]> 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 #3598
[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 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
512 err = dsp->dsa_err;
513 goto out;
514 }
515
516 dsl_dataset_long_hold(ds, FTAG);
517 dsl_pool_rele(dp, tag);
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 } dmu_recv_begin_arg_t;
686
687 static int
688 recv_begin_check_existing_impl(dmu_recv_begin_arg_t *drba, dsl_dataset_t *ds,
689 uint64_t fromguid)
690 {
691 uint64_t val;
692 int error;
693 dsl_pool_t *dp = ds->ds_dir->dd_pool;
694
695 /* must not have any changes since most recent snapshot */
696 if (!drba->drba_cookie->drc_force &&
697 dsl_dataset_modified_since_lastsnap(ds))
698 return (SET_ERROR(ETXTBSY));
699
700 /* temporary clone name must not exist */
701 error = zap_lookup(dp->dp_meta_objset,
702 ds->ds_dir->dd_phys->dd_child_dir_zapobj, recv_clone_name,
703 8, 1, &val);
704 if (error != ENOENT)
705 return (error == 0 ? EBUSY : error);
706
707 /* new snapshot name must not exist */
708 error = zap_lookup(dp->dp_meta_objset,
709 ds->ds_phys->ds_snapnames_zapobj, drba->drba_cookie->drc_tosnap,
710 8, 1, &val);
711 if (error != ENOENT)
712 return (error == 0 ? EEXIST : error);
713
714 if (fromguid != 0) {
715 /* if incremental, most recent snapshot must match fromguid */
716 if (ds->ds_prev == NULL)
717 return (SET_ERROR(ENODEV));
718
719 /*
720 * most recent snapshot must match fromguid, or there are no
721 * changes since the fromguid one
722 */
723 if (ds->ds_prev->ds_phys->ds_guid != fromguid) {
724 uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth;
725 uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj;
726 while (obj != 0) {
727 dsl_dataset_t *snap;
728 error = dsl_dataset_hold_obj(dp, obj, FTAG,
729 &snap);
730 if (error != 0)
731 return (SET_ERROR(ENODEV));
732 if (snap->ds_phys->ds_creation_txg < birth) {
733 dsl_dataset_rele(snap, FTAG);
734 return (SET_ERROR(ENODEV));
735 }
736 if (snap->ds_phys->ds_guid == fromguid) {
737 dsl_dataset_rele(snap, FTAG);
738 break; /* it's ok */
739 }
740 obj = snap->ds_phys->ds_prev_snap_obj;
741 dsl_dataset_rele(snap, FTAG);
742 }
743 if (obj == 0)
744 return (SET_ERROR(ENODEV));
745 }
746 } else {
747 /* if full, most recent snapshot must be $ORIGIN */
748 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
749 return (SET_ERROR(ENODEV));
750 }
751
752 return (0);
753
754 }
755
756 static int
757 dmu_recv_begin_check(void *arg, dmu_tx_t *tx)
758 {
759 dmu_recv_begin_arg_t *drba = arg;
760 dsl_pool_t *dp = dmu_tx_pool(tx);
761 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
762 uint64_t fromguid = drrb->drr_fromguid;
763 int flags = drrb->drr_flags;
764 int error;
765 dsl_dataset_t *ds;
766 const char *tofs = drba->drba_cookie->drc_tofs;
767
768 /* already checked */
769 ASSERT3U(drrb->drr_magic, ==, DMU_BACKUP_MAGIC);
770
771 if (DMU_GET_STREAM_HDRTYPE(drrb->drr_versioninfo) ==
772 DMU_COMPOUNDSTREAM ||
773 drrb->drr_type >= DMU_OST_NUMTYPES ||
774 ((flags & DRR_FLAG_CLONE) && drba->drba_origin == NULL))
775 return (SET_ERROR(EINVAL));
776
777 /* Verify pool version supports SA if SA_SPILL feature set */
778 if ((DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo) &
779 DMU_BACKUP_FEATURE_SA_SPILL) &&
780 spa_version(dp->dp_spa) < SPA_VERSION_SA) {
781 return (SET_ERROR(ENOTSUP));
782 }
783
784 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
785 if (error == 0) {
786 /* target fs already exists; recv into temp clone */
787
788 /* Can't recv a clone into an existing fs */
789 if (flags & DRR_FLAG_CLONE) {
790 dsl_dataset_rele(ds, FTAG);
791 return (SET_ERROR(EINVAL));
792 }
793
794 error = recv_begin_check_existing_impl(drba, ds, fromguid);
795 dsl_dataset_rele(ds, FTAG);
796 } else if (error == ENOENT) {
797 /* target fs does not exist; must be a full backup or clone */
798 char buf[MAXNAMELEN];
799
800 /*
801 * If it's a non-clone incremental, we are missing the
802 * target fs, so fail the recv.
