Print this page
OS-1566 filesystem limits for ZFS datasets
| 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 *
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 23 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
24 24 * Copyright (c) 2012 by Delphix. All rights reserved.
25 25 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
26 26 */
27 27
28 28 #include <sys/dmu.h>
29 29 #include <sys/dmu_impl.h>
30 30 #include <sys/dmu_tx.h>
31 31 #include <sys/dbuf.h>
32 32 #include <sys/dnode.h>
33 33 #include <sys/zfs_context.h>
34 34 #include <sys/dmu_objset.h>
35 35 #include <sys/dmu_traverse.h>
36 36 #include <sys/dsl_dataset.h>
37 37 #include <sys/dsl_dir.h>
38 38 #include <sys/dsl_prop.h>
39 39 #include <sys/dsl_pool.h>
40 40 #include <sys/dsl_synctask.h>
41 41 #include <sys/zfs_ioctl.h>
42 42 #include <sys/zap.h>
43 43 #include <sys/zio_checksum.h>
44 44 #include <sys/zfs_znode.h>
45 45 #include <zfs_fletcher.h>
46 46 #include <sys/avl.h>
47 47 #include <sys/ddt.h>
48 48 #include <sys/zfs_onexit.h>
49 49
50 50 /* Set this tunable to TRUE to replace corrupt data with 0x2f5baddb10c */
51 51 int zfs_send_corrupt_data = B_FALSE;
52 52
53 53 static char *dmu_recv_tag = "dmu_recv_tag";
54 54
55 55 static int
56 56 dump_bytes(dmu_sendarg_t *dsp, void *buf, int len)
57 57 {
58 58 dsl_dataset_t *ds = dsp->dsa_os->os_dsl_dataset;
59 59 ssize_t resid; /* have to get resid to get detailed errno */
60 60 ASSERT0(len % 8);
61 61
62 62 fletcher_4_incremental_native(buf, len, &dsp->dsa_zc);
63 63 dsp->dsa_err = vn_rdwr(UIO_WRITE, dsp->dsa_vp,
64 64 (caddr_t)buf, len,
65 65 0, UIO_SYSSPACE, FAPPEND, RLIM64_INFINITY, CRED(), &resid);
66 66
67 67 mutex_enter(&ds->ds_sendstream_lock);
68 68 *dsp->dsa_off += len;
69 69 mutex_exit(&ds->ds_sendstream_lock);
70 70
71 71 return (dsp->dsa_err);
72 72 }
73 73
74 74 static int
75 75 dump_free(dmu_sendarg_t *dsp, uint64_t object, uint64_t offset,
76 76 uint64_t length)
77 77 {
78 78 struct drr_free *drrf = &(dsp->dsa_drr->drr_u.drr_free);
79 79
80 80 if (length != -1ULL && offset + length < offset)
81 81 length = -1ULL;
82 82
83 83 /*
84 84 * If there is a pending op, but it's not PENDING_FREE, push it out,
85 85 * since free block aggregation can only be done for blocks of the
86 86 * same type (i.e., DRR_FREE records can only be aggregated with
87 87 * other DRR_FREE records. DRR_FREEOBJECTS records can only be
88 88 * aggregated with other DRR_FREEOBJECTS records.
89 89 */
90 90 if (dsp->dsa_pending_op != PENDING_NONE &&
91 91 dsp->dsa_pending_op != PENDING_FREE) {
92 92 if (dump_bytes(dsp, dsp->dsa_drr,
93 93 sizeof (dmu_replay_record_t)) != 0)
94 94 return (EINTR);
95 95 dsp->dsa_pending_op = PENDING_NONE;
96 96 }
97 97
98 98 if (dsp->dsa_pending_op == PENDING_FREE) {
99 99 /*
100 100 * There should never be a PENDING_FREE if length is -1
101 101 * (because dump_dnode is the only place where this
102 102 * function is called with a -1, and only after flushing
103 103 * any pending record).
104 104 */
105 105 ASSERT(length != -1ULL);
106 106 /*
107 107 * Check to see whether this free block can be aggregated
108 108 * with pending one.
109 109 */
110 110 if (drrf->drr_object == object && drrf->drr_offset +
111 111 drrf->drr_length == offset) {
112 112 drrf->drr_length += length;
113 113 return (0);
114 114 } else {
115 115 /* not a continuation. Push out pending record */
116 116 if (dump_bytes(dsp, dsp->dsa_drr,
117 117 sizeof (dmu_replay_record_t)) != 0)
118 118 return (EINTR);
119 119 dsp->dsa_pending_op = PENDING_NONE;
120 120 }
121 121 }
122 122 /* create a FREE record and make it pending */
123 123 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
124 124 dsp->dsa_drr->drr_type = DRR_FREE;
125 125 drrf->drr_object = object;
126 126 drrf->drr_offset = offset;
127 127 drrf->drr_length = length;
128 128 drrf->drr_toguid = dsp->dsa_toguid;
129 129 if (length == -1ULL) {
130 130 if (dump_bytes(dsp, dsp->dsa_drr,
131 131 sizeof (dmu_replay_record_t)) != 0)
132 132 return (EINTR);
133 133 } else {
134 134 dsp->dsa_pending_op = PENDING_FREE;
135 135 }
136 136
137 137 return (0);
138 138 }
139 139
140 140 static int
141 141 dump_data(dmu_sendarg_t *dsp, dmu_object_type_t type,
142 142 uint64_t object, uint64_t offset, int blksz, const blkptr_t *bp, void *data)
143 143 {
144 144 struct drr_write *drrw = &(dsp->dsa_drr->drr_u.drr_write);
145 145
146 146
147 147 /*
148 148 * If there is any kind of pending aggregation (currently either
149 149 * a grouping of free objects or free blocks), push it out to
150 150 * the stream, since aggregation can't be done across operations
151 151 * of different types.
152 152 */
153 153 if (dsp->dsa_pending_op != PENDING_NONE) {
154 154 if (dump_bytes(dsp, dsp->dsa_drr,
155 155 sizeof (dmu_replay_record_t)) != 0)
156 156 return (EINTR);
157 157 dsp->dsa_pending_op = PENDING_NONE;
158 158 }
159 159 /* write a DATA record */
160 160 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
161 161 dsp->dsa_drr->drr_type = DRR_WRITE;
162 162 drrw->drr_object = object;
163 163 drrw->drr_type = type;
164 164 drrw->drr_offset = offset;
165 165 drrw->drr_length = blksz;
166 166 drrw->drr_toguid = dsp->dsa_toguid;
167 167 drrw->drr_checksumtype = BP_GET_CHECKSUM(bp);
168 168 if (zio_checksum_table[drrw->drr_checksumtype].ci_dedup)
169 169 drrw->drr_checksumflags |= DRR_CHECKSUM_DEDUP;
170 170 DDK_SET_LSIZE(&drrw->drr_key, BP_GET_LSIZE(bp));
171 171 DDK_SET_PSIZE(&drrw->drr_key, BP_GET_PSIZE(bp));
172 172 DDK_SET_COMPRESS(&drrw->drr_key, BP_GET_COMPRESS(bp));
173 173 drrw->drr_key.ddk_cksum = bp->blk_cksum;
174 174
175 175 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
176 176 return (EINTR);
177 177 if (dump_bytes(dsp, data, blksz) != 0)
178 178 return (EINTR);
179 179 return (0);
180 180 }
181 181
182 182 static int
183 183 dump_spill(dmu_sendarg_t *dsp, uint64_t object, int blksz, void *data)
184 184 {
185 185 struct drr_spill *drrs = &(dsp->dsa_drr->drr_u.drr_spill);
186 186
187 187 if (dsp->dsa_pending_op != PENDING_NONE) {
188 188 if (dump_bytes(dsp, dsp->dsa_drr,
189 189 sizeof (dmu_replay_record_t)) != 0)
190 190 return (EINTR);
191 191 dsp->dsa_pending_op = PENDING_NONE;
192 192 }
193 193
194 194 /* write a SPILL record */
195 195 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
196 196 dsp->dsa_drr->drr_type = DRR_SPILL;
197 197 drrs->drr_object = object;
198 198 drrs->drr_length = blksz;
199 199 drrs->drr_toguid = dsp->dsa_toguid;
200 200
201 201 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)))
202 202 return (EINTR);
203 203 if (dump_bytes(dsp, data, blksz))
204 204 return (EINTR);
205 205 return (0);
206 206 }
207 207
208 208 static int
209 209 dump_freeobjects(dmu_sendarg_t *dsp, uint64_t firstobj, uint64_t numobjs)
210 210 {
211 211 struct drr_freeobjects *drrfo = &(dsp->dsa_drr->drr_u.drr_freeobjects);
212 212
213 213 /*
214 214 * If there is a pending op, but it's not PENDING_FREEOBJECTS,
215 215 * push it out, since free block aggregation can only be done for
216 216 * blocks of the same type (i.e., DRR_FREE records can only be
217 217 * aggregated with other DRR_FREE records. DRR_FREEOBJECTS records
218 218 * can only be aggregated with other DRR_FREEOBJECTS records.
