid)
1274 return (-1);
1275 else if (gmep1->guid > gmep2->guid)
1276 return (1);
1277 return (0);
1278 }
1279
1280 static void
free_guid_map_onexit(void * arg)1281 free_guid_map_onexit(void *arg)
1282 {
1283 avl_tree_t *ca = arg;
1284 void *cookie = NULL;
1285 guid_map_entry_t *gmep;
1286
1287 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
1288 dsl_dataset_long_rele(gmep->gme_ds, gmep);
1289 dsl_dataset_rele(gmep->gme_ds, gmep);
1290 kmem_free(gmep, sizeof (guid_map_entry_t));
1291 }
1292 avl_destroy(ca);
1293 kmem_free(ca, sizeof (avl_tree_t));
1294 }
1295
1296 static int
restore_bytes(struct restorearg * ra,void * buf,int len,off_t off,ssize_t * resid)1297 restore_bytes(struct restorearg *ra, void *buf, int len, off_t off, ssize_t *resid)
1298 {
1299 struct uio auio;
1300 struct iovec aiov;
1301 int error;
1302
1303 aiov.iov_base = buf;
1304 aiov.iov_len = len;
1305 auio.uio_iov = &aiov;
1306 auio.uio_iovcnt = 1;
1307 auio.uio_resid = len;
1308 auio.uio_segflg = UIO_SYSSPACE;
1309 auio.uio_rw = UIO_READ;
1310 auio.uio_offset = off;
1311 auio.uio_td = ra->td;
1312 #ifdef _KERNEL
1313 error = fo_read(ra->fp, &auio, ra->td->td_ucred, FOF_OFFSET, ra->td);
1314 #else
1315 fprintf(stderr, "%s: returning EOPNOTSUPP\n", __func__);
1316 error = EOPNOTSUPP;
1317 #endif
1318 *resid = auio.uio_resid;
1319 return (error);
1320 }
1321
1322 static void *
restore_read(struct restorearg * ra,int len,char * buf)1323 restore_read(struct restorearg *ra, int len, char *buf)
1324 {
1325 int done = 0;
1326
1327 if (buf == NULL)
1328 buf = ra->buf;
1329
1330 /* some things will require 8-byte alignment, so everything must */
1331 ASSERT0(len % 8);
1332 ASSERT3U(len, <=, ra->bufsize);
1333
1334 while (done < len) {
1335 ssize_t resid;
1336
1337 ra->err = restore_bytes(ra, buf + done,
1338 len - done, ra->voff, &resid);
1339
1340 if (resid == len - done)
1341 ra->err = SET_ERROR(EINVAL);
1342 ra->voff += len - done - resid;
1343 done = len - resid;
1344 if (ra->err != 0)
1345 return (NULL);
1346 }
1347
1348 ASSERT3U(done, ==, len);
1349 if (ra->byteswap)
1350 fletcher_4_incremental_byteswap(buf, len, &ra->cksum);
1351 else
1352 fletcher_4_incremental_native(buf, len, &ra->cksum);
1353 return (buf);
1354 }
1355
1356 static void
backup_byteswap(dmu_replay_record_t * drr)1357 backup_byteswap(dmu_replay_record_t *drr)
1358 {
1359 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1360 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1361 drr->drr_type = BSWAP_32(drr->drr_type);
1362 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1363 switch (drr->drr_type) {
1364 case DRR_BEGIN:
1365 DO64(drr_begin.drr_magic);
1366 DO64(drr_begin.drr_versioninfo);
1367 DO64(drr_begin.drr_creation_time);
1368 DO32(drr_begin.drr_type);
1369 DO32(drr_begin.drr_flags);
1370 DO64(drr_begin.drr_toguid);
1371 DO64(drr_begin.drr_fromguid);
1372 break;
1373 case DRR_OBJECT:
1374 DO64(drr_object.drr_object);
1375 DO32(drr_object.drr_type);
1376 DO32(drr_object.drr_bonustype);
1377 DO32(drr_object.drr_blksz);
1378 DO32(drr_object.drr_bonuslen);
1379 DO64(drr_object.drr_toguid);
1380 break;
1381 case DRR_FREEOBJECTS:
1382 DO64(drr_freeobjects.drr_firstobj);
1383 DO64(drr_freeobjects.drr_numobjs);
1384 DO64(drr_freeobjects.drr_toguid);
1385 break;
1386 case DRR_WRITE:
1387 DO64(drr_write.drr_object);
1388 DO32(drr_write.drr_type);
1389 DO64(drr_write.drr_offset);
1390 DO64(drr_write.drr_length);
1391 DO64(drr_write.drr_toguid);