803 */
804 if (fromguid != 0 && !(flags & DRR_FLAG_CLONE))
805 return (SET_ERROR(ENOENT));
806
807 /* Open the parent of tofs */
808 ASSERT3U(strlen(tofs), <, MAXNAMELEN);
809 (void) strlcpy(buf, tofs, strrchr(tofs, '/') - tofs + 1);
810 error = dsl_dataset_hold(dp, buf, FTAG, &ds);
811 if (error != 0)
812 return (error);
813
814 if (drba->drba_origin != NULL) {
815 dsl_dataset_t *origin;
816 error = dsl_dataset_hold(dp, drba->drba_origin,
817 FTAG, &origin);
818 if (error != 0) {
819 dsl_dataset_rele(ds, FTAG);
820 return (error);
821 }
822 if (!dsl_dataset_is_snapshot(origin)) {
823 dsl_dataset_rele(origin, FTAG);
824 dsl_dataset_rele(ds, FTAG);
825 return (SET_ERROR(EINVAL));
826 }
827 if (origin->ds_phys->ds_guid != fromguid) {
828 dsl_dataset_rele(origin, FTAG);
829 dsl_dataset_rele(ds, FTAG);
830 return (SET_ERROR(ENODEV));
831 }
832 dsl_dataset_rele(origin, FTAG);
833 }
834 dsl_dataset_rele(ds, FTAG);
835 error = 0;
836 }
837 return (error);
838 }
839
840 static void
841 dmu_recv_begin_sync(void *arg, dmu_tx_t *tx)
842 {
843 dmu_recv_begin_arg_t *drba = arg;
844 dsl_pool_t *dp = dmu_tx_pool(tx);
845 struct drr_begin *drrb = drba->drba_cookie->drc_drrb;
846 const char *tofs = drba->drba_cookie->drc_tofs;
847 dsl_dataset_t *ds, *newds;
848 uint64_t dsobj;
849 int error;
850 uint64_t crflags;
851
852 crflags = (drrb->drr_flags & DRR_FLAG_CI_DATA) ?
853 DS_FLAG_CI_DATASET : 0;
854
855 error = dsl_dataset_hold(dp, tofs, FTAG, &ds);
856 if (error == 0) {
857 /* create temporary clone */
858 dsobj = dsl_dataset_create_sync(ds->ds_dir, recv_clone_name,
859 ds->ds_prev, crflags, drba->drba_cred, tx);
860 dsl_dataset_rele(ds, FTAG);
861 } else {
862 dsl_dir_t *dd;
863 const char *tail;
864 dsl_dataset_t *origin = NULL;
865
866 VERIFY0(dsl_dir_hold(dp, tofs, FTAG, &dd, &tail));
867
868 if (drba->drba_origin != NULL) {
869 VERIFY0(dsl_dataset_hold(dp, drba->drba_origin,
870 FTAG, &origin));
871 }
872
873 /* Create new dataset. */
874 dsobj = dsl_dataset_create_sync(dd,
875 strrchr(tofs, '/') + 1,
876 origin, crflags, drba->drba_cred, tx);
877 if (origin != NULL)
878 dsl_dataset_rele(origin, FTAG);
879 dsl_dir_rele(dd, FTAG);
880 drba->drba_cookie->drc_newfs = B_TRUE;
881 }
882 VERIFY0(dsl_dataset_own_obj(dp, dsobj, dmu_recv_tag, &newds));
883
884 dmu_buf_will_dirty(newds->ds_dbuf, tx);
885 newds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
886
887 /*
888 * If we actually created a non-clone, we need to create the
889 * objset in our new dataset.
890 */
891 if (BP_IS_HOLE(dsl_dataset_get_blkptr(newds))) {
892 (void) dmu_objset_create_impl(dp->dp_spa,
893 newds, dsl_dataset_get_blkptr(newds), drrb->drr_type, tx);
894 }
895
896 drba->drba_cookie->drc_ds = newds;
897
898 spa_history_log_internal_ds(newds, "receive", tx, "");
899 }
900
901 /*
902 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
903 * succeeds; otherwise we will leak the holds on the datasets.