219 219 */
220 220 if (dsp->dsa_pending_op != PENDING_NONE &&
221 221 dsp->dsa_pending_op != PENDING_FREEOBJECTS) {
222 222 if (dump_bytes(dsp, dsp->dsa_drr,
223 223 sizeof (dmu_replay_record_t)) != 0)
224 224 return (EINTR);
225 225 dsp->dsa_pending_op = PENDING_NONE;
226 226 }
227 227 if (dsp->dsa_pending_op == PENDING_FREEOBJECTS) {
228 228 /*
229 229 * See whether this free object array can be aggregated
230 230 * with pending one
231 231 */
232 232 if (drrfo->drr_firstobj + drrfo->drr_numobjs == firstobj) {
233 233 drrfo->drr_numobjs += numobjs;
234 234 return (0);
235 235 } else {
236 236 /* can't be aggregated. Push out pending record */
237 237 if (dump_bytes(dsp, dsp->dsa_drr,
238 238 sizeof (dmu_replay_record_t)) != 0)
239 239 return (EINTR);
240 240 dsp->dsa_pending_op = PENDING_NONE;
241 241 }
242 242 }
243 243
244 244 /* write a FREEOBJECTS record */
245 245 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
246 246 dsp->dsa_drr->drr_type = DRR_FREEOBJECTS;
247 247 drrfo->drr_firstobj = firstobj;
248 248 drrfo->drr_numobjs = numobjs;
249 249 drrfo->drr_toguid = dsp->dsa_toguid;
250 250
251 251 dsp->dsa_pending_op = PENDING_FREEOBJECTS;
252 252
253 253 return (0);
254 254 }
255 255
256 256 static int
257 257 dump_dnode(dmu_sendarg_t *dsp, uint64_t object, dnode_phys_t *dnp)
258 258 {
259 259 struct drr_object *drro = &(dsp->dsa_drr->drr_u.drr_object);
260 260
261 261 if (dnp == NULL || dnp->dn_type == DMU_OT_NONE)
262 262 return (dump_freeobjects(dsp, object, 1));
263 263
264 264 if (dsp->dsa_pending_op != PENDING_NONE) {
265 265 if (dump_bytes(dsp, dsp->dsa_drr,
266 266 sizeof (dmu_replay_record_t)) != 0)
267 267 return (EINTR);
268 268 dsp->dsa_pending_op = PENDING_NONE;
269 269 }
270 270
271 271 /* write an OBJECT record */
272 272 bzero(dsp->dsa_drr, sizeof (dmu_replay_record_t));
273 273 dsp->dsa_drr->drr_type = DRR_OBJECT;
274 274 drro->drr_object = object;
275 275 drro->drr_type = dnp->dn_type;
276 276 drro->drr_bonustype = dnp->dn_bonustype;
277 277 drro->drr_blksz = dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT;
278 278 drro->drr_bonuslen = dnp->dn_bonuslen;
279 279 drro->drr_checksumtype = dnp->dn_checksum;
280 280 drro->drr_compress = dnp->dn_compress;
281 281 drro->drr_toguid = dsp->dsa_toguid;
282 282
283 283 if (dump_bytes(dsp, dsp->dsa_drr, sizeof (dmu_replay_record_t)) != 0)
284 284 return (EINTR);
285 285
286 286 if (dump_bytes(dsp, DN_BONUS(dnp), P2ROUNDUP(dnp->dn_bonuslen, 8)) != 0)
287 287 return (EINTR);
288 288
289 289 /* free anything past the end of the file */
290 290 if (dump_free(dsp, object, (dnp->dn_maxblkid + 1) *
291 291 (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT), -1ULL))
292 292 return (EINTR);
293 293 if (dsp->dsa_err)
294 294 return (EINTR);
295 295 return (0);
296 296 }
297 297
298 298 #define BP_SPAN(dnp, level) \
299 299 (((uint64_t)dnp->dn_datablkszsec) << (SPA_MINBLOCKSHIFT + \
300 300 (level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)))
301 301
302 302 /* ARGSUSED */
303 303 static int
304 304 backup_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
305 305 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
306 306 {
307 307 dmu_sendarg_t *dsp = arg;
308 308 dmu_object_type_t type = bp ? BP_GET_TYPE(bp) : DMU_OT_NONE;
309 309 int err = 0;
310 310
311 311 if (issig(JUSTLOOKING) && issig(FORREAL))
312 312 return (EINTR);
313 313
314 314 if (zb->zb_object != DMU_META_DNODE_OBJECT &&
315 315 DMU_OBJECT_IS_SPECIAL(zb->zb_object)) {
316 316 return (0);
317 317 } else if (bp == NULL && zb->zb_object == DMU_META_DNODE_OBJECT) {
318 318 uint64_t span = BP_SPAN(dnp, zb->zb_level);
319 319 uint64_t dnobj = (zb->zb_blkid * span) >> DNODE_SHIFT;
320 320 err = dump_freeobjects(dsp, dnobj, span >> DNODE_SHIFT);
321 321 } else if (bp == NULL) {
322 322 uint64_t span = BP_SPAN(dnp, zb->zb_level);
323 323 err = dump_free(dsp, zb->zb_object, zb->zb_blkid * span, span);
324 324 } else if (zb->zb_level > 0 || type == DMU_OT_OBJSET) {
325 325 return (0);
326 326 } else if (type == DMU_OT_DNODE) {
327 327 dnode_phys_t *blk;
328 328 int i;
329 329 int blksz = BP_GET_LSIZE(bp);
330 330 uint32_t aflags = ARC_WAIT;
331 331 arc_buf_t *abuf;
332 332
333 333 if (dsl_read(NULL, spa, bp, pbuf,
334 334 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
335 335 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
336 336 return (EIO);
337 337
338 338 blk = abuf->b_data;
339 339 for (i = 0; i < blksz >> DNODE_SHIFT; i++) {
340 340 uint64_t dnobj = (zb->zb_blkid <<
341 341 (DNODE_BLOCK_SHIFT - DNODE_SHIFT)) + i;
342 342 err = dump_dnode(dsp, dnobj, blk+i);
343 343 if (err)
344 344 break;
345 345 }
346 346 (void) arc_buf_remove_ref(abuf, &abuf);
347 347 } else if (type == DMU_OT_SA) {
348 348 uint32_t aflags = ARC_WAIT;
349 349 arc_buf_t *abuf;
350 350 int blksz = BP_GET_LSIZE(bp);
351 351
352 352 if (arc_read_nolock(NULL, spa, bp,
353 353 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
354 354 ZIO_FLAG_CANFAIL, &aflags, zb) != 0)
355 355 return (EIO);
356 356
357 357 err = dump_spill(dsp, zb->zb_object, blksz, abuf->b_data);
358 358 (void) arc_buf_remove_ref(abuf, &abuf);
359 359 } else { /* it's a level-0 block of a regular object */
360 360 uint32_t aflags = ARC_WAIT;
361 361 arc_buf_t *abuf;
362 362 int blksz = BP_GET_LSIZE(bp);
363 363
364 364 if (dsl_read(NULL, spa, bp, pbuf,
365 365 arc_getbuf_func, &abuf, ZIO_PRIORITY_ASYNC_READ,
366 366 ZIO_FLAG_CANFAIL, &aflags, zb) != 0) {
367 367 if (zfs_send_corrupt_data) {
368 368 /* Send a block filled with 0x"zfs badd bloc" */
369 369 abuf = arc_buf_alloc(spa, blksz, &abuf,
370 370 ARC_BUFC_DATA);
371 371 uint64_t *ptr;
372 372 for (ptr = abuf->b_data;
373 373 (char *)ptr < (char *)abuf->b_data + blksz;
374 374 ptr++)
375 375 *ptr = 0x2f5baddb10c;
376 376 } else {
377 377 return (EIO);
378 378 }
379 379 }
380 380
381 381 err = dump_data(dsp, type, zb->zb_object, zb->zb_blkid * blksz,
382 382 blksz, bp, abuf->b_data);
383 383 (void) arc_buf_remove_ref(abuf, &abuf);
384 384 }
385 385
386 386 ASSERT(err == 0 || err == EINTR);
387 387 return (err);
388 388 }
389 389
390 390 /*
391 391 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
392 392 * For example, they could both be snapshots of the same filesystem, and
393 393 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
394 394 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
395 395 * filesystem. Or 'earlier' could be the origin's origin.
396 396 */
397 397 static boolean_t
398 398 is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
399 399 {
400 400 dsl_pool_t *dp = later->ds_dir->dd_pool;
401 401 int error;
402 402 boolean_t ret;
403 403 dsl_dataset_t *origin;
404 404
405 405 if (earlier->ds_phys->ds_creation_txg >=
406 406 later->ds_phys->ds_creation_txg)
407 407 return (B_FALSE);
408 408
409 409 if (later->ds_dir == earlier->ds_dir)
410 410 return (B_TRUE);
411 411 if (!dsl_dir_is_clone(later->ds_dir))
412 412 return (B_FALSE);
413 413
414 414 rw_enter(&dp->dp_config_rwlock, RW_READER);
415 415 if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object) {
416 416 rw_exit(&dp->dp_config_rwlock);
417 417 return (B_TRUE);
418 418 }
419 419 error = dsl_dataset_hold_obj(dp,
420 420 later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
421 421 rw_exit(&dp->dp_config_rwlock);
422 422 if (error != 0)
423 423 return (B_FALSE);
424 424 ret = is_before(origin, earlier);
425 425 dsl_dataset_rele(origin, FTAG);
426 426 return (ret);
427 427 }
428 428
429 429 int
430 430 dmu_send(objset_t *tosnap, objset_t *fromsnap, int outfd, vnode_t *vp,
431 431 offset_t *off)
432 432 {
433 433 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
434 434 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
435 435 dmu_replay_record_t *drr;
436 436 dmu_sendarg_t *dsp;
437 437 int err;
438 438 uint64_t fromtxg = 0;
439 439
440 440 /* tosnap must be a snapshot */
441 441 if (ds->ds_phys->ds_next_snap_obj == 0)
442 442 return (EINVAL);
443 443
444 444 /*
445 445 * fromsnap must be an earlier snapshot from the same fs as tosnap,
446 446 * or the origin's fs.