1392 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1393 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1394 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1395 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1396 DO64(drr_write.drr_key.ddk_prop);
1397 break;
1398 case DRR_WRITE_BYREF:
1399 DO64(drr_write_byref.drr_object);
1400 DO64(drr_write_byref.drr_offset);
1401 DO64(drr_write_byref.drr_length);
1402 DO64(drr_write_byref.drr_toguid);
1403 DO64(drr_write_byref.drr_refguid);
1404 DO64(drr_write_byref.drr_refobject);
1405 DO64(drr_write_byref.drr_refoffset);
1406 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1407 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1408 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1409 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1410 DO64(drr_write_byref.drr_key.ddk_prop);
1411 break;
1412 case DRR_WRITE_EMBEDDED:
1413 DO64(drr_write_embedded.drr_object);
1414 DO64(drr_write_embedded.drr_offset);
1415 DO64(drr_write_embedded.drr_length);
1416 DO64(drr_write_embedded.drr_toguid);
1417 DO32(drr_write_embedded.drr_lsize);
1418 DO32(drr_write_embedded.drr_psize);
1419 break;
1420 case DRR_FREE:
1421 DO64(drr_free.drr_object);
1422 DO64(drr_free.drr_offset);
1423 DO64(drr_free.drr_length);
1424 DO64(drr_free.drr_toguid);
1425 break;
1426 case DRR_SPILL:
1427 DO64(drr_spill.drr_object);
1428 DO64(drr_spill.drr_length);
1429 DO64(drr_spill.drr_toguid);
1430 break;
1431 case DRR_END:
1432 DO64(drr_end.drr_checksum.zc_word[0]);
1433 DO64(drr_end.drr_checksum.zc_word[1]);
1434 DO64(drr_end.drr_checksum.zc_word[2]);
1435 DO64(drr_end.drr_checksum.zc_word[3]);
1436 DO64(drr_end.drr_toguid);
1437 break;
1438 }
1439 #undef DO64
1440 #undef DO32
1441 }
1442
1443 static inline uint8_t
deduce_nblkptr(dmu_object_type_t bonus_type,uint64_t bonus_size)1444 deduce_nblkptr(dmu_object_type_t bonus_type, uint64_t bonus_size)
1445 {
1446 if (bonus_type == DMU_OT_SA) {
1447 return (1);
1448 } else {
1449 return (1 +
1450 ((DN_MAX_BONUSLEN - bonus_size) >> SPA_BLKPTRSHIFT));
1451 }
1452 }
1453
1454 static int
restore_object(struct restorearg * ra,objset_t * os,struct drr_object * drro)1455 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1456 {
1457 dmu_object_info_t doi;
1458 dmu_tx_t *tx;
1459 void *data = NULL;
1460 uint64_t object;
1461 int err;
1462
1463 if (drro->drr_type == DMU_OT_NONE ||
1464 !DMU_OT_IS_VALID(drro->drr_type) ||
1465 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1466 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1467 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1468 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1469 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1470 drro->drr_blksz > spa_maxblocksize(dmu_objset_spa(os)) ||
1471 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1472 return (SET_ERROR(EINVAL));
1473 }
1474
1475 err = dmu_object_info(os, drro->drr_object, &doi);
1476
1477 if (err != 0 && err != ENOENT)
1478 return (SET_ERROR(EINVAL));
1479 object = err == 0 ? drro->drr_object : DMU_NEW_OBJECT;
1480
1481 if (drro->drr_bonuslen) {
1482 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8), NULL);
1483 if (ra->err != 0)
1484 return (ra->err);
1485 }
1486
1487 /*
1488 * If we are losing blkptrs or changing the block size this must
1489 * be a new file instance. We must clear out the previous file
1490 * contents before we can change this type of metadata in the dnode.