904 */
905 int
906 dmu_recv_begin(char *tofs, char *tosnap, struct drr_begin *drrb,
907 boolean_t force, char *origin, dmu_recv_cookie_t *drc)
908 {
909 dmu_recv_begin_arg_t drba = { 0 };
910 dmu_replay_record_t *drr;
911
912 bzero(drc, sizeof (dmu_recv_cookie_t));
913 drc->drc_drrb = drrb;
914 drc->drc_tosnap = tosnap;
915 drc->drc_tofs = tofs;
916 drc->drc_force = force;
917
918 if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
919 drc->drc_byteswap = B_TRUE;
920 else if (drrb->drr_magic != DMU_BACKUP_MAGIC)
921 return (SET_ERROR(EINVAL));
922
923 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
924 drr->drr_type = DRR_BEGIN;
925 drr->drr_u.drr_begin = *drc->drc_drrb;
926 if (drc->drc_byteswap) {
927 fletcher_4_incremental_byteswap(drr,
928 sizeof (dmu_replay_record_t), &drc->drc_cksum);
929 } else {
930 fletcher_4_incremental_native(drr,
931 sizeof (dmu_replay_record_t), &drc->drc_cksum);
932 }
933 kmem_free(drr, sizeof (dmu_replay_record_t));
934
935 if (drc->drc_byteswap) {
936 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
937 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
938 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
939 drrb->drr_type = BSWAP_32(drrb->drr_type);
940 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
941 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
942 }
943
944 drba.drba_origin = origin;
945 drba.drba_cookie = drc;
946 drba.drba_cred = CRED();
947
948 return (dsl_sync_task(tofs, dmu_recv_begin_check, dmu_recv_begin_sync,
949 &drba, 5));
950 }
951
952 struct restorearg {
953 int err;
954 boolean_t byteswap;
955 vnode_t *vp;
956 char *buf;
957 uint64_t voff;
958 int bufsize; /* amount of memory allocated for buf */
959 zio_cksum_t cksum;
960 avl_tree_t *guid_to_ds_map;
961 };
962
963 typedef struct guid_map_entry {
964 uint64_t guid;
965 dsl_dataset_t *gme_ds;
966 avl_node_t avlnode;
967 } guid_map_entry_t;
968
969 static int
970 guid_compare(const void *arg1, const void *arg2)
971 {
972 const guid_map_entry_t *gmep1 = arg1;
973 const guid_map_entry_t *gmep2 = arg2;
974
975 if (gmep1->guid < gmep2->guid)
976 return (-1);
977 else if (gmep1->guid > gmep2->guid)
978 return (1);
979 return (0);
980 }
981
982 static void
983 free_guid_map_onexit(void *arg)
984 {
985 avl_tree_t *ca = arg;
986 void *cookie = NULL;
987 guid_map_entry_t *gmep;
988
989 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
990 dsl_dataset_long_rele(gmep->gme_ds, gmep);
991 kmem_free(gmep, sizeof (guid_map_entry_t));
992 }
993 avl_destroy(ca);
994 kmem_free(ca, sizeof (avl_tree_t));
995 }
996
997 static void *
998 restore_read(struct restorearg *ra, int len)
999 {
1000 void *rv;
1001 int done = 0;
1002
1003 /* some things will require 8-byte alignment, so everything must */
1004 ASSERT0(len % 8);
1005
1006 while (done < len) {
1007 ssize_t resid;
1008
1009 ra->err = vn_rdwr(UIO_READ, ra->vp,
1010 (caddr_t)ra->buf + done, len - done,
1011 ra->voff, UIO_SYSSPACE, FAPPEND,
1012 RLIM64_INFINITY, CRED(), &resid);
1013
1014 if (resid == len - done)
1015 ra->err = SET_ERROR(EINVAL);
1016 ra->voff += len - done - resid;
1017 done = len - resid;
1018 if (ra->err != 0)
1019 return (NULL);
1020 }
1021
1022 ASSERT3U(done, ==, len);
1023 rv = ra->buf;
1024 if (ra->byteswap)
1025 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1026 else
1027 fletcher_4_incremental_native(rv, len, &ra->cksum);
1028 return (rv);
1029 }
1030
1031 noinline static void
1032 backup_byteswap(dmu_replay_record_t *drr)
1033 {
1034 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1035 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1036 drr->drr_type = BSWAP_32(drr->drr_type);
1037 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1038 switch (drr->drr_type) {