447 447 */
448 448 if (fromds != NULL && !is_before(ds, fromds))
449 449 return (EXDEV);
450 450
451 451 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
452 452 drr->drr_type = DRR_BEGIN;
453 453 drr->drr_u.drr_begin.drr_magic = DMU_BACKUP_MAGIC;
454 454 DMU_SET_STREAM_HDRTYPE(drr->drr_u.drr_begin.drr_versioninfo,
455 455 DMU_SUBSTREAM);
456 456
457 457 #ifdef _KERNEL
458 458 if (dmu_objset_type(tosnap) == DMU_OST_ZFS) {
459 459 uint64_t version;
460 460 if (zfs_get_zplprop(tosnap, ZFS_PROP_VERSION, &version) != 0) {
461 461 kmem_free(drr, sizeof (dmu_replay_record_t));
462 462 return (EINVAL);
463 463 }
464 464 if (version == ZPL_VERSION_SA) {
465 465 DMU_SET_FEATUREFLAGS(
466 466 drr->drr_u.drr_begin.drr_versioninfo,
467 467 DMU_BACKUP_FEATURE_SA_SPILL);
468 468 }
469 469 }
470 470 #endif
471 471
472 472 drr->drr_u.drr_begin.drr_creation_time =
473 473 ds->ds_phys->ds_creation_time;
474 474 drr->drr_u.drr_begin.drr_type = tosnap->os_phys->os_type;
475 475 if (fromds != NULL && ds->ds_dir != fromds->ds_dir)
476 476 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CLONE;
477 477 drr->drr_u.drr_begin.drr_toguid = ds->ds_phys->ds_guid;
478 478 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
479 479 drr->drr_u.drr_begin.drr_flags |= DRR_FLAG_CI_DATA;
480 480
481 481 if (fromds)
482 482 drr->drr_u.drr_begin.drr_fromguid = fromds->ds_phys->ds_guid;
483 483 dsl_dataset_name(ds, drr->drr_u.drr_begin.drr_toname);
484 484
485 485 if (fromds)
486 486 fromtxg = fromds->ds_phys->ds_creation_txg;
487 487
488 488 dsp = kmem_zalloc(sizeof (dmu_sendarg_t), KM_SLEEP);
489 489
490 490 dsp->dsa_drr = drr;
491 491 dsp->dsa_vp = vp;
492 492 dsp->dsa_outfd = outfd;
493 493 dsp->dsa_proc = curproc;
494 494 dsp->dsa_os = tosnap;
495 495 dsp->dsa_off = off;
496 496 dsp->dsa_toguid = ds->ds_phys->ds_guid;
497 497 ZIO_SET_CHECKSUM(&dsp->dsa_zc, 0, 0, 0, 0);
498 498 dsp->dsa_pending_op = PENDING_NONE;
499 499
500 500 mutex_enter(&ds->ds_sendstream_lock);
501 501 list_insert_head(&ds->ds_sendstreams, dsp);
502 502 mutex_exit(&ds->ds_sendstream_lock);
503 503
504 504 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
505 505 err = dsp->dsa_err;
506 506 goto out;
507 507 }
508 508
509 509 err = traverse_dataset(ds, fromtxg, TRAVERSE_PRE | TRAVERSE_PREFETCH,
510 510 backup_cb, dsp);
511 511
512 512 if (dsp->dsa_pending_op != PENDING_NONE)
513 513 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0)
514 514 err = EINTR;
515 515
516 516 if (err) {
517 517 if (err == EINTR && dsp->dsa_err)
518 518 err = dsp->dsa_err;
519 519 goto out;
520 520 }
521 521
522 522 bzero(drr, sizeof (dmu_replay_record_t));
523 523 drr->drr_type = DRR_END;
524 524 drr->drr_u.drr_end.drr_checksum = dsp->dsa_zc;
525 525 drr->drr_u.drr_end.drr_toguid = dsp->dsa_toguid;
526 526
527 527 if (dump_bytes(dsp, drr, sizeof (dmu_replay_record_t)) != 0) {
528 528 err = dsp->dsa_err;
529 529 goto out;
530 530 }
531 531
532 532 out:
533 533 mutex_enter(&ds->ds_sendstream_lock);
534 534 list_remove(&ds->ds_sendstreams, dsp);
535 535 mutex_exit(&ds->ds_sendstream_lock);
536 536
537 537 kmem_free(drr, sizeof (dmu_replay_record_t));
538 538 kmem_free(dsp, sizeof (dmu_sendarg_t));
539 539
540 540 return (err);
541 541 }
542 542
543 543 int
544 544 dmu_send_estimate(objset_t *tosnap, objset_t *fromsnap, uint64_t *sizep)
545 545 {
546 546 dsl_dataset_t *ds = tosnap->os_dsl_dataset;
547 547 dsl_dataset_t *fromds = fromsnap ? fromsnap->os_dsl_dataset : NULL;
548 548 dsl_pool_t *dp = ds->ds_dir->dd_pool;
549 549 int err;
550 550 uint64_t size;
551 551
552 552 /* tosnap must be a snapshot */
553 553 if (ds->ds_phys->ds_next_snap_obj == 0)
554 554 return (EINVAL);
555 555
556 556 /*
557 557 * fromsnap must be an earlier snapshot from the same fs as tosnap,
558 558 * or the origin's fs.
559 559 */
560 560 if (fromds != NULL && !is_before(ds, fromds))
561 561 return (EXDEV);
562 562
563 563 /* Get uncompressed size estimate of changed data. */
564 564 if (fromds == NULL) {
565 565 size = ds->ds_phys->ds_uncompressed_bytes;
566 566 } else {
567 567 uint64_t used, comp;
568 568 err = dsl_dataset_space_written(fromds, ds,
569 569 &used, &comp, &size);
570 570 if (err)
571 571 return (err);
572 572 }
573 573
574 574 /*
575 575 * Assume that space (both on-disk and in-stream) is dominated by
576 576 * data. We will adjust for indirect blocks and the copies property,
577 577 * but ignore per-object space used (eg, dnodes and DRR_OBJECT records).
578 578 */
579 579
580 580 /*
581 581 * Subtract out approximate space used by indirect blocks.
582 582 * Assume most space is used by data blocks (non-indirect, non-dnode).
583 583 * Assume all blocks are recordsize. Assume ditto blocks and
584 584 * internal fragmentation counter out compression.
585 585 *
586 586 * Therefore, space used by indirect blocks is sizeof(blkptr_t) per
587 587 * block, which we observe in practice.
588 588 */
589 589 uint64_t recordsize;
590 590 rw_enter(&dp->dp_config_rwlock, RW_READER);
591 591 err = dsl_prop_get_ds(ds, "recordsize",
592 592 sizeof (recordsize), 1, &recordsize, NULL);
593 593 rw_exit(&dp->dp_config_rwlock);
594 594 if (err)
595 595 return (err);
596 596 size -= size / recordsize * sizeof (blkptr_t);
597 597
598 598 /* Add in the space for the record associated with each block. */
599 599 size += size / recordsize * sizeof (dmu_replay_record_t);
600 600
601 601 *sizep = size;
602 602
603 603 return (0);
604 604 }
605 605
606 606 struct recvbeginsyncarg {
607 607 const char *tofs;
608 608 const char *tosnap;
609 609 dsl_dataset_t *origin;
610 610 uint64_t fromguid;
611 611 dmu_objset_type_t type;
612 612 void *tag;
613 613 boolean_t force;
614 614 uint64_t dsflags;
615 615 char clonelastname[MAXNAMELEN];
616 616 dsl_dataset_t *ds; /* the ds to recv into; returned from the syncfunc */
617 617 cred_t *cr;
618 618 };
619 619
620 620 /* ARGSUSED */
621 621 static int
622 622 recv_new_check(void *arg1, void *arg2, dmu_tx_t *tx)
623 623 {
624 624 dsl_dir_t *dd = arg1;
625 625 struct recvbeginsyncarg *rbsa = arg2;
626 626 objset_t *mos = dd->dd_pool->dp_meta_objset;
627 627 uint64_t val;
628 628 int err;
629 629
630 630 err = zap_lookup(mos, dd->dd_phys->dd_child_dir_zapobj,
631 631 strrchr(rbsa->tofs, '/') + 1, sizeof (uint64_t), 1, &val);
632 632
633 633 if (err != ENOENT)
634 634 return (err ? err : EEXIST);
635 635
|
↓ open down ↓ |
635 lines elided |
↑ open up ↑ |
636 636 if (rbsa->origin) {
637 637 /* make sure it's a snap in the same pool */
638 638 if (rbsa->origin->ds_dir->dd_pool != dd->dd_pool)
639 639 return (EXDEV);
640 640 if (!dsl_dataset_is_snapshot(rbsa->origin))
641 641 return (EINVAL);
642 642 if (rbsa->origin->ds_phys->ds_guid != rbsa->fromguid)
643 643 return (ENODEV);
644 644 }
645 645
646 + /*
647 + * Check filesystem and snapshot limits before receiving. We'll recheck
648 + * again at the end, but might as well abort before receiving if we're
649 + * already over the limit.
