]> 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 #2882, #2883, #2900
[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) 2011 by Delphix. All rights reserved.
26 * Copyright (c) 2012, Joyent, Inc. 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
52 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
53 int zfs_send_corrupt_data = B_FALSE;
54
55 static char *dmu_recv_tag = "dmu_recv_tag";
56
57 typedef struct dump_bytes_io {
58 dmu_sendarg_t *dbi_dsp;
59 void *dbi_buf;
60 int dbi_len;
61 } dump_bytes_io_t;
62
63 static void
64 dump_bytes_strategy(void *arg)
65 {
66 dump_bytes_io_t *dbi = (dump_bytes_io_t *)arg;
67 dmu_sendarg_t *dsp = dbi->dbi_dsp;
68 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
69 ssize_t resid; /* have to get resid to get detailed errno */
70 ASSERT0(dbi->dbi_len % 8);
71
72 fletcher_4_incremental_native(dbi->dbi_buf, dbi->dbi_len, &dsp->dsa_zc);
73 dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
74 (caddr_t)dbi->dbi_buf, dbi->dbi_len,
75 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
76
77 mutex_enter(&ds->ds_sendstream_lock);
78 *dsp->dsa_off += dbi->dbi_len;
79 mutex_exit(&ds->ds_sendstream_lock);
80 }
81
82 static int
83 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
84 {
85 dump_bytes_io_t dbi;
86
87 dbi.dbi_dsp = dsp;
88 dbi.dbi_buf = buf;
89 dbi.dbi_len = len;
90
91 /*
92 * The vn_rdwr() call is performed in a taskq to ensure that there is
93 * always enough stack space to write safely to the target filesystem.
94 * The ZIO_TYPE_FREE threads are used because there can be a lot of
95 * them and they are used in vdev_file.c for a similar purpose.
96 */
97 spa_taskq_dispatch_sync(dmu_objset_spa(dsp->dsa_os), ZIO_TYPE_FREE,
98 ZIO_TASKQ_ISSUE, dump_bytes_strategy, &dbi, TQ_SLEEP);
99
100 return (dsp->dsa_err);
101 }
102
103 static int
104 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
105 uint64_t length)
106 {
107 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
108
109 if (length != -1ULL && offset + length < offset)
110 length = -1ULL;
111
112 /*
113 * If there is a pending op, but it's not PENDING_FREE, push it out,
114 * since free block aggregation can only be done for blocks of the
115 * same type (i.e., DRR_FREE records can only be aggregated with
116 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
117 * aggregated with other DRR_FREEOBJECTS records.
118 */
119 if (dsp->dsa_pending_op != PENDING_NONE &&
120 dsp->dsa_pending_op != PENDING_FREE) {
121 if (dump_bytes(dsp, dsp->dsa_drr,
122 sizeof (dmu_replay_record_t)) != 0)
123 return (EINTR);
124 dsp->dsa_pending_op = PENDING_NONE;
125 }
126
127 if (dsp->dsa_pending_op == PENDING_FREE) {
128 /*
129 * There should never be a PENDING_FREE if length is -1
130 * (because dump_dnode is the only place where this
131 * function is called with a -1, and only after flushing
132 * any pending record).
133 */
134 ASSERT(length != -1ULL);
135 /*
136 * Check to see whether this free block can be aggregated
137 * with pending one.
138 */
139 if (drrf->drr_object == object && drrf->drr_offset +
140 drrf->drr_length == offset) {
141 drrf->drr_length += length;
142 return (0);
143 } else {
144 /* not a continuation. Push out pending record */
145 if (dump_bytes(dsp, dsp->dsa_drr,
146 sizeof (dmu_replay_record_t)) != 0)
147 return (EINTR);
148 dsp->dsa_pending_op = PENDING_NONE;
149 }
150 }
151 /* create a FREE record and make it pending */
152 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
153 dsp->dsa_drr->drr_type = DRR_FREE;
154 drrf->drr_object = object;
155 drrf->drr_offset = offset;
156 drrf->drr_length = length;
157 drrf->drr_toguid = dsp->dsa_toguid;
158 if (length == -1ULL) {
159 if (dump_bytes(dsp, dsp->dsa_drr,
160 sizeof (dmu_replay_record_t)) != 0)
161 return (EINTR);
162 } else {
163 dsp->dsa_pending_op = PENDING_FREE;
164 }
165
166 return (0);
167 }
168
169 static int
170 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
171 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
172 {
173 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
174
175
176 /*
177 * If there is any kind of pending aggregation (currently either
178 * a grouping of free objects or free blocks), push it out to
179 * the stream, since aggregation can't be done across operations
180 * of different types.
181 */
182 if (dsp->dsa_pending_op != PENDING_NONE) {
183 if (dump_bytes(dsp, dsp->dsa_drr,
184 sizeof (dmu_replay_record_t)) != 0)
185 return (EINTR);
186 dsp->dsa_pending_op = PENDING_NONE;
187 }
188 /* write a DATA record */
189 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
190 dsp->dsa_drr->drr_type = DRR_WRITE;
191 drrw->drr_object = object;
192 drrw->drr_type = type;
193 drrw->drr_offset = offset;
194 drrw->drr_length = blksz;
195 drrw->drr_toguid = dsp->dsa_toguid;
196 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
197 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
198 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
199 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
200 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
201 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
202 drrw->drr_key.ddk_cksum = bp->blk_cksum;
203
204 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
205 return (EINTR);
206 if (dump_bytes(dsp, data, blksz) != 0)
207 return (EINTR);
208 return (0);
209 }
210
211 static int
212 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
213 {
214 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
215
216 if (dsp->dsa_pending_op != PENDING_NONE) {
217 if (dump_bytes(dsp, dsp->dsa_drr,
218 sizeof (dmu_replay_record_t)) != 0)
219 return (EINTR);
220 dsp->dsa_pending_op = PENDING_NONE;
221 }
222
223 /* write a SPILL record */
224 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
225 dsp->dsa_drr->drr_type = DRR_SPILL;
226 drrs->drr_object = object;
227 drrs->drr_length = blksz;
228 drrs->drr_toguid = dsp->dsa_toguid;
229
230 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
231 return (EINTR);
232 if (dump_bytes(dsp, data, blksz))
233 return (EINTR);
234 return (0);
235 }
236
237 static int
238 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
239 {
240 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
241
242 /*
243 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
244 * push it out, since free block aggregation can only be done for
245 * blocks of the same type (i.e., DRR_FREE records can only be
246 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
247 * can only be aggregated with other DRR_FREEOBJECTS records.