1491 */
1492 if (err == 0) {
1493 int nblkptr;
1494
1495 nblkptr = deduce_nblkptr(drro->drr_bonustype,
1496 drro->drr_bonuslen);
1497
1498 if (drro->drr_blksz != doi.doi_data_block_size ||
1499 nblkptr < doi.doi_nblkptr) {
1500 err = dmu_free_long_range(os, drro->drr_object,
1501 0, DMU_OBJECT_END);
1502 if (err != 0)
1503 return (SET_ERROR(EINVAL));
1504 }
1505 }
1506
1507 tx = dmu_tx_create(os);
1508 dmu_tx_hold_bonus(tx, object);
1509 err = dmu_tx_assign(tx, TXG_WAIT);
1510 if (err != 0) {
1511 dmu_tx_abort(tx);
1512 return (err);
1513 }
1514
1515 if (object == DMU_NEW_OBJECT) {
1516 /* currently free, want to be allocated */
1517 err = dmu_object_claim(os, drro->drr_object,
1518 drro->drr_type, drro->drr_blksz,
1519 drro->drr_bonustype, drro->drr_bonuslen, tx);
1520 } else if (drro->drr_type != doi.doi_type ||
1521 drro->drr_blksz != doi.doi_data_block_size ||
1522 drro->drr_bonustype != doi.doi_bonus_type ||
1523 drro->drr_bonuslen != doi.doi_bonus_size) {
1524 /* currently allocated, but with different properties */
1525 err = dmu_object_reclaim(os, drro->drr_object,
1526 drro->drr_type, drro->drr_blksz,
1527 drro->drr_bonustype, drro->drr_bonuslen, tx);
1528 }
1529 if (err != 0) {
1530 dmu_tx_commit(tx);
1531 return (SET_ERROR(EINVAL));
1532 }
1533
1534 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1535 tx);
1536 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1537
1538 if (data != NULL) {
1539 dmu_buf_t *db;
1540
1541 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1542 dmu_buf_will_dirty(db, tx);
1543
1544 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1545 bcopy(data, db->db_data, drro->drr_bonuslen);
1546 if (ra->byteswap) {
1547 dmu_object_byteswap_t byteswap =
1548 DMU_OT_BYTESWAP(drro->drr_bonustype);
1549 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1550 drro->drr_bonuslen);
1551 }
1552 dmu_buf_rele(db, FTAG);
1553 }
1554 dmu_tx_commit(tx);
1555 return (0);
1556 }
1557
1558 /* ARGSUSED */
1559 static int
restore_freeobjects(struct restorearg * ra,objset_t * os,struct drr_freeobjects * drrfo)1560 restore_freeobjects(struct restorearg *ra, objset_t *os,
1561 struct drr_freeobjects *drrfo)
1562 {
1563 uint64_t obj;
1564
1565 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1566 return (SET_ERROR(EINVAL));
1567
1568 for (obj = drrfo->drr_firstobj;
1569 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1570 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1571 int err;
1572
1573 if (dmu_object_info(os, obj, NULL) != 0)
1574 continue;
1575
1576 err = dmu_free_long_object(os, obj);
1577 if (err != 0)
1578 return (err);
1579 }
1580 return (0);
1581 }
1582
1583 static int
restore_write(struct restorearg * ra,objset_t * os,struct drr_write * drrw)1584 restore_write(struct restorearg *ra, objset_t *os,
1585 struct drr_write *drrw)
1586 {
1587 dmu_tx_t *tx;
1588 void *data;
1589 int err;
1590
1591 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1592 !DMU_OT_IS_VALID(drrw->drr_type))
1593 return (SET_ERROR(EINVAL));
1594
1595 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1596 return (SET_ERROR(EINVAL));
1597
1598 dmu_buf_t *bonus;
1599 if (dmu_bonus_hold(os, drrw->drr_object, FTAG, &bonus) != 0)
1600 return (SET_ERROR(EINVAL));
1601
1602 arc_buf_t *abuf = dmu_request_arcbuf(bonus, drrw->drr_length);
1603
1604 data = restore_read(ra, drrw->drr_length, abuf->b_data);
1605 if (data == NULL) {
1606 dmu_return_arcbuf(abuf);
1607 dmu_buf_rele(bonus, FTAG);
1608 return (ra->err);
1609 }
1610
1611 tx = dmu_tx_create(os);
1612
1613 dmu_tx_hold_write(tx, drrw->drr_object,
1614 drrw->drr_offset, drrw->drr_length);
1615 err = dmu_tx_assign(tx, TXG_WAIT);
1616 if (err != 0) {
1617 dmu_return_arcbuf(abuf);
1618 dmu_buf_rele(bonus, FTAG);
1619 dmu_tx_abort(tx);
1620 return (err);
1621 }
1622 if (ra->byteswap) {
1623 dmu_object_byteswap_t byteswap =
1624 DMU_OT_BYTESWAP(drrw->drr_type);
1625 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1626 }
1627 dmu_assign_arcbuf(bonus, drrw->drr_offset, abuf, tx);
1628 dmu_tx_commit(tx);
1629 dmu_buf_rele(bonus, FTAG);
1630 return (0);
1631 }
1632
1633 /*
1634 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1635 * streams to refer to a copy of the data that is already on the
1636 * system because it came in earlier in the stream. This function
1637 * finds the earlier copy of the data, and uses that copy instead of
1638 * data from the stream to fulfill this write.
