]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/dmu_send.c
projects / mirror_zfs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Illumos #3888
[mirror_zfs.git] / module / zfs / dmu_send.c
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 by Delphix. All rights reserved.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2013 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
54 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
55 int zfs_send_corrupt_data = B_FALSE;
56
57 static char *dmu_recv_tag = "dmu_recv_tag";
58 static const char *recv_clone_name = "%recv";
59
60 typedef struct dump_bytes_io {
61 dmu_sendarg_t *dbi_dsp;
62 void *dbi_buf;
63 int dbi_len;
64 } dump_bytes_io_t;
65
66 static void
67 dump_bytes_strategy(void *arg)
68 {
69 dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg;
70 dmu_sendarg_t *dsp = dbi->dbi_dsp;
71 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
72 ssize_t resid; /* have to get resid to get detailed errno */
73 ASSERT0(dbi->dbi_len % 8);
74
75 fletcher_4_incremental_native(dbi->dbi_buf, dbi->dbi_len, &dsp->dsa_zc);
76 dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
77 (caddr_t)dbi->dbi_buf, dbi->dbi_len,
78 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
79
80 mutex_enter(&ds->ds_sendstream_lock);
81 *dsp->dsa_off += dbi->dbi_len;
82 mutex_exit(&ds->ds_sendstream_lock);
83 }
84
85 static int
86 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
87 {
88 dump_bytes_io_t dbi;
89
90 dbi.dbi_dsp = dsp;
91 dbi.dbi_buf = buf;
92 dbi.dbi_len = len;
93
94 /*
95 * The vn_rdwr() call is performed in a taskq to ensure that there is
96 * always enough stack space to write safely to the target filesystem.
97 * The ZIO_TYPE_FREE threads are used because there can be a lot of
98 * them and they are used in vdev_file.c for a similar purpose.
99 */
100 spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE,
101 ZIO_TASKQ_ISSUE, dump_bytes_strategy, &dbi, TQ_SLEEP);
102
103 return (dsp->dsa_err);
104 }
105
106 static int
107 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
108 uint64_t length)
109 {
110 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
111
112 if (length != -1ULL && offset + length < offset)
113 length = -1ULL;
114
115 /*
116 * If there is a pending op, but it's not PENDING_FREE, push it out,
117 * since free block aggregation can only be done for blocks of the
118 * same type (i.e., DRR_FREE records can only be aggregated with
119 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
120 * aggregated with other DRR_FREEOBJECTS records.
121 */
122 if (dsp->dsa_pending_op != PENDING_NONE &&
123 dsp->dsa_pending_op != PENDING_FREE) {
124 if (dump_bytes(dsp, dsp->dsa_drr,
125 sizeof (dmu_replay_record_t)) != 0)
126 return (SET_ERROR(EINTR));
127 dsp->dsa_pending_op = PENDING_NONE;
128 }
129
130 if (dsp->dsa_pending_op == PENDING_FREE) {
131 /*
132 * There should never be a PENDING_FREE if length is -1
133 * (because dump_dnode is the only place where this
134 * function is called with a -1, and only after flushing
135 * any pending record).
136 */
137 ASSERT(length != -1ULL);
138 /*
139 * Check to see whether this free block can be aggregated
140 * with pending one.
141 */
142 if (drrf->drr_object == object && drrf->drr_offset +
143 drrf->drr_length == offset) {
144 drrf->drr_length += length;
145 return (0);
146 } else {
147 /* not a continuation. Push out pending record */
148 if (dump_bytes(dsp, dsp->dsa_drr,
149 sizeof (dmu_replay_record_t)) != 0)
150 return (SET_ERROR(EINTR));
151 dsp->dsa_pending_op = PENDING_NONE;
152 }
153 }
154 /* create a FREE record and make it pending */
155 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
156 dsp->dsa_drr->drr_type = DRR_FREE;
157 drrf->drr_object = object;
158 drrf->drr_offset = offset;
159 drrf->drr_length = length;
160 drrf->drr_toguid = dsp->dsa_toguid;
161 if (length == -1ULL) {
162 if (dump_bytes(dsp, dsp->dsa_drr,
163 sizeof (dmu_replay_record_t)) != 0)
164 return (SET_ERROR(EINTR));
165 } else {
166 dsp->dsa_pending_op = PENDING_FREE;
167 }
168
169 return (0);
170 }
171
172 static int
173 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
174 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
175 {
176 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
177
178
179 /*
180 * If there is any kind of pending aggregation (currently either
181 * a grouping of free objects or free blocks), push it out to
182 * the stream, since aggregation can't be done across operations
183 * of different types.
184 */
185 if (dsp->dsa_pending_op != PENDING_NONE) {
186 if (dump_bytes(dsp, dsp->dsa_drr,
187 sizeof (dmu_replay_record_t)) != 0)
188 return (SET_ERROR(EINTR));
189 dsp->dsa_pending_op = PENDING_NONE;
190 }
191 /* write a DATA record */
192 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
193 dsp->dsa_drr->drr_type = DRR_WRITE;
194 drrw->drr_object = object;
195 drrw->drr_type = type;
196 drrw->drr_offset = offset;
197 drrw->drr_length = blksz;
198 drrw->drr_toguid = dsp->dsa_toguid;
199 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
200 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
201 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
202 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
203 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
204 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
205 drrw->drr_key.ddk_cksum = bp->blk_cksum;
206
207 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
208 return (SET_ERROR(EINTR));
209 if (dump_bytes(dsp, data, blksz) != 0)
210 return (SET_ERROR(EINTR));
211 return (0);
212 }
213
214 static int
215 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
216 {
217 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
218
219 if (dsp->dsa_pending_op != PENDING_NONE) {
220 if (dump_bytes(dsp, dsp->dsa_drr,
221 sizeof (dmu_replay_record_t)) != 0)
222 return (SET_ERROR(EINTR));
223 dsp->dsa_pending_op = PENDING_NONE;
224 }
225
226 /* write a SPILL record */
227 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
228 dsp->dsa_drr->drr_type = DRR_SPILL;
229 drrs->drr_object = object;
230 drrs->drr_length = blksz;
231 drrs->drr_toguid = dsp->dsa_toguid;
232
233 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
234 return (SET_ERROR(EINTR));
235 if (dump_bytes(dsp, data, blksz))
236 return (SET_ERROR(EINTR));
237 return (0);
238 }
239
240 static int
241 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
242 {
243 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
244
245 /*
246 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
247 * push it out, since free block aggregation can only be done for
248 * blocks of the same type (i.e., DRR_FREE records can only be
249 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
250 * can only be aggregated with other DRR_FREEOBJECTS records.