248 */
249 if (dsp->dsa_pending_op != PENDING_NONE &&
250 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
251 if (dump_bytes(dsp, dsp->dsa_drr,
252 sizeof (dmu_replay_record_t)) != 0)
253 return (EINTR);
254 dsp->dsa_pending_op = PENDING_NONE;
255 }
256 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
257 /*
258 * See whether this free object array can be aggregated
259 * with pending one
260 */
261 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
262 drrfo->drr_numobjs += numobjs;
263 return (0);
264 } else {
265 /* can't be aggregated. Push out pending record */
266 if (dump_bytes(dsp, dsp->dsa_drr,
267 sizeof (dmu_replay_record_t)) != 0)
268 return (EINTR);
269 dsp->dsa_pending_op = PENDING_NONE;
270 }
271 }
272
273 /* write a FREEOBJECTS record */
274 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
275 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
276 drrfo->drr_firstobj = firstobj;
277 drrfo->drr_numobjs = numobjs;
278 drrfo->drr_toguid = dsp->dsa_toguid;
279
280 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
281
282 return (0);
283 }
284
285 static int
286 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
287 {
288 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
289
290 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
291 return (dump_freeobjects(dsp, object, 1));
292
293 if (dsp->dsa_pending_op != PENDING_NONE) {
294 if (dump_bytes(dsp, dsp->dsa_drr,
295 sizeof (dmu_replay_record_t)) != 0)
296 return (EINTR);
297 dsp->dsa_pending_op = PENDING_NONE;
298 }
299
300 /* write an OBJECT record */
301 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
302 dsp->dsa_drr->drr_type = DRR_OBJECT;
303 drro->drr_object = object;
304 drro->drr_type = dnp->dn_type;
305 drro->drr_bonustype = dnp->dn_bonustype;
306 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
307 drro->drr_bonuslen = dnp->dn_bonuslen;
308 drro->drr_checksumtype = dnp->dn_checksum;
309 drro->drr_compress = dnp->dn_compress;
310 drro->drr_toguid = dsp->dsa_toguid;
311
312 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
313 return (EINTR);
314
315 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
316 return (EINTR);
317
318 /* free anything past the end of the file */
319 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
320 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
321 return (EINTR);
322 if (dsp->dsa_err)
323 return (EINTR);
324 return (0);
325 }
326
327 #define BP_SPAN(dnp, level) \
328 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
329 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
330
331 /* ARGSUSED */
332 static int
333 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
334 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
335 {
336 dmu_sendarg_t *dsp = arg;
337 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
338 int err = 0;
339
340 if (issig(JUSTLOOKING) && issig(FORREAL))
341 return (EINTR);
342
343 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
344 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
345 return (0);
346 } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) {
347 uint64_t span = BP_SPAN(dnp, zb->zb_level);
348 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
349 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
350 } else if (bp == NULL) {
351 uint64_t span = BP_SPAN(dnp, zb->zb_level);
352 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
353 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
354 return (0);
355 } else if (type == DMU_OT_DNODE) {
356 dnode_phys_t *blk;
357 int i;
358 int blksz = BP_GET_LSIZE(bp);
359 uint32_t aflags = ARC_WAIT;
360 arc_buf_t *abuf;
361
362 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
363 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
364 &aflags, zb) != 0)
365 return (EIO);
366
367 blk = abuf->b_data;
368 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
369 uint64_t dnobj = (zb->zb_blkid <<
370 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
371 err = dump_dnode(dsp, dnobj, blk+i);
372 if (err)
373 break;
374 }
375 (void) arc_buf_remove_ref(abuf, &abuf);
376 } else if (type == DMU_OT_SA) {
377 uint32_t aflags = ARC_WAIT;
378 arc_buf_t *abuf;
379 int blksz = BP_GET_LSIZE(bp);
380
381 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
382 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
383 &aflags, zb) != 0)
384 return (EIO);
385
386 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
387 (void) arc_buf_remove_ref(abuf, &abuf);
388 } else { /* it's a level-0 block of a regular object */
389 uint32_t aflags = ARC_WAIT;
390 arc_buf_t *abuf;
391 int blksz = BP_GET_LSIZE(bp);
392
393 if (arc_read(NULL, spa, bp, arc_getbuf_func, &abuf,
394 ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL,
395 &aflags, zb) != 0) {
396 if (zfs_send_corrupt_data) {
397 uint64_t *ptr;
398 /* Send a block filled with 0x"zfs badd bloc" */
399 abuf = arc_buf_alloc(spa, blksz, &abuf,
400 ARC_BUFC_DATA);
401 for (ptr = abuf->b_data;
402 (char *)ptr < (char *)abuf->b_data + blksz;
403 ptr++)
404 *ptr = 0x2f5baddb10cULL;
405 } else {
406 return (EIO);
407 }
408 }
409
410 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
411 blksz, bp, abuf->b_data);
412 (void) arc_buf_remove_ref(abuf, &abuf);
413 }
414
415 ASSERT(err == 0 || err == EINTR);
416 return (err);
417 }
418
419 /*
420 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
421 * For example, they could both be snapshots of the same filesystem, and
422 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
423 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
424 * filesystem. Or 'earlier' could be the origin's origin.
425 */
426 static boolean_t
427 is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
428 {
429 dsl_pool_t *dp = later->ds_dir->dd_pool;
430 int error;
431 boolean_t ret;
432 dsl_dataset_t *origin;
433
434 if (earlier->ds_phys->ds_creation_txg >=
435 later->ds_phys->ds_creation_txg)
436 return (B_FALSE);
437
438 if (later->ds_dir == earlier->ds_dir)
439 return (B_TRUE);
440 if (!dsl_dir_is_clone(later->ds_dir))
441 return (B_FALSE);
442
443 rw_enter(&dp->dp_config_rwlock, RW_READER);
444 if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object) {
445 rw_exit(&dp->dp_config_rwlock);
446 return (B_TRUE);
447 }
448 error = dsl_dataset_hold_obj(dp,
449 later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
450 rw_exit(&dp->dp_config_rwlock);
451 if (error != 0)
452 return (B_FALSE);
453 ret = is_before(origin, earlier);
454 dsl_dataset_rele(origin, FTAG);
455 return (ret);
456 }
457
458 int
459 dmu_send(objset_t *tosnap, objset_t *fromsnap, int outfd, vnode_t *vp,
460 offset_t *off)
461 {
462 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
463 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
464 dmu_replay_record_t *drr;
465 dmu_sendarg_t *dsp;
466 int err;
467 uint64_t fromtxg = 0;
468
469 /* tosnap must be a snapshot */
470 if (ds->ds_phys->ds_next_snap_obj == 0)
471 return (EINVAL);
472
473 /*
474 * fromsnap must be an earlier snapshot from the same fs as tosnap,
475 * or the origin's fs.
476 */
477 if (fromds != NULL && !is_before(ds, fromds))
478 return (EXDEV);
479
480 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
481 drr->drr_type = DRR_BEGIN;
482 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
483 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
484 DMU_SUBSTREAM);
485
486 #ifdef _KERNEL
487 if (dmu_objset_type(tosnap) == DMU_OST_ZFS) {
488 uint64_t version;
489 if (zfs_get_zplprop(tosnap, ZFS_PROP_VERSION, &version) != 0) {
490 kmem_free(drr, sizeof (dmu_replay_record_t));
491 return (EINVAL);
492 }
493 if (version == ZPL_VERSION_SA) {
494 DMU_SET_FEATUREFLAGS(
495 drr->drr_u.drr_begin.drr_versioninfo,
496 DMU_BACKUP_FEATURE_SA_SPILL);
497 }
498 }
499 #endif
500
501 drr->drr_u.drr_begin.drr_creation_time =
502 ds->ds_phys->ds_creation_time;
503 drr->drr_u.drr_begin.drr_type = tosnap->os_phys->os_type;
504 if (fromds != NULL && ds->ds_dir != fromds->ds_dir)
505 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
506 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
507 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
508 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
509
510 if (fromds)
511 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
512 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
513
514 if (fromds)
515 fromtxg = fromds->ds_phys->ds_creation_txg;
516
517 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
518
519 dsp->dsa_drr = drr;
520 dsp->dsa_vp = vp;
521 dsp->dsa_outfd = outfd;
522 dsp->dsa_proc = curproc;
523 dsp->dsa_os = tosnap;
524 dsp->dsa_off = off;
525 dsp->dsa_toguid = ds->ds_phys->ds_guid;
526 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
527 dsp->dsa_pending_op = PENDING_NONE;
528
529 mutex_enter(&ds->ds_sendstream_lock);
530 list_insert_head(&ds->ds_sendstreams, dsp);
531 mutex_exit(&ds->ds_sendstream_lock);
532
533 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
534 err = dsp->dsa_err;
535 goto out;
536 }
537
538 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
539 backup_cb, dsp);
540
541 if (dsp->dsa_pending_op != PENDING_NONE)
542 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
543 err = EINTR;
544
545 if (err) {
546 if (err == EINTR && dsp->dsa_err)
547 err = dsp->dsa_err;
548 goto out;
549 }
550
551 bzero(drr, sizeof (dmu_replay_record_t));
552 drr->drr_type = DRR_END;
553 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
554 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
555
556 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
557 err = dsp->dsa_err;
558 goto out;
559 }
560
561 out:
562 mutex_enter(&ds->ds_sendstream_lock);
563 list_remove(&ds->ds_sendstreams, dsp);
564 mutex_exit(&ds->ds_sendstream_lock);
565
566 kmem_free(drr, sizeof (dmu_replay_record_t));
567 kmem_free(dsp, sizeof (dmu_sendarg_t));
568
569 return (err);
570 }
571
572 int
573 dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, uint64_t *sizep)
574 {
575 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
576 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
577 dsl_pool_t *dp = ds->ds_dir->dd_pool;
578 int err;
579 uint64_t size, recordsize;
580
581 /* tosnap must be a snapshot */
582 if (ds->ds_phys->ds_next_snap_obj == 0)
583 return (EINVAL);
584
585 /*
586 * fromsnap must be an earlier snapshot from the same fs as tosnap,
587 * or the origin's fs.
588 */
589 if (fromds != NULL && !is_before(ds, fromds))
590 return (EXDEV);
591
592 /* Get uncompressed size estimate of changed data. */
593 if (fromds == NULL) {
594 size = ds->ds_phys->ds_uncompressed_bytes;
595 } else {
596 uint64_t used, comp;
597 err = dsl_dataset_space_written(fromds, ds,
598 &used, &comp, &size);
599 if (err)
600 return (err);
601 }
602
603 /*
604 * Assume that space (both on-disk and in-stream) is dominated by
605 * data. We will adjust for indirect blocks and the copies property,
606 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
607 */
608
609 /*
610 * Subtract out approximate space used by indirect blocks.
611 * Assume most space is used by data blocks (non-indirect, non-dnode).
612 * Assume all blocks are recordsize. Assume ditto blocks and
613 * internal fragmentation counter out compression.
614 *
615 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
616 * block, which we observe in practice.
617 */
618 rw_enter(&dp->dp_config_rwlock, RW_READER);
619 err = dsl_prop_get_ds(ds, "recordsize",
620 sizeof (recordsize), 1, &recordsize, NULL);
621 rw_exit(&dp->dp_config_rwlock);
622 if (err)
623 return (err);
624 size -= size / recordsize * sizeof (blkptr_t);
625
626 /* Add in the space for the record associated with each block. */
627 size += size / recordsize * sizeof (dmu_replay_record_t);
628
629 *sizep = size;
630
631 return (0);
632 }
633
634 struct recvbeginsyncarg {
635 const char *tofs;
636 const char *tosnap;
637 dsl_dataset_t *origin;
638 uint64_t fromguid;
639 dmu_objset_type_t type;
640 void *tag;
641 boolean_t force;
642 uint64_t dsflags;
643 char clonelastname[MAXNAMELEN];
644 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
645 cred_t *cr;
646 };
647
648 /* ARGSUSED */
649 static int
650 recv_new_check(void *arg1, void *arg2, dmu_tx_t *tx)
651 {
652 dsl_dir_t *dd = arg1;
653 struct recvbeginsyncarg *rbsa = arg2;
654 objset_t *mos = dd->dd_pool->dp_meta_objset;
655 uint64_t val;
656 int err;
657
658 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
659 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
660
661 if (err != ENOENT)
662 return (err ? err : EEXIST);
663
664 if (rbsa->origin) {
665 /* make sure it's a snap in the same pool */
666 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
667 return (EXDEV);
668 if (!dsl_dataset_is_snapshot(rbsa->origin))
669 return (EINVAL);
670 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
671 return (ENODEV);
672 }
673
674 return (0);
675 }
676
677 static void
678 recv_new_sync(void *arg1, void *arg2, dmu_tx_t *tx)
679 {
680 dsl_dir_t *dd = arg1;
681 struct recvbeginsyncarg *rbsa = arg2;
682 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
683 uint64_t dsobj;
684
685 /* Create and open new dataset. */
686 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
687 rbsa->origin, flags, rbsa->cr, tx);
688 VERIFY(0 == dsl_dataset_own_obj(dd->dd_pool, dsobj,
689 B_TRUE, dmu_recv_tag, &rbsa->ds));
690
691 if (rbsa->origin == NULL) {
692 (void) dmu_objset_create_impl(dd->dd_pool->dp_spa,
693 rbsa->ds, &rbsa->ds->ds_phys->ds_bp, rbsa->type, tx);
694 }
695
696 spa_history_log_internal_ds(rbsa->ds, "receive new", tx, "");
697 }
698
699 /* ARGSUSED */
700 static int
701 recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
702 {
703 dsl_dataset_t *ds = arg1;
704 struct recvbeginsyncarg *rbsa = arg2;
705 int err;
706 uint64_t val;
707
708 /* must not have any changes since most recent snapshot */
709 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
710 return (ETXTBSY);
711
712 /* new snapshot name must not exist */
713 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
714 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
715 if (err == 0)
716 return (EEXIST);
717 if (err != ENOENT)
718 return (err);
719
720 if (rbsa->fromguid) {
721 /* if incremental, most recent snapshot must match fromguid */
722 if (ds->ds_prev == NULL)
723 return (ENODEV);
724
725 /*
726 * most recent snapshot must match fromguid, or there are no
727 * changes since the fromguid one
728 */
729 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid) {
730 uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth;
731 uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj;
732 while (obj != 0) {
733 dsl_dataset_t *snap;
734 err = dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
735 obj, FTAG, &snap);
736 if (err)
737 return (ENODEV);
738 if (snap->ds_phys->ds_creation_txg < birth) {
739 dsl_dataset_rele(snap, FTAG);
740 return (ENODEV);
741 }
742 if (snap->ds_phys->ds_guid == rbsa->fromguid) {
743 dsl_dataset_rele(snap, FTAG);
744 break; /* it's ok */
745 }
746 obj = snap->ds_phys->ds_prev_snap_obj;
747 dsl_dataset_rele(snap, FTAG);
748 }
749 if (obj == 0)
750 return (ENODEV);
751 }
752 } else {
753 /* if full, most recent snapshot must be $ORIGIN */
754 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
755 return (ENODEV);
756 }
757
758 /* temporary clone name must not exist */
759 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
760 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
761 rbsa->clonelastname, 8, 1, &val);
762 if (err == 0)
763 return (EEXIST);
764 if (err != ENOENT)
765 return (err);
766
767 return (0);
768 }
769
770 /* ARGSUSED */
771 static void
772 recv_existing_sync(void *arg1, void *arg2, dmu_tx_t *tx)
773 {
774 dsl_dataset_t *ohds = arg1;
775 struct recvbeginsyncarg *rbsa = arg2;
776 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
777 dsl_dataset_t *cds;
778 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
779 uint64_t dsobj;
780
781 /* create and open the temporary clone */
782 dsobj = dsl_dataset_create_sync(ohds->ds_dir, rbsa->clonelastname,
783 ohds->ds_prev, flags, rbsa->cr, tx);
784 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, B_TRUE, dmu_recv_tag, &cds));
785
786 /*
787 * If we actually created a non-clone, we need to create the
788 * objset in our new dataset.
789 */
790 if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds))) {
791 (void) dmu_objset_create_impl(dp->dp_spa,
792 cds, dsl_dataset_get_blkptr(cds), rbsa->type, tx);
793 }
794
795 rbsa->ds = cds;
796
797 spa_history_log_internal_ds(cds, "receive over existing", tx, "");
798 }
799
800 static boolean_t
801 dmu_recv_verify_features(dsl_dataset_t *ds, struct drr_begin *drrb)
802 {
803 int featureflags;
804
805 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
806
807 /* Verify pool version supports SA if SA_SPILL feature set */
808 return ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
809 (spa_version(dsl_dataset_get_spa(ds)) < SPA_VERSION_SA));
810 }
811
812 /*
813 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
814 * succeeds; otherwise we will leak the holds on the datasets.
815 */
816 int
817 dmu_recv_begin(char *tofs, char *tosnap, char *top_ds, struct drr_begin *drrb,
818 boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
819 {
820 int err = 0;
821 boolean_t byteswap;
822 struct recvbeginsyncarg rbsa = { 0 };
823 uint64_t versioninfo;
824 int flags;
825 dsl_dataset_t *ds;
826
827 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
828 byteswap = FALSE;
829 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
830 byteswap = TRUE;
831 else
832 return (EINVAL);
833
834 rbsa.tofs = tofs;
835 rbsa.tosnap = tosnap;
836 rbsa.origin = origin ? origin->os_dsl_dataset : NULL;
837 rbsa.fromguid = drrb->drr_fromguid;
838 rbsa.type = drrb->drr_type;
839 rbsa.tag = FTAG;
840 rbsa.dsflags = 0;
841 rbsa.cr = CRED();
842 versioninfo = drrb->drr_versioninfo;
843 flags = drrb->drr_flags;
844
845 if (byteswap) {
846 rbsa.type = BSWAP_32(rbsa.type);
847 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
848 versioninfo = BSWAP_64(versioninfo);
849 flags = BSWAP_32(flags);
850 }
851
852 if (DMU_GET_STREAM_HDRTYPE(versioninfo) == DMU_COMPOUNDSTREAM ||
853 rbsa.type >= DMU_OST_NUMTYPES ||
854 ((flags & DRR_FLAG_CLONE) && origin == NULL))
855 return (EINVAL);
856
857 if (flags & DRR_FLAG_CI_DATA)
858 rbsa.dsflags = DS_FLAG_CI_DATASET;
859
860 bzero(drc, sizeof (dmu_recv_cookie_t));
861 drc->drc_drrb = drrb;
862 drc->drc_tosnap = tosnap;
863 drc->drc_top_ds = top_ds;
864 drc->drc_force = force;
865
866 /*
867 * Process the begin in syncing context.
868 */
869
870 /* open the dataset we are logically receiving into */
871 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
872 if (err == 0) {
873 if (dmu_recv_verify_features(ds, drrb)) {
874 dsl_dataset_rele(ds, dmu_recv_tag);
875 return (ENOTSUP);
876 }
877 /* target fs already exists; recv into temp clone */
878
879 /* Can't recv a clone into an existing fs */
880 if (flags & DRR_FLAG_CLONE) {
881 dsl_dataset_rele(ds, dmu_recv_tag);
882 return (EINVAL);
883 }
884
885 /* must not have an incremental recv already in progress */
886 if (!mutex_tryenter(&ds->ds_recvlock)) {
887 dsl_dataset_rele(ds, dmu_recv_tag);
888 return (EBUSY);
889 }
890
891 /* tmp clone name is: tofs/%tosnap" */
892 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
893 "%%%s", tosnap);
894 rbsa.force = force;
895 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
896 recv_existing_check, recv_existing_sync, ds, &rbsa, 5);
897 if (err) {
898 mutex_exit(&ds->ds_recvlock);
899 dsl_dataset_rele(ds, dmu_recv_tag);
900 return (err);
901 }
902 drc->drc_logical_ds = ds;
903 drc->drc_real_ds = rbsa.ds;
904 } else if (err == ENOENT) {
905 /* target fs does not exist; must be a full backup or clone */
906 char *cp;
907
908 /*
909 * If it's a non-clone incremental, we are missing the
910 * target fs, so fail the recv.
911 */
912 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE))
913 return (ENOENT);
914
915 /* Open the parent of tofs */
916 cp = strrchr(tofs, '/');
917 *cp = '\0';
918 err = dsl_dataset_hold(tofs, FTAG, &ds);
919 *cp = '/';
920 if (err)
921 return (err);
922
923 if (dmu_recv_verify_features(ds, drrb)) {
924 dsl_dataset_rele(ds, FTAG);
925 return (ENOTSUP);
926 }
927
928 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
929 recv_new_check, recv_new_sync, ds->ds_dir, &rbsa, 5);
930 dsl_dataset_rele(ds, FTAG);
931 if (err)
932 return (err);
933 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
934 drc->drc_newfs = B_TRUE;
935 }
936
937 return (err);
938 }
939
940 struct restorearg {
941 int err;
942 int byteswap;
943 vnode_t *vp;
944 char *buf;
945 uint64_t voff;
946 int bufsize; /* amount of memory allocated for buf */
947 zio_cksum_t cksum;
948 avl_tree_t *guid_to_ds_map;
949 };
950
951 typedef struct guid_map_entry {
952 uint64_t guid;
953 dsl_dataset_t *gme_ds;
954 avl_node_t avlnode;
955 } guid_map_entry_t;
956
957 static int
958 guid_compare(const void *arg1, const void *arg2)
959 {
960 const guid_map_entry_t *gmep1 = arg1;
961 const guid_map_entry_t *gmep2 = arg2;
962
963 if (gmep1->guid < gmep2->guid)
964 return (-1);
965 else if (gmep1->guid > gmep2->guid)
966 return (1);
967 return (0);
968 }
969
970 static void
971 free_guid_map_onexit(void *arg)
972 {
973 avl_tree_t *ca = arg;
974 void *cookie = NULL;
975 guid_map_entry_t *gmep;
976
977 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
978 dsl_dataset_rele(gmep->gme_ds, ca);
979 kmem_free(gmep, sizeof (guid_map_entry_t));
980 }
981 avl_destroy(ca);
982 kmem_free(ca, sizeof (avl_tree_t));
983 }
984
985 static void *
986 restore_read(struct restorearg *ra, int len)
987 {
988 void *rv;
989 int done = 0;
990
991 /* some things will require 8-byte alignment, so everything must */
992 ASSERT0(len % 8);
993
994 while (done < len) {
995 ssize_t resid;
996
997 ra->err = vn_rdwr(UIO_READ, ra->vp,
998 (caddr_t)ra->buf + done, len - done,
999 ra->voff, UIO_SYSSPACE, FAPPEND,
1000 RLIM64_INFINITY, CRED(), &resid);
1001
1002 if (resid == len - done)
1003 ra->err = EINVAL;
1004 ra->voff += len - done - resid;
1005 done = len - resid;
1006 if (ra->err)
1007 return (NULL);
1008 }
1009
1010 ASSERT3U(done, ==, len);
1011 rv = ra->buf;
1012 if (ra->byteswap)
1013 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
1014 else
1015 fletcher_4_incremental_native(rv, len, &ra->cksum);
1016 return (rv);
1017 }
1018
1019 noinline static void
1020 backup_byteswap(dmu_replay_record_t *drr)
1021 {
1022 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
1023 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
1024 drr->drr_type = BSWAP_32(drr->drr_type);
1025 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
1026 switch (drr->drr_type) {
1027 case DRR_BEGIN:
1028 DO64(drr_begin.drr_magic);
1029 DO64(drr_begin.drr_versioninfo);
1030 DO64(drr_begin.drr_creation_time);
1031 DO32(drr_begin.drr_type);
1032 DO32(drr_begin.drr_flags);
1033 DO64(drr_begin.drr_toguid);
1034 DO64(drr_begin.drr_fromguid);
1035 break;
1036 case DRR_OBJECT:
1037 DO64(drr_object.drr_object);
1038 /* DO64(drr_object.drr_allocation_txg); */
1039 DO32(drr_object.drr_type);
1040 DO32(drr_object.drr_bonustype);
1041 DO32(drr_object.drr_blksz);
1042 DO32(drr_object.drr_bonuslen);
1043 DO64(drr_object.drr_toguid);
1044 break;
1045 case DRR_FREEOBJECTS:
1046 DO64(drr_freeobjects.drr_firstobj);
1047 DO64(drr_freeobjects.drr_numobjs);
1048 DO64(drr_freeobjects.drr_toguid);
1049 break;
1050 case DRR_WRITE:
1051 DO64(drr_write.drr_object);
1052 DO32(drr_write.drr_type);
1053 DO64(drr_write.drr_offset);
1054 DO64(drr_write.drr_length);
1055 DO64(drr_write.drr_toguid);
1056 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1057 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1058 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1059 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1060 DO64(drr_write.drr_key.ddk_prop);
1061 break;
1062 case DRR_WRITE_BYREF:
1063 DO64(drr_write_byref.drr_object);
1064 DO64(drr_write_byref.drr_offset);
1065 DO64(drr_write_byref.drr_length);
1066 DO64(drr_write_byref.drr_toguid);
1067 DO64(drr_write_byref.drr_refguid);
1068 DO64(drr_write_byref.drr_refobject);
1069 DO64(drr_write_byref.drr_refoffset);
1070 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1071 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1072 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1073 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1074 DO64(drr_write_byref.drr_key.ddk_prop);
1075 break;
1076 case DRR_FREE:
1077 DO64(drr_free.drr_object);
1078 DO64(drr_free.drr_offset);
1079 DO64(drr_free.drr_length);
1080 DO64(drr_free.drr_toguid);
1081 break;
1082 case DRR_SPILL:
1083 DO64(drr_spill.drr_object);
1084 DO64(drr_spill.drr_length);
1085 DO64(drr_spill.drr_toguid);
1086 break;
1087 case DRR_END:
1088 DO64(drr_end.drr_checksum.zc_word[0]);
1089 DO64(drr_end.drr_checksum.zc_word[1]);
1090 DO64(drr_end.drr_checksum.zc_word[2]);
1091 DO64(drr_end.drr_checksum.zc_word[3]);
1092 DO64(drr_end.drr_toguid);
1093 break;
1094 default:
1095 break;
1096 }
1097 #undef DO64
1098 #undef DO32
1099 }
1100
1101 noinline static int
1102 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1103 {
1104 int err;
1105 dmu_tx_t *tx;
1106 void *data = NULL;
1107
1108 if (drro->drr_type == DMU_OT_NONE ||
1109 !DMU_OT_IS_VALID(drro->drr_type) ||
1110 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1111 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1112 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1113 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1114 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1115 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1116 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1117 return (EINVAL);
1118 }
1119
1120 err = dmu_object_info(os, drro->drr_object, NULL);
1121
1122 if (err != 0 && err != ENOENT)
1123 return (EINVAL);
1124
1125 if (drro->drr_bonuslen) {
1126 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1127 if (ra->err)
1128 return (ra->err);
1129 }
1130
1131 if (err == ENOENT) {
1132 /* currently free, want to be allocated */
1133 tx = dmu_tx_create(os);
1134 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1135 err = dmu_tx_assign(tx, TXG_WAIT);
1136 if (err) {
1137 dmu_tx_abort(tx);
1138 return (err);
1139 }
1140 err = dmu_object_claim(os, drro->drr_object,
1141 drro->drr_type, drro->drr_blksz,
1142 drro->drr_bonustype, drro->drr_bonuslen, tx);
1143 dmu_tx_commit(tx);
1144 } else {
1145 /* currently allocated, want to be allocated */
1146 err = dmu_object_reclaim(os, drro->drr_object,
1147 drro->drr_type, drro->drr_blksz,
1148 drro->drr_bonustype, drro->drr_bonuslen);
1149 }
1150 if (err) {
1151 return (EINVAL);
1152 }
1153
1154 tx = dmu_tx_create(os);
1155 dmu_tx_hold_bonus(tx, drro->drr_object);
1156 err = dmu_tx_assign(tx, TXG_WAIT);
1157 if (err) {
1158 dmu_tx_abort(tx);
1159 return (err);
1160 }
1161
1162 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1163 tx);
1164 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1165
1166 if (data != NULL) {
1167 dmu_buf_t *db;
1168
1169 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1170 dmu_buf_will_dirty(db, tx);
1171
1172 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1173 bcopy(data, db->db_data, drro->drr_bonuslen);
1174 if (ra->byteswap) {
1175 dmu_object_byteswap_t byteswap =
1176 DMU_OT_BYTESWAP(drro->drr_bonustype);
1177 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1178 drro->drr_bonuslen);
1179 }
1180 dmu_buf_rele(db, FTAG);
1181 }
1182 dmu_tx_commit(tx);
1183 return (0);
1184 }
1185
1186 /* ARGSUSED */
1187 noinline static int
1188 restore_freeobjects(struct restorearg *ra, objset_t *os,
1189 struct drr_freeobjects *drrfo)
1190 {
1191 uint64_t obj;
1192
1193 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1194 return (EINVAL);
1195
1196 for (obj = drrfo->drr_firstobj;
1197 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1198 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1199 int err;
1200
1201 if (dmu_object_info(os, obj, NULL) != 0)
1202 continue;
1203
1204 err = dmu_free_object(os, obj);
1205 if (err)
1206 return (err);
1207 }
1208 return (0);
1209 }
1210
1211 noinline static int
1212 restore_write(struct restorearg *ra, objset_t *os,
1213 struct drr_write *drrw)
1214 {
1215 dmu_tx_t *tx;
1216 void *data;
1217 int err;
1218
1219 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1220 !DMU_OT_IS_VALID(drrw->drr_type))
1221 return (EINVAL);
1222
1223 data = restore_read(ra, drrw->drr_length);
1224 if (data == NULL)
1225 return (ra->err);
1226
1227 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1228 return (EINVAL);
1229
1230 tx = dmu_tx_create(os);
1231
1232 dmu_tx_hold_write(tx, drrw->drr_object,
1233 drrw->drr_offset, drrw->drr_length);
1234 err = dmu_tx_assign(tx, TXG_WAIT);
1235 if (err) {
1236 dmu_tx_abort(tx);
1237 return (err);
1238 }
1239 if (ra->byteswap) {
1240 dmu_object_byteswap_t byteswap =
1241 DMU_OT_BYTESWAP(drrw->drr_type);
1242 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1243 }
1244 dmu_write(os, drrw->drr_object,
1245 drrw->drr_offset, drrw->drr_length, data, tx);
1246 dmu_tx_commit(tx);
1247 return (0);
1248 }
1249
1250 /*
1251 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1252 * streams to refer to a copy of the data that is already on the
1253 * system because it came in earlier in the stream. This function
1254 * finds the earlier copy of the data, and uses that copy instead of
1255 * data from the stream to fulfill this write.
1256 */
1257 static int
1258 restore_write_byref(struct restorearg *ra, objset_t *os,
1259 struct drr_write_byref *drrwbr)
1260 {
1261 dmu_tx_t *tx;
1262 int err;
1263 guid_map_entry_t gmesrch;
1264 guid_map_entry_t *gmep;
1265 avl_index_t where;
1266 objset_t *ref_os = NULL;
1267 dmu_buf_t *dbp;
1268
1269 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1270 return (EINVAL);
1271
1272 /*
1273 * If the GUID of the referenced dataset is different from the
1274 * GUID of the target dataset, find the referenced dataset.
1275 */
1276 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1277 gmesrch.guid = drrwbr->drr_refguid;
1278 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1279 &where)) == NULL) {
1280 return (EINVAL);
1281 }
1282 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1283 return (EINVAL);
1284 } else {
1285 ref_os = os;
1286 }
1287
1288 err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1289 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH);
1290 if (err)
1291 return (err);
1292
1293 tx = dmu_tx_create(os);
1294
1295 dmu_tx_hold_write(tx, drrwbr->drr_object,
1296 drrwbr->drr_offset, drrwbr->drr_length);
1297 err = dmu_tx_assign(tx, TXG_WAIT);
1298 if (err) {
1299 dmu_tx_abort(tx);
1300 return (err);
1301 }
1302 dmu_write(os, drrwbr->drr_object,
1303 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1304 dmu_buf_rele(dbp, FTAG);
1305 dmu_tx_commit(tx);
1306 return (0);
1307 }
1308
1309 static int
1310 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1311 {
1312 dmu_tx_t *tx;
1313 void *data;
1314 dmu_buf_t *db, *db_spill;
1315 int err;
1316
1317 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1318 drrs->drr_length > SPA_MAXBLOCKSIZE)
1319 return (EINVAL);
1320
1321 data = restore_read(ra, drrs->drr_length);
1322 if (data == NULL)
1323 return (ra->err);
1324
1325 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1326 return (EINVAL);
1327
1328 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1329 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1330 dmu_buf_rele(db, FTAG);
1331 return (err);
1332 }
1333
1334 tx = dmu_tx_create(os);
1335
1336 dmu_tx_hold_spill(tx, db->db_object);
1337
1338 err = dmu_tx_assign(tx, TXG_WAIT);
1339 if (err) {
1340 dmu_buf_rele(db, FTAG);
1341 dmu_buf_rele(db_spill, FTAG);
1342 dmu_tx_abort(tx);
1343 return (err);
1344 }
1345 dmu_buf_will_dirty(db_spill, tx);
1346
1347 if (db_spill->db_size < drrs->drr_length)
1348 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1349 drrs->drr_length, tx));
1350 bcopy(data, db_spill->db_data, drrs->drr_length);
1351
1352 dmu_buf_rele(db, FTAG);
1353 dmu_buf_rele(db_spill, FTAG);
1354
1355 dmu_tx_commit(tx);
1356 return (0);
1357 }
1358
1359 /* ARGSUSED */
1360 noinline static int
1361 restore_free(struct restorearg *ra, objset_t *os,
1362 struct drr_free *drrf)
1363 {
1364 int err;
1365
1366 if (drrf->drr_length != -1ULL &&
1367 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1368 return (EINVAL);
1369
1370 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1371 return (EINVAL);
1372
1373 err = dmu_free_long_range(os, drrf->drr_object,
1374 drrf->drr_offset, drrf->drr_length);
1375 return (err);
1376 }
1377
1378 /*
1379 * NB: callers *must* call dmu_recv_end() if this succeeds.
1380 */
1381 int
1382 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1383 int cleanup_fd, uint64_t *action_handlep)
1384 {
1385 struct restorearg ra = { 0 };
1386 dmu_replay_record_t *drr;
1387 objset_t *os;
1388 zio_cksum_t pcksum;
1389 int featureflags;
1390
1391 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1392 ra.byteswap = TRUE;
1393
1394 {
1395 /* compute checksum of drr_begin record */
1396 dmu_replay_record_t *drr;
1397 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1398
1399 drr->drr_type = DRR_BEGIN;
1400 drr->drr_u.drr_begin = *drc->drc_drrb;
1401 if (ra.byteswap) {
1402 fletcher_4_incremental_byteswap(drr,
1403 sizeof (dmu_replay_record_t), &ra.cksum);
1404 } else {
1405 fletcher_4_incremental_native(drr,
1406 sizeof (dmu_replay_record_t), &ra.cksum);
1407 }
1408 kmem_free(drr, sizeof (dmu_replay_record_t));
1409 }
1410
1411 if (ra.byteswap) {
1412 struct drr_begin *drrb = drc->drc_drrb;
1413 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1414 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1415 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1416 drrb->drr_type = BSWAP_32(drrb->drr_type);
1417 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1418 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1419 }
1420
1421 ra.vp = vp;
1422 ra.voff = *voffp;
1423 ra.bufsize = 1<<20;
1424 ra.buf = vmem_alloc(ra.bufsize, KM_SLEEP);
1425
1426 /* these were verified in dmu_recv_begin */
1427 ASSERT(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo) ==
1428 DMU_SUBSTREAM);
1429 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1430
1431 /*
1432 * Open the objset we are modifying.
1433 */
1434 VERIFY(dmu_objset_from_ds(drc->drc_real_ds, &os) == 0);
1435
1436 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1437
1438 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1439
1440 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1441 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1442 minor_t minor;
1443
1444 if (cleanup_fd == -1) {
1445 ra.err = EBADF;
1446 goto out;
1447 }
1448 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1449 if (ra.err) {
1450 cleanup_fd = -1;
1451 goto out;
1452 }
1453
1454 if (*action_handlep == 0) {
1455 ra.guid_to_ds_map =
1456 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1457 avl_create(ra.guid_to_ds_map, guid_compare,
1458 sizeof (guid_map_entry_t),
1459 offsetof(guid_map_entry_t, avlnode));
1460 ra.err = zfs_onexit_add_cb(minor,
1461 free_guid_map_onexit, ra.guid_to_ds_map,
1462 action_handlep);
1463 if (ra.err)
1464 goto out;
1465 } else {
1466 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1467 (void **)&ra.guid_to_ds_map);
1468 if (ra.err)
1469 goto out;
1470 }
1471
1472 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1473 }
1474
1475 /*
1476 * Read records and process them.
1477 */
1478 pcksum = ra.cksum;
1479 while (ra.err == 0 &&
1480 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1481 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1482 ra.err = EINTR;
1483 goto out;
1484 }
1485
1486 if (ra.byteswap)
1487 backup_byteswap(drr);
1488
1489 switch (drr->drr_type) {
1490 case DRR_OBJECT:
1491 {
1492 /*
1493 * We need to make a copy of the record header,
1494 * because restore_{object,write} may need to
1495 * restore_read(), which will invalidate drr.
1496 */
1497 struct drr_object drro = drr->drr_u.drr_object;
1498 ra.err = restore_object(&ra, os, &drro);
1499 break;
1500 }
1501 case DRR_FREEOBJECTS:
1502 {
1503 struct drr_freeobjects drrfo =
1504 drr->drr_u.drr_freeobjects;
1505 ra.err = restore_freeobjects(&ra, os, &drrfo);
1506 break;
1507 }
1508 case DRR_WRITE:
1509 {
1510 struct drr_write drrw = drr->drr_u.drr_write;
1511 ra.err = restore_write(&ra, os, &drrw);
1512 break;
1513 }
1514 case DRR_WRITE_BYREF:
1515 {
1516 struct drr_write_byref drrwbr =
1517 drr->drr_u.drr_write_byref;
1518 ra.err = restore_write_byref(&ra, os, &drrwbr);
1519 break;
1520 }
1521 case DRR_FREE:
1522 {
1523 struct drr_free drrf = drr->drr_u.drr_free;
1524 ra.err = restore_free(&ra, os, &drrf);
1525 break;
1526 }
1527 case DRR_END:
1528 {
1529 struct drr_end drre = drr->drr_u.drr_end;
1530 /*
1531 * We compare against the *previous* checksum
1532 * value, because the stored checksum is of
1533 * everything before the DRR_END record.
1534 */
1535 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1536 ra.err = ECKSUM;
1537 goto out;
1538 }
1539 case DRR_SPILL:
1540 {
1541 struct drr_spill drrs = drr->drr_u.drr_spill;
1542 ra.err = restore_spill(&ra, os, &drrs);
1543 break;
1544 }
1545 default:
1546 ra.err = EINVAL;
1547 goto out;
1548 }
1549 pcksum = ra.cksum;
1550 }
1551 ASSERT(ra.err != 0);
1552
1553 out:
1554 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1555 zfs_onexit_fd_rele(cleanup_fd);
1556
1557 if (ra.err != 0) {
1558 /*
1559 * destroy what we created, so we don't leave it in the
1560 * inconsistent restoring state.
1561 */
1562 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1563
1564 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1565 B_FALSE);
1566 if (drc->drc_real_ds != drc->drc_logical_ds) {
1567 mutex_exit(&drc->drc_logical_ds->ds_recvlock);
1568 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
1569 }
1570 }
1571
1572 vmem_free(ra.buf, ra.bufsize);
1573 *voffp = ra.voff;
1574 return (ra.err);
1575 }
1576
1577 struct recvendsyncarg {
1578 char *tosnap;
1579 uint64_t creation_time;
1580 uint64_t toguid;
1581 };
1582
1583 static int
1584 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1585 {
1586 dsl_dataset_t *ds = arg1;
1587 struct recvendsyncarg *resa = arg2;
1588
1589 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1590 }
1591
1592 static void
1593 recv_end_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1594 {
1595 dsl_dataset_t *ds = arg1;
1596 struct recvendsyncarg *resa = arg2;
1597
1598 dsl_dataset_snapshot_sync(ds, resa->tosnap, tx);
1599
1600 /* set snapshot's creation time and guid */
1601 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1602 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1603 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1604 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1605
1606 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1607 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1608 spa_history_log_internal_ds(ds, "finished receiving", tx, "");
1609 }
1610
1611 static int
1612 add_ds_to_guidmap(avl_tree_t *guid_map, dsl_dataset_t *ds)
1613 {
1614 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1615 uint64_t snapobj = ds->ds_phys->ds_prev_snap_obj;
1616 dsl_dataset_t *snapds;
1617 guid_map_entry_t *gmep;
1618 int err;
1619
1620 ASSERT(guid_map != NULL);
1621
1622 rw_enter(&dp->dp_config_rwlock, RW_READER);
1623 err = dsl_dataset_hold_obj(dp, snapobj, guid_map, &snapds);
1624 if (err == 0) {
1625 gmep = kmem_alloc(sizeof (guid_map_entry_t), KM_SLEEP);
1626 gmep->guid = snapds->ds_phys->ds_guid;
1627 gmep->gme_ds = snapds;
1628 avl_add(guid_map, gmep);
1629 }
1630
1631 rw_exit(&dp->dp_config_rwlock);
1632 return (err);
1633 }
1634
1635 static int
1636 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1637 {
1638 struct recvendsyncarg resa;
1639 dsl_dataset_t *ds = drc->drc_logical_ds;
1640 int err, myerr;
1641
1642 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1643 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1644 drc->drc_force);
1645 if (err)
1646 goto out;
1647 } else {
1648 mutex_exit(&ds->ds_recvlock);
1649 dsl_dataset_rele(ds, dmu_recv_tag);
1650 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1651 B_FALSE);
1652 return (EBUSY);
1653 }
1654
1655 resa.creation_time = drc->drc_drrb->drr_creation_time;
1656 resa.toguid = drc->drc_drrb->drr_toguid;
1657 resa.tosnap = drc->drc_tosnap;
1658
1659 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1660 recv_end_check, recv_end_sync, ds, &resa, 3);
1661 if (err) {
1662 /* swap back */
1663 (void) dsl_dataset_clone_swap(drc->drc_real_ds, ds, B_TRUE);
1664 }
1665
1666 out:
1667 mutex_exit(&ds->ds_recvlock);
1668 if (err == 0 && drc->drc_guid_to_ds_map != NULL)
1669 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1670 dsl_dataset_disown(ds, dmu_recv_tag);
1671 myerr = dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, B_FALSE);
1672 ASSERT0(myerr);
1673 return (err);
1674 }
1675
1676 static int
1677 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1678 {
1679 struct recvendsyncarg resa;
1680 dsl_dataset_t *ds = drc->drc_logical_ds;
1681 int err;
1682
1683 /*
1684 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1685 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1686 * can close it.
1687 */
1688 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1689
1690 resa.creation_time = drc->drc_drrb->drr_creation_time;
1691 resa.toguid = drc->drc_drrb->drr_toguid;
1692 resa.tosnap = drc->drc_tosnap;
1693
1694 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1695 recv_end_check, recv_end_sync, ds, &resa, 3);
1696 if (err) {
1697 /* clean up the fs we just recv'd into */
1698 (void) dsl_dataset_destroy(ds, dmu_recv_tag, B_FALSE);
1699 } else {
1700 if (drc->drc_guid_to_ds_map != NULL)
1701 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1702 /* release the hold from dmu_recv_begin */
1703 dsl_dataset_disown(ds, dmu_recv_tag);
1704 }
1705 return (err);
1706 }
1707
1708 int
1709 dmu_recv_end(dmu_recv_cookie_t *drc)
1710 {
1711 if (drc->drc_logical_ds != drc->drc_real_ds)
1712 return (dmu_recv_existing_end(drc));
1713 else
1714 return (dmu_recv_new_end(drc));
1715 }
ZFS On Linux mirror for PVE