1639 */
1640 static int
restore_write_byref(struct restorearg * ra,objset_t * os,struct drr_write_byref * drrwbr)1641 restore_write_byref(struct restorearg *ra, objset_t *os,
1642 struct drr_write_byref *drrwbr)
1643 {
1644 dmu_tx_t *tx;
1645 int err;
1646 guid_map_entry_t gmesrch;
1647 guid_map_entry_t *gmep;
1648 avl_index_t where;
1649 objset_t *ref_os = NULL;
1650 dmu_buf_t *dbp;
1651
1652 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1653 return (SET_ERROR(EINVAL));
1654
1655 /*
1656 * If the GUID of the referenced dataset is different from the
1657 * GUID of the target dataset, find the referenced dataset.
1658 */
1659 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1660 gmesrch.guid = drrwbr->drr_refguid;
1661 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1662 &where)) == NULL) {
1663 return (SET_ERROR(EINVAL));
1664 }
1665 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1666 return (SET_ERROR(EINVAL));
1667 } else {
1668 ref_os = os;
1669 }
1670
1671 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1672 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1673 if (err != 0)
1674 return (err);
1675
1676 tx = dmu_tx_create(os);
1677
1678 dmu_tx_hold_write(tx, drrwbr->drr_object,
1679 drrwbr->drr_offset, drrwbr->drr_length);
1680 err = dmu_tx_assign(tx, TXG_WAIT);
1681 if (err != 0) {
1682 dmu_tx_abort(tx);
1683 return (err);
1684 }
1685 dmu_write(os, drrwbr->drr_object,
1686 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1687 dmu_buf_rele(dbp, FTAG);
1688 dmu_tx_commit(tx);
1689 return (0);
1690 }
1691
1692 static int
restore_write_embedded(struct restorearg * ra,objset_t * os,struct drr_write_embedded * drrwnp)1693 restore_write_embedded(struct restorearg *ra, objset_t *os,
1694 struct drr_write_embedded *drrwnp)
1695 {
1696 dmu_tx_t *tx;
1697 int err;
1698 void *data;
1699
1700 if (drrwnp->drr_offset + drrwnp->drr_length < drrwnp->drr_offset)
1701 return (EINVAL);
1702
1703 if (drrwnp->drr_psize > BPE_PAYLOAD_SIZE)
1704 return (EINVAL);
1705
1706 if (drrwnp->drr_etype >= NUM_BP_EMBEDDED_TYPES)
1707 return (EINVAL);
1708 if (drrwnp->drr_compression >= ZIO_COMPRESS_FUNCTIONS)
1709 return (EINVAL);
1710
1711 data = restore_read(ra, P2ROUNDUP(drrwnp->drr_psize, 8), NULL);
1712 if (data == NULL)
1713 return (ra->err);
1714
1715 tx = dmu_tx_create(os);
1716
1717 dmu_tx_hold_write(tx, drrwnp->drr_object,
1718 drrwnp->drr_offset, drrwnp->drr_length);
1719 err = dmu_tx_assign(tx, TXG_WAIT);
1720 if (err != 0) {
1721 dmu_tx_abort(tx);
1722 return (err);
1723 }
1724
1725 dmu_write_embedded(os, drrwnp->drr_object,
1726 drrwnp->drr_offset, data, drrwnp->drr_etype,
1727 drrwnp->drr_compression, drrwnp->drr_lsize, drrwnp->drr_psize,
1728 ra->byteswap ^ ZFS_HOST_BYTEORDER, tx);
1729
1730 dmu_tx_commit(tx);
1731 return (0);
1732 }
1733
1734 static int
restore_spill(struct restorearg * ra,objset_t * os,struct drr_spill * drrs)1735 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1736 {
1737 dmu_tx_t *tx;
1738 void *data;
1739 dmu_buf_t *db, *db_spill;