251 */
252 if (dsp->dsa_pending_op != PENDING_NONE &&
253 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
254 if (dump_bytes(dsp, dsp->dsa_drr,
255 sizeof (dmu_replay_record_t)) != 0)
256 return (SET_ERROR(EINTR));
257 dsp->dsa_pending_op = PENDING_NONE;
258 }
259 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
260 /*
261 * See whether this free object array can be aggregated
262 * with pending one
263 */
264 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
265 drrfo->drr_numobjs += numobjs;
266 return (0);
267 } else {
268 /* can't be aggregated. Push out pending record */
269 if (dump_bytes(dsp, dsp->dsa_drr,
270 sizeof (dmu_replay_record_t)) != 0)
271 return (SET_ERROR(EINTR));
272 dsp->dsa_pending_op = PENDING_NONE;
273 }
274 }
275
276 /* write a FREEOBJECTS record */
277 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
278 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
279 drrfo->drr_firstobj = firstobj;
280 drrfo->drr_numobjs = numobjs;
281 drrfo->drr_toguid = dsp->dsa_toguid;
282
283 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
284
285 return (0);
286 }
287
288 static int
289 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
290 {
291 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
292
293 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
294 return (dump_freeobjects(dsp, object, 1));
295
296 if (dsp->dsa_pending_op != PENDING_NONE) {
297 if (dump_bytes(dsp, dsp->dsa_drr,
298 sizeof (dmu_replay_record_t)) != 0)
299 return (SET_ERROR(EINTR));
300 dsp->dsa_pending_op = PENDING_NONE;
301 }
302
303 /* write an OBJECT record */
304 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
305 dsp->dsa_drr->drr_type = DRR_OBJECT;
306 drro->drr_object = object;
307 drro->drr_type = dnp->dn_type;
308 drro->drr_bonustype = dnp->dn_bonustype;
309 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
310 drro->drr_bonuslen = dnp->dn_bonuslen;
311 drro->drr_checksumtype = dnp->dn_checksum;
312 drro->drr_compress = dnp->dn_compress;
313 drro->drr_toguid = dsp->dsa_toguid;
314
315 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
316 return (SET_ERROR(EINTR));
317
318 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
319 return (SET_ERROR(EINTR));
320
321 /* free anything past the end of the file */
322 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
323 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
324 return (SET_ERROR(EINTR));
325 if (dsp->dsa_err != 0)
326 return (SET_ERROR(EINTR));
327 return (0);
328 }
329
330 #define BP_SPAN(dnp, level) \
331 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
332 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
333
334 /* ARGSUSED */
335 static int
336 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
337 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
338 {
339 dmu_sendarg_t *dsp = arg;
340 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
341 int err = 0;
342
343 if (issig(JUSTLOOKING) && issig(FORREAL))
344 return (SET_ERROR(EINTR));
345
346 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
347 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
348 return (0);
349 } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) {
350 uint64_t span = BP_SPAN(dnp, zb->zb_level);
351 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
352 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
353 } else if (bp == NULL) {
354 uint64_t span = BP_SPAN(dnp, zb->zb_level);
355 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
356 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
357 return (0);
358 } else if (type == DMU_OT_DNODE) {
359 dnode_phys_t *blk;
360 int i;
361 int blksz = BP_GET_LSIZE(bp);
362 uint32_t aflags = ARC_WAIT;
363 arc_buf_t *abuf;
364
365 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
366 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
367 &aflags, zb) != 0)
368 return (SET_ERROR(EIO));
369
370 blk = abuf->b_data;
371 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
372 uint64_t dnobj = (zb->zb_blkid <<
373 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
374 err = dump_dnode(dsp, dnobj, blk+i);
375 if (err != 0)
376 break;
377 }
378 (void) arc_buf_remove_ref(abuf, &abuf);
379 } else if (type == DMU_OT_SA) {
380 uint32_t aflags = ARC_WAIT;
381 arc_buf_t *abuf;
382 int blksz = BP_GET_LSIZE(bp);
383
384 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
385 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
386 &aflags, zb) != 0)
387 return (SET_ERROR(EIO));
388
389 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
390 (void) arc_buf_remove_ref(abuf, &abuf);
391 } else { /* it's a level-0 block of a regular object */
392 uint32_t aflags = ARC_WAIT;
393 arc_buf_t *abuf;
394 int blksz = BP_GET_LSIZE(bp);
395
396 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
397 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
398 &aflags, zb) != 0) {
399 if (zfs_send_corrupt_data) {
400 uint64_t *ptr;
401 /* Send a block filled with 0x"zfs badd bloc" */
402 abuf = arc_buf_alloc(spa, blksz, &abuf,
403 ARC_BUFC_DATA);
404 for (ptr = abuf->b_data;
405 (char *)ptr < (char *)abuf->b_data + blksz;
406 ptr++)
407 *ptr = 0x2f5baddb10cULL;
408 } else {
409 return (SET_ERROR(EIO));
410 }
411 }
412
413 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
414 blksz, bp, abuf->b_data);
415 (void) arc_buf_remove_ref(abuf, &abuf);
416 }
417
418 ASSERT(err == 0 || err == EINTR);
419 return (err);
420 }
421
422 /*
423 * Releases dp, ds, and fromds, using the specified tag.