1039 case DRR_BEGIN:
1040 DO64(drr_begin.drr_magic);
1041 DO64(drr_begin.drr_versioninfo);
1042 DO64(drr_begin.drr_creation_time);
1043 DO32(drr_begin.drr_type);
1044 DO32(drr_begin.drr_flags);
1045 DO64(drr_begin.drr_toguid);
1046 DO64(drr_begin.drr_fromguid);
1047 break;
1048 case DRR_OBJECT:
1049 DO64(drr_object.drr_object);
1050 /* DO64(drr_object.drr_allocation_txg); */
1051 DO32(drr_object.drr_type);
1052 DO32(drr_object.drr_bonustype);
1053 DO32(drr_object.drr_blksz);
1054 DO32(drr_object.drr_bonuslen);
1055 DO64(drr_object.drr_toguid);
1056 break;
1057 case DRR_FREEOBJECTS:
1058 DO64(drr_freeobjects.drr_firstobj);
1059 DO64(drr_freeobjects.drr_numobjs);
1060 DO64(drr_freeobjects.drr_toguid);
1061 break;
1062 case DRR_WRITE:
1063 DO64(drr_write.drr_object);
1064 DO32(drr_write.drr_type);
1065 DO64(drr_write.drr_offset);
1066 DO64(drr_write.drr_length);
1067 DO64(drr_write.drr_toguid);
1068 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1069 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1070 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1071 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1072 DO64(drr_write.drr_key.ddk_prop);
1073 break;
1074 case DRR_WRITE_BYREF:
1075 DO64(drr_write_byref.drr_object);
1076 DO64(drr_write_byref.drr_offset);
1077 DO64(drr_write_byref.drr_length);
1078 DO64(drr_write_byref.drr_toguid);
1079 DO64(drr_write_byref.drr_refguid);
1080 DO64(drr_write_byref.drr_refobject);
1081 DO64(drr_write_byref.drr_refoffset);
1082 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1083 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1084 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1085 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1086 DO64(drr_write_byref.drr_key.ddk_prop);
1087 break;
1088 case DRR_FREE:
1089 DO64(drr_free.drr_object);
1090 DO64(drr_free.drr_offset);
1091 DO64(drr_free.drr_length);
1092 DO64(drr_free.drr_toguid);
1093 break;
1094 case DRR_SPILL:
1095 DO64(drr_spill.drr_object);
1096 DO64(drr_spill.drr_length);
1097 DO64(drr_spill.drr_toguid);
1098 break;
1099 case DRR_END:
1100 DO64(drr_end.drr_checksum.zc_word[0]);
1101 DO64(drr_end.drr_checksum.zc_word[1]);
1102 DO64(drr_end.drr_checksum.zc_word[2]);
1103 DO64(drr_end.drr_checksum.zc_word[3]);
1104 DO64(drr_end.drr_toguid);
1105 break;
1106 default:
1107 break;
1108 }
1109 #undef DO64
1110 #undef DO32
1111 }
1112
1113 noinline static int
1114 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1115 {
1116 int err;
1117 dmu_tx_t *tx;
1118 void *data = NULL;
1119
1120 if (drro->drr_type == DMU_OT_NONE ||
1121 !DMU_OT_IS_VALID(drro->drr_type) ||
1122 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1123 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1124 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1125 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1126 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1127 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1128 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1129 return (SET_ERROR(EINVAL));
1130 }
1131
1132 err = dmu_object_info(os, drro->drr_object, NULL);
1133
1134 if (err != 0 && err != ENOENT)
1135 return (SET_ERROR(EINVAL));
1136
1137 if (drro->drr_bonuslen) {
1138 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1139 if (ra->err != 0)
1140 return (ra->err);
1141 }
1142
1143 if (err == ENOENT) {
1144 /* currently free, want to be allocated */
1145 tx = dmu_tx_create(os);
1146 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1147 err = dmu_tx_assign(tx, TXG_WAIT);
1148 if (err != 0) {
1149 dmu_tx_abort(tx);
1150 return (err);
1151 }
1152 err = dmu_object_claim(os, drro->drr_object,
1153 drro->drr_type, drro->drr_blksz,
1154 drro->drr_bonustype, drro->drr_bonuslen, tx);
1155 dmu_tx_commit(tx);
1156 } else {
1157 /* currently allocated, want to be allocated */
1158 err = dmu_object_reclaim(os, drro->drr_object,
1159 drro->drr_type, drro->drr_blksz,
1160 drro->drr_bonustype, drro->drr_bonuslen);
1161 }
1162 if (err != 0) {
1163 return (SET_ERROR(EINVAL));
1164 }
1165
1166 tx = dmu_tx_create(os);
1167 dmu_tx_hold_bonus(tx, drro->drr_object);
1168 err = dmu_tx_assign(tx, TXG_WAIT);
1169 if (err != 0) {
1170 dmu_tx_abort(tx);
1171 return (err);
1172 }
1173
1174 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1175 tx);
1176 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1177
1178 if (data != NULL) {
1179 dmu_buf_t *db;
1180
1181 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1182 dmu_buf_will_dirty(db, tx);
1183
1184 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1185 bcopy(data, db->db_data, drro->drr_bonuslen);
1186 if (ra->byteswap) {
1187 dmu_object_byteswap_t byteswap =
1188 DMU_OT_BYTESWAP(drro->drr_bonustype);
1189 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1190 drro->drr_bonuslen);
1191 }
1192 dmu_buf_rele(db, FTAG);
1193 }
1194 dmu_tx_commit(tx);
1195 return (0);
1196 }
1197
1198 /* ARGSUSED */
1199 noinline static int
1200 restore_freeobjects(struct restorearg *ra, objset_t *os,
1201 struct drr_freeobjects *drrfo)
1202 {
1203 uint64_t obj;
1204
1205 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1206 return (SET_ERROR(EINVAL));
1207
1208 for (obj = drrfo->drr_firstobj;
1209 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1210 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1211 int err;
1212
1213 if (dmu_object_info(os, obj, NULL) != 0)
1214 continue;
1215
1216 err = dmu_free_object(os, obj);
1217 if (err != 0)
1218 return (err);
1219 }
1220 return (0);
1221 }
1222
1223 noinline static int
1224 restore_write(struct restorearg *ra, objset_t *os,
1225 struct drr_write *drrw)
1226 {
1227 dmu_tx_t *tx;
1228 void *data;
1229 int err;
1230
1231 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1232 !DMU_OT_IS_VALID(drrw->drr_type))
1233 return (SET_ERROR(EINVAL));
1234
1235 data = restore_read(ra, drrw->drr_length);
1236 if (data == NULL)
1237 return (ra->err);
1238
1239 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1240 return (SET_ERROR(EINVAL));
1241
1242 tx = dmu_tx_create(os);
1243
1244 dmu_tx_hold_write(tx, drrw->drr_object,
1245 drrw->drr_offset, drrw->drr_length);
1246 err = dmu_tx_assign(tx, TXG_WAIT);
1247 if (err != 0) {
1248 dmu_tx_abort(tx);
1249 return (err);
1250 }
1251 if (ra->byteswap) {
1252 dmu_object_byteswap_t byteswap =
1253 DMU_OT_BYTESWAP(drrw->drr_type);
1254 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1255 }
1256 dmu_write(os, drrw->drr_object,
1257 drrw->drr_offset, drrw->drr_length, data, tx);
1258 dmu_tx_commit(tx);
1259 return (0);
1260 }
1261
1262 /*
1263 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1264 * streams to refer to a copy of the data that is already on the
1265 * system because it came in earlier in the stream. This function
1266 * finds the earlier copy of the data, and uses that copy instead of
1267 * data from the stream to fulfill this write.
1268 */
1269 static int
1270 restore_write_byref(struct restorearg *ra, objset_t *os,
1271 struct drr_write_byref *drrwbr)
1272 {
1273 dmu_tx_t *tx;
1274 int err;
1275 guid_map_entry_t gmesrch;
1276 guid_map_entry_t *gmep;
1277 avl_index_t where;
1278 objset_t *ref_os = NULL;
1279 dmu_buf_t *dbp;
1280
1281 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1282 return (SET_ERROR(EINVAL));
1283
1284 /*
1285 * If the GUID of the referenced dataset is different from the
1286 * GUID of the target dataset, find the referenced dataset.