1740 int err;
1741
1742 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1743 drrs->drr_length > spa_maxblocksize(dmu_objset_spa(os)))
1744 return (SET_ERROR(EINVAL));
1745
1746 data = restore_read(ra, drrs->drr_length, NULL);
1747 if (data == NULL)
1748 return (ra->err);
1749
1750 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1751 return (SET_ERROR(EINVAL));
1752
1753 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1754 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1755 dmu_buf_rele(db, FTAG);
1756 return (err);
1757 }
1758
1759 tx = dmu_tx_create(os);
1760
1761 dmu_tx_hold_spill(tx, db->db_object);
1762
1763 err = dmu_tx_assign(tx, TXG_WAIT);
1764 if (err != 0) {
1765 dmu_buf_rele(db, FTAG);
1766 dmu_buf_rele(db_spill, FTAG);
1767 dmu_tx_abort(tx);
1768 return (err);
1769 }
1770 dmu_buf_will_dirty(db_spill, tx);
1771
1772 if (db_spill->db_size < drrs->drr_length)
1773 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1774 drrs->drr_length, tx));
1775 bcopy(data, db_spill->db_data, drrs->drr_length);
1776
1777 dmu_buf_rele(db, FTAG);
1778 dmu_buf_rele(db_spill, FTAG);
1779
1780 dmu_tx_commit(tx);
1781 return (0);
1782 }
1783
1784 /* ARGSUSED */
1785 static int
restore_free(struct restorearg * ra,objset_t * os,struct drr_free * drrf)1786 restore_free(struct restorearg *ra, objset_t *os,
1787 struct drr_free *drrf)
1788 {
1789 int err;
1790
1791 if (drrf->drr_length != -1ULL &&
1792 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1793 return (SET_ERROR(EINVAL));
1794
1795 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1796 return (SET_ERROR(EINVAL));
1797
1798 err = dmu_free_long_range(os, drrf->drr_object,
1799 drrf->drr_offset, drrf->drr_length);
1800 return (err);
1801 }
1802
1803 /* used to destroy the drc_ds on error */
1804 static void
dmu_recv_cleanup_ds(dmu_recv_cookie_t * drc)1805 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1806 {
1807 char name[MAXNAMELEN];
1808 dsl_dataset_name(drc->drc_ds, name);
1809 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1810 (void) dsl_destroy_head(name);
1811 }
1812
1813 /*
1814 * NB: callers *must* call dmu_recv_end() if this succeeds.
1815 */
1816 int
dmu_recv_stream(dmu_recv_cookie_t * drc,struct file * fp,offset_t * voffp,int cleanup_fd,uint64_t * action_handlep)1817 dmu_recv_stream(dmu_recv_cookie_t *drc, struct file *fp, offset_t *voffp,
1818 int cleanup_fd, uint64_t *action_handlep)
1819 {
1820 struct restorearg ra = { 0 };
1821 dmu_replay_record_t *drr;
1822 objset_t *os;
1823 zio_cksum_t pcksum;
1824 int featureflags;
1825
1826 ra.byteswap = drc->drc_byteswap;
1827 ra.cksum = drc->drc_cksum;
1828 ra.td = curthread;
1829 ra.fp = fp;
1830 ra.voff = *voffp;
1831 ra.bufsize = SPA_MAXBLOCKSIZE;
1832 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1833
1834 /* these were verified in dmu_recv_begin */
1835 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1836 DMU_SUBSTREAM);
1837 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1838
1839 /*
1840 * Open the objset we are modifying.