424 */
425 static int
426 dmu_send_impl(void *tag, dsl_pool_t *dp, dsl_dataset_t *ds,
427 dsl_dataset_t *fromds, int outfd, vnode_t *vp, offset_t *off)
428 {
429 objset_t *os;
430 dmu_replay_record_t *drr;
431 dmu_sendarg_t *dsp;
432 int err;
433 uint64_t fromtxg = 0;
434
435 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds)) {
436 dsl_dataset_rele(fromds, tag);
437 dsl_dataset_rele(ds, tag);
438 dsl_pool_rele(dp, tag);
439 return (SET_ERROR(EXDEV));
440 }
441
442 err = dmu_objset_from_ds(ds, &os);
443 if (err != 0) {
444 if (fromds != NULL)
445 dsl_dataset_rele(fromds, tag);
446 dsl_dataset_rele(ds, tag);
447 dsl_pool_rele(dp, tag);
448 return (err);
449 }
450
451 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
452 drr->drr_type = DRR_BEGIN;
453 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
454 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
455 DMU_SUBSTREAM);
456
457 #ifdef _KERNEL
458 if (dmu_objset_type(os) == DMU_OST_ZFS) {
459 uint64_t version;
460 if (zfs_get_zplprop(os, ZFS_PROP_VERSION, &version) != 0) {
461 kmem_free(drr, sizeof (dmu_replay_record_t));
462 if (fromds != NULL)
463 dsl_dataset_rele(fromds, tag);
464 dsl_dataset_rele(ds, tag);
465 dsl_pool_rele(dp, tag);
466 return (SET_ERROR(EINVAL));
467 }
468 if (version >= ZPL_VERSION_SA) {
469 DMU_SET_FEATUREFLAGS(
470 drr->drr_u.drr_begin.drr_versioninfo,
471 DMU_BACKUP_FEATURE_SA_SPILL);
472 }
473 }
474 #endif
475
476 drr->drr_u.drr_begin.drr_creation_time =
477 ds->ds_phys->ds_creation_time;
478 drr->drr_u.drr_begin.drr_type = dmu_objset_type(os);
479 if (fromds != NULL && ds->ds_dir != fromds->ds_dir)
480 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
481 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
482 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
483 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
484
485 if (fromds != NULL)
486 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
487 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
488
489 if (fromds != NULL) {
490 fromtxg = fromds->ds_phys->ds_creation_txg;
491 dsl_dataset_rele(fromds, tag);
492 fromds = NULL;
493 }
494
495 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
496
497 dsp->dsa_drr = drr;
498 dsp->dsa_vp = vp;
499 dsp->dsa_outfd = outfd;
500 dsp->dsa_proc = curproc;
501 dsp->dsa_os = os;
502 dsp->dsa_off = off;
503 dsp->dsa_toguid = ds->ds_phys->ds_guid;
504 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
505 dsp->dsa_pending_op = PENDING_NONE;
506
507 mutex_enter(&ds->ds_sendstream_lock);
508 list_insert_head(&ds->ds_sendstreams, dsp);
509 mutex_exit(&ds->ds_sendstream_lock);
510
511 dsl_dataset_long_hold(ds, FTAG);
512 dsl_pool_rele(dp, tag);
513
514 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
515 err = dsp->dsa_err;
516 goto out;
517 }
518
519 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
520 backup_cb, dsp);
521
522 if (dsp->dsa_pending_op != PENDING_NONE)
523 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
524 err = SET_ERROR(EINTR);
525
526 if (err != 0) {
527 if (err == EINTR && dsp->dsa_err != 0)
528 err = dsp->dsa_err;
529 goto out;
530 }
531
532 bzero(drr, sizeof (dmu_replay_record_t));
533 drr->drr_type = DRR_END;
534 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
535 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
536
537 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
538 err = dsp->dsa_err;
539 goto out;
540 }
541
542 out:
543 mutex_enter(&ds->ds_sendstream_lock);
544 list_remove(&ds->ds_sendstreams, dsp);
545 mutex_exit(&ds->ds_sendstream_lock);
546
547 kmem_free(drr, sizeof (dmu_replay_record_t));
548 kmem_free(dsp, sizeof (dmu_sendarg_t));
549
550 dsl_dataset_long_rele(ds, FTAG);
551 dsl_dataset_rele(ds, tag);
552
553 return (err);
554 }
555
556 int
557 dmu_send_obj(const char *pool, uint64_t tosnap, uint64_t fromsnap,
558 int outfd, vnode_t *vp, offset_t *off)
559 {
560 dsl_pool_t *dp;
561 dsl_dataset_t *ds;
562 dsl_dataset_t *fromds = NULL;
563 int err;
564
565 err = dsl_pool_hold(pool, FTAG, &dp);
566 if (err != 0)
567 return (err);
568
569 err = dsl_dataset_hold_obj(dp, tosnap, FTAG, &ds);
570 if (err != 0) {
571 dsl_pool_rele(dp, FTAG);
572 return (err);
573 }
574
575 if (fromsnap != 0) {
576 err = dsl_dataset_hold_obj(dp, fromsnap, FTAG, &fromds);
577 if (err != 0) {
578 dsl_dataset_rele(ds, FTAG);
579 dsl_pool_rele(dp, FTAG);
580 return (err);
581 }
582 }
583
584 return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, vp, off));
585 }
586
587 int
588 dmu_send(const char *tosnap, const char *fromsnap,
589 int outfd, vnode_t *vp, offset_t *off)
590 {
591 dsl_pool_t *dp;
592 dsl_dataset_t *ds;
593 dsl_dataset_t *fromds = NULL;
594 int err;
595
596 if (strchr(tosnap, '@') == NULL)
597 return (SET_ERROR(EINVAL));
598 if (fromsnap != NULL && strchr(fromsnap, '@') == NULL)
599 return (SET_ERROR(EINVAL));
600
601 err = dsl_pool_hold(tosnap, FTAG, &dp);
602 if (err != 0)
603 return (err);
604
605 err = dsl_dataset_hold(dp, tosnap, FTAG, &ds);
606 if (err != 0) {
607 dsl_pool_rele(dp, FTAG);
608 return (err);
609 }
610
611 if (fromsnap != NULL) {
612 err = dsl_dataset_hold(dp, fromsnap, FTAG, &fromds);
613 if (err != 0) {
614 dsl_dataset_rele(ds, FTAG);
615 dsl_pool_rele(dp, FTAG);
616 return (err);
617 }
618 }
619 return (dmu_send_impl(FTAG, dp, ds, fromds, outfd, vp, off));
620 }
621
622 int
623 dmu_send_estimate(dsl_dataset_t *ds, dsl_dataset_t *fromds, uint64_t *sizep)
624 {
625 int err;
626 uint64_t size, recordsize;
627 ASSERTV(dsl_pool_t *dp = ds->ds_dir->dd_pool);
628
629 ASSERT(dsl_pool_config_held(dp));
630
631 /* tosnap must be a snapshot */
632 if (!dsl_dataset_is_snapshot(ds))
633 return (SET_ERROR(EINVAL));
634
635 /*
636 * fromsnap must be an earlier snapshot from the same fs as tosnap,
637 * or the origin's fs.
638 */
639 if (fromds != NULL && !dsl_dataset_is_before(ds, fromds))
640 return (SET_ERROR(EXDEV));
641
642 /* Get uncompressed size estimate of changed data. */
643 if (fromds == NULL) {
644 size = ds->ds_phys->ds_uncompressed_bytes;
645 } else {
646 uint64_t used, comp;
647 err = dsl_dataset_space_written(fromds, ds,
648 &used, &comp, &size);
649 if (err != 0)
650 return (err);
651 }
652
653 /*
654 * Assume that space (both on-disk and in-stream) is dominated by
655 * data. We will adjust for indirect blocks and the copies property,
656 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
657 */
658
659 /*
660 * Subtract out approximate space used by indirect blocks.
661 * Assume most space is used by data blocks (non-indirect, non-dnode).
662 * Assume all blocks are recordsize. Assume ditto blocks and
663 * internal fragmentation counter out compression.
664 *
665 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
666 * block, which we observe in practice.
667 */
668 err = dsl_prop_get_int_ds(ds, "recordsize", &recordsize);
669 if (err != 0)
670 return (err);
671 size -= size / recordsize * sizeof (blkptr_t);
672
673 /* Add in the space for the record associated with each block. */
674 size += size / recordsize * sizeof (dmu_replay_record_t);
675
676 *sizep = size;
677
678 return (0);
679 }
680
681 typedef struct dmu_recv_begin_arg {
682 const char *drba_origin;
683 dmu_recv_cookie_t *drba_cookie;
684 cred_t *drba_cred;
685 uint64_t drba_snapobj;
686 } dmu_recv_begin_arg_t;
687
688 static int
689 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
690 uint64_t fromguid)
691 {
692 uint64_t val;
693 int error;
694 dsl_pool_t *dp = ds->ds_dir->dd_pool;
695
696 /* temporary clone name must not exist */
697 error = zap_lookup(dp->dp_meta_objset,
698 ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name,
699 8, 1, &val);
700 if (error != ENOENT)
701 return (error == 0 ? EBUSY : error);
702
703 /* new snapshot name must not exist */
704 error = zap_lookup(dp->dp_meta_objset,
705 ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap,
706 8, 1, &val);
707 if (error != ENOENT)
708 return (error == 0 ? EEXIST : error);
709
710 if (fromguid != 0) {
711 dsl_dataset_t *snap;
712 uint64_t obj = ds->ds_phys->ds_prev_snap_obj;
713
714 /* Find snapshot in this dir that matches fromguid. */
715 while (obj != 0) {
716 error = dsl_dataset_hold_obj(dp, obj, FTAG,
717 &snap);
718 if (error != 0)
719 return (SET_ERROR(ENODEV));
720 if (snap->ds_dir != ds->ds_dir) {
721 dsl_dataset_rele(snap, FTAG);
722 return (SET_ERROR(ENODEV));
723 }
724 if (snap->ds_phys->ds_guid == fromguid)
725 break;
726 obj = snap->ds_phys->ds_prev_snap_obj;
727 dsl_dataset_rele(snap, FTAG);
728 }
729 if (obj == 0)
730 return (SET_ERROR(ENODEV));
731
732 if (drba->drba_cookie->drc_force) {
733 drba->drba_snapobj = obj;
734 } else {
735 /*
736 * If we are not forcing, there must be no
737 * changes since fromsnap.
738 */
739 if (dsl_dataset_modified_since_snap(ds, snap)) {
740 dsl_dataset_rele(snap, FTAG);
741 return (SET_ERROR(ETXTBSY));
742 }
743 drba->drba_snapobj = ds->ds_prev->ds_object;
744 }
745
746 dsl_dataset_rele(snap, FTAG);
747 } else {
748 /* if full, most recent snapshot must be $ORIGIN */
749 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
750 return (SET_ERROR(ENODEV));
751 drba->drba_snapobj = ds->ds_phys->ds_prev_snap_obj;
752 }
753
754 return (0);
755
756 }
757
758 static int
759 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
760 {
761 dmu_recv_begin_arg_t *drba = arg;
762 dsl_pool_t *dp = dmu_tx_pool(tx);
763 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
764 uint64_t fromguid = drrb->drr_fromguid;
765 int flags = drrb->drr_flags;
766 int error;
767 dsl_dataset_t *ds;
768 const char *tofs = drba->drba_cookie->drc_tofs;
769
770 /* already checked */
771 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
772
773 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
774 DMU_COMPOUNDSTREAM ||
775 drrb->drr_type >= DMU_OST_NUMTYPES ||
776 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
777 return (SET_ERROR(EINVAL));
778
779 /* Verify pool version supports SA if SA_SPILL feature set */
780 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
781 DMU_BACKUP_FEATURE_SA_SPILL) &&
782 spa_version(dp->dp_spa) < SPA_VERSION_SA) {
783 return (SET_ERROR(ENOTSUP));
784 }
785
786 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
787 if (error == 0) {
788 /* target fs already exists; recv into temp clone */
789
790 /* Can't recv a clone into an existing fs */
791 if (flags & DRR_FLAG_CLONE) {
792 dsl_dataset_rele(ds, FTAG);
793 return (SET_ERROR(EINVAL));
794 }
795
796 error = recv_begin_check_existing_impl(drba, ds, fromguid);
797 dsl_dataset_rele(ds, FTAG);
798 } else if (error == ENOENT) {
799 /* target fs does not exist; must be a full backup or clone */
800 char buf[MAXNAMELEN];
801
802 /*
803 * If it's a non-clone incremental, we are missing the
804 * target fs, so fail the recv.
805 */
806 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
807 return (SET_ERROR(ENOENT));
808
809 /* Open the parent of tofs */
810 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
811 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
812 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
813 if (error != 0)
814 return (error);
815
816 if (drba->drba_origin != NULL) {
817 dsl_dataset_t *origin;
818 error = dsl_dataset_hold(dp, drba->drba_origin,
819 FTAG, &origin);
820 if (error != 0) {
821 dsl_dataset_rele(ds, FTAG);
822 return (error);
823 }
824 if (!dsl_dataset_is_snapshot(origin)) {
825 dsl_dataset_rele(origin, FTAG);
826 dsl_dataset_rele(ds, FTAG);
827 return (SET_ERROR(EINVAL));
828 }
829 if (origin->ds_phys->ds_guid != fromguid) {
830 dsl_dataset_rele(origin, FTAG);
831 dsl_dataset_rele(ds, FTAG);
832 return (SET_ERROR(ENODEV));
833 }
834 dsl_dataset_rele(origin, FTAG);
835 }
836 dsl_dataset_rele(ds, FTAG);
837 error = 0;
838 }
839 return (error);
840 }
841
842 static void
843 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
844 {
845 dmu_recv_begin_arg_t *drba = arg;
846 dsl_pool_t *dp = dmu_tx_pool(tx);
847 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
848 const char *tofs = drba->drba_cookie->drc_tofs;
849 dsl_dataset_t *ds, *newds;
850 uint64_t dsobj;
851 int error;
852 uint64_t crflags;
853
854 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
855 DS_FLAG_CI_DATASET : 0;
856
857 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
858 if (error == 0) {
859 /* create temporary clone */
860 dsl_dataset_t *snap = NULL;
861 if (drba->drba_snapobj != 0) {
862 VERIFY0(dsl_dataset_hold_obj(dp,
863 drba->drba_snapobj, FTAG, &snap));
864 }
865 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
866 snap, crflags, drba->drba_cred, tx);
867 dsl_dataset_rele(snap, FTAG);
868 dsl_dataset_rele(ds, FTAG);
869 } else {
870 dsl_dir_t *dd;
871 const char *tail;
872 dsl_dataset_t *origin = NULL;
873
874 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
875
876 if (drba->drba_origin != NULL) {
877 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
878 FTAG, &origin));
879 }
880
881 /* Create new dataset. */
882 dsobj = dsl_dataset_create_sync(dd,
883 strrchr(tofs, '/') + 1,
884 origin, crflags, drba->drba_cred, tx);
885 if (origin != NULL)
886 dsl_dataset_rele(origin, FTAG);
887 dsl_dir_rele(dd, FTAG);
888 drba->drba_cookie->drc_newfs = B_TRUE;
889 }
890 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
891
892 dmu_buf_will_dirty(newds->ds_dbuf, tx);
893 newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
894
895 /*
896 * If we actually created a non-clone, we need to create the
897 * objset in our new dataset.
898 */
899 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
900 (void) dmu_objset_create_impl(dp->dp_spa,
901 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
902 }
903
904 drba->drba_cookie->drc_ds = newds;
905
906 spa_history_log_internal_ds(newds, "receive", tx, "");
907 }
908
909 /*
910 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
911 * succeeds; otherwise we will leak the holds on the datasets.
912 */
913 int
914 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
915 boolean_t force, char *origin, dmu_recv_cookie_t *drc)
916 {
917 dmu_recv_begin_arg_t drba = { 0 };
918 dmu_replay_record_t *drr;
919
920 bzero(drc, sizeof (dmu_recv_cookie_t));
921 drc->drc_drrb = drrb;
922 drc->drc_tosnap = tosnap;
923 drc->drc_tofs = tofs;
924 drc->drc_force = force;
925
926 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
927 drc->drc_byteswap = B_TRUE;
928 else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
929 return (SET_ERROR(EINVAL));
930
931 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
932 drr->drr_type = DRR_BEGIN;
933 drr->drr_u.drr_begin = *drc->drc_drrb;
934 if (drc->drc_byteswap) {
935 fletcher_4_incremental_byteswap(drr,
936 sizeof (dmu_replay_record_t), &drc->drc_cksum);
937 } else {
938 fletcher_4_incremental_native(drr,
939 sizeof (dmu_replay_record_t), &drc->drc_cksum);
940 }
941 kmem_free(drr, sizeof (dmu_replay_record_t));
942
943 if (drc->drc_byteswap) {
944 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
945 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
946 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
947 drrb->drr_type = BSWAP_32(drrb->drr_type);
948 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
949 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
950 }
951
952 drba.drba_origin = origin;
953 drba.drba_cookie = drc;
954 drba.drba_cred = CRED();
955
956 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
957 &drba, 5));
958 }
959
960 struct restorearg {
961 int err;
962 boolean_t byteswap;
963 vnode_t *vp;
964 char *buf;
965 uint64_t voff;
966 int bufsize; /* amount of memory allocated for buf */
967 zio_cksum_t cksum;
968 avl_tree_t *guid_to_ds_map;
969 };
970
971 typedef struct guid_map_entry {
972 uint64_t guid;
973 dsl_dataset_t *gme_ds;
974 avl_node_t avlnode;
975 } guid_map_entry_t;
976
977 static int
978 guid_compare(const void *arg1, const void *arg2)
979 {
980 const guid_map_entry_t *gmep1 = arg1;
981 const guid_map_entry_t *gmep2 = arg2;
982
983 if (gmep1->guid < gmep2->guid)
984 return (-1);
985 else if (gmep1->guid > gmep2->guid)
986 return (1);
987 return (0);
988 }
989
990 static void
991 free_guid_map_onexit(void *arg)
992 {
993 avl_tree_t *ca = arg;
994 void *cookie = NULL;
995 guid_map_entry_t *gmep;
996
997 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
998 dsl_dataset_long_rele(gmep->gme_ds, gmep);
999 dsl_dataset_rele(gmep->gme_ds, gmep);
1000 kmem_free(gmep, sizeof (guid_map_entry_t));
1001 }
1002 avl_destroy(ca);
1003 kmem_free(ca, sizeof (avl_tree_t));
1004 }
1005
1006 static void *
1007 restore_read(struct restorearg *ra, int len)
1008 {
1009 void *rv;
1010 int done = 0;
1011
1012 /* some things will require 8-byte alignment, so everything must */
1013 ASSERT0(len % 8);
1014
1015 while (done < len) {
1016 ssize_t resid;
1017
1018 ra->err = vn_rdwr(UIO_READ, ra->vp,
1019 (caddr_t)ra->buf + done, len - done,
1020 ra->voff, UIO_SYSSPACE, FAPPEND,
1021 RLIM64_INFINITY, CRED(), &resid);
1022
1023 if (resid == len - done)
1024 ra->err = SET_ERROR(EINVAL);
1025 ra->voff += len - done - resid;
1026 done = len - resid;
1027 if (ra->err != 0)
1028 return (NULL);
1029 }
1030
1031 ASSERT3U(done, ==, len);
1032 rv = ra->buf;
1033 if (ra->byteswap)
1034 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1035 else
1036 fletcher_4_incremental_native(rv, len, &ra->cksum);
1037 return (rv);
1038 }
1039
1040 noinline static void
1041 backup_byteswap(dmu_replay_record_t *drr)
1042 {
1043 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1044 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1045 drr->drr_type = BSWAP_32(drr->drr_type);
1046 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1047 switch (drr->drr_type) {
1048 case DRR_BEGIN:
1049 DO64(drr_begin.drr_magic);
1050 DO64(drr_begin.drr_versioninfo);
1051 DO64(drr_begin.drr_creation_time);
1052 DO32(drr_begin.drr_type);
1053 DO32(drr_begin.drr_flags);
1054 DO64(drr_begin.drr_toguid);
1055 DO64(drr_begin.drr_fromguid);
1056 break;
1057 case DRR_OBJECT:
1058 DO64(drr_object.drr_object);
1059 /* DO64(drr_object.drr_allocation_txg); */
1060 DO32(drr_object.drr_type);
1061 DO32(drr_object.drr_bonustype);
1062 DO32(drr_object.drr_blksz);
1063 DO32(drr_object.drr_bonuslen);
1064 DO64(drr_object.drr_toguid);
1065 break;
1066 case DRR_FREEOBJECTS:
1067 DO64(drr_freeobjects.drr_firstobj);
1068 DO64(drr_freeobjects.drr_numobjs);
1069 DO64(drr_freeobjects.drr_toguid);
1070 break;
1071 case DRR_WRITE:
1072 DO64(drr_write.drr_object);
1073 DO32(drr_write.drr_type);
1074 DO64(drr_write.drr_offset);
1075 DO64(drr_write.drr_length);
1076 DO64(drr_write.drr_toguid);
1077 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1078 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1079 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1080 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1081 DO64(drr_write.drr_key.ddk_prop);
1082 break;
1083 case DRR_WRITE_BYREF:
1084 DO64(drr_write_byref.drr_object);
1085 DO64(drr_write_byref.drr_offset);
1086 DO64(drr_write_byref.drr_length);
1087 DO64(drr_write_byref.drr_toguid);
1088 DO64(drr_write_byref.drr_refguid);
1089 DO64(drr_write_byref.drr_refobject);
1090 DO64(drr_write_byref.drr_refoffset);
1091 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1092 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1093 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1094 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1095 DO64(drr_write_byref.drr_key.ddk_prop);
1096 break;
1097 case DRR_FREE:
1098 DO64(drr_free.drr_object);
1099 DO64(drr_free.drr_offset);
1100 DO64(drr_free.drr_length);
1101 DO64(drr_free.drr_toguid);
1102 break;
1103 case DRR_SPILL:
1104 DO64(drr_spill.drr_object);
1105 DO64(drr_spill.drr_length);
1106 DO64(drr_spill.drr_toguid);
1107 break;
1108 case DRR_END:
1109 DO64(drr_end.drr_checksum.zc_word[0]);
1110 DO64(drr_end.drr_checksum.zc_word[1]);
1111 DO64(drr_end.drr_checksum.zc_word[2]);
1112 DO64(drr_end.drr_checksum.zc_word[3]);
1113 DO64(drr_end.drr_toguid);
1114 break;
1115 default:
1116 break;
1117 }
1118 #undef DO64
1119 #undef DO32
1120 }
1121
1122 noinline static int
1123 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1124 {
1125 int err;
1126 dmu_tx_t *tx;
1127 void *data = NULL;
1128
1129 if (drro->drr_type == DMU_OT_NONE ||
1130 !DMU_OT_IS_VALID(drro->drr_type) ||
1131 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1132 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1133 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1134 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1135 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1136 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1137 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1138 return (SET_ERROR(EINVAL));
1139 }
1140
1141 err = dmu_object_info(os, drro->drr_object, NULL);
1142
1143 if (err != 0 && err != ENOENT)
1144 return (SET_ERROR(EINVAL));
1145
1146 if (drro->drr_bonuslen) {
1147 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1148 if (ra->err != 0)
1149 return (ra->err);
1150 }
1151
1152 if (err == ENOENT) {
1153 /* currently free, want to be allocated */
1154 tx = dmu_tx_create(os);
1155 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1156 err = dmu_tx_assign(tx, TXG_WAIT);
1157 if (err != 0) {
1158 dmu_tx_abort(tx);
1159 return (err);
1160 }
1161 err = dmu_object_claim(os, drro->drr_object,
1162 drro->drr_type, drro->drr_blksz,
1163 drro->drr_bonustype, drro->drr_bonuslen, tx);
1164 dmu_tx_commit(tx);
1165 } else {
1166 /* currently allocated, want to be allocated */
1167 err = dmu_object_reclaim(os, drro->drr_object,
1168 drro->drr_type, drro->drr_blksz,
1169 drro->drr_bonustype, drro->drr_bonuslen);
1170 }
1171 if (err != 0) {
1172 return (SET_ERROR(EINVAL));
1173 }
1174
1175 tx = dmu_tx_create(os);
1176 dmu_tx_hold_bonus(tx, drro->drr_object);
1177 err = dmu_tx_assign(tx, TXG_WAIT);
1178 if (err != 0) {
1179 dmu_tx_abort(tx);
1180 return (err);
1181 }
1182
1183 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1184 tx);
1185 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1186
1187 if (data != NULL) {
1188 dmu_buf_t *db;
1189
1190 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1191 dmu_buf_will_dirty(db, tx);
1192
1193 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1194 bcopy(data, db->db_data, drro->drr_bonuslen);
1195 if (ra->byteswap) {
1196 dmu_object_byteswap_t byteswap =
1197 DMU_OT_BYTESWAP(drro->drr_bonustype);
1198 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1199 drro->drr_bonuslen);
1200 }
1201 dmu_buf_rele(db, FTAG);
1202 }
1203 dmu_tx_commit(tx);
1204 return (0);
1205 }
1206
1207 /* ARGSUSED */
1208 noinline static int
1209 restore_freeobjects(struct restorearg *ra, objset_t *os,
1210 struct drr_freeobjects *drrfo)
1211 {
1212 uint64_t obj;
1213
1214 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1215 return (SET_ERROR(EINVAL));
1216
1217 for (obj = drrfo->drr_firstobj;
1218 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1219 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1220 int err;
1221
1222 if (dmu_object_info(os, obj, NULL) != 0)
1223 continue;
1224
1225 err = dmu_free_object(os, obj);
1226 if (err != 0)
1227 return (err);
1228 }
1229 return (0);
1230 }
1231
1232 noinline static int
1233 restore_write(struct restorearg *ra, objset_t *os,
1234 struct drr_write *drrw)
1235 {
1236 dmu_tx_t *tx;
1237 void *data;
1238 int err;
1239
1240 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1241 !DMU_OT_IS_VALID(drrw->drr_type))
1242 return (SET_ERROR(EINVAL));
1243
1244 data = restore_read(ra, drrw->drr_length);
1245 if (data == NULL)
1246 return (ra->err);
1247
1248 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1249 return (SET_ERROR(EINVAL));
1250
1251 tx = dmu_tx_create(os);
1252
1253 dmu_tx_hold_write(tx, drrw->drr_object,
1254 drrw->drr_offset, drrw->drr_length);
1255 err = dmu_tx_assign(tx, TXG_WAIT);
1256 if (err != 0) {
1257 dmu_tx_abort(tx);
1258 return (err);
1259 }
1260 if (ra->byteswap) {
1261 dmu_object_byteswap_t byteswap =
1262 DMU_OT_BYTESWAP(drrw->drr_type);
1263 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1264 }
1265 dmu_write(os, drrw->drr_object,
1266 drrw->drr_offset, drrw->drr_length, data, tx);
1267 dmu_tx_commit(tx);
1268 return (0);
1269 }
1270
1271 /*
1272 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1273 * streams to refer to a copy of the data that is already on the
1274 * system because it came in earlier in the stream. This function
1275 * finds the earlier copy of the data, and uses that copy instead of
1276 * data from the stream to fulfill this write.
1277 */
1278 static int
1279 restore_write_byref(struct restorearg *ra, objset_t *os,
1280 struct drr_write_byref *drrwbr)
1281 {
1282 dmu_tx_t *tx;
1283 int err;
1284 guid_map_entry_t gmesrch;
1285 guid_map_entry_t *gmep;
1286 avl_index_t where;
1287 objset_t *ref_os = NULL;
1288 dmu_buf_t *dbp;
1289
1290 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1291 return (SET_ERROR(EINVAL));
1292
1293 /*
1294 * If the GUID of the referenced dataset is different from the
1295 * GUID of the target dataset, find the referenced dataset.
1296 */
1297 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1298 gmesrch.guid = drrwbr->drr_refguid;
1299 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1300 &where)) == NULL) {
1301 return (SET_ERROR(EINVAL));
1302 }
1303 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1304 return (SET_ERROR(EINVAL));
1305 } else {
1306 ref_os = os;
1307 }
1308
1309 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1310 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1311 if (err)
1312 return (err);
1313
1314 tx = dmu_tx_create(os);
1315
1316 dmu_tx_hold_write(tx, drrwbr->drr_object,
1317 drrwbr->drr_offset, drrwbr->drr_length);
1318 err = dmu_tx_assign(tx, TXG_WAIT);
1319 if (err != 0) {
1320 dmu_tx_abort(tx);
1321 return (err);
1322 }
1323 dmu_write(os, drrwbr->drr_object,
1324 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1325 dmu_buf_rele(dbp, FTAG);
1326 dmu_tx_commit(tx);
1327 return (0);
1328 }
1329
1330 static int
1331 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1332 {
1333 dmu_tx_t *tx;
1334 void *data;
1335 dmu_buf_t *db, *db_spill;
1336 int err;
1337
1338 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1339 drrs->drr_length > SPA_MAXBLOCKSIZE)
1340 return (SET_ERROR(EINVAL));
1341
1342 data = restore_read(ra, drrs->drr_length);
1343 if (data == NULL)
1344 return (ra->err);
1345
1346 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1347 return (SET_ERROR(EINVAL));
1348
1349 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1350 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1351 dmu_buf_rele(db, FTAG);
1352 return (err);
1353 }
1354
1355 tx = dmu_tx_create(os);
1356
1357 dmu_tx_hold_spill(tx, db->db_object);
1358
1359 err = dmu_tx_assign(tx, TXG_WAIT);
1360 if (err != 0) {
1361 dmu_buf_rele(db, FTAG);
1362 dmu_buf_rele(db_spill, FTAG);
1363 dmu_tx_abort(tx);
1364 return (err);
1365 }
1366 dmu_buf_will_dirty(db_spill, tx);
1367
1368 if (db_spill->db_size < drrs->drr_length)
1369 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1370 drrs->drr_length, tx));
1371 bcopy(data, db_spill->db_data, drrs->drr_length);
1372
1373 dmu_buf_rele(db, FTAG);
1374 dmu_buf_rele(db_spill, FTAG);
1375
1376 dmu_tx_commit(tx);
1377 return (0);
1378 }
1379
1380 /* ARGSUSED */
1381 noinline static int
1382 restore_free(struct restorearg *ra, objset_t *os,
1383 struct drr_free *drrf)
1384 {
1385 int err;
1386
1387 if (drrf->drr_length != -1ULL &&
1388 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1389 return (SET_ERROR(EINVAL));
1390
1391 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1392 return (SET_ERROR(EINVAL));
1393
1394 err = dmu_free_long_range(os, drrf->drr_object,
1395 drrf->drr_offset, drrf->drr_length);
1396 return (err);
1397 }
1398
1399 /* used to destroy the drc_ds on error */
1400 static void
1401 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1402 {
1403 char name[MAXNAMELEN];
1404 dsl_dataset_name(drc->drc_ds, name);
1405 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1406 (void) dsl_destroy_head(name);
1407 }
1408
1409 /*
1410 * NB: callers *must* call dmu_recv_end() if this succeeds.
1411 */
1412 int
1413 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1414 int cleanup_fd, uint64_t *action_handlep)
1415 {
1416 struct restorearg ra = { 0 };
1417 dmu_replay_record_t *drr;
1418 objset_t *os;
1419 zio_cksum_t pcksum;
1420 int featureflags;
1421
1422 ra.byteswap = drc->drc_byteswap;
1423 ra.cksum = drc->drc_cksum;
1424 ra.vp = vp;
1425 ra.voff = *voffp;
1426 ra.bufsize = 1<<20;
1427 ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
1428
1429 /* these were verified in dmu_recv_begin */
1430 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1431 DMU_SUBSTREAM);
1432 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1433
1434 /*
1435 * Open the objset we are modifying.
1436 */
1437 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1438
1439 ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1440
1441 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1442
1443 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1444 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1445 minor_t minor;
1446
1447 if (cleanup_fd == -1) {
1448 ra.err = SET_ERROR(EBADF);
1449 goto out;
1450 }
1451 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1452 if (ra.err != 0) {
1453 cleanup_fd = -1;
1454 goto out;
1455 }
1456
1457 if (*action_handlep == 0) {
1458 ra.guid_to_ds_map =
1459 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1460 avl_create(ra.guid_to_ds_map, guid_compare,
1461 sizeof (guid_map_entry_t),
1462 offsetof(guid_map_entry_t, avlnode));
1463 ra.err = zfs_onexit_add_cb(minor,
1464 free_guid_map_onexit, ra.guid_to_ds_map,
1465 action_handlep);
1466 if (ra.err != 0)
1467 goto out;
1468 } else {
1469 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1470 (void **)&ra.guid_to_ds_map);
1471 if (ra.err != 0)
1472 goto out;
1473 }
1474
1475 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1476 }
1477
1478 /*
1479 * Read records and process them.
1480 */
1481 pcksum = ra.cksum;
1482 while (ra.err == 0 &&
1483 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1484 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1485 ra.err = SET_ERROR(EINTR);
1486 goto out;
1487 }
1488
1489 if (ra.byteswap)
1490 backup_byteswap(drr);
1491
1492 switch (drr->drr_type) {
1493 case DRR_OBJECT:
1494 {
1495 /*
1496 * We need to make a copy of the record header,
1497 * because restore_{object,write} may need to
1498 * restore_read(), which will invalidate drr.
1499 */
1500 struct drr_object drro = drr->drr_u.drr_object;
1501 ra.err = restore_object(&ra, os, &drro);
1502 break;
1503 }
1504 case DRR_FREEOBJECTS:
1505 {
1506 struct drr_freeobjects drrfo =
1507 drr->drr_u.drr_freeobjects;
1508 ra.err = restore_freeobjects(&ra, os, &drrfo);
1509 break;
1510 }
1511 case DRR_WRITE:
1512 {
1513 struct drr_write drrw = drr->drr_u.drr_write;
1514 ra.err = restore_write(&ra, os, &drrw);
1515 break;
1516 }
1517 case DRR_WRITE_BYREF:
1518 {
1519 struct drr_write_byref drrwbr =
1520 drr->drr_u.drr_write_byref;
1521 ra.err = restore_write_byref(&ra, os, &drrwbr);
1522 break;
1523 }
1524 case DRR_FREE:
1525 {
1526 struct drr_free drrf = drr->drr_u.drr_free;
1527 ra.err = restore_free(&ra, os, &drrf);
1528 break;
1529 }
1530 case DRR_END:
1531 {
1532 struct drr_end drre = drr->drr_u.drr_end;
1533 /*
1534 * We compare against the *previous* checksum
1535 * value, because the stored checksum is of
1536 * everything before the DRR_END record.
1537 */
1538 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1539 ra.err = SET_ERROR(ECKSUM);
1540 goto out;
1541 }
1542 case DRR_SPILL:
1543 {
1544 struct drr_spill drrs = drr->drr_u.drr_spill;
1545 ra.err = restore_spill(&ra, os, &drrs);
1546 break;
1547 }
1548 default:
1549 ra.err = SET_ERROR(EINVAL);
1550 goto out;
1551 }
1552 pcksum = ra.cksum;
1553 }
1554 ASSERT(ra.err != 0);
1555
1556 out:
1557 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1558 zfs_onexit_fd_rele(cleanup_fd);
1559
1560 if (ra.err != 0) {
1561 /*
1562 * destroy what we created, so we don't leave it in the
1563 * inconsistent restoring state.
1564 */
1565 dmu_recv_cleanup_ds(drc);
1566 }
1567
1568 vmem_free(ra.buf, ra.bufsize);
1569 *voffp = ra.voff;
1570 return (ra.err);
1571 }
1572
1573 static int
1574 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1575 {
1576 dmu_recv_cookie_t *drc = arg;
1577 dsl_pool_t *dp = dmu_tx_pool(tx);
1578 int error;
1579
1580 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1581
1582 if (!drc->drc_newfs) {
1583 dsl_dataset_t *origin_head;
1584
1585 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1586 if (error != 0)
1587 return (error);
1588 if (drc->drc_force) {
1589 /*
1590 * We will destroy any snapshots in tofs (i.e. before
1591 * origin_head) that are after the origin (which is
1592 * the snap before drc_ds, because drc_ds can not
1593 * have any snaps of its own).
1594 */
1595 uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1596 while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1597 dsl_dataset_t *snap;
1598 error = dsl_dataset_hold_obj(dp, obj, FTAG,
1599 &snap);
1600 if (error != 0)
1601 return (error);
1602 if (snap->ds_dir != origin_head->ds_dir)
1603 error = SET_ERROR(EINVAL);
1604 if (error == 0) {
1605 error = dsl_destroy_snapshot_check_impl(
1606 snap, B_FALSE);
1607 }
1608 obj = snap->ds_phys->ds_prev_snap_obj;
1609 dsl_dataset_rele(snap, FTAG);
1610 if (error != 0)
1611 return (error);
1612 }
1613 }
1614 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1615 origin_head, drc->drc_force);
1616 if (error != 0) {
1617 dsl_dataset_rele(origin_head, FTAG);
1618 return (error);
1619 }
1620 error = dsl_dataset_snapshot_check_impl(origin_head,
1621 drc->drc_tosnap, tx, B_TRUE);
1622 dsl_dataset_rele(origin_head, FTAG);
1623 if (error != 0)
1624 return (error);
1625
1626 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1627 } else {
1628 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1629 drc->drc_tosnap, tx, B_TRUE);
1630 }
1631 return (error);
1632 }
1633
1634 static void
1635 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1636 {
1637 dmu_recv_cookie_t *drc = arg;
1638 dsl_pool_t *dp = dmu_tx_pool(tx);
1639
1640 spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1641 tx, "snap=%s", drc->drc_tosnap);
1642
1643 if (!drc->drc_newfs) {
1644 dsl_dataset_t *origin_head;
1645
1646 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1647 &origin_head));
1648
1649 if (drc->drc_force) {
1650 /*
1651 * Destroy any snapshots of drc_tofs (origin_head)
1652 * after the origin (the snap before drc_ds).
1653 */
1654 uint64_t obj = origin_head->ds_phys->ds_prev_snap_obj;
1655 while (obj != drc->drc_ds->ds_phys->ds_prev_snap_obj) {
1656 dsl_dataset_t *snap;
1657 VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG,
1658 &snap));
1659 ASSERT3P(snap->ds_dir, ==, origin_head->ds_dir);
1660 obj = snap->ds_phys->ds_prev_snap_obj;
1661 dsl_destroy_snapshot_sync_impl(snap,
1662 B_FALSE, tx);
1663 dsl_dataset_rele(snap, FTAG);
1664 }
1665 }
1666 VERIFY3P(drc->drc_ds->ds_prev, ==,
1667 origin_head->ds_prev);
1668
1669 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1670 origin_head, tx);
1671 dsl_dataset_snapshot_sync_impl(origin_head,
1672 drc->drc_tosnap, tx);
1673
1674 /* set snapshot's creation time and guid */
1675 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1676 origin_head->ds_prev->ds_phys->ds_creation_time =
1677 drc->drc_drrb->drr_creation_time;
1678 origin_head->ds_prev->ds_phys->ds_guid =
1679 drc->drc_drrb->drr_toguid;
1680 origin_head->ds_prev->ds_phys->ds_flags &=
1681 ~DS_FLAG_INCONSISTENT;
1682
1683 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1684 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1685
1686 dsl_dataset_rele(origin_head, FTAG);
1687 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1688 } else {
1689 dsl_dataset_t *ds = drc->drc_ds;
1690
1691 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1692
1693 /* set snapshot's creation time and guid */
1694 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1695 ds->ds_prev->ds_phys->ds_creation_time =
1696 drc->drc_drrb->drr_creation_time;
1697 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1698 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1699
1700 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1701 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1702 }
1703 drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1704 /*
1705 * Release the hold from dmu_recv_begin. This must be done before
1706 * we return to open context, so that when we free the dataset's dnode,
1707 * we can evict its bonus buffer.
1708 */
1709 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1710 drc->drc_ds = NULL;
1711 }
1712
1713 static int
1714 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
1715 {
1716 dsl_pool_t *dp;
1717 dsl_dataset_t *snapds;
1718 guid_map_entry_t *gmep;
1719 int err;
1720
1721 ASSERT(guid_map != NULL);
1722
1723 err = dsl_pool_hold(name, FTAG, &dp);
1724 if (err != 0)
1725 return (err);
1726 gmep = kmem_alloc(sizeof (*gmep), KM_SLEEP);
1727 err = dsl_dataset_hold_obj(dp, snapobj, gmep, &snapds);
1728 if (err == 0) {
1729 gmep->guid = snapds->ds_phys->ds_guid;
1730 gmep->gme_ds = snapds;
1731 avl_add(guid_map, gmep);
1732 dsl_dataset_long_hold(snapds, gmep);
1733 } else {
1734 kmem_free(gmep, sizeof (*gmep));
1735 }
1736
1737 dsl_pool_rele(dp, FTAG);
1738 return (err);
1739 }
1740
1741 static int dmu_recv_end_modified_blocks = 3;
1742
1743 static int
1744 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1745 {
1746 int error;
1747
1748 #ifdef _KERNEL
1749 char *name;
1750
1751 /*
1752 * We will be destroying the ds; make sure its origin is unmounted if
1753 * necessary.
1754 */
1755 name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
1756 dsl_dataset_name(drc->drc_ds, name);
1757 zfs_destroy_unmount_origin(name);
1758 kmem_free(name, MAXNAMELEN);
1759 #endif
1760
1761 error = dsl_sync_task(drc->drc_tofs,
1762 dmu_recv_end_check, dmu_recv_end_sync, drc,
1763 dmu_recv_end_modified_blocks);
1764
1765 if (error != 0)
1766 dmu_recv_cleanup_ds(drc);
1767 return (error);
1768 }
1769
1770 static int
1771 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1772 {
1773 int error;
1774
1775 error = dsl_sync_task(drc->drc_tofs,
1776 dmu_recv_end_check, dmu_recv_end_sync, drc,
1777 dmu_recv_end_modified_blocks);
1778
1779 if (error != 0) {
1780 dmu_recv_cleanup_ds(drc);
1781 } else if (drc->drc_guid_to_ds_map != NULL) {
1782 (void) add_ds_to_guidmap(drc->drc_tofs,
1783 drc->drc_guid_to_ds_map,
1784 drc->drc_newsnapobj);
1785 }
1786 return (error);
1787 }
1788
1789 int
1790 dmu_recv_end(dmu_recv_cookie_t *drc)
1791 {
1792 if (drc->drc_newfs)
1793 return (dmu_recv_new_end(drc));
1794 else
1795 return (dmu_recv_existing_end(drc));
1796 }
ZFS On Linux mirror for PVE