1287 */
1288 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1289 gmesrch.guid = drrwbr->drr_refguid;
1290 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1291 &where)) == NULL) {
1292 return (SET_ERROR(EINVAL));
1293 }
1294 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1295 return (SET_ERROR(EINVAL));
1296 } else {
1297 ref_os = os;
1298 }
1299
1300 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1301 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1302 if (err)
1303 return (err);
1304
1305 tx = dmu_tx_create(os);
1306
1307 dmu_tx_hold_write(tx, drrwbr->drr_object,
1308 drrwbr->drr_offset, drrwbr->drr_length);
1309 err = dmu_tx_assign(tx, TXG_WAIT);
1310 if (err != 0) {
1311 dmu_tx_abort(tx);
1312 return (err);
1313 }
1314 dmu_write(os, drrwbr->drr_object,
1315 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1316 dmu_buf_rele(dbp, FTAG);
1317 dmu_tx_commit(tx);
1318 return (0);
1319 }
1320
1321 static int
1322 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1323 {
1324 dmu_tx_t *tx;
1325 void *data;
1326 dmu_buf_t *db, *db_spill;
1327 int err;
1328
1329 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1330 drrs->drr_length > SPA_MAXBLOCKSIZE)
1331 return (SET_ERROR(EINVAL));
1332
1333 data = restore_read(ra, drrs->drr_length);
1334 if (data == NULL)
1335 return (ra->err);
1336
1337 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1338 return (SET_ERROR(EINVAL));
1339
1340 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1341 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1342 dmu_buf_rele(db, FTAG);
1343 return (err);
1344 }
1345
1346 tx = dmu_tx_create(os);
1347
1348 dmu_tx_hold_spill(tx, db->db_object);
1349
1350 err = dmu_tx_assign(tx, TXG_WAIT);
1351 if (err != 0) {
1352 dmu_buf_rele(db, FTAG);
1353 dmu_buf_rele(db_spill, FTAG);
1354 dmu_tx_abort(tx);
1355 return (err);
1356 }
1357 dmu_buf_will_dirty(db_spill, tx);
1358
1359 if (db_spill->db_size < drrs->drr_length)
1360 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1361 drrs->drr_length, tx));
1362 bcopy(data, db_spill->db_data, drrs->drr_length);
1363
1364 dmu_buf_rele(db, FTAG);
1365 dmu_buf_rele(db_spill, FTAG);
1366
1367 dmu_tx_commit(tx);
1368 return (0);
1369 }
1370
1371 /* ARGSUSED */
1372 noinline static int
1373 restore_free(struct restorearg *ra, objset_t *os,
1374 struct drr_free *drrf)
1375 {
1376 int err;
1377
1378 if (drrf->drr_length != -1ULL &&
1379 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1380 return (SET_ERROR(EINVAL));
1381
1382 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1383 return (SET_ERROR(EINVAL));
1384
1385 err = dmu_free_long_range(os, drrf->drr_object,
1386 drrf->drr_offset, drrf->drr_length);
1387 return (err);
1388 }
1389
1390 /* used to destroy the drc_ds on error */
1391 static void
1392 dmu_recv_cleanup_ds(dmu_recv_cookie_t *drc)
1393 {
1394 char name[MAXNAMELEN];
1395 dsl_dataset_name(drc->drc_ds, name);
1396 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1397 (void) dsl_destroy_head(name);
1398 }
1399
1400 /*
1401 * NB: callers *must* call dmu_recv_end() if this succeeds.
1402 */
1403 int
1404 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1405 int cleanup_fd, uint64_t *action_handlep)
1406 {
1407 struct restorearg ra = { 0 };
1408 dmu_replay_record_t *drr;
1409 objset_t *os;
1410 zio_cksum_t pcksum;
1411 int featureflags;
1412
1413 ra.byteswap = drc->drc_byteswap;
1414 ra.cksum = drc->drc_cksum;
1415 ra.vp = vp;
1416 ra.voff = *voffp;
1417 ra.bufsize = 1<<20;
1418 ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
1419
1420 /* these were verified in dmu_recv_begin */
1421 ASSERT3U(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo), ==,
1422 DMU_SUBSTREAM);
1423 ASSERT3U(drc->drc_drrb->drr_type, <, DMU_OST_NUMTYPES);
1424
1425 /*
1426 * Open the objset we are modifying.
1427 */
1428 VERIFY0(dmu_objset_from_ds(drc->drc_ds, &os));
1429
1430 ASSERT(drc->drc_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1431
1432 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1433
1434 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1435 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1436 minor_t minor;
1437
1438 if (cleanup_fd == -1) {
1439 ra.err = SET_ERROR(EBADF);
1440 goto out;
1441 }
1442 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1443 if (ra.err != 0) {
1444 cleanup_fd = -1;
1445 goto out;
1446 }
1447
1448 if (*action_handlep == 0) {
1449 ra.guid_to_ds_map =
1450 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1451 avl_create(ra.guid_to_ds_map, guid_compare,
1452 sizeof (guid_map_entry_t),
1453 offsetof(guid_map_entry_t, avlnode));
1454 ra.err = zfs_onexit_add_cb(minor,
1455 free_guid_map_onexit, ra.guid_to_ds_map,
1456 action_handlep);
1457 if (ra.err != 0)
1458 goto out;
1459 } else {
1460 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1461 (void **)&ra.guid_to_ds_map);
1462 if (ra.err != 0)
1463 goto out;
1464 }
1465
1466 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1467 }
1468
1469 /*
1470 * Read records and process them.
1471 */
1472 pcksum = ra.cksum;
1473 while (ra.err == 0 &&
1474 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1475 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1476 ra.err = SET_ERROR(EINTR);
1477 goto out;
1478 }
1479
1480 if (ra.byteswap)
1481 backup_byteswap(drr);
1482
1483 switch (drr->drr_type) {
1484 case DRR_OBJECT:
1485 {
1486 /*
1487 * We need to make a copy of the record header,
1488 * because restore_{object,write} may need to
1489 * restore_read(), which will invalidate drr.
1490 */
1491 struct drr_object drro = drr->drr_u.drr_object;
1492 ra.err = restore_object(&ra, os, &drro);
1493 break;
1494 }
1495 case DRR_FREEOBJECTS:
1496 {
1497 struct drr_freeobjects drrfo =
1498 drr->drr_u.drr_freeobjects;
1499 ra.err = restore_freeobjects(&ra, os, &drrfo);
1500 break;
1501 }
1502 case DRR_WRITE:
1503 {
1504 struct drr_write drrw = drr->drr_u.drr_write;
1505 ra.err = restore_write(&ra, os, &drrw);
1506 break;
1507 }
1508 case DRR_WRITE_BYREF:
1509 {
1510 struct drr_write_byref drrwbr =
1511 drr->drr_u.drr_write_byref;
1512 ra.err = restore_write_byref(&ra, os, &drrwbr);
1513 break;
1514 }
1515 case DRR_FREE:
1516 {
1517 struct drr_free drrf = drr->drr_u.drr_free;
1518 ra.err = restore_free(&ra, os, &drrf);
1519 break;
1520 }
1521 case DRR_END:
1522 {
1523 struct drr_end drre = drr->drr_u.drr_end;
1524 /*
1525 * We compare against the *previous* checksum
1526 * value, because the stored checksum is of
1527 * everything before the DRR_END record.
1528 */
1529 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1530 ra.err = SET_ERROR(ECKSUM);
1531 goto out;
1532 }
1533 case DRR_SPILL:
1534 {
1535 struct drr_spill drrs = drr->drr_u.drr_spill;
1536 ra.err = restore_spill(&ra, os, &drrs);
1537 break;
1538 }
1539 default:
1540 ra.err = SET_ERROR(EINVAL);
1541 goto out;
1542 }
1543 pcksum = ra.cksum;
1544 }
1545 ASSERT(ra.err != 0);
1546
1547 out:
1548 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1549 zfs_onexit_fd_rele(cleanup_fd);
1550
1551 if (ra.err != 0) {
1552 /*
1553 * destroy what we created, so we don't leave it in the
1554 * inconsistent restoring state.
1555 */
1556 dmu_recv_cleanup_ds(drc);
1557 }
1558
1559 vmem_free(ra.buf, ra.bufsize);
1560 *voffp = ra.voff;
1561 return (ra.err);
1562 }
1563
1564 static int
1565 dmu_recv_end_check(void *arg, dmu_tx_t *tx)
1566 {
1567 dmu_recv_cookie_t *drc = arg;
1568 dsl_pool_t *dp = dmu_tx_pool(tx);
1569 int error;
1570
1571 ASSERT3P(drc->drc_ds->ds_owner, ==, dmu_recv_tag);
1572
1573 if (!drc->drc_newfs) {
1574 dsl_dataset_t *origin_head;
1575
1576 error = dsl_dataset_hold(dp, drc->drc_tofs, FTAG, &origin_head);
1577 if (error != 0)
1578 return (error);
1579 error = dsl_dataset_clone_swap_check_impl(drc->drc_ds,
1580 origin_head, drc->drc_force);
1581 if (error != 0) {
1582 dsl_dataset_rele(origin_head, FTAG);
1583 return (error);
1584 }
1585 error = dsl_dataset_snapshot_check_impl(origin_head,
1586 drc->drc_tosnap, tx);
1587 dsl_dataset_rele(origin_head, FTAG);
1588 if (error != 0)
1589 return (error);
1590
1591 error = dsl_destroy_head_check_impl(drc->drc_ds, 1);
1592 } else {
1593 error = dsl_dataset_snapshot_check_impl(drc->drc_ds,
1594 drc->drc_tosnap, tx);
1595 }
1596 return (error);
1597 }
1598
1599 static void
1600 dmu_recv_end_sync(void *arg, dmu_tx_t *tx)
1601 {
1602 dmu_recv_cookie_t *drc = arg;
1603 dsl_pool_t *dp = dmu_tx_pool(tx);
1604
1605 spa_history_log_internal_ds(drc->drc_ds, "finish receiving",
1606 tx, "snap=%s", drc->drc_tosnap);
1607
1608 if (!drc->drc_newfs) {
1609 dsl_dataset_t *origin_head;
1610
1611 VERIFY0(dsl_dataset_hold(dp, drc->drc_tofs, FTAG,
1612 &origin_head));
1613 dsl_dataset_clone_swap_sync_impl(drc->drc_ds,
1614 origin_head, tx);
1615 dsl_dataset_snapshot_sync_impl(origin_head,
1616 drc->drc_tosnap, tx);
1617
1618 /* set snapshot's creation time and guid */
1619 dmu_buf_will_dirty(origin_head->ds_prev->ds_dbuf, tx);
1620 origin_head->ds_prev->ds_phys->ds_creation_time =
1621 drc->drc_drrb->drr_creation_time;
1622 origin_head->ds_prev->ds_phys->ds_guid =
1623 drc->drc_drrb->drr_toguid;
1624 origin_head->ds_prev->ds_phys->ds_flags &=
1625 ~DS_FLAG_INCONSISTENT;
1626
1627 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
1628 origin_head->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1629
1630 dsl_dataset_rele(origin_head, FTAG);
1631 dsl_destroy_head_sync_impl(drc->drc_ds, tx);
1632 } else {
1633 dsl_dataset_t *ds = drc->drc_ds;
1634
1635 dsl_dataset_snapshot_sync_impl(ds, drc->drc_tosnap, tx);
1636
1637 /* set snapshot's creation time and guid */
1638 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1639 ds->ds_prev->ds_phys->ds_creation_time =
1640 drc->drc_drrb->drr_creation_time;
1641 ds->ds_prev->ds_phys->ds_guid = drc->drc_drrb->drr_toguid;
1642 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1643
1644 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1645 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1646 }
1647 drc->drc_newsnapobj = drc->drc_ds->ds_phys->ds_prev_snap_obj;
1648 /*
1649 * Release the hold from dmu_recv_begin. This must be done before
1650 * we return to open context, so that when we free the dataset's dnode,
1651 * we can evict its bonus buffer.
1652 */
1653 dsl_dataset_disown(drc->drc_ds, dmu_recv_tag);
1654 drc->drc_ds = NULL;
1655 }
1656
1657 static int
1658 add_ds_to_guidmap(const char *name, avl_tree_t *guid_map, uint64_t snapobj)
1659 {
1660 dsl_pool_t *dp;
1661 dsl_dataset_t *snapds;
1662 guid_map_entry_t *gmep;
1663 int err;
1664
1665 ASSERT(guid_map != NULL);
1666
1667 err = dsl_pool_hold(name, FTAG, &dp);
1668 if (err != 0)
1669 return (err);
1670 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snapds);
1671 if (err == 0) {
1672 gmep = kmem_alloc(sizeof (guid_map_entry_t), KM_SLEEP);
1673 gmep->guid = snapds->ds_phys->ds_guid;
1674 gmep->gme_ds = snapds;
1675 avl_add(guid_map, gmep);
1676 dsl_dataset_long_hold(snapds, gmep);
1677 dsl_dataset_rele(snapds, FTAG);
1678 }
1679
1680 dsl_pool_rele(dp, FTAG);
1681 return (err);
1682 }
1683
1684 static int dmu_recv_end_modified_blocks = 3;
1685
1686 static int
1687 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1688 {
1689 int error;
1690
1691 #ifdef _KERNEL
1692 char *name;
1693
1694 /*
1695 * We will be destroying the ds; make sure its origin is unmounted if
1696 * necessary.
1697 */
1698 name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
1699 dsl_dataset_name(drc->drc_ds, name);
1700 zfs_destroy_unmount_origin(name);
1701 kmem_free(name, MAXNAMELEN);
1702 #endif
1703
1704 error = dsl_sync_task(drc->drc_tofs,
1705 dmu_recv_end_check, dmu_recv_end_sync, drc,
1706 dmu_recv_end_modified_blocks);
1707
1708 if (error != 0)
1709 dmu_recv_cleanup_ds(drc);
1710 return (error);
1711 }
1712
1713 static int
1714 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1715 {
1716 int error;
1717
1718 error = dsl_sync_task(drc->drc_tofs,
1719 dmu_recv_end_check, dmu_recv_end_sync, drc,
1720 dmu_recv_end_modified_blocks);
1721
1722 if (error != 0) {
1723 dmu_recv_cleanup_ds(drc);
1724 } else if (drc->drc_guid_to_ds_map != NULL) {
1725 (void) add_ds_to_guidmap(drc->drc_tofs,
1726 drc->drc_guid_to_ds_map,
1727 drc->drc_newsnapobj);
1728 }
1729 return (error);
1730 }
1731
1732 int
1733 dmu_recv_end(dmu_recv_cookie_t *drc)
1734 {
1735 if (drc->drc_newfs)
1736 return (dmu_recv_new_end(drc));
1737 else
1738 return (dmu_recv_existing_end(drc));
1739 }
ZFS On Linux mirror for PVE