1841 */
1842 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1843
1844 ASSERT(dsl_dataset_phys(drc->drc_ds)->ds_flags & DS_FLAG_INCONSISTENT);
1845
1846 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1847
1848 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1849 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1850 minor_t minor;
1851
1852 if (cleanup_fd == -1) {
1853 ra.err = SET_ERROR(EBADF);
1854 goto out;
1855 }
1856 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1857 if (ra.err != 0) {
1858 cleanup_fd = -1;
1859 goto out;
1860 }
1861
1862 if (*action_handlep == 0) {
1863 ra.guid_to_ds_map =
1864 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1865 avl_create(ra.guid_to_ds_map, guid_compare,
1866 sizeof (guid_map_entry_t),
1867 offsetof(guid_map_entry_t, avlnode));
1868 ra.err = zfs_onexit_add_cb(minor,
1869 free_guid_map_onexit, ra.guid_to_ds_map,
1870 action_handlep);
1871 if (ra.err != 0)
1872 goto out;
1873 } else {
1874 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1875 (void **)&ra.guid_to_ds_map);
1876 if (ra.err != 0)
1877 goto out;
1878 }
1879
1880 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1881 }
1882
1883 /*
1884 * Read records and process them.
1885 */
1886 pcksum = ra.cksum;
1887 while (ra.err == 0 &&
1888 NULL != (drr = restore_read(&ra, sizeof (*drr), NULL))) {
1889 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1890 ra.err = SET_ERROR(EINTR);
1891 goto out;
1892 }
1893
1894 if (ra.byteswap)
1895 backup_byteswap(drr);
1896
1897 switch (drr->drr_type) {
1898 case DRR_OBJECT:
1899 {
1900 /*
1901 * We need to make a copy of the record header,
1902 * because restore_{object,write} may need to
1903 * restore_read(), which will invalidate drr.
1904 */
1905 struct drr_object drro = drr->drr_u.drr_object;
1906 ra.err = restore_object(&ra, os, &drro);
1907 break;
1908 }
1909 case DRR_FREEOBJECTS:
1910 {
1911 struct drr_freeobjects drrfo =
1912 drr->drr_u.drr_freeobjects;
1913 ra.err = restore_freeobjects(&ra, os, &drrfo);
1914 break;
1915 }
1916 case DRR_WRITE:
1917 {
1918 struct drr_write drrw = drr->drr_u.drr_write;
1919 ra.err = restore_write(&ra, os, &drrw);
1920 break;
1921 }
1922 case DRR_WRITE_BYREF:
1923 {
1924 struct drr_write_byref drrwbr =
1925 drr->drr_u.drr_write_byref;
1926 ra.err = restore_write_byref(&ra, os, &drrwbr);
1927 break;
1928 }
1929 case DRR_WRITE_EMBEDDED:
1930 {
1931 struct drr_write_embedded drrwe =
1932 drr->drr_u.drr_write_embedded;
1933 ra.err = restore_write_embedded(&ra, os, &drrwe);
1934 break;
1935 }
1936 case DRR_FREE:
1937 {
1938 struct drr_free drrf = drr->drr_u.drr_free;
1939 ra.err = restore_free(&ra, os, &drrf);
1940 break;
1941 }
1942 case DRR_END:
1943 {
1944 struct drr_end drre = drr->drr_u.drr_end;
1945 /*
1946 * We compare against the *previous* checksum
1947 * value, because the stored checksum is of
1948 * everything before the DRR_END record.
1949 */
1950 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1951 ra.err = SET_ERROR(ECKSUM);
1952 goto out;
1953 }
1954 case DRR_SPILL:
1955 {
1956 struct drr_spill drrs = drr->drr_u.drr_spill;
1957 ra.err = restore_spill(&ra, os, &drrs);
1958 break;
1959 }
1960 default:
1961 ra.err = SET_ERROR(EINVAL);
1962 goto out;
1963 }
1964 pcksum = ra.cksum;
1965 }
1966 ASSERT(ra.err != 0);
1967
1968 out:
1969 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1970 zfs_onexit_fd_rele(cleanup_fd);
1971
1972 if (ra.err != 0) {
1973 /*
1974 * destroy what we created, so we don't leave it in the
1975 * inconsistent restoring state.
1976 */
1977 dmu_recv_cleanup_ds(drc);
1978 }
1979
1980 kmem_free(ra.buf, ra.bufsize);
1981 *voffp = ra.voff;
1982 return (ra.err);
1983 }
1984
1985 static int
dmu_recv_end_check(void * arg,dmu_tx_t * tx)1986 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1987 {
1988 dmu_recv_cookie_t *drc = arg;
1989 dsl_pool_t *dp = dmu_tx_pool(tx);
1990 int error;
1991
1992 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1993
1994 if (!drc->drc_newfs) {
1995 dsl_dataset_t *origin_head;
1996
1997 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1998 if (error != 0)
1999 return (error);
2000 if (drc->drc_force) {
2001 /*
2002 * We will destroy any snapshots in tofs (i.e. before
2003 * origin_head) that are after the origin (which is
2004 * the snap before drc_ds, because drc_ds can not
2005 * have any snaps of its own).
2006 */
2007 uint64_t obj;
2008
2009 obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj;
2010 while (obj !=
2011 dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) {
2012 dsl_dataset_t *snap;
2013 error = dsl_dataset_hold_obj(dp, obj, FTAG,
2014 &snap);
2015 if (error != 0)
2016 break;
2017 if (snap->ds_dir != origin_head->ds_dir)
2018 error = SET_ERROR(EINVAL);
2019 if (error == 0) {
2020 error = dsl_destroy_snapshot_check_impl(
2021 snap, B_FALSE);
2022 }
2023 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
2024 dsl_dataset_rele(snap, FTAG);
2025 if (error != 0)
2026 break;
2027 }
2028 if (error != 0) {
2029 dsl_dataset_rele(origin_head, FTAG);
2030 return (error);
2031 }
2032 }
2033 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
2034 origin_head, drc->drc_force, drc->drc_owner, tx);
2035 if (error != 0) {
2036 dsl_dataset_rele(origin_head, FTAG);
2037 return (error);
2038 }
2039 error = dsl_dataset_snapshot_check_impl(origin_head,
2040 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
2041 dsl_dataset_rele(origin_head, FTAG);
2042 if (error != 0)
2043 return (error);
2044
2045 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
2046 } else {
2047 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
2048 drc->drc_tosnap, tx, B_TRUE, 1, drc->drc_cred);
2049 }
2050 return (error);
2051 }
2052
2053 static void
dmu_recv_end_sync(void * arg,dmu_tx_t * tx)2054 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
2055 {
2056 dmu_recv_cookie_t *drc = arg;
2057 dsl_pool_t *dp = dmu_tx_pool(tx);
2058
2059 spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
2060 tx, "snap=%s", drc->drc_tosnap);
2061
2062 if (!drc->drc_newfs) {
2063 dsl_dataset_t *origin_head;
2064
2065 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
2066 &origin_head));
2067
2068 if (drc->drc_force) {
2069 /*
2070 * Destroy any snapshots of drc_tofs (origin_head)
2071 * after the origin (the snap before drc_ds).
2072 */
2073 uint64_t obj;
2074
2075 obj = dsl_dataset_phys(origin_head)->ds_prev_snap_obj;
2076 while (obj !=
2077 dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj) {
2078 dsl_dataset_t *snap;
2079 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
2080 &snap));
2081 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
2082 obj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
2083 dsl_destroy_snapshot_sync_impl(snap,
2084 B_FALSE, tx);
2085 dsl_dataset_rele(snap, FTAG);
2086 }
2087 }
2088 VERIFY3P(drc->drc_ds->ds_prev, ==,
2089 origin_head->ds_prev);
2090
2091 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
2092 origin_head, tx);
2093 dsl_dataset_snapshot_sync_impl(origin_head,
2094 drc->drc_tosnap, tx);
2095
2096 /* set snapshot's creation time and guid */
2097 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
2098 dsl_dataset_phys(origin_head->ds_prev)->ds_creation_time =
2099 drc->drc_drrb->drr_creation_time;
2100 dsl_dataset_phys(origin_head->ds_prev)->ds_guid =
2101 drc->drc_drrb->drr_toguid;
2102 dsl_dataset_phys(origin_head->ds_prev)->ds_flags &=
2103 ~DS_FLAG_INCONSISTENT;
2104
2105 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2106 dsl_dataset_phys(origin_head)->ds_flags &=
2107 ~DS_FLAG_INCONSISTENT;
2108
2109 dsl_dataset_rele(origin_head, FTAG);
2110 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
2111
2112 if (drc->drc_owner != NULL)
2113 VERIFY3P(origin_head->ds_owner, ==, drc->drc_owner);
2114 } else {
2115 dsl_dataset_t *ds = drc->drc_ds;
2116
2117 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
2118
2119 /* set snapshot's creation time and guid */
2120 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2121 dsl_dataset_phys(ds->ds_prev)->ds_creation_time =
2122 drc->drc_drrb->drr_creation_time;
2123 dsl_dataset_phys(ds->ds_prev)->ds_guid =
2124 drc->drc_drrb->drr_toguid;
2125 dsl_dataset_phys(ds->ds_prev)->ds_flags &=
2126 ~DS_FLAG_INCONSISTENT;
2127
2128 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2129 dsl_dataset_phys(ds)->ds_flags &= ~DS_FLAG_INCONSISTENT;
2130 }
2131 drc->drc_newsnapobj = dsl_dataset_phys(drc->drc_ds)->ds_prev_snap_obj;
2132 /*
2133 * Release the hold from dmu_recv_begin. This must be done before
2134 * we return to open context, so that when we free the dataset's dnode,
2135 * we can evict its bonus buffer.
2136 */
2137 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
2138 drc->drc_ds = NULL;
2139 }
2140
2141 static int
add_ds_to_guidmap(const char * name,avl_tree_t * guid_map,uint64_t snapobj)2142 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
2143 {
2144 dsl_pool_t *dp;
2145 dsl_dataset_t *snapds;
2146 guid_map_entry_t *gmep;
2147 int err;
2148
2149 ASSERT(guid_map != NULL);
2150
2151 err = dsl_pool_hold(name, FTAG, &dp);
2152 if (err != 0)
2153 return (err);
2154 gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
2155 err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
2156 if (err == 0) {
2157 gmep->guid = dsl_dataset_phys(snapds)->ds_guid;
2158 gmep->gme_ds = snapds;
2159 avl_add(guid_map, gmep);
2160 dsl_dataset_long_hold(snapds, gmep);
2161 } else
2162 kmem_free(gmep, sizeof (*gmep));
2163
2164 dsl_pool_rele(dp, FTAG);
2165 return (err);
2166 }
2167
2168 static int dmu_recv_end_modified_blocks = 3;
2169
2170 static int
dmu_recv_existing_end(dmu_recv_cookie_t * drc)2171 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
2172 {
2173 int error;
2174 char name[MAXNAMELEN];
2175
2176 #ifdef _KERNEL
2177 /*
2178 * We will be destroying the ds; make sure its origin is unmounted if
2179 * necessary.
2180 */
2181 dsl_dataset_name(drc->drc_ds, name);
2182 zfs_destroy_unmount_origin(name);
2183 #endif
2184
2185 error = dsl_sync_task(drc->drc_tofs,
2186 dmu_recv_end_check, dmu_recv_end_sync, drc,
2187 dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2188
2189 if (error != 0)
2190 dmu_recv_cleanup_ds(drc);
2191 return (error);
2192 }
2193
2194 static int
dmu_recv_new_end(dmu_recv_cookie_t * drc)2195 dmu_recv_new_end(dmu_recv_cookie_t *drc)
2196 {
2197 int error;
2198
2199 error = dsl_sync_task(drc->drc_tofs,
2200 dmu_recv_end_check, dmu_recv_end_sync, drc,
2201 dmu_recv_end_modified_blocks, ZFS_SPACE_CHECK_NORMAL);
2202
2203 if (error != 0) {
2204 dmu_recv_cleanup_ds(drc);
2205 } else if (drc->drc_guid_to_ds_map != NULL) {
2206 (void) add_ds_to_guidmap(drc->drc_tofs,
2207 drc->drc_guid_to_ds_map,
2208 drc->drc_newsnapobj);
2209 }
2210 return (error);
2211 }
2212
2213 int
dmu_recv_end(dmu_recv_cookie_t * drc,void * owner)2214 dmu_recv_end(dmu_recv_cookie_t *drc, void *owner)
2215 {
2216 drc->drc_owner = owner;
2217
2218 if (drc->drc_newfs)
2219 return (dmu_recv_new_end(drc));
2220 else
2221 return (dmu_recv_existing_end(drc));
2222 }
2223
2224 /*
2225 * Return TRUE if this objset is currently being received into.
2226 */
2227 boolean_t
dmu_objset_is_receiving(objset_t * os)2228 dmu_objset_is_receiving(objset_t *os)
2229 {
2230 return (os->os_dsl_dataset != NULL &&
2231 os->os_dsl_dataset->ds_owner == dmu_recv_tag);
2232 }
2233