650 + */
651 + err = dsl_dir_fscount_check(dd, 1, NULL, rbsa->cr);
652 + if (err != 0)
653 + return (err);
654 +
655 + err = dsl_snapcount_check(dd, 1, NULL, rbsa->cr);
656 + if (err != 0)
657 + return (err);
658 +
646 659 return (0);
647 660 }
648 661
649 662 static void
650 663 recv_new_sync(void *arg1, void *arg2, dmu_tx_t *tx)
651 664 {
652 665 dsl_dir_t *dd = arg1;
653 666 struct recvbeginsyncarg *rbsa = arg2;
654 667 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
655 668 uint64_t dsobj;
656 669
657 670 /* Create and open new dataset. */
658 671 dsobj = dsl_dataset_create_sync(dd, strrchr(rbsa->tofs, '/') + 1,
659 672 rbsa->origin, flags, rbsa->cr, tx);
660 673 VERIFY(0 == dsl_dataset_own_obj(dd->dd_pool, dsobj,
|
↓ open down ↓ |
5 lines elided |
↑ open up ↑ |
661 674 B_TRUE, dmu_recv_tag, &rbsa->ds));
662 675
663 676 if (rbsa->origin == NULL) {
664 677 (void) dmu_objset_create_impl(dd->dd_pool->dp_spa,
665 678 rbsa->ds, &rbsa->ds->ds_phys->ds_bp, rbsa->type, tx);
666 679 }
667 680
668 681 spa_history_log_internal_ds(rbsa->ds, "receive new", tx, "");
669 682 }
670 683
684 +/*
685 + * Note that we do not check the file system limit with dsl_dir_fscount_check
686 + * because the temporary %clones don't count against that limit.
687 + */
671 688 /* ARGSUSED */
672 689 static int
673 690 recv_existing_check(void *arg1, void *arg2, dmu_tx_t *tx)
674 691 {
675 692 dsl_dataset_t *ds = arg1;
676 693 struct recvbeginsyncarg *rbsa = arg2;
677 694 int err;
678 695 uint64_t val;
679 696
680 697 /* must not have any changes since most recent snapshot */
681 698 if (!rbsa->force && dsl_dataset_modified_since_lastsnap(ds))
682 699 return (ETXTBSY);
683 700
684 701 /* new snapshot name must not exist */
685 702 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
686 703 ds->ds_phys->ds_snapnames_zapobj, rbsa->tosnap, 8, 1, &val);
687 704 if (err == 0)
688 705 return (EEXIST);
689 706 if (err != ENOENT)
690 707 return (err);
691 708
692 709 if (rbsa->fromguid) {
693 710 /* if incremental, most recent snapshot must match fromguid */
694 711 if (ds->ds_prev == NULL)
695 712 return (ENODEV);
696 713
697 714 /*
698 715 * most recent snapshot must match fromguid, or there are no
699 716 * changes since the fromguid one
700 717 */
701 718 if (ds->ds_prev->ds_phys->ds_guid != rbsa->fromguid) {
702 719 uint64_t birth = ds->ds_prev->ds_phys->ds_bp.blk_birth;
703 720 uint64_t obj = ds->ds_prev->ds_phys->ds_prev_snap_obj;
704 721 while (obj != 0) {
705 722 dsl_dataset_t *snap;
706 723 err = dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
707 724 obj, FTAG, &snap);
708 725 if (err)
709 726 return (ENODEV);
710 727 if (snap->ds_phys->ds_creation_txg < birth) {
711 728 dsl_dataset_rele(snap, FTAG);
712 729 return (ENODEV);
713 730 }
714 731 if (snap->ds_phys->ds_guid == rbsa->fromguid) {
715 732 dsl_dataset_rele(snap, FTAG);
716 733 break; /* it's ok */
717 734 }
|
↓ open down ↓ |
37 lines elided |
↑ open up ↑ |
718 735 obj = snap->ds_phys->ds_prev_snap_obj;
719 736 dsl_dataset_rele(snap, FTAG);
720 737 }
721 738 if (obj == 0)
722 739 return (ENODEV);
723 740 }
724 741 } else {
725 742 /* if full, most recent snapshot must be $ORIGIN */
726 743 if (ds->ds_phys->ds_prev_snap_txg >= TXG_INITIAL)
727 744 return (ENODEV);
745 +
746 + /* Check snapshot limit before receiving */
747 + err = dsl_snapcount_check(ds->ds_dir, 1, NULL, rbsa->cr);
748 + if (err != 0)
749 + return (err);
728 750 }
729 751
730 752 /* temporary clone name must not exist */
731 753 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset,
732 754 ds->ds_dir->dd_phys->dd_child_dir_zapobj,
733 755 rbsa->clonelastname, 8, 1, &val);
734 756 if (err == 0)
735 757 return (EEXIST);
736 758 if (err != ENOENT)
737 759 return (err);
738 760
739 761 return (0);
740 762 }
741 763
742 764 /* ARGSUSED */
743 765 static void
744 766 recv_existing_sync(void *arg1, void *arg2, dmu_tx_t *tx)
745 767 {
746 768 dsl_dataset_t *ohds = arg1;
747 769 struct recvbeginsyncarg *rbsa = arg2;
748 770 dsl_pool_t *dp = ohds->ds_dir->dd_pool;
749 771 dsl_dataset_t *cds;
750 772 uint64_t flags = DS_FLAG_INCONSISTENT | rbsa->dsflags;
751 773 uint64_t dsobj;
752 774
753 775 /* create and open the temporary clone */
754 776 dsobj = dsl_dataset_create_sync(ohds->ds_dir, rbsa->clonelastname,
755 777 ohds->ds_prev, flags, rbsa->cr, tx);
756 778 VERIFY(0 == dsl_dataset_own_obj(dp, dsobj, B_TRUE, dmu_recv_tag, &cds));
757 779
758 780 /*
759 781 * If we actually created a non-clone, we need to create the
760 782 * objset in our new dataset.
761 783 */
762 784 if (BP_IS_HOLE(dsl_dataset_get_blkptr(cds))) {
763 785 (void) dmu_objset_create_impl(dp->dp_spa,
764 786 cds, dsl_dataset_get_blkptr(cds), rbsa->type, tx);
765 787 }
766 788
767 789 rbsa->ds = cds;
768 790
769 791 spa_history_log_internal_ds(cds, "receive over existing", tx, "");
770 792 }
771 793
772 794 static boolean_t
773 795 dmu_recv_verify_features(dsl_dataset_t *ds, struct drr_begin *drrb)
774 796 {
775 797 int featureflags;
776 798
777 799 featureflags = DMU_GET_FEATUREFLAGS(drrb->drr_versioninfo);
778 800
779 801 /* Verify pool version supports SA if SA_SPILL feature set */
780 802 return ((featureflags & DMU_BACKUP_FEATURE_SA_SPILL) &&
781 803 (spa_version(dsl_dataset_get_spa(ds)) < SPA_VERSION_SA));
782 804 }
783 805
784 806 /*
785 807 * NB: callers *MUST* call dmu_recv_stream() if dmu_recv_begin()
786 808 * succeeds; otherwise we will leak the holds on the datasets.
787 809 */
788 810 int
789 811 dmu_recv_begin(char *tofs, char *tosnap, char *top_ds, struct drr_begin *drrb,
790 812 boolean_t force, objset_t *origin, dmu_recv_cookie_t *drc)
791 813 {
792 814 int err = 0;
793 815 boolean_t byteswap;
794 816 struct recvbeginsyncarg rbsa = { 0 };
795 817 uint64_t versioninfo;
796 818 int flags;
797 819 dsl_dataset_t *ds;
798 820
799 821 if (drrb->drr_magic == DMU_BACKUP_MAGIC)
800 822 byteswap = FALSE;
801 823 else if (drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
802 824 byteswap = TRUE;
803 825 else
804 826 return (EINVAL);
805 827
806 828 rbsa.tofs = tofs;
807 829 rbsa.tosnap = tosnap;
808 830 rbsa.origin = origin ? origin->os_dsl_dataset : NULL;
809 831 rbsa.fromguid = drrb->drr_fromguid;
810 832 rbsa.type = drrb->drr_type;
811 833 rbsa.tag = FTAG;
812 834 rbsa.dsflags = 0;
813 835 rbsa.cr = CRED();
814 836 versioninfo = drrb->drr_versioninfo;
815 837 flags = drrb->drr_flags;
816 838
817 839 if (byteswap) {
818 840 rbsa.type = BSWAP_32(rbsa.type);
819 841 rbsa.fromguid = BSWAP_64(rbsa.fromguid);
820 842 versioninfo = BSWAP_64(versioninfo);
821 843 flags = BSWAP_32(flags);
822 844 }
823 845
824 846 if (DMU_GET_STREAM_HDRTYPE(versioninfo) == DMU_COMPOUNDSTREAM ||
825 847 rbsa.type >= DMU_OST_NUMTYPES ||
826 848 ((flags & DRR_FLAG_CLONE) && origin == NULL))
827 849 return (EINVAL);
828 850
829 851 if (flags & DRR_FLAG_CI_DATA)
830 852 rbsa.dsflags = DS_FLAG_CI_DATASET;
831 853
832 854 bzero(drc, sizeof (dmu_recv_cookie_t));
833 855 drc->drc_drrb = drrb;
834 856 drc->drc_tosnap = tosnap;
835 857 drc->drc_top_ds = top_ds;
836 858 drc->drc_force = force;
837 859
838 860 /*
839 861 * Process the begin in syncing context.
840 862 */
841 863
842 864 /* open the dataset we are logically receiving into */
843 865 err = dsl_dataset_hold(tofs, dmu_recv_tag, &ds);
844 866 if (err == 0) {
845 867 if (dmu_recv_verify_features(ds, drrb)) {
846 868 dsl_dataset_rele(ds, dmu_recv_tag);
847 869 return (ENOTSUP);
848 870 }
849 871 /* target fs already exists; recv into temp clone */
850 872
851 873 /* Can't recv a clone into an existing fs */
852 874 if (flags & DRR_FLAG_CLONE) {
853 875 dsl_dataset_rele(ds, dmu_recv_tag);
854 876 return (EINVAL);
855 877 }
856 878
857 879 /* must not have an incremental recv already in progress */
858 880 if (!mutex_tryenter(&ds->ds_recvlock)) {
859 881 dsl_dataset_rele(ds, dmu_recv_tag);
860 882 return (EBUSY);
861 883 }
862 884
863 885 /* tmp clone name is: tofs/%tosnap" */
864 886 (void) snprintf(rbsa.clonelastname, sizeof (rbsa.clonelastname),
865 887 "%%%s", tosnap);
866 888 rbsa.force = force;
867 889 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
868 890 recv_existing_check, recv_existing_sync, ds, &rbsa, 5);
869 891 if (err) {
870 892 mutex_exit(&ds->ds_recvlock);
871 893 dsl_dataset_rele(ds, dmu_recv_tag);
872 894 return (err);
873 895 }
874 896 drc->drc_logical_ds = ds;
875 897 drc->drc_real_ds = rbsa.ds;
876 898 } else if (err == ENOENT) {
877 899 /* target fs does not exist; must be a full backup or clone */
878 900 char *cp;
879 901
880 902 /*
881 903 * If it's a non-clone incremental, we are missing the
882 904 * target fs, so fail the recv.
883 905 */
884 906 if (rbsa.fromguid && !(flags & DRR_FLAG_CLONE))
885 907 return (ENOENT);
886 908
887 909 /* Open the parent of tofs */
888 910 cp = strrchr(tofs, '/');
889 911 *cp = '\0';
890 912 err = dsl_dataset_hold(tofs, FTAG, &ds);
891 913 *cp = '/';
892 914 if (err)
893 915 return (err);
894 916
895 917 if (dmu_recv_verify_features(ds, drrb)) {
896 918 dsl_dataset_rele(ds, FTAG);
897 919 return (ENOTSUP);
898 920 }
899 921
900 922 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
901 923 recv_new_check, recv_new_sync, ds->ds_dir, &rbsa, 5);
902 924 dsl_dataset_rele(ds, FTAG);
903 925 if (err)
904 926 return (err);
905 927 drc->drc_logical_ds = drc->drc_real_ds = rbsa.ds;
906 928 drc->drc_newfs = B_TRUE;
907 929 }
908 930
909 931 return (err);
910 932 }
911 933
912 934 struct restorearg {
913 935 int err;
914 936 int byteswap;
915 937 vnode_t *vp;
916 938 char *buf;
917 939 uint64_t voff;
918 940 int bufsize; /* amount of memory allocated for buf */
919 941 zio_cksum_t cksum;
920 942 avl_tree_t *guid_to_ds_map;
921 943 };
922 944
923 945 typedef struct guid_map_entry {
924 946 uint64_t guid;
925 947 dsl_dataset_t *gme_ds;
926 948 avl_node_t avlnode;
927 949 } guid_map_entry_t;
928 950
929 951 static int
930 952 guid_compare(const void *arg1, const void *arg2)
931 953 {
932 954 const guid_map_entry_t *gmep1 = arg1;
933 955 const guid_map_entry_t *gmep2 = arg2;
934 956
935 957 if (gmep1->guid < gmep2->guid)
936 958 return (-1);
937 959 else if (gmep1->guid > gmep2->guid)
938 960 return (1);
939 961 return (0);
940 962 }
941 963
942 964 static void
943 965 free_guid_map_onexit(void *arg)
944 966 {
945 967 avl_tree_t *ca = arg;
946 968 void *cookie = NULL;
947 969 guid_map_entry_t *gmep;
948 970
949 971 while ((gmep = avl_destroy_nodes(ca, &cookie)) != NULL) {
950 972 dsl_dataset_rele(gmep->gme_ds, ca);
951 973 kmem_free(gmep, sizeof (guid_map_entry_t));
952 974 }
953 975 avl_destroy(ca);
954 976 kmem_free(ca, sizeof (avl_tree_t));
955 977 }
956 978
957 979 static void *
958 980 restore_read(struct restorearg *ra, int len)
959 981 {
960 982 void *rv;
961 983 int done = 0;
962 984
963 985 /* some things will require 8-byte alignment, so everything must */
964 986 ASSERT0(len % 8);
965 987
966 988 while (done < len) {
967 989 ssize_t resid;
968 990
969 991 ra->err = vn_rdwr(UIO_READ, ra->vp,
970 992 (caddr_t)ra->buf + done, len - done,
971 993 ra->voff, UIO_SYSSPACE, FAPPEND,
972 994 RLIM64_INFINITY, CRED(), &resid);
973 995
974 996 if (resid == len - done)
975 997 ra->err = EINVAL;
976 998 ra->voff += len - done - resid;
977 999 done = len - resid;
978 1000 if (ra->err)
979 1001 return (NULL);
980 1002 }
981 1003
982 1004 ASSERT3U(done, ==, len);
983 1005 rv = ra->buf;
984 1006 if (ra->byteswap)
985 1007 fletcher_4_incremental_byteswap(rv, len, &ra->cksum);
986 1008 else
987 1009 fletcher_4_incremental_native(rv, len, &ra->cksum);
988 1010 return (rv);
989 1011 }
990 1012
991 1013 static void
992 1014 backup_byteswap(dmu_replay_record_t *drr)
993 1015 {
994 1016 #define DO64(X) (drr->drr_u.X = BSWAP_64(drr->drr_u.X))
995 1017 #define DO32(X) (drr->drr_u.X = BSWAP_32(drr->drr_u.X))
996 1018 drr->drr_type = BSWAP_32(drr->drr_type);
997 1019 drr->drr_payloadlen = BSWAP_32(drr->drr_payloadlen);
998 1020 switch (drr->drr_type) {
999 1021 case DRR_BEGIN:
1000 1022 DO64(drr_begin.drr_magic);
1001 1023 DO64(drr_begin.drr_versioninfo);
1002 1024 DO64(drr_begin.drr_creation_time);
1003 1025 DO32(drr_begin.drr_type);
1004 1026 DO32(drr_begin.drr_flags);
1005 1027 DO64(drr_begin.drr_toguid);
1006 1028 DO64(drr_begin.drr_fromguid);
1007 1029 break;
1008 1030 case DRR_OBJECT:
1009 1031 DO64(drr_object.drr_object);
1010 1032 /* DO64(drr_object.drr_allocation_txg); */
1011 1033 DO32(drr_object.drr_type);
1012 1034 DO32(drr_object.drr_bonustype);
1013 1035 DO32(drr_object.drr_blksz);
1014 1036 DO32(drr_object.drr_bonuslen);
1015 1037 DO64(drr_object.drr_toguid);
1016 1038 break;
1017 1039 case DRR_FREEOBJECTS:
1018 1040 DO64(drr_freeobjects.drr_firstobj);
1019 1041 DO64(drr_freeobjects.drr_numobjs);
1020 1042 DO64(drr_freeobjects.drr_toguid);
1021 1043 break;
1022 1044 case DRR_WRITE:
1023 1045 DO64(drr_write.drr_object);
1024 1046 DO32(drr_write.drr_type);
1025 1047 DO64(drr_write.drr_offset);
1026 1048 DO64(drr_write.drr_length);
1027 1049 DO64(drr_write.drr_toguid);
1028 1050 DO64(drr_write.drr_key.ddk_cksum.zc_word[0]);
1029 1051 DO64(drr_write.drr_key.ddk_cksum.zc_word[1]);
1030 1052 DO64(drr_write.drr_key.ddk_cksum.zc_word[2]);
1031 1053 DO64(drr_write.drr_key.ddk_cksum.zc_word[3]);
1032 1054 DO64(drr_write.drr_key.ddk_prop);
1033 1055 break;
1034 1056 case DRR_WRITE_BYREF:
1035 1057 DO64(drr_write_byref.drr_object);
1036 1058 DO64(drr_write_byref.drr_offset);
1037 1059 DO64(drr_write_byref.drr_length);
1038 1060 DO64(drr_write_byref.drr_toguid);
1039 1061 DO64(drr_write_byref.drr_refguid);
1040 1062 DO64(drr_write_byref.drr_refobject);
1041 1063 DO64(drr_write_byref.drr_refoffset);
1042 1064 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[0]);
1043 1065 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[1]);
1044 1066 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[2]);
1045 1067 DO64(drr_write_byref.drr_key.ddk_cksum.zc_word[3]);
1046 1068 DO64(drr_write_byref.drr_key.ddk_prop);
1047 1069 break;
1048 1070 case DRR_FREE:
1049 1071 DO64(drr_free.drr_object);
1050 1072 DO64(drr_free.drr_offset);
1051 1073 DO64(drr_free.drr_length);
1052 1074 DO64(drr_free.drr_toguid);
1053 1075 break;
1054 1076 case DRR_SPILL:
1055 1077 DO64(drr_spill.drr_object);
1056 1078 DO64(drr_spill.drr_length);
1057 1079 DO64(drr_spill.drr_toguid);
1058 1080 break;
1059 1081 case DRR_END:
1060 1082 DO64(drr_end.drr_checksum.zc_word[0]);
1061 1083 DO64(drr_end.drr_checksum.zc_word[1]);
1062 1084 DO64(drr_end.drr_checksum.zc_word[2]);
1063 1085 DO64(drr_end.drr_checksum.zc_word[3]);
1064 1086 DO64(drr_end.drr_toguid);
1065 1087 break;
1066 1088 }
1067 1089 #undef DO64
1068 1090 #undef DO32
1069 1091 }
1070 1092
1071 1093 static int
1072 1094 restore_object(struct restorearg *ra, objset_t *os, struct drr_object *drro)
1073 1095 {
1074 1096 int err;
1075 1097 dmu_tx_t *tx;
1076 1098 void *data = NULL;
1077 1099
1078 1100 if (drro->drr_type == DMU_OT_NONE ||
1079 1101 !DMU_OT_IS_VALID(drro->drr_type) ||
1080 1102 !DMU_OT_IS_VALID(drro->drr_bonustype) ||
1081 1103 drro->drr_checksumtype >= ZIO_CHECKSUM_FUNCTIONS ||
1082 1104 drro->drr_compress >= ZIO_COMPRESS_FUNCTIONS ||
1083 1105 P2PHASE(drro->drr_blksz, SPA_MINBLOCKSIZE) ||
1084 1106 drro->drr_blksz < SPA_MINBLOCKSIZE ||
1085 1107 drro->drr_blksz > SPA_MAXBLOCKSIZE ||
1086 1108 drro->drr_bonuslen > DN_MAX_BONUSLEN) {
1087 1109 return (EINVAL);
1088 1110 }
1089 1111
1090 1112 err = dmu_object_info(os, drro->drr_object, NULL);
1091 1113
1092 1114 if (err != 0 && err != ENOENT)
1093 1115 return (EINVAL);
1094 1116
1095 1117 if (drro->drr_bonuslen) {
1096 1118 data = restore_read(ra, P2ROUNDUP(drro->drr_bonuslen, 8));
1097 1119 if (ra->err)
1098 1120 return (ra->err);
1099 1121 }
1100 1122
1101 1123 if (err == ENOENT) {
1102 1124 /* currently free, want to be allocated */
1103 1125 tx = dmu_tx_create(os);
1104 1126 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT);
1105 1127 err = dmu_tx_assign(tx, TXG_WAIT);
1106 1128 if (err) {
1107 1129 dmu_tx_abort(tx);
1108 1130 return (err);
1109 1131 }
1110 1132 err = dmu_object_claim(os, drro->drr_object,
1111 1133 drro->drr_type, drro->drr_blksz,
1112 1134 drro->drr_bonustype, drro->drr_bonuslen, tx);
1113 1135 dmu_tx_commit(tx);
1114 1136 } else {
1115 1137 /* currently allocated, want to be allocated */
1116 1138 err = dmu_object_reclaim(os, drro->drr_object,
1117 1139 drro->drr_type, drro->drr_blksz,
1118 1140 drro->drr_bonustype, drro->drr_bonuslen);
1119 1141 }
1120 1142 if (err) {
1121 1143 return (EINVAL);
1122 1144 }
1123 1145
1124 1146 tx = dmu_tx_create(os);
1125 1147 dmu_tx_hold_bonus(tx, drro->drr_object);
1126 1148 err = dmu_tx_assign(tx, TXG_WAIT);
1127 1149 if (err) {
1128 1150 dmu_tx_abort(tx);
1129 1151 return (err);
1130 1152 }
1131 1153
1132 1154 dmu_object_set_checksum(os, drro->drr_object, drro->drr_checksumtype,
1133 1155 tx);
1134 1156 dmu_object_set_compress(os, drro->drr_object, drro->drr_compress, tx);
1135 1157
1136 1158 if (data != NULL) {
1137 1159 dmu_buf_t *db;
1138 1160
1139 1161 VERIFY(0 == dmu_bonus_hold(os, drro->drr_object, FTAG, &db));
1140 1162 dmu_buf_will_dirty(db, tx);
1141 1163
1142 1164 ASSERT3U(db->db_size, >=, drro->drr_bonuslen);
1143 1165 bcopy(data, db->db_data, drro->drr_bonuslen);
1144 1166 if (ra->byteswap) {
1145 1167 dmu_object_byteswap_t byteswap =
1146 1168 DMU_OT_BYTESWAP(drro->drr_bonustype);
1147 1169 dmu_ot_byteswap[byteswap].ob_func(db->db_data,
1148 1170 drro->drr_bonuslen);
1149 1171 }
1150 1172 dmu_buf_rele(db, FTAG);
1151 1173 }
1152 1174 dmu_tx_commit(tx);
1153 1175 return (0);
1154 1176 }
1155 1177
1156 1178 /* ARGSUSED */
1157 1179 static int
1158 1180 restore_freeobjects(struct restorearg *ra, objset_t *os,
1159 1181 struct drr_freeobjects *drrfo)
1160 1182 {
1161 1183 uint64_t obj;
1162 1184
1163 1185 if (drrfo->drr_firstobj + drrfo->drr_numobjs < drrfo->drr_firstobj)
1164 1186 return (EINVAL);
1165 1187
1166 1188 for (obj = drrfo->drr_firstobj;
1167 1189 obj < drrfo->drr_firstobj + drrfo->drr_numobjs;
1168 1190 (void) dmu_object_next(os, &obj, FALSE, 0)) {
1169 1191 int err;
1170 1192
1171 1193 if (dmu_object_info(os, obj, NULL) != 0)
1172 1194 continue;
1173 1195
1174 1196 err = dmu_free_object(os, obj);
1175 1197 if (err)
1176 1198 return (err);
1177 1199 }
1178 1200 return (0);
1179 1201 }
1180 1202
1181 1203 static int
1182 1204 restore_write(struct restorearg *ra, objset_t *os,
1183 1205 struct drr_write *drrw)
1184 1206 {
1185 1207 dmu_tx_t *tx;
1186 1208 void *data;
1187 1209 int err;
1188 1210
1189 1211 if (drrw->drr_offset + drrw->drr_length < drrw->drr_offset ||
1190 1212 !DMU_OT_IS_VALID(drrw->drr_type))
1191 1213 return (EINVAL);
1192 1214
1193 1215 data = restore_read(ra, drrw->drr_length);
1194 1216 if (data == NULL)
1195 1217 return (ra->err);
1196 1218
1197 1219 if (dmu_object_info(os, drrw->drr_object, NULL) != 0)
1198 1220 return (EINVAL);
1199 1221
1200 1222 tx = dmu_tx_create(os);
1201 1223
1202 1224 dmu_tx_hold_write(tx, drrw->drr_object,
1203 1225 drrw->drr_offset, drrw->drr_length);
1204 1226 err = dmu_tx_assign(tx, TXG_WAIT);
1205 1227 if (err) {
1206 1228 dmu_tx_abort(tx);
1207 1229 return (err);
1208 1230 }
1209 1231 if (ra->byteswap) {
1210 1232 dmu_object_byteswap_t byteswap =
1211 1233 DMU_OT_BYTESWAP(drrw->drr_type);
1212 1234 dmu_ot_byteswap[byteswap].ob_func(data, drrw->drr_length);
1213 1235 }
1214 1236 dmu_write(os, drrw->drr_object,
1215 1237 drrw->drr_offset, drrw->drr_length, data, tx);
1216 1238 dmu_tx_commit(tx);
1217 1239 return (0);
1218 1240 }
1219 1241
1220 1242 /*
1221 1243 * Handle a DRR_WRITE_BYREF record. This record is used in dedup'ed
1222 1244 * streams to refer to a copy of the data that is already on the
1223 1245 * system because it came in earlier in the stream. This function
1224 1246 * finds the earlier copy of the data, and uses that copy instead of
1225 1247 * data from the stream to fulfill this write.
1226 1248 */
1227 1249 static int
1228 1250 restore_write_byref(struct restorearg *ra, objset_t *os,
1229 1251 struct drr_write_byref *drrwbr)
1230 1252 {
1231 1253 dmu_tx_t *tx;
1232 1254 int err;
1233 1255 guid_map_entry_t gmesrch;
1234 1256 guid_map_entry_t *gmep;
1235 1257 avl_index_t where;
1236 1258 objset_t *ref_os = NULL;
1237 1259 dmu_buf_t *dbp;
1238 1260
1239 1261 if (drrwbr->drr_offset + drrwbr->drr_length < drrwbr->drr_offset)
1240 1262 return (EINVAL);
1241 1263
1242 1264 /*
1243 1265 * If the GUID of the referenced dataset is different from the
1244 1266 * GUID of the target dataset, find the referenced dataset.
1245 1267 */
1246 1268 if (drrwbr->drr_toguid != drrwbr->drr_refguid) {
1247 1269 gmesrch.guid = drrwbr->drr_refguid;
1248 1270 if ((gmep = avl_find(ra->guid_to_ds_map, &gmesrch,
1249 1271 &where)) == NULL) {
1250 1272 return (EINVAL);
1251 1273 }
1252 1274 if (dmu_objset_from_ds(gmep->gme_ds, &ref_os))
1253 1275 return (EINVAL);
1254 1276 } else {
1255 1277 ref_os = os;
1256 1278 }
1257 1279
1258 1280 if (err = dmu_buf_hold(ref_os, drrwbr->drr_refobject,
1259 1281 drrwbr->drr_refoffset, FTAG, &dbp, DMU_READ_PREFETCH))
1260 1282 return (err);
1261 1283
1262 1284 tx = dmu_tx_create(os);
1263 1285
1264 1286 dmu_tx_hold_write(tx, drrwbr->drr_object,
1265 1287 drrwbr->drr_offset, drrwbr->drr_length);
1266 1288 err = dmu_tx_assign(tx, TXG_WAIT);
1267 1289 if (err) {
1268 1290 dmu_tx_abort(tx);
1269 1291 return (err);
1270 1292 }
1271 1293 dmu_write(os, drrwbr->drr_object,
1272 1294 drrwbr->drr_offset, drrwbr->drr_length, dbp->db_data, tx);
1273 1295 dmu_buf_rele(dbp, FTAG);
1274 1296 dmu_tx_commit(tx);
1275 1297 return (0);
1276 1298 }
1277 1299
1278 1300 static int
1279 1301 restore_spill(struct restorearg *ra, objset_t *os, struct drr_spill *drrs)
1280 1302 {
1281 1303 dmu_tx_t *tx;
1282 1304 void *data;
1283 1305 dmu_buf_t *db, *db_spill;
1284 1306 int err;
1285 1307
1286 1308 if (drrs->drr_length < SPA_MINBLOCKSIZE ||
1287 1309 drrs->drr_length > SPA_MAXBLOCKSIZE)
1288 1310 return (EINVAL);
1289 1311
1290 1312 data = restore_read(ra, drrs->drr_length);
1291 1313 if (data == NULL)
1292 1314 return (ra->err);
1293 1315
1294 1316 if (dmu_object_info(os, drrs->drr_object, NULL) != 0)
1295 1317 return (EINVAL);
1296 1318
1297 1319 VERIFY(0 == dmu_bonus_hold(os, drrs->drr_object, FTAG, &db));
1298 1320 if ((err = dmu_spill_hold_by_bonus(db, FTAG, &db_spill)) != 0) {
1299 1321 dmu_buf_rele(db, FTAG);
1300 1322 return (err);
1301 1323 }
1302 1324
1303 1325 tx = dmu_tx_create(os);
1304 1326
1305 1327 dmu_tx_hold_spill(tx, db->db_object);
1306 1328
1307 1329 err = dmu_tx_assign(tx, TXG_WAIT);
1308 1330 if (err) {
1309 1331 dmu_buf_rele(db, FTAG);
1310 1332 dmu_buf_rele(db_spill, FTAG);
1311 1333 dmu_tx_abort(tx);
1312 1334 return (err);
1313 1335 }
1314 1336 dmu_buf_will_dirty(db_spill, tx);
1315 1337
1316 1338 if (db_spill->db_size < drrs->drr_length)
1317 1339 VERIFY(0 == dbuf_spill_set_blksz(db_spill,
1318 1340 drrs->drr_length, tx));
1319 1341 bcopy(data, db_spill->db_data, drrs->drr_length);
1320 1342
1321 1343 dmu_buf_rele(db, FTAG);
1322 1344 dmu_buf_rele(db_spill, FTAG);
1323 1345
1324 1346 dmu_tx_commit(tx);
1325 1347 return (0);
1326 1348 }
1327 1349
1328 1350 /* ARGSUSED */
1329 1351 static int
1330 1352 restore_free(struct restorearg *ra, objset_t *os,
1331 1353 struct drr_free *drrf)
1332 1354 {
1333 1355 int err;
1334 1356
1335 1357 if (drrf->drr_length != -1ULL &&
1336 1358 drrf->drr_offset + drrf->drr_length < drrf->drr_offset)
1337 1359 return (EINVAL);
1338 1360
1339 1361 if (dmu_object_info(os, drrf->drr_object, NULL) != 0)
1340 1362 return (EINVAL);
1341 1363
1342 1364 err = dmu_free_long_range(os, drrf->drr_object,
1343 1365 drrf->drr_offset, drrf->drr_length);
1344 1366 return (err);
1345 1367 }
1346 1368
1347 1369 /*
1348 1370 * NB: callers *must* call dmu_recv_end() if this succeeds.
1349 1371 */
1350 1372 int
1351 1373 dmu_recv_stream(dmu_recv_cookie_t *drc, vnode_t *vp, offset_t *voffp,
1352 1374 int cleanup_fd, uint64_t *action_handlep)
1353 1375 {
1354 1376 struct restorearg ra = { 0 };
1355 1377 dmu_replay_record_t *drr;
1356 1378 objset_t *os;
1357 1379 zio_cksum_t pcksum;
1358 1380 int featureflags;
1359 1381
1360 1382 if (drc->drc_drrb->drr_magic == BSWAP_64(DMU_BACKUP_MAGIC))
1361 1383 ra.byteswap = TRUE;
1362 1384
1363 1385 {
1364 1386 /* compute checksum of drr_begin record */
1365 1387 dmu_replay_record_t *drr;
1366 1388 drr = kmem_zalloc(sizeof (dmu_replay_record_t), KM_SLEEP);
1367 1389
1368 1390 drr->drr_type = DRR_BEGIN;
1369 1391 drr->drr_u.drr_begin = *drc->drc_drrb;
1370 1392 if (ra.byteswap) {
1371 1393 fletcher_4_incremental_byteswap(drr,
1372 1394 sizeof (dmu_replay_record_t), &ra.cksum);
1373 1395 } else {
1374 1396 fletcher_4_incremental_native(drr,
1375 1397 sizeof (dmu_replay_record_t), &ra.cksum);
1376 1398 }
1377 1399 kmem_free(drr, sizeof (dmu_replay_record_t));
1378 1400 }
1379 1401
1380 1402 if (ra.byteswap) {
1381 1403 struct drr_begin *drrb = drc->drc_drrb;
1382 1404 drrb->drr_magic = BSWAP_64(drrb->drr_magic);
1383 1405 drrb->drr_versioninfo = BSWAP_64(drrb->drr_versioninfo);
1384 1406 drrb->drr_creation_time = BSWAP_64(drrb->drr_creation_time);
1385 1407 drrb->drr_type = BSWAP_32(drrb->drr_type);
1386 1408 drrb->drr_toguid = BSWAP_64(drrb->drr_toguid);
1387 1409 drrb->drr_fromguid = BSWAP_64(drrb->drr_fromguid);
1388 1410 }
1389 1411
1390 1412 ra.vp = vp;
1391 1413 ra.voff = *voffp;
1392 1414 ra.bufsize = 1<<20;
1393 1415 ra.buf = kmem_alloc(ra.bufsize, KM_SLEEP);
1394 1416
1395 1417 /* these were verified in dmu_recv_begin */
1396 1418 ASSERT(DMU_GET_STREAM_HDRTYPE(drc->drc_drrb->drr_versioninfo) ==
1397 1419 DMU_SUBSTREAM);
1398 1420 ASSERT(drc->drc_drrb->drr_type < DMU_OST_NUMTYPES);
1399 1421
1400 1422 /*
1401 1423 * Open the objset we are modifying.
1402 1424 */
1403 1425 VERIFY(dmu_objset_from_ds(drc->drc_real_ds, &os) == 0);
1404 1426
1405 1427 ASSERT(drc->drc_real_ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT);
1406 1428
1407 1429 featureflags = DMU_GET_FEATUREFLAGS(drc->drc_drrb->drr_versioninfo);
1408 1430
1409 1431 /* if this stream is dedup'ed, set up the avl tree for guid mapping */
1410 1432 if (featureflags & DMU_BACKUP_FEATURE_DEDUP) {
1411 1433 minor_t minor;
1412 1434
1413 1435 if (cleanup_fd == -1) {
1414 1436 ra.err = EBADF;
1415 1437 goto out;
1416 1438 }
1417 1439 ra.err = zfs_onexit_fd_hold(cleanup_fd, &minor);
1418 1440 if (ra.err) {
1419 1441 cleanup_fd = -1;
1420 1442 goto out;
1421 1443 }
1422 1444
1423 1445 if (*action_handlep == 0) {
1424 1446 ra.guid_to_ds_map =
1425 1447 kmem_alloc(sizeof (avl_tree_t), KM_SLEEP);
1426 1448 avl_create(ra.guid_to_ds_map, guid_compare,
1427 1449 sizeof (guid_map_entry_t),
1428 1450 offsetof(guid_map_entry_t, avlnode));
1429 1451 ra.err = zfs_onexit_add_cb(minor,
1430 1452 free_guid_map_onexit, ra.guid_to_ds_map,
1431 1453 action_handlep);
1432 1454 if (ra.err)
1433 1455 goto out;
1434 1456 } else {
1435 1457 ra.err = zfs_onexit_cb_data(minor, *action_handlep,
1436 1458 (void **)&ra.guid_to_ds_map);
1437 1459 if (ra.err)
1438 1460 goto out;
1439 1461 }
1440 1462
1441 1463 drc->drc_guid_to_ds_map = ra.guid_to_ds_map;
1442 1464 }
1443 1465
1444 1466 /*
1445 1467 * Read records and process them.
1446 1468 */
1447 1469 pcksum = ra.cksum;
1448 1470 while (ra.err == 0 &&
1449 1471 NULL != (drr = restore_read(&ra, sizeof (*drr)))) {
1450 1472 if (issig(JUSTLOOKING) && issig(FORREAL)) {
1451 1473 ra.err = EINTR;
1452 1474 goto out;
1453 1475 }
1454 1476
1455 1477 if (ra.byteswap)
1456 1478 backup_byteswap(drr);
1457 1479
1458 1480 switch (drr->drr_type) {
1459 1481 case DRR_OBJECT:
1460 1482 {
1461 1483 /*
1462 1484 * We need to make a copy of the record header,
1463 1485 * because restore_{object,write} may need to
1464 1486 * restore_read(), which will invalidate drr.
1465 1487 */
1466 1488 struct drr_object drro = drr->drr_u.drr_object;
1467 1489 ra.err = restore_object(&ra, os, &drro);
1468 1490 break;
1469 1491 }
1470 1492 case DRR_FREEOBJECTS:
1471 1493 {
1472 1494 struct drr_freeobjects drrfo =
1473 1495 drr->drr_u.drr_freeobjects;
1474 1496 ra.err = restore_freeobjects(&ra, os, &drrfo);
1475 1497 break;
1476 1498 }
1477 1499 case DRR_WRITE:
1478 1500 {
1479 1501 struct drr_write drrw = drr->drr_u.drr_write;
1480 1502 ra.err = restore_write(&ra, os, &drrw);
1481 1503 break;
1482 1504 }
1483 1505 case DRR_WRITE_BYREF:
1484 1506 {
1485 1507 struct drr_write_byref drrwbr =
1486 1508 drr->drr_u.drr_write_byref;
1487 1509 ra.err = restore_write_byref(&ra, os, &drrwbr);
1488 1510 break;
1489 1511 }
1490 1512 case DRR_FREE:
1491 1513 {
1492 1514 struct drr_free drrf = drr->drr_u.drr_free;
1493 1515 ra.err = restore_free(&ra, os, &drrf);
1494 1516 break;
1495 1517 }
1496 1518 case DRR_END:
1497 1519 {
1498 1520 struct drr_end drre = drr->drr_u.drr_end;
1499 1521 /*
1500 1522 * We compare against the *previous* checksum
1501 1523 * value, because the stored checksum is of
1502 1524 * everything before the DRR_END record.
1503 1525 */
1504 1526 if (!ZIO_CHECKSUM_EQUAL(drre.drr_checksum, pcksum))
1505 1527 ra.err = ECKSUM;
1506 1528 goto out;
1507 1529 }
1508 1530 case DRR_SPILL:
1509 1531 {
1510 1532 struct drr_spill drrs = drr->drr_u.drr_spill;
1511 1533 ra.err = restore_spill(&ra, os, &drrs);
1512 1534 break;
1513 1535 }
1514 1536 default:
1515 1537 ra.err = EINVAL;
1516 1538 goto out;
1517 1539 }
1518 1540 pcksum = ra.cksum;
1519 1541 }
1520 1542 ASSERT(ra.err != 0);
1521 1543
1522 1544 out:
1523 1545 if ((featureflags & DMU_BACKUP_FEATURE_DEDUP) && (cleanup_fd != -1))
1524 1546 zfs_onexit_fd_rele(cleanup_fd);
1525 1547
1526 1548 if (ra.err != 0) {
1527 1549 /*
1528 1550 * destroy what we created, so we don't leave it in the
1529 1551 * inconsistent restoring state.
1530 1552 */
1531 1553 txg_wait_synced(drc->drc_real_ds->ds_dir->dd_pool, 0);
1532 1554
1533 1555 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1534 1556 B_FALSE);
1535 1557 if (drc->drc_real_ds != drc->drc_logical_ds) {
1536 1558 mutex_exit(&drc->drc_logical_ds->ds_recvlock);
1537 1559 dsl_dataset_rele(drc->drc_logical_ds, dmu_recv_tag);
1538 1560 }
1539 1561 }
|
↓ open down ↓ |
802 lines elided |
↑ open up ↑ |
1540 1562
1541 1563 kmem_free(ra.buf, ra.bufsize);
1542 1564 *voffp = ra.voff;
1543 1565 return (ra.err);
1544 1566 }
1545 1567
1546 1568 struct recvendsyncarg {
1547 1569 char *tosnap;
1548 1570 uint64_t creation_time;
1549 1571 uint64_t toguid;
1572 + boolean_t is_new;
1573 + cred_t *cr;
1550 1574 };
1551 1575
1552 1576 static int
1553 1577 recv_end_check(void *arg1, void *arg2, dmu_tx_t *tx)
1554 1578 {
1555 1579 dsl_dataset_t *ds = arg1;
1556 1580 struct recvendsyncarg *resa = arg2;
1557 1581
1558 - return (dsl_dataset_snapshot_check(ds, resa->tosnap, tx));
1582 + if (resa->is_new) {
1583 + /* re-check the filesystem limit now that recv is complete */
1584 + int err;
1585 +
1586 + err = dsl_dir_fscount_check(ds->ds_dir, 1, NULL, resa->cr);
1587 + if (err != 0)
1588 + return (err);
1589 + }
1590 +
1591 + return (dsl_dataset_snapshot_check(ds, resa->tosnap, 1, tx, resa->cr));
1559 1592 }
1560 1593
1561 1594 static void
1562 1595 recv_end_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1563 1596 {
1564 1597 dsl_dataset_t *ds = arg1;
1565 1598 struct recvendsyncarg *resa = arg2;
1566 1599
1600 + if (resa->is_new) {
1601 + /* update the filesystem counts */
1602 + dsl_dir_fscount_adjust(ds->ds_dir->dd_parent, tx, 1, B_TRUE);
1603 + }
1604 +
1567 1605 dsl_dataset_snapshot_sync(ds, resa->tosnap, tx);
1568 1606
1569 1607 /* set snapshot's creation time and guid */
1570 1608 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1571 1609 ds->ds_prev->ds_phys->ds_creation_time = resa->creation_time;
1572 1610 ds->ds_prev->ds_phys->ds_guid = resa->toguid;
1573 1611 ds->ds_prev->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1574 1612
1575 1613 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1576 1614 ds->ds_phys->ds_flags &= ~DS_FLAG_INCONSISTENT;
1577 1615 spa_history_log_internal_ds(ds, "finished receiving", tx, "");
1578 1616 }
1579 1617
1580 1618 static int
1581 1619 add_ds_to_guidmap(avl_tree_t *guid_map, dsl_dataset_t *ds)
1582 1620 {
1583 1621 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1584 1622 uint64_t snapobj = ds->ds_phys->ds_prev_snap_obj;
1585 1623 dsl_dataset_t *snapds;
1586 1624 guid_map_entry_t *gmep;
1587 1625 int err;
1588 1626
1589 1627 ASSERT(guid_map != NULL);
1590 1628
1591 1629 rw_enter(&dp->dp_config_rwlock, RW_READER);
1592 1630 err = dsl_dataset_hold_obj(dp, snapobj, guid_map, &snapds);
1593 1631 if (err == 0) {
1594 1632 gmep = kmem_alloc(sizeof (guid_map_entry_t), KM_SLEEP);
1595 1633 gmep->guid = snapds->ds_phys->ds_guid;
1596 1634 gmep->gme_ds = snapds;
1597 1635 avl_add(guid_map, gmep);
1598 1636 }
1599 1637
1600 1638 rw_exit(&dp->dp_config_rwlock);
1601 1639 return (err);
1602 1640 }
1603 1641
1604 1642 static int
1605 1643 dmu_recv_existing_end(dmu_recv_cookie_t *drc)
1606 1644 {
1607 1645 struct recvendsyncarg resa;
1608 1646 dsl_dataset_t *ds = drc->drc_logical_ds;
1609 1647 int err, myerr;
1610 1648
1611 1649 if (dsl_dataset_tryown(ds, FALSE, dmu_recv_tag)) {
1612 1650 err = dsl_dataset_clone_swap(drc->drc_real_ds, ds,
1613 1651 drc->drc_force);
1614 1652 if (err)
1615 1653 goto out;
1616 1654 } else {
|
↓ open down ↓ |
40 lines elided |
↑ open up ↑ |
1617 1655 mutex_exit(&ds->ds_recvlock);
1618 1656 dsl_dataset_rele(ds, dmu_recv_tag);
1619 1657 (void) dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag,
1620 1658 B_FALSE);
1621 1659 return (EBUSY);
1622 1660 }
1623 1661
1624 1662 resa.creation_time = drc->drc_drrb->drr_creation_time;
1625 1663 resa.toguid = drc->drc_drrb->drr_toguid;
1626 1664 resa.tosnap = drc->drc_tosnap;
1665 + resa.is_new = B_FALSE;
1666 + resa.cr = CRED();
1627 1667
1628 1668 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1629 1669 recv_end_check, recv_end_sync, ds, &resa, 3);
1630 1670 if (err) {
1631 1671 /* swap back */
1632 1672 (void) dsl_dataset_clone_swap(drc->drc_real_ds, ds, B_TRUE);
1633 1673 }
1634 1674
1635 1675 out:
1636 1676 mutex_exit(&ds->ds_recvlock);
1637 1677 if (err == 0 && drc->drc_guid_to_ds_map != NULL)
1638 1678 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1639 1679 dsl_dataset_disown(ds, dmu_recv_tag);
1640 1680 myerr = dsl_dataset_destroy(drc->drc_real_ds, dmu_recv_tag, B_FALSE);
1641 1681 ASSERT0(myerr);
1642 1682 return (err);
1643 1683 }
1644 1684
1645 1685 static int
1646 1686 dmu_recv_new_end(dmu_recv_cookie_t *drc)
1647 1687 {
1648 1688 struct recvendsyncarg resa;
1649 1689 dsl_dataset_t *ds = drc->drc_logical_ds;
1650 1690 int err;
1651 1691
|
↓ open down ↓ |
15 lines elided |
↑ open up ↑ |
1652 1692 /*
1653 1693 * XXX hack; seems the ds is still dirty and dsl_pool_zil_clean()
1654 1694 * expects it to have a ds_user_ptr (and zil), but clone_swap()
1655 1695 * can close it.
1656 1696 */
1657 1697 txg_wait_synced(ds->ds_dir->dd_pool, 0);
1658 1698
1659 1699 resa.creation_time = drc->drc_drrb->drr_creation_time;
1660 1700 resa.toguid = drc->drc_drrb->drr_toguid;
1661 1701 resa.tosnap = drc->drc_tosnap;
1702 + resa.is_new = B_TRUE;
1703 + resa.cr = CRED();
1662 1704
1663 1705 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1664 1706 recv_end_check, recv_end_sync, ds, &resa, 3);
1665 1707 if (err) {
1666 1708 /* clean up the fs we just recv'd into */
1667 1709 (void) dsl_dataset_destroy(ds, dmu_recv_tag, B_FALSE);
1668 1710 } else {
1669 1711 if (drc->drc_guid_to_ds_map != NULL)
1670 1712 (void) add_ds_to_guidmap(drc->drc_guid_to_ds_map, ds);
1671 1713 /* release the hold from dmu_recv_begin */
1672 1714 dsl_dataset_disown(ds, dmu_recv_tag);
1673 1715 }
1674 1716 return (err);
1675 1717 }
1676 1718
1677 1719 int
1678 1720 dmu_recv_end(dmu_recv_cookie_t *drc)
1679 1721 {
1680 1722 if (drc->drc_logical_ds != drc->drc_real_ds)
1681 1723 return (dmu_recv_existing_end(drc));
1682 1724 else
1683 1725 return (dmu_recv_new_end(drc));
1684 1726 }
|
↓ open down ↓ |
13 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX