Print this page
2619 asynchronous destruction of ZFS file systems
2747 SPA versioning with zfs feature flags
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <gwilson@delphix.com>
Reviewed by: Richard Lowe <richlowe@richlowe.net>
Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com>
Approved by: Dan McDonald <danmcd@nexenta.com>
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/fs/zfs/dmu_send.c
+++ new/usr/src/uts/common/fs/zfs/dmu_send.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
|
↓ open down ↓ |
12 lines elided |
↑ open up ↑ |
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 - * Copyright (c) 2011 by Delphix. All rights reserved.
24 - */
25 -/*
26 23 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
27 - * Copyright (c) 2011 by Delphix. All rights reserved.
24 + * Copyright (c) 2012 by Delphix. All rights reserved.
28 25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
29 26 */
30 27
31 28 #include <sys/dmu.h>
32 29 #include <sys/dmu_impl.h>
33 30 #include <sys/dmu_tx.h>
34 31 #include <sys/dbuf.h>
35 32 #include <sys/dnode.h>
36 33 #include <sys/zfs_context.h>
37 34 #include <sys/dmu_objset.h>
38 35 #include <sys/dmu_traverse.h>
39 36 #include <sys/dsl_dataset.h>
40 37 #include <sys/dsl_dir.h>
41 38 #include <sys/dsl_prop.h>
42 39 #include <sys/dsl_pool.h>
43 40 #include <sys/dsl_synctask.h>
44 41 #include <sys/zfs_ioctl.h>
45 42 #include <sys/zap.h>
46 43 #include <sys/zio_checksum.h>
47 44 #include <sys/zfs_znode.h>
48 45 #include <zfs_fletcher.h>
49 46 #include <sys/avl.h>
50 47 #include <sys/ddt.h>
51 48 #include <sys/zfs_onexit.h>
52 49
53 50 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
54 51 int zfs_send_corrupt_data = B_FALSE;
55 52
56 53 static char *dmu_recv_tag = "dmu_recv_tag";
57 54
58 55 static int
59 56 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
60 57 {
61 58 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
62 59 ssize_t resid; /* have to get resid to get detailed errno */
63 60 ASSERT3U(len % 8, ==, 0);
64 61
65 62 fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
66 63 dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
67 64 (caddr_t)buf, len,
68 65 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
69 66
70 67 mutex_enter(&ds->ds_sendstream_lock);
71 68 *dsp->dsa_off += len;
72 69 mutex_exit(&ds->ds_sendstream_lock);
73 70
74 71 return (dsp->dsa_err);
75 72 }
76 73
77 74 static int
78 75 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
79 76 uint64_t length)
80 77 {
81 78 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
82 79
83 80 /*
84 81 * If there is a pending op, but it's not PENDING_FREE, push it out,
85 82 * since free block aggregation can only be done for blocks of the
86 83 * same type (i.e., DRR_FREE records can only be aggregated with
87 84 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
88 85 * aggregated with other DRR_FREEOBJECTS records.
89 86 */
90 87 if (dsp->dsa_pending_op != PENDING_NONE &&
91 88 dsp->dsa_pending_op != PENDING_FREE) {
92 89 if (dump_bytes(dsp, dsp->dsa_drr,
93 90 sizeof (dmu_replay_record_t)) != 0)
94 91 return (EINTR);
95 92 dsp->dsa_pending_op = PENDING_NONE;
96 93 }
97 94
98 95 if (dsp->dsa_pending_op == PENDING_FREE) {
99 96 /*
100 97 * There should never be a PENDING_FREE if length is -1
101 98 * (because dump_dnode is the only place where this
102 99 * function is called with a -1, and only after flushing
103 100 * any pending record).
104 101 */
105 102 ASSERT(length != -1ULL);
106 103 /*
107 104 * Check to see whether this free block can be aggregated
108 105 * with pending one.
109 106 */
110 107 if (drrf->drr_object == object && drrf->drr_offset +
111 108 drrf->drr_length == offset) {
112 109 drrf->drr_length += length;
113 110 return (0);
114 111 } else {
115 112 /* not a continuation. Push out pending record */
116 113 if (dump_bytes(dsp, dsp->dsa_drr,
117 114 sizeof (dmu_replay_record_t)) != 0)
118 115 return (EINTR);
119 116 dsp->dsa_pending_op = PENDING_NONE;
120 117 }
121 118 }
122 119 /* create a FREE record and make it pending */
123 120 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
124 121 dsp->dsa_drr->drr_type = DRR_FREE;
125 122 drrf->drr_object = object;
126 123 drrf->drr_offset = offset;
127 124 drrf->drr_length = length;
128 125 drrf->drr_toguid = dsp->dsa_toguid;
129 126 if (length == -1ULL) {
130 127 if (dump_bytes(dsp, dsp->dsa_drr,
131 128 sizeof (dmu_replay_record_t)) != 0)
132 129 return (EINTR);
133 130 } else {
134 131 dsp->dsa_pending_op = PENDING_FREE;
135 132 }
136 133
137 134 return (0);
138 135 }
139 136
140 137 static int
141 138 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
142 139 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
143 140 {
144 141 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
145 142
146 143
147 144 /*
148 145 * If there is any kind of pending aggregation (currently either
149 146 * a grouping of free objects or free blocks), push it out to
150 147 * the stream, since aggregation can't be done across operations
151 148 * of different types.
152 149 */
153 150 if (dsp->dsa_pending_op != PENDING_NONE) {
154 151 if (dump_bytes(dsp, dsp->dsa_drr,
155 152 sizeof (dmu_replay_record_t)) != 0)
156 153 return (EINTR);
157 154 dsp->dsa_pending_op = PENDING_NONE;
158 155 }
159 156 /* write a DATA record */
160 157 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
161 158 dsp->dsa_drr->drr_type = DRR_WRITE;
162 159 drrw->drr_object = object;
163 160 drrw->drr_type = type;
164 161 drrw->drr_offset = offset;
165 162 drrw->drr_length = blksz;
166 163 drrw->drr_toguid = dsp->dsa_toguid;
167 164 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
168 165 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
169 166 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
170 167 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
171 168 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
172 169 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
173 170 drrw->drr_key.ddk_cksum = bp->blk_cksum;
174 171
175 172 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
176 173 return (EINTR);
177 174 if (dump_bytes(dsp, data, blksz) != 0)
178 175 return (EINTR);
179 176 return (0);
180 177 }
181 178
182 179 static int
183 180 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
184 181 {
185 182 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
186 183
187 184 if (dsp->dsa_pending_op != PENDING_NONE) {
188 185 if (dump_bytes(dsp, dsp->dsa_drr,
189 186 sizeof (dmu_replay_record_t)) != 0)
190 187 return (EINTR);
191 188 dsp->dsa_pending_op = PENDING_NONE;
192 189 }
193 190
194 191 /* write a SPILL record */
195 192 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
196 193 dsp->dsa_drr->drr_type = DRR_SPILL;
197 194 drrs->drr_object = object;
198 195 drrs->drr_length = blksz;
199 196 drrs->drr_toguid = dsp->dsa_toguid;
200 197
201 198 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
202 199 return (EINTR);
203 200 if (dump_bytes(dsp, data, blksz))
204 201 return (EINTR);
205 202 return (0);
206 203 }
207 204
208 205 static int
209 206 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
210 207 {
211 208 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
212 209
213 210 /*
214 211 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
215 212 * push it out, since free block aggregation can only be done for
216 213 * blocks of the same type (i.e., DRR_FREE records can only be
217 214 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
218 215 * can only be aggregated with other DRR_FREEOBJECTS records.
219 216 */
220 217 if (dsp->dsa_pending_op != PENDING_NONE &&
221 218 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
222 219 if (dump_bytes(dsp, dsp->dsa_drr,
223 220 sizeof (dmu_replay_record_t)) != 0)
224 221 return (EINTR);
225 222 dsp->dsa_pending_op = PENDING_NONE;
226 223 }
227 224 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
228 225 /*
229 226 * See whether this free object array can be aggregated
230 227 * with pending one
231 228 */
232 229 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
233 230 drrfo->drr_numobjs += numobjs;
234 231 return (0);
235 232 } else {
236 233 /* can't be aggregated. Push out pending record */
237 234 if (dump_bytes(dsp, dsp->dsa_drr,
238 235 sizeof (dmu_replay_record_t)) != 0)
239 236 return (EINTR);
240 237 dsp->dsa_pending_op = PENDING_NONE;
241 238 }
242 239 }
243 240
244 241 /* write a FREEOBJECTS record */
245 242 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
246 243 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
247 244 drrfo->drr_firstobj = firstobj;
248 245 drrfo->drr_numobjs = numobjs;
249 246 drrfo->drr_toguid = dsp->dsa_toguid;
250 247
251 248 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
252 249
253 250 return (0);
254 251 }
255 252
256 253 static int
257 254 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
258 255 {
259 256 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
260 257
261 258 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
262 259 return (dump_freeobjects(dsp, object, 1));
263 260
264 261 if (dsp->dsa_pending_op != PENDING_NONE) {
265 262 if (dump_bytes(dsp, dsp->dsa_drr,
266 263 sizeof (dmu_replay_record_t)) != 0)
267 264 return (EINTR);
268 265 dsp->dsa_pending_op = PENDING_NONE;
269 266 }
270 267
271 268 /* write an OBJECT record */
272 269 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
273 270 dsp->dsa_drr->drr_type = DRR_OBJECT;
274 271 drro->drr_object = object;
275 272 drro->drr_type = dnp->dn_type;
276 273 drro->drr_bonustype = dnp->dn_bonustype;
277 274 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
278 275 drro->drr_bonuslen = dnp->dn_bonuslen;
279 276 drro->drr_checksumtype = dnp->dn_checksum;
280 277 drro->drr_compress = dnp->dn_compress;
281 278 drro->drr_toguid = dsp->dsa_toguid;
282 279
283 280 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
284 281 return (EINTR);
285 282
286 283 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
287 284 return (EINTR);
288 285
289 286 /* free anything past the end of the file */
290 287 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
291 288 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
292 289 return (EINTR);
293 290 if (dsp->dsa_err)
294 291 return (EINTR);
295 292 return (0);
296 293 }
297 294
298 295 #define BP_SPAN(dnp, level) \
299 296 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
300 297 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
301 298
302 299 /* ARGSUSED */
303 300 static int
304 301 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
305 302 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
306 303 {
307 304 dmu_sendarg_t *dsp = arg;
308 305 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
309 306 int err = 0;
310 307
311 308 if (issig(JUSTLOOKING) && issig(FORREAL))
312 309 return (EINTR);
313 310
314 311 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
315 312 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
316 313 return (0);
317 314 } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) {
318 315 uint64_t span = BP_SPAN(dnp, zb->zb_level);
319 316 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
320 317 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
321 318 } else if (bp == NULL) {
322 319 uint64_t span = BP_SPAN(dnp, zb->zb_level);
323 320 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
324 321 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
325 322 return (0);
326 323 } else if (type == DMU_OT_DNODE) {
327 324 dnode_phys_t *blk;
328 325 int i;
329 326 int blksz = BP_GET_LSIZE(bp);
330 327 uint32_t aflags = ARC_WAIT;
331 328 arc_buf_t *abuf;
332 329
333 330 if (dsl_read(NULL, spa, bp, pbuf,
334 331 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
335 332 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
336 333 return (EIO);
337 334
338 335 blk = abuf->b_data;
339 336 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
340 337 uint64_t dnobj = (zb->zb_blkid <<
341 338 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
342 339 err = dump_dnode(dsp, dnobj, blk+i);
343 340 if (err)
344 341 break;
345 342 }
346 343 (void) arc_buf_remove_ref(abuf, &abuf);
347 344 } else if (type == DMU_OT_SA) {
348 345 uint32_t aflags = ARC_WAIT;
349 346 arc_buf_t *abuf;
350 347 int blksz = BP_GET_LSIZE(bp);
351 348
352 349 if (arc_read_nolock(NULL, spa, bp,
353 350 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
354 351 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
355 352 return (EIO);
356 353
357 354 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
358 355 (void) arc_buf_remove_ref(abuf, &abuf);
359 356 } else { /* it's a level-0 block of a regular object */
360 357 uint32_t aflags = ARC_WAIT;
361 358 arc_buf_t *abuf;
362 359 int blksz = BP_GET_LSIZE(bp);
363 360
364 361 if (dsl_read(NULL, spa, bp, pbuf,
365 362 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
366 363 ZIO_FLAG_CANFAIL, &aflags, zb) != 0) {
367 364 if (zfs_send_corrupt_data) {
368 365 /* Send a block filled with 0x"zfs badd bloc" */
369 366 abuf = arc_buf_alloc(spa, blksz, &abuf,
370 367 ARC_BUFC_DATA);
371 368 uint64_t *ptr;
372 369 for (ptr = abuf->b_data;
373 370 (char *)ptr < (char *)abuf->b_data + blksz;
374 371 ptr++)
375 372 *ptr = 0x2f5baddb10c;
376 373 } else {
377 374 return (EIO);
378 375 }
379 376 }
380 377
381 378 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
382 379 blksz, bp, abuf->b_data);
383 380 (void) arc_buf_remove_ref(abuf, &abuf);
384 381 }
385 382
386 383 ASSERT(err == 0 || err == EINTR);
387 384 return (err);
388 385 }
389 386
390 387 int
391 388 dmu_send(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
392 389 int outfd, vnode_t *vp, offset_t *off)
393 390 {
394 391 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
395 392 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
396 393 dmu_replay_record_t *drr;
397 394 dmu_sendarg_t *dsp;
398 395 int err;
399 396 uint64_t fromtxg = 0;
400 397
401 398 /* tosnap must be a snapshot */
402 399 if (ds->ds_phys->ds_next_snap_obj == 0)
403 400 return (EINVAL);
404 401
405 402 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
406 403 if (fromds && (ds->ds_dir != fromds->ds_dir ||
407 404 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
408 405 return (EXDEV);
409 406
410 407 if (fromorigin) {
411 408 dsl_pool_t *dp = ds->ds_dir->dd_pool;
412 409
413 410 if (fromsnap)
414 411 return (EINVAL);
415 412
416 413 if (dsl_dir_is_clone(ds->ds_dir)) {
417 414 rw_enter(&dp->dp_config_rwlock, RW_READER);
418 415 err = dsl_dataset_hold_obj(dp,
419 416 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
420 417 rw_exit(&dp->dp_config_rwlock);
421 418 if (err)
422 419 return (err);
423 420 } else {
424 421 fromorigin = B_FALSE;
425 422 }
426 423 }
427 424
428 425
429 426 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
430 427 drr->drr_type = DRR_BEGIN;
431 428 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
432 429 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
433 430 DMU_SUBSTREAM);
434 431
435 432 #ifdef _KERNEL
436 433 if (dmu_objset_type(tosnap) == DMU_OST_ZFS) {
437 434 uint64_t version;
438 435 if (zfs_get_zplprop(tosnap, ZFS_PROP_VERSION, &version) != 0) {
439 436 kmem_free(drr, sizeof (dmu_replay_record_t));
440 437 return (EINVAL);
441 438 }
442 439 if (version == ZPL_VERSION_SA) {
443 440 DMU_SET_FEATUREFLAGS(
444 441 drr->drr_u.drr_begin.drr_versioninfo,
445 442 DMU_BACKUP_FEATURE_SA_SPILL);
446 443 }
447 444 }
448 445 #endif
449 446
450 447 drr->drr_u.drr_begin.drr_creation_time =
451 448 ds->ds_phys->ds_creation_time;
452 449 drr->drr_u.drr_begin.drr_type = tosnap->os_phys->os_type;
453 450 if (fromorigin)
454 451 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
455 452 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
456 453 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
457 454 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
458 455
459 456 if (fromds)
460 457 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
461 458 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
462 459
463 460 if (fromds)
464 461 fromtxg = fromds->ds_phys->ds_creation_txg;
465 462 if (fromorigin)
466 463 dsl_dataset_rele(fromds, FTAG);
467 464
468 465 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
469 466
470 467 dsp->dsa_drr = drr;
471 468 dsp->dsa_vp = vp;
472 469 dsp->dsa_outfd = outfd;
473 470 dsp->dsa_proc = curproc;
474 471 dsp->dsa_os = tosnap;
475 472 dsp->dsa_off = off;
476 473 dsp->dsa_toguid = ds->ds_phys->ds_guid;
477 474 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
478 475 dsp->dsa_pending_op = PENDING_NONE;
479 476
480 477 mutex_enter(&ds->ds_sendstream_lock);
481 478 list_insert_head(&ds->ds_sendstreams, dsp);
482 479 mutex_exit(&ds->ds_sendstream_lock);
483 480
484 481 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
485 482 err = dsp->dsa_err;
486 483 goto out;
487 484 }
488 485
489 486 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
490 487 backup_cb, dsp);
491 488
492 489 if (dsp->dsa_pending_op != PENDING_NONE)
493 490 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
494 491 err = EINTR;
495 492
496 493 if (err) {
497 494 if (err == EINTR && dsp->dsa_err)
498 495 err = dsp->dsa_err;
499 496 goto out;
500 497 }
501 498
502 499 bzero(drr, sizeof (dmu_replay_record_t));
503 500 drr->drr_type = DRR_END;
504 501 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
505 502 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
506 503
507 504 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
508 505 err = dsp->dsa_err;
509 506 goto out;
510 507 }
511 508
512 509 out:
513 510 mutex_enter(&ds->ds_sendstream_lock);
514 511 list_remove(&ds->ds_sendstreams, dsp);
515 512 mutex_exit(&ds->ds_sendstream_lock);
516 513
517 514 kmem_free(drr, sizeof (dmu_replay_record_t));
518 515 kmem_free(dsp, sizeof (dmu_sendarg_t));
519 516
520 517 return (err);
521 518 }
522 519
523 520 int
524 521 dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, boolean_t fromorigin,
525 522 uint64_t *sizep)
526 523 {
527 524 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
528 525 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
529 526 dsl_pool_t *dp = ds->ds_dir->dd_pool;
530 527 int err;
531 528 uint64_t size;
532 529
533 530 /* tosnap must be a snapshot */
534 531 if (ds->ds_phys->ds_next_snap_obj == 0)
535 532 return (EINVAL);
536 533
537 534 /* fromsnap must be an earlier snapshot from the same fs as tosnap */
538 535 if (fromds && (ds->ds_dir != fromds->ds_dir ||
539 536 fromds->ds_phys->ds_creation_txg >= ds->ds_phys->ds_creation_txg))
540 537 return (EXDEV);
541 538
542 539 if (fromorigin) {
543 540 if (fromsnap)
544 541 return (EINVAL);
545 542
546 543 if (dsl_dir_is_clone(ds->ds_dir)) {
547 544 rw_enter(&dp->dp_config_rwlock, RW_READER);
548 545 err = dsl_dataset_hold_obj(dp,
549 546 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &fromds);
550 547 rw_exit(&dp->dp_config_rwlock);
551 548 if (err)
552 549 return (err);
553 550 } else {
554 551 fromorigin = B_FALSE;
555 552 }
556 553 }
557 554
558 555 /* Get uncompressed size estimate of changed data. */
559 556 if (fromds == NULL) {
560 557 size = ds->ds_phys->ds_uncompressed_bytes;
561 558 } else {
562 559 uint64_t used, comp;
563 560 err = dsl_dataset_space_written(fromds, ds,
564 561 &used, &comp, &size);
565 562 if (fromorigin)
566 563 dsl_dataset_rele(fromds, FTAG);
567 564 if (err)
568 565 return (err);
569 566 }
570 567
571 568 /*
572 569 * Assume that space (both on-disk and in-stream) is dominated by
573 570 * data. We will adjust for indirect blocks and the copies property,
574 571 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
575 572 */
576 573
577 574 /*
578 575 * Subtract out approximate space used by indirect blocks.
579 576 * Assume most space is used by data blocks (non-indirect, non-dnode).
580 577 * Assume all blocks are recordsize. Assume ditto blocks and
581 578 * internal fragmentation counter out compression.
582 579 *
583 580 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
584 581 * block, which we observe in practice.
585 582 */
586 583 uint64_t recordsize;
587 584 rw_enter(&dp->dp_config_rwlock, RW_READER);
588 585 err = dsl_prop_get_ds(ds, "recordsize",
589 586 sizeof (recordsize), 1, &recordsize, NULL);
590 587 rw_exit(&dp->dp_config_rwlock);
591 588 if (err)
592 589 return (err);
593 590 size -= size / recordsize * sizeof (blkptr_t);
594 591
595 592 /* Add in the space for the record associated with each block. */
596 593 size += size / recordsize * sizeof (dmu_replay_record_t);
597 594
598 595 *sizep = size;
599 596
600 597 return (0);
601 598 }
602 599
603 600 struct recvbeginsyncarg {
604 601 const char *tofs;
605 602 const char *tosnap;
606 603 dsl_dataset_t *origin;
607 604 uint64_t fromguid;
608 605 dmu_objset_type_t type;
609 606 void *tag;
610 607 boolean_t force;
611 608 uint64_t dsflags;
612 609 char clonelastname[MAXNAMELEN];
613 610 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
614 611 cred_t *cr;
615 612 };
616 613
617 614 /* ARGSUSED */
618 615 static int
619 616 recv_new_check(void *arg1, void *arg2, dmu_tx_t *tx)
620 617 {
621 618 dsl_dir_t *dd = arg1;
622 619 struct recvbeginsyncarg *rbsa = arg2;
623 620 objset_t *mos = dd->dd_pool->dp_meta_objset;
624 621 uint64_t val;
625 622 int err;
626 623
627 624 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
628 625 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
629 626
630 627 if (err != ENOENT)
631 628 return (err ? err : EEXIST);
632 629
633 630 if (rbsa->origin) {
634 631 /* make sure it's a snap in the same pool */
635 632 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
636 633 return (EXDEV);
637 634 if (!dsl_dataset_is_snapshot(rbsa->origin))
638 635 return (EINVAL);
639 636 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
640 637 return (ENODEV);
641 638 }
642 639
643 640 return (0);
644 641 }
645 642
646 643 static void
647 644 recv_new_sync(void *arg1, void *arg2, dmu_tx_t *tx)
648 645 {
649 646 dsl_dir_t *dd = arg1;
650 647 struct recvbeginsyncarg *rbsa = arg2;
651 648 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
652 649 uint64_t dsobj;
653 650
654 651 /* Create and open new dataset. */
655 652 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
656 653 rbsa->origin, flags, rbsa->cr, tx);
657 654 VERIFY(0 == dsl_dataset_own_obj(dd->dd_pool, dsobj,
658 655 B_TRUE, dmu_recv_tag, &rbsa->ds));
659 656
660 657 if (rbsa->origin == NULL) {
661 658 (void) dmu_objset_create_impl(dd->dd_pool->dp_spa,
662 659 rbsa->ds, &rbsa->ds->ds_phys->ds_bp, rbsa->type, tx);
663 660 }
664 661
665 662 spa_history_log_internal(LOG_DS_REPLAY_FULL_SYNC,
666 663 dd->dd_pool->dp_spa, tx, "dataset = %lld", dsobj);
667 664 }
668 665
669 666 /* ARGSUSED */
670 667 static int
671 668 recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
672 669 {
673 670 dsl_dataset_t *ds = arg1;
674 671 struct recvbeginsyncarg *rbsa = arg2;
675 672 int err;
676 673 uint64_t val;
677 674
678 675 /* must not have any changes since most recent snapshot */
679 676 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
680 677 return (ETXTBSY);
681 678
682 679 /* new snapshot name must not exist */
683 680 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
684 681 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
685 682 if (err == 0)
686 683 return (EEXIST);
687 684 if (err != ENOENT)
688 685 return (err);
689 686
690 687 if (rbsa->fromguid) {
691 688 /* if incremental, most recent snapshot must match fromguid */
692 689 if (ds->ds_prev == NULL)
693 690 return (ENODEV);
694 691
695 692 /*
696 693 * most recent snapshot must match fromguid, or there are no
697 694 * changes since the fromguid one
698 695 */
699 696 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid) {
700 697 uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth;
701 698 uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj;
702 699 while (obj != 0) {
703 700 dsl_dataset_t *snap;
704 701 err = dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
705 702 obj, FTAG, &snap);
706 703 if (err)
707 704 return (ENODEV);
708 705 if (snap->ds_phys->ds_creation_txg < birth) {
709 706 dsl_dataset_rele(snap, FTAG);
710 707 return (ENODEV);
711 708 }
712 709 if (snap->ds_phys->ds_guid == rbsa->fromguid) {
713 710 dsl_dataset_rele(snap, FTAG);
714 711 break; /* it's ok */
715 712 }
716 713 obj = snap->ds_phys->ds_prev_snap_obj;
717 714 dsl_dataset_rele(snap, FTAG);
718 715 }
719 716 if (obj == 0)
720 717 return (ENODEV);
721 718 }
722 719 } else {
723 720 /* if full, most recent snapshot must be $ORIGIN */
724 721 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
725 722 return (ENODEV);
726 723 }
727 724
728 725 /* temporary clone name must not exist */
729 726 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
730 727 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
731 728 rbsa->clonelastname, 8, 1, &val);
732 729 if (err == 0)
733 730 return (EEXIST);
734 731 if (err != ENOENT)
735 732 return (err);
736 733
737 734 return (0);
738 735 }
739 736
740 737 /* ARGSUSED */
741 738 static void
742 739 recv_existing_sync(void *arg1, void *arg2, dmu_tx_t *tx)
743 740 {
744 741 dsl_dataset_t *ohds = arg1;
745 742 struct recvbeginsyncarg *rbsa = arg2;
746 743 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
747 744 dsl_dataset_t *cds;
748 745 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
749 746 uint64_t dsobj;
750 747
751 748 /* create and open the temporary clone */
752 749 dsobj = dsl_dataset_create_sync(ohds->ds_dir, rbsa->clonelastname,
753 750 ohds->ds_prev, flags, rbsa->cr, tx);
754 751 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, B_TRUE, dmu_recv_tag, &cds));
755 752
756 753 /*
757 754 * If we actually created a non-clone, we need to create the
758 755 * objset in our new dataset.
759 756 */
760 757 if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds))) {
761 758 (void) dmu_objset_create_impl(dp->dp_spa,
762 759 cds, dsl_dataset_get_blkptr(cds), rbsa->type, tx);
763 760 }
764 761
765 762 rbsa->ds = cds;
766 763
767 764 spa_history_log_internal(LOG_DS_REPLAY_INC_SYNC,
768 765 dp->dp_spa, tx, "dataset = %lld", dsobj);
769 766 }
770 767
771 768 static boolean_t
772 769 dmu_recv_verify_features(dsl_dataset_t *ds, struct drr_begin *drrb)
773 770 {
774 771 int featureflags;
775 772
776 773 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
777 774
778 775 /* Verify pool version supports SA if SA_SPILL feature set */
779 776 return ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
780 777 (spa_version(dsl_dataset_get_spa(ds)) < SPA_VERSION_SA));
781 778 }
782 779
783 780 /*
784 781 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
785 782 * succeeds; otherwise we will leak the holds on the datasets.
786 783 */
787 784 int
788 785 dmu_recv_begin(char *tofs, char *tosnap, char *top_ds, struct drr_begin *drrb,
789 786 boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
790 787 {
791 788 int err = 0;
792 789 boolean_t byteswap;
793 790 struct recvbeginsyncarg rbsa = { 0 };
794 791 uint64_t versioninfo;
795 792 int flags;
796 793 dsl_dataset_t *ds;
797 794
798 795 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
799 796 byteswap = FALSE;
800 797 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
801 798 byteswap = TRUE;
802 799 else
803 800 return (EINVAL);
804 801
805 802 rbsa.tofs = tofs;
806 803 rbsa.tosnap = tosnap;
807 804 rbsa.origin = origin ? origin->os_dsl_dataset : NULL;
808 805 rbsa.fromguid = drrb->drr_fromguid;
809 806 rbsa.type = drrb->drr_type;
810 807 rbsa.tag = FTAG;
811 808 rbsa.dsflags = 0;
812 809 rbsa.cr = CRED();
813 810 versioninfo = drrb->drr_versioninfo;
814 811 flags = drrb->drr_flags;
815 812
816 813 if (byteswap) {
817 814 rbsa.type = BSWAP_32(rbsa.type);
818 815 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
819 816 versioninfo = BSWAP_64(versioninfo);
820 817 flags = BSWAP_32(flags);
821 818 }
822 819
823 820 if (DMU_GET_STREAM_HDRTYPE(versioninfo) == DMU_COMPOUNDSTREAM ||
824 821 rbsa.type >= DMU_OST_NUMTYPES ||
825 822 ((flags & DRR_FLAG_CLONE) && origin == NULL))
826 823 return (EINVAL);
827 824
828 825 if (flags & DRR_FLAG_CI_DATA)
829 826 rbsa.dsflags = DS_FLAG_CI_DATASET;
830 827
831 828 bzero(drc, sizeof (dmu_recv_cookie_t));
832 829 drc->drc_drrb = drrb;
833 830 drc->drc_tosnap = tosnap;
834 831 drc->drc_top_ds = top_ds;
835 832 drc->drc_force = force;
836 833
837 834 /*
838 835 * Process the begin in syncing context.
839 836 */
840 837
841 838 /* open the dataset we are logically receiving into */
842 839 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
843 840 if (err == 0) {
844 841 if (dmu_recv_verify_features(ds, drrb)) {
845 842 dsl_dataset_rele(ds, dmu_recv_tag);
846 843 return (ENOTSUP);
847 844 }
848 845 /* target fs already exists; recv into temp clone */
849 846
850 847 /* Can't recv a clone into an existing fs */
851 848 if (flags & DRR_FLAG_CLONE) {
852 849 dsl_dataset_rele(ds, dmu_recv_tag);
853 850 return (EINVAL);
854 851 }
855 852
856 853 /* must not have an incremental recv already in progress */
857 854 if (!mutex_tryenter(&ds->ds_recvlock)) {
858 855 dsl_dataset_rele(ds, dmu_recv_tag);
859 856 return (EBUSY);
860 857 }
861 858
862 859 /* tmp clone name is: tofs/%tosnap" */
863 860 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
864 861 "%%%s", tosnap);
865 862 rbsa.force = force;
866 863 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
867 864 recv_existing_check, recv_existing_sync, ds, &rbsa, 5);
868 865 if (err) {
869 866 mutex_exit(&ds->ds_recvlock);
870 867 dsl_dataset_rele(ds, dmu_recv_tag);
871 868 return (err);
872 869 }
873 870 drc->drc_logical_ds = ds;
874 871 drc->drc_real_ds = rbsa.ds;
875 872 } else if (err == ENOENT) {
876 873 /* target fs does not exist; must be a full backup or clone */
877 874 char *cp;
878 875
879 876 /*
880 877 * If it's a non-clone incremental, we are missing the
881 878 * target fs, so fail the recv.
882 879 */
883 880 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE))
884 881 return (ENOENT);
885 882
886 883 /* Open the parent of tofs */
887 884 cp = strrchr(tofs, '/');
888 885 *cp = '\0';
889 886 err = dsl_dataset_hold(tofs, FTAG, &ds);
890 887 *cp = '/';
891 888 if (err)
892 889 return (err);
893 890
894 891 if (dmu_recv_verify_features(ds, drrb)) {
895 892 dsl_dataset_rele(ds, FTAG);
896 893 return (ENOTSUP);
897 894 }
898 895
899 896 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
900 897 recv_new_check, recv_new_sync, ds->ds_dir, &rbsa, 5);
901 898 dsl_dataset_rele(ds, FTAG);
902 899 if (err)
903 900 return (err);
904 901 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
905 902 drc->drc_newfs = B_TRUE;
906 903 }
907 904
908 905 return (err);
909 906 }
910 907
911 908 struct restorearg {
912 909 int err;
913 910 int byteswap;
914 911 vnode_t *vp;
915 912 char *buf;
916 913 uint64_t voff;
917 914 int bufsize; /* amount of memory allocated for buf */
918 915 zio_cksum_t cksum;
919 916 avl_tree_t *guid_to_ds_map;
920 917 };
921 918
922 919 typedef struct guid_map_entry {
923 920 uint64_t guid;
924 921 dsl_dataset_t *gme_ds;
925 922 avl_node_t avlnode;
926 923 } guid_map_entry_t;
927 924
928 925 static int
929 926 guid_compare(const void *arg1, const void *arg2)
930 927 {
931 928 const guid_map_entry_t *gmep1 = arg1;
932 929 const guid_map_entry_t *gmep2 = arg2;
933 930
934 931 if (gmep1->guid < gmep2->guid)
935 932 return (-1);
936 933 else if (gmep1->guid > gmep2->guid)
937 934 return (1);
938 935 return (0);
939 936 }
940 937
941 938 static void
942 939 free_guid_map_onexit(void *arg)
943 940 {
944 941 avl_tree_t *ca = arg;
945 942 void *cookie = NULL;
946 943 guid_map_entry_t *gmep;
947 944
948 945 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
949 946 dsl_dataset_rele(gmep->gme_ds, ca);
950 947 kmem_free(gmep, sizeof (guid_map_entry_t));
951 948 }
952 949 avl_destroy(ca);
953 950 kmem_free(ca, sizeof (avl_tree_t));
954 951 }
955 952
956 953 static void *
957 954 restore_read(struct restorearg *ra, int len)
958 955 {
959 956 void *rv;
960 957 int done = 0;
961 958
962 959 /* some things will require 8-byte alignment, so everything must */
963 960 ASSERT3U(len % 8, ==, 0);
964 961
965 962 while (done < len) {
966 963 ssize_t resid;
967 964
968 965 ra->err = vn_rdwr(UIO_READ, ra->vp,
969 966 (caddr_t)ra->buf + done, len - done,
970 967 ra->voff, UIO_SYSSPACE, FAPPEND,
971 968 RLIM64_INFINITY, CRED(), &resid);
972 969
973 970 if (resid == len - done)
974 971 ra->err = EINVAL;
975 972 ra->voff += len - done - resid;
976 973 done = len - resid;
977 974 if (ra->err)
978 975 return (NULL);
979 976 }
980 977
981 978 ASSERT3U(done, ==, len);
982 979 rv = ra->buf;
983 980 if (ra->byteswap)
984 981 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
985 982 else
986 983 fletcher_4_incremental_native(rv, len, &ra->cksum);
987 984 return (rv);
988 985 }
989 986
990 987 static void
991 988 backup_byteswap(dmu_replay_record_t *drr)
992 989 {
993 990 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
994 991 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
995 992 drr->drr_type = BSWAP_32(drr->drr_type);
996 993 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
997 994 switch (drr->drr_type) {
998 995 case DRR_BEGIN:
999 996 DO64(drr_begin.drr_magic);
1000 997 DO64(drr_begin.drr_versioninfo);
1001 998 DO64(drr_begin.drr_creation_time);
1002 999 DO32(drr_begin.drr_type);
1003 1000 DO32(drr_begin.drr_flags);
1004 1001 DO64(drr_begin.drr_toguid);
1005 1002 DO64(drr_begin.drr_fromguid);
1006 1003 break;
1007 1004 case DRR_OBJECT:
1008 1005 DO64(drr_object.drr_object);
1009 1006 /* DO64(drr_object.drr_allocation_txg); */
1010 1007 DO32(drr_object.drr_type);
1011 1008 DO32(drr_object.drr_bonustype);
1012 1009 DO32(drr_object.drr_blksz);
1013 1010 DO32(drr_object.drr_bonuslen);
1014 1011 DO64(drr_object.drr_toguid);
1015 1012 break;
1016 1013 case DRR_FREEOBJECTS:
1017 1014 DO64(drr_freeobjects.drr_firstobj);
1018 1015 DO64(drr_freeobjects.drr_numobjs);
1019 1016 DO64(drr_freeobjects.drr_toguid);
1020 1017 break;
1021 1018 case DRR_WRITE:
1022 1019 DO64(drr_write.drr_object);
1023 1020 DO32(drr_write.drr_type);
1024 1021 DO64(drr_write.drr_offset);
1025 1022 DO64(drr_write.drr_length);
1026 1023 DO64(drr_write.drr_toguid);
1027 1024 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1028 1025 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1029 1026 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1030 1027 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1031 1028 DO64(drr_write.drr_key.ddk_prop);
1032 1029 break;
1033 1030 case DRR_WRITE_BYREF:
1034 1031 DO64(drr_write_byref.drr_object);
1035 1032 DO64(drr_write_byref.drr_offset);
1036 1033 DO64(drr_write_byref.drr_length);
1037 1034 DO64(drr_write_byref.drr_toguid);
1038 1035 DO64(drr_write_byref.drr_refguid);
1039 1036 DO64(drr_write_byref.drr_refobject);
1040 1037 DO64(drr_write_byref.drr_refoffset);
1041 1038 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1042 1039 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1043 1040 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1044 1041 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1045 1042 DO64(drr_write_byref.drr_key.ddk_prop);
1046 1043 break;
1047 1044 case DRR_FREE:
1048 1045 DO64(drr_free.drr_object);
1049 1046 DO64(drr_free.drr_offset);
1050 1047 DO64(drr_free.drr_length);
1051 1048 DO64(drr_free.drr_toguid);
1052 1049 break;
1053 1050 case DRR_SPILL:
1054 1051 DO64(drr_spill.drr_object);
1055 1052 DO64(drr_spill.drr_length);
1056 1053 DO64(drr_spill.drr_toguid);
1057 1054 break;
1058 1055 case DRR_END:
1059 1056 DO64(drr_end.drr_checksum.zc_word[0]);
1060 1057 DO64(drr_end.drr_checksum.zc_word[1]);
1061 1058 DO64(drr_end.drr_checksum.zc_word[2]);
1062 1059 DO64(drr_end.drr_checksum.zc_word[3]);
1063 1060 DO64(drr_end.drr_toguid);
1064 1061 break;
1065 1062 }
1066 1063 #undef DO64
1067 1064 #undef DO32
|
↓ open down ↓ |
1030 lines elided |
↑ open up ↑ |
1068 1065 }
1069 1066
1070 1067 static int
1071 1068 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1072 1069 {
1073 1070 int err;
1074 1071 dmu_tx_t *tx;
1075 1072 void *data = NULL;
1076 1073
1077 1074 if (drro->drr_type == DMU_OT_NONE ||
1078 - drro->drr_type >= DMU_OT_NUMTYPES ||
1079 - drro->drr_bonustype >= DMU_OT_NUMTYPES ||
1075 + !DMU_OT_IS_VALID(drro->drr_type) ||
1076 + !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1080 1077 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1081 1078 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1082 1079 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1083 1080 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1084 1081 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1085 1082 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1086 1083 return (EINVAL);
1087 1084 }
1088 1085
1089 1086 err = dmu_object_info(os, drro->drr_object, NULL);
1090 1087
1091 1088 if (err != 0 && err != ENOENT)
1092 1089 return (EINVAL);
1093 1090
1094 1091 if (drro->drr_bonuslen) {
1095 1092 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1096 1093 if (ra->err)
1097 1094 return (ra->err);
1098 1095 }
1099 1096
1100 1097 if (err == ENOENT) {
1101 1098 /* currently free, want to be allocated */
1102 1099 tx = dmu_tx_create(os);
1103 1100 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1104 1101 err = dmu_tx_assign(tx, TXG_WAIT);
1105 1102 if (err) {
1106 1103 dmu_tx_abort(tx);
1107 1104 return (err);
1108 1105 }
1109 1106 err = dmu_object_claim(os, drro->drr_object,
1110 1107 drro->drr_type, drro->drr_blksz,
1111 1108 drro->drr_bonustype, drro->drr_bonuslen, tx);
1112 1109 dmu_tx_commit(tx);
1113 1110 } else {
1114 1111 /* currently allocated, want to be allocated */
1115 1112 err = dmu_object_reclaim(os, drro->drr_object,
1116 1113 drro->drr_type, drro->drr_blksz,
1117 1114 drro->drr_bonustype, drro->drr_bonuslen);
1118 1115 }
1119 1116 if (err) {
1120 1117 return (EINVAL);
1121 1118 }
1122 1119
1123 1120 tx = dmu_tx_create(os);
1124 1121 dmu_tx_hold_bonus(tx, drro->drr_object);
1125 1122 err = dmu_tx_assign(tx, TXG_WAIT);
1126 1123 if (err) {
1127 1124 dmu_tx_abort(tx);
1128 1125 return (err);
1129 1126 }
1130 1127
1131 1128 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1132 1129 tx);
1133 1130 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
|
↓ open down ↓ |
44 lines elided |
↑ open up ↑ |
1134 1131
1135 1132 if (data != NULL) {
1136 1133 dmu_buf_t *db;
1137 1134
1138 1135 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1139 1136 dmu_buf_will_dirty(db, tx);
1140 1137
1141 1138 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1142 1139 bcopy(data, db->db_data, drro->drr_bonuslen);
1143 1140 if (ra->byteswap) {
1144 - dmu_ot[drro->drr_bonustype].ot_byteswap(db->db_data,
1141 + dmu_object_byteswap_t byteswap =
1142 + DMU_OT_BYTESWAP(drro->drr_bonustype);
1143 + dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1145 1144 drro->drr_bonuslen);
1146 1145 }
1147 1146 dmu_buf_rele(db, FTAG);
1148 1147 }
1149 1148 dmu_tx_commit(tx);
1150 1149 return (0);
1151 1150 }
1152 1151
1153 1152 /* ARGSUSED */
1154 1153 static int
1155 1154 restore_freeobjects(struct restorearg *ra, objset_t *os,
1156 1155 struct drr_freeobjects *drrfo)
1157 1156 {
1158 1157 uint64_t obj;
1159 1158
1160 1159 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1161 1160 return (EINVAL);
1162 1161
1163 1162 for (obj = drrfo->drr_firstobj;
1164 1163 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1165 1164 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1166 1165 int err;
1167 1166
1168 1167 if (dmu_object_info(os, obj, NULL) != 0)
1169 1168 continue;
1170 1169
1171 1170 err = dmu_free_object(os, obj);
1172 1171 if (err)
1173 1172 return (err);
1174 1173 }
1175 1174 return (0);
1176 1175 }
|
↓ open down ↓ |
22 lines elided |
↑ open up ↑ |
1177 1176
1178 1177 static int
1179 1178 restore_write(struct restorearg *ra, objset_t *os,
1180 1179 struct drr_write *drrw)
1181 1180 {
1182 1181 dmu_tx_t *tx;
1183 1182 void *data;
1184 1183 int err;
1185 1184
1186 1185 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1187 - drrw->drr_type >= DMU_OT_NUMTYPES)
1186 + !DMU_OT_IS_VALID(drrw->drr_type))
1188 1187 return (EINVAL);
1189 1188
1190 1189 data = restore_read(ra, drrw->drr_length);
1191 1190 if (data == NULL)
1192 1191 return (ra->err);
1193 1192
1194 1193 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1195 1194 return (EINVAL);
1196 1195
1197 1196 tx = dmu_tx_create(os);
1198 1197
1199 1198 dmu_tx_hold_write(tx, drrw->drr_object,
1200 1199 drrw->drr_offset, drrw->drr_length);
1201 1200 err = dmu_tx_assign(tx, TXG_WAIT);
1202 1201 if (err) {
1203 1202 dmu_tx_abort(tx);
1204 1203 return (err);
1205 1204 }
1206 - if (ra->byteswap)
1207 - dmu_ot[drrw->drr_type].ot_byteswap(data, drrw->drr_length);
1205 + if (ra->byteswap) {
1206 + dmu_object_byteswap_t byteswap =
1207 + DMU_OT_BYTESWAP(drrw->drr_type);
1208 + dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1209 + }
1208 1210 dmu_write(os, drrw->drr_object,
1209 1211 drrw->drr_offset, drrw->drr_length, data, tx);
1210 1212 dmu_tx_commit(tx);
1211 1213 return (0);
1212 1214 }
1213 1215
1214 1216 /*
1215 1217 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1216 1218 * streams to refer to a copy of the data that is already on the
1217 1219 * system because it came in earlier in the stream. This function
1218 1220 * finds the earlier copy of the data, and uses that copy instead of
1219 1221 * data from the stream to fulfill this write.
1220 1222 */
1221 1223 static int
1222 1224 restore_write_byref(struct restorearg *ra, objset_t *os,
1223 1225 struct drr_write_byref *drrwbr)
1224 1226 {
1225 1227 dmu_tx_t *tx;
1226 1228 int err;
1227 1229 guid_map_entry_t gmesrch;
1228 1230 guid_map_entry_t *gmep;
1229 1231 avl_index_t where;
1230 1232 objset_t *ref_os = NULL;
1231 1233 dmu_buf_t *dbp;
1232 1234
1233 1235 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1234 1236 return (EINVAL);
1235 1237
1236 1238 /*
1237 1239 * If the GUID of the referenced dataset is different from the
1238 1240 * GUID of the target dataset, find the referenced dataset.
1239 1241 */
1240 1242 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1241 1243 gmesrch.guid = drrwbr->drr_refguid;
1242 1244 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1243 1245 &where)) == NULL) {
1244 1246 return (EINVAL);
1245 1247 }
1246 1248 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1247 1249 return (EINVAL);
1248 1250 } else {
1249 1251 ref_os = os;
1250 1252 }
1251 1253
1252 1254 if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1253 1255 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1254 1256 return (err);
1255 1257
1256 1258 tx = dmu_tx_create(os);
1257 1259
1258 1260 dmu_tx_hold_write(tx, drrwbr->drr_object,
1259 1261 drrwbr->drr_offset, drrwbr->drr_length);
1260 1262 err = dmu_tx_assign(tx, TXG_WAIT);
1261 1263 if (err) {
1262 1264 dmu_tx_abort(tx);
1263 1265 return (err);
1264 1266 }
1265 1267 dmu_write(os, drrwbr->drr_object,
1266 1268 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1267 1269 dmu_buf_rele(dbp, FTAG);
1268 1270 dmu_tx_commit(tx);
1269 1271 return (0);
1270 1272 }
1271 1273
1272 1274 static int
1273 1275 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1274 1276 {
1275 1277 dmu_tx_t *tx;
1276 1278 void *data;
1277 1279 dmu_buf_t *db, *db_spill;
1278 1280 int err;
1279 1281
1280 1282 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1281 1283 drrs->drr_length > SPA_MAXBLOCKSIZE)
1282 1284 return (EINVAL);
1283 1285
1284 1286 data = restore_read(ra, drrs->drr_length);
1285 1287 if (data == NULL)
1286 1288 return (ra->err);
1287 1289
1288 1290 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1289 1291 return (EINVAL);
1290 1292
1291 1293 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1292 1294 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1293 1295 dmu_buf_rele(db, FTAG);
1294 1296 return (err);
1295 1297 }
1296 1298
1297 1299 tx = dmu_tx_create(os);
1298 1300
1299 1301 dmu_tx_hold_spill(tx, db->db_object);
1300 1302
1301 1303 err = dmu_tx_assign(tx, TXG_WAIT);
1302 1304 if (err) {
1303 1305 dmu_buf_rele(db, FTAG);
1304 1306 dmu_buf_rele(db_spill, FTAG);
1305 1307 dmu_tx_abort(tx);
1306 1308 return (err);
1307 1309 }
1308 1310 dmu_buf_will_dirty(db_spill, tx);
1309 1311
1310 1312 if (db_spill->db_size < drrs->drr_length)
1311 1313 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1312 1314 drrs->drr_length, tx));
1313 1315 bcopy(data, db_spill->db_data, drrs->drr_length);
1314 1316
1315 1317 dmu_buf_rele(db, FTAG);
1316 1318 dmu_buf_rele(db_spill, FTAG);
1317 1319
1318 1320 dmu_tx_commit(tx);
1319 1321 return (0);
1320 1322 }
1321 1323
1322 1324 /* ARGSUSED */
1323 1325 static int
1324 1326 restore_free(struct restorearg *ra, objset_t *os,
1325 1327 struct drr_free *drrf)
1326 1328 {
1327 1329 int err;
1328 1330
1329 1331 if (drrf->drr_length != -1ULL &&
1330 1332 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1331 1333 return (EINVAL);
1332 1334
1333 1335 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1334 1336 return (EINVAL);
1335 1337
1336 1338 err = dmu_free_long_range(os, drrf->drr_object,
1337 1339 drrf->drr_offset, drrf->drr_length);
1338 1340 return (err);
1339 1341 }
1340 1342
1341 1343 /*
1342 1344 * NB: callers *must* call dmu_recv_end() if this succeeds.
1343 1345 */
1344 1346 int
1345 1347 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1346 1348 int cleanup_fd, uint64_t *action_handlep)
1347 1349 {
1348 1350 struct restorearg ra = { 0 };
1349 1351 dmu_replay_record_t *drr;
1350 1352 objset_t *os;
1351 1353 zio_cksum_t pcksum;
1352 1354 int featureflags;
1353 1355
1354 1356 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1355 1357 ra.byteswap = TRUE;
1356 1358
1357 1359 {
1358 1360 /* compute checksum of drr_begin record */
1359 1361 dmu_replay_record_t *drr;
1360 1362 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1361 1363
1362 1364 drr->drr_type = DRR_BEGIN;
1363 1365 drr->drr_u.drr_begin = *drc->drc_drrb;
1364 1366 if (ra.byteswap) {
1365 1367 fletcher_4_incremental_byteswap(drr,
1366 1368 sizeof (dmu_replay_record_t), &ra.cksum);
1367 1369 } else {
1368 1370 fletcher_4_incremental_native(drr,
1369 1371 sizeof (dmu_replay_record_t), &ra.cksum);
1370 1372 }
1371 1373 kmem_free(drr, sizeof (dmu_replay_record_t));
1372 1374 }
1373 1375
1374 1376 if (ra.byteswap) {
1375 1377 struct drr_begin *drrb = drc->drc_drrb;
1376 1378 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1377 1379 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1378 1380 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1379 1381 drrb->drr_type = BSWAP_32(drrb->drr_type);
1380 1382 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1381 1383 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1382 1384 }
1383 1385
1384 1386 ra.vp = vp;
1385 1387 ra.voff = *voffp;
1386 1388 ra.bufsize = 1<<20;
1387 1389 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1388 1390
1389 1391 /* these were verified in dmu_recv_begin */
1390 1392 ASSERT(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo) ==
1391 1393 DMU_SUBSTREAM);
1392 1394 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1393 1395
1394 1396 /*
1395 1397 * Open the objset we are modifying.
1396 1398 */
1397 1399 VERIFY(dmu_objset_from_ds(drc->drc_real_ds, &os) == 0);
1398 1400
1399 1401 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1400 1402
1401 1403 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1402 1404
1403 1405 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1404 1406 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1405 1407 minor_t minor;
1406 1408
1407 1409 if (cleanup_fd == -1) {
1408 1410 ra.err = EBADF;
1409 1411 goto out;
1410 1412 }
1411 1413 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1412 1414 if (ra.err) {
1413 1415 cleanup_fd = -1;
1414 1416 goto out;
1415 1417 }
1416 1418
1417 1419 if (*action_handlep == 0) {
1418 1420 ra.guid_to_ds_map =
1419 1421 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1420 1422 avl_create(ra.guid_to_ds_map, guid_compare,
1421 1423 sizeof (guid_map_entry_t),
1422 1424 offsetof(guid_map_entry_t, avlnode));
1423 1425 ra.err = zfs_onexit_add_cb(minor,
1424 1426 free_guid_map_onexit, ra.guid_to_ds_map,
1425 1427 action_handlep);
1426 1428 if (ra.err)
1427 1429 goto out;
1428 1430 } else {
1429 1431 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1430 1432 (void **)&ra.guid_to_ds_map);
1431 1433 if (ra.err)
1432 1434 goto out;
1433 1435 }
1434 1436
1435 1437 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1436 1438 }
1437 1439
1438 1440 /*
1439 1441 * Read records and process them.
1440 1442 */
1441 1443 pcksum = ra.cksum;
1442 1444 while (ra.err == 0 &&
1443 1445 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1444 1446 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1445 1447 ra.err = EINTR;
1446 1448 goto out;
1447 1449 }
1448 1450
1449 1451 if (ra.byteswap)
1450 1452 backup_byteswap(drr);
1451 1453
1452 1454 switch (drr->drr_type) {
1453 1455 case DRR_OBJECT:
1454 1456 {
1455 1457 /*
1456 1458 * We need to make a copy of the record header,
1457 1459 * because restore_{object,write} may need to
1458 1460 * restore_read(), which will invalidate drr.
1459 1461 */
1460 1462 struct drr_object drro = drr->drr_u.drr_object;
1461 1463 ra.err = restore_object(&ra, os, &drro);
1462 1464 break;
1463 1465 }
1464 1466 case DRR_FREEOBJECTS:
1465 1467 {
1466 1468 struct drr_freeobjects drrfo =
1467 1469 drr->drr_u.drr_freeobjects;
1468 1470 ra.err = restore_freeobjects(&ra, os, &drrfo);
1469 1471 break;
1470 1472 }
1471 1473 case DRR_WRITE:
1472 1474 {
1473 1475 struct drr_write drrw = drr->drr_u.drr_write;
1474 1476 ra.err = restore_write(&ra, os, &drrw);
1475 1477 break;
1476 1478 }
1477 1479 case DRR_WRITE_BYREF:
1478 1480 {
1479 1481 struct drr_write_byref drrwbr =
1480 1482 drr->drr_u.drr_write_byref;
1481 1483 ra.err = restore_write_byref(&ra, os, &drrwbr);
1482 1484 break;
1483 1485 }
1484 1486 case DRR_FREE:
1485 1487 {
1486 1488 struct drr_free drrf = drr->drr_u.drr_free;
1487 1489 ra.err = restore_free(&ra, os, &drrf);
1488 1490 break;
1489 1491 }
1490 1492 case DRR_END:
1491 1493 {
1492 1494 struct drr_end drre = drr->drr_u.drr_end;
1493 1495 /*
1494 1496 * We compare against the *previous* checksum
1495 1497 * value, because the stored checksum is of
1496 1498 * everything before the DRR_END record.
1497 1499 */
1498 1500 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1499 1501 ra.err = ECKSUM;
1500 1502 goto out;
1501 1503 }
1502 1504 case DRR_SPILL:
1503 1505 {
1504 1506 struct drr_spill drrs = drr->drr_u.drr_spill;
1505 1507 ra.err = restore_spill(&ra, os, &drrs);
1506 1508 break;
1507 1509 }
1508 1510 default:
1509 1511 ra.err = EINVAL;
1510 1512 goto out;
1511 1513 }
1512 1514 pcksum = ra.cksum;
1513 1515 }
1514 1516 ASSERT(ra.err != 0);
1515 1517
1516 1518 out:
1517 1519 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1518 1520 zfs_onexit_fd_rele(cleanup_fd);
1519 1521
1520 1522 if (ra.err != 0) {
1521 1523 /*
1522 1524 * destroy what we created, so we don't leave it in the
1523 1525 * inconsistent restoring state.
1524 1526 */
1525 1527 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1526 1528
1527 1529 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1528 1530 B_FALSE);
1529 1531 if (drc->drc_real_ds != drc->drc_logical_ds) {
1530 1532 mutex_exit(&drc->drc_logical_ds->ds_recvlock);
1531 1533 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
1532 1534 }
1533 1535 }
1534 1536
1535 1537 kmem_free(ra.buf, ra.bufsize);
1536 1538 *voffp = ra.voff;
1537 1539 return (ra.err);
1538 1540 }
1539 1541
1540 1542 struct recvendsyncarg {
1541 1543 char *tosnap;
1542 1544 uint64_t creation_time;
1543 1545 uint64_t toguid;
1544 1546 };
1545 1547
1546 1548 static int
1547 1549 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1548 1550 {
1549 1551 dsl_dataset_t *ds = arg1;
1550 1552 struct recvendsyncarg *resa = arg2;
1551 1553
1552 1554 return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1553 1555 }
1554 1556
1555 1557 static void
1556 1558 recv_end_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1557 1559 {
1558 1560 dsl_dataset_t *ds = arg1;
1559 1561 struct recvendsyncarg *resa = arg2;
1560 1562
1561 1563 dsl_dataset_snapshot_sync(ds, resa->tosnap, tx);
1562 1564
1563 1565 /* set snapshot's creation time and guid */
1564 1566 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1565 1567 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1566 1568 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1567 1569 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1568 1570
1569 1571 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1570 1572 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1571 1573 }
1572 1574
1573 1575 static int
1574 1576 add_ds_to_guidmap(avl_tree_t *guid_map, dsl_dataset_t *ds)
1575 1577 {
1576 1578 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1577 1579 uint64_t snapobj = ds->ds_phys->ds_prev_snap_obj;
1578 1580 dsl_dataset_t *snapds;
1579 1581 guid_map_entry_t *gmep;
1580 1582 int err;
1581 1583
1582 1584 ASSERT(guid_map != NULL);
1583 1585
1584 1586 rw_enter(&dp->dp_config_rwlock, RW_READER);
1585 1587 err = dsl_dataset_hold_obj(dp, snapobj, guid_map, &snapds);
1586 1588 if (err == 0) {
1587 1589 gmep = kmem_alloc(sizeof (guid_map_entry_t), KM_SLEEP);
1588 1590 gmep->guid = snapds->ds_phys->ds_guid;
1589 1591 gmep->gme_ds = snapds;
1590 1592 avl_add(guid_map, gmep);
1591 1593 }
1592 1594
1593 1595 rw_exit(&dp->dp_config_rwlock);
1594 1596 return (err);
1595 1597 }
1596 1598
1597 1599 static int
1598 1600 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1599 1601 {
1600 1602 struct recvendsyncarg resa;
1601 1603 dsl_dataset_t *ds = drc->drc_logical_ds;
1602 1604 int err, myerr;
1603 1605
1604 1606 /*
1605 1607 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1606 1608 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1607 1609 * can close it.
1608 1610 */
1609 1611 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1610 1612
1611 1613 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1612 1614 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1613 1615 drc->drc_force);
1614 1616 if (err)
1615 1617 goto out;
1616 1618 } else {
1617 1619 mutex_exit(&ds->ds_recvlock);
1618 1620 dsl_dataset_rele(ds, dmu_recv_tag);
1619 1621 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1620 1622 B_FALSE);
1621 1623 return (EBUSY);
1622 1624 }
1623 1625
1624 1626 resa.creation_time = drc->drc_drrb->drr_creation_time;
1625 1627 resa.toguid = drc->drc_drrb->drr_toguid;
1626 1628 resa.tosnap = drc->drc_tosnap;
1627 1629
1628 1630 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1629 1631 recv_end_check, recv_end_sync, ds, &resa, 3);
1630 1632 if (err) {
1631 1633 /* swap back */
1632 1634 (void) dsl_dataset_clone_swap(drc->drc_real_ds, ds, B_TRUE);
1633 1635 }
1634 1636
1635 1637 out:
1636 1638 mutex_exit(&ds->ds_recvlock);
1637 1639 if (err == 0 && drc->drc_guid_to_ds_map != NULL)
1638 1640 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1639 1641 dsl_dataset_disown(ds, dmu_recv_tag);
1640 1642 myerr = dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, B_FALSE);
1641 1643 ASSERT3U(myerr, ==, 0);
1642 1644 return (err);
1643 1645 }
1644 1646
1645 1647 static int
1646 1648 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1647 1649 {
1648 1650 struct recvendsyncarg resa;
1649 1651 dsl_dataset_t *ds = drc->drc_logical_ds;
1650 1652 int err;
1651 1653
1652 1654 /*
1653 1655 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1654 1656 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1655 1657 * can close it.
1656 1658 */
1657 1659 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1658 1660
1659 1661 resa.creation_time = drc->drc_drrb->drr_creation_time;
1660 1662 resa.toguid = drc->drc_drrb->drr_toguid;
1661 1663 resa.tosnap = drc->drc_tosnap;
1662 1664
1663 1665 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1664 1666 recv_end_check, recv_end_sync, ds, &resa, 3);
1665 1667 if (err) {
1666 1668 /* clean up the fs we just recv'd into */
1667 1669 (void) dsl_dataset_destroy(ds, dmu_recv_tag, B_FALSE);
1668 1670 } else {
1669 1671 if (drc->drc_guid_to_ds_map != NULL)
1670 1672 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1671 1673 /* release the hold from dmu_recv_begin */
1672 1674 dsl_dataset_disown(ds, dmu_recv_tag);
1673 1675 }
1674 1676 return (err);
1675 1677 }
1676 1678
1677 1679 int
1678 1680 dmu_recv_end(dmu_recv_cookie_t *drc)
1679 1681 {
1680 1682 if (drc->drc_logical_ds != drc->drc_real_ds)
1681 1683 return (dmu_recv_existing_end(drc));
1682 1684 else
1683 1685 return (dmu_recv_new_end(drc));
1684 1686 }
|
↓ open down ↓ |
467 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX