1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
26 #include <sys/dmu.h>
27 #include <sys/dmu_impl.h>
28 #include <sys/dmu_tx.h>
29 #include <sys/dbuf.h>
30 #include <sys/dnode.h>
31 #include <sys/zfs_context.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dmu_traverse.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dsl_prop.h>
39 #include <sys/dmu_zfetch.h>
40 #include <sys/zfs_ioctl.h>
41 #include <sys/zap.h>
42 #include <sys/zio_checksum.h>
43 #ifdef _KERNEL
44 #include <sys/vmsystm.h>
45 #include <sys/zfs_znode.h>
46 #endif
47
48 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = {
49 { byteswap_uint8_array, TRUE, "unallocated" },
50 { zap_byteswap, TRUE, "object directory" },
51 { byteswap_uint64_array, TRUE, "object array" },
52 { byteswap_uint8_array, TRUE, "packed nvlist" },
53 { byteswap_uint64_array, TRUE, "packed nvlist size" },
54 { byteswap_uint64_array, TRUE, "bplist" },
55 { byteswap_uint64_array, TRUE, "bplist header" },
56 { byteswap_uint64_array, TRUE, "SPA space map header" },
57 { byteswap_uint64_array, TRUE, "SPA space map" },
58 { byteswap_uint64_array, TRUE, "ZIL intent log" },
59 { dnode_buf_byteswap, TRUE, "DMU dnode" },
60 { dmu_objset_byteswap, TRUE, "DMU objset" },
61 { byteswap_uint64_array, TRUE, "DSL directory" },
62 { zap_byteswap, TRUE, "DSL directory child map"},
63 { zap_byteswap, TRUE, "DSL dataset snap map" },
64 { zap_byteswap, TRUE, "DSL props" },
65 { byteswap_uint64_array, TRUE, "DSL dataset" },
66 { zfs_znode_byteswap, TRUE, "ZFS znode" },
67 { zfs_oldacl_byteswap, TRUE, "ZFS V0 ACL" },
68 { byteswap_uint8_array, FALSE, "ZFS plain file" },
69 { zap_byteswap, TRUE, "ZFS directory" },
70 { zap_byteswap, TRUE, "ZFS master node" },
71 { zap_byteswap, TRUE, "ZFS delete queue" },
72 { byteswap_uint8_array, FALSE, "zvol object" },
73 { zap_byteswap, TRUE, "zvol prop" },
74 { byteswap_uint8_array, FALSE, "other uint8[]" },
75 { byteswap_uint64_array, FALSE, "other uint64[]" },
76 { zap_byteswap, TRUE, "other ZAP" },
77 { zap_byteswap, TRUE, "persistent error log" },
78 { byteswap_uint8_array, TRUE, "SPA history" },
79 { byteswap_uint64_array, TRUE, "SPA history offsets" },
80 { zap_byteswap, TRUE, "Pool properties" },
81 { zap_byteswap, TRUE, "DSL permissions" },
82 { zfs_acl_byteswap, TRUE, "ZFS ACL" },
83 { byteswap_uint8_array, TRUE, "ZFS SYSACL" },
84 { byteswap_uint8_array, TRUE, "FUID table" },
85 { byteswap_uint64_array, TRUE, "FUID table size" },
86 { zap_byteswap, TRUE, "DSL dataset next clones"},
87 { zap_byteswap, TRUE, "scrub work queue" },
88 { zap_byteswap, TRUE, "ZFS user/group used" },
89 { zap_byteswap, TRUE, "ZFS user/group quota" },
90 { zap_byteswap, TRUE, "snapshot refcount tags"},
91 { zap_byteswap, TRUE, "DDT ZAP algorithm" },
92 { zap_byteswap, TRUE, "DDT statistics" },
93 };
94
95 int
dmu_buf_hold(objset_t * os,uint64_t object,uint64_t offset,void * tag,dmu_buf_t ** dbp)96 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset,
97 void *tag, dmu_buf_t **dbp)
98 {
99 dnode_t *dn;
100 uint64_t blkid;
101 dmu_buf_impl_t *db;
102 int err;
103
104 err = dnode_hold(os, object, FTAG, &dn);
105 if (err)
106 return (err);
107 blkid = dbuf_whichblock(dn, offset);
108 rw_enter(&dn->dn_struct_rwlock, RW_READER);
109 db = dbuf_hold(dn, blkid, tag);
110 rw_exit(&dn->dn_struct_rwlock);
111 if (db == NULL) {
112 err = EIO;
113 } else {
114 err = dbuf_read(db, NULL, DB_RF_CANFAIL);
115 if (err) {
116 dbuf_rele(db, tag);
117 db = NULL;
118 }
119 }
120
121 dnode_rele(dn, FTAG);
122 *dbp = &db->db;
123 return (err);
124 }
125
126 int
dmu_bonus_max(void)127 dmu_bonus_max(void)
128 {
129 return (DN_MAX_BONUSLEN);
130 }
131
132 int
dmu_set_bonus(dmu_buf_t * db,int newsize,dmu_tx_t * tx)133 dmu_set_bonus(dmu_buf_t *db, int newsize, dmu_tx_t *tx)
134 {
135 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode;
136
137 if (dn->dn_bonus != (dmu_buf_impl_t *)db)
138 return (EINVAL);
139 if (newsize < 0 || newsize > db->db_size)
140 return (EINVAL);
141 dnode_setbonuslen(dn, newsize, tx);
142 return (0);
143 }
144
145 /*
146 * returns ENOENT, EIO, or 0.
147 */
148 int
dmu_bonus_hold(objset_t * os,uint64_t object,void * tag,dmu_buf_t ** dbp)149 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp)
150 {
151 dnode_t *dn;
152 dmu_buf_impl_t *db;
153 int error;
154
155 error = dnode_hold(os, object, FTAG, &dn);
156 if (error)
157 return (error);
158
159 rw_enter(&dn->dn_struct_rwlock, RW_READER);
160 if (dn->dn_bonus == NULL) {
161 rw_exit(&dn->dn_struct_rwlock);
162 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
163 if (dn->dn_bonus == NULL)
164 dbuf_create_bonus(dn);
165 }
166 db = dn->dn_bonus;
167 rw_exit(&dn->dn_struct_rwlock);
168
169 /* as long as the bonus buf is held, the dnode will be held */
170 if (refcount_add(&db->db_holds, tag) == 1)
171 VERIFY(dnode_add_ref(dn, db));
172
173 dnode_rele(dn, FTAG);
174
175 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED));
176
177 *dbp = &db->db;
178 return (0);
179 }
180
181 /*
182 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces
183 * to take a held dnode rather than <os, object> -- the lookup is wasteful,
184 * and can induce severe lock contention when writing to several files
185 * whose dnodes are in the same block.
186 */
187 static int
dmu_buf_hold_array_by_dnode(dnode_t * dn,uint64_t offset,uint64_t length,int read,void * tag,int * numbufsp,dmu_buf_t *** dbpp,uint32_t flags)188 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length,
189 int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags)
190 {
191 dsl_pool_t *dp = NULL;
192 dmu_buf_t **dbp;
193 uint64_t blkid, nblks, i;
194 uint32_t dbuf_flags;
195 int err;
196 zio_t *zio;
197 hrtime_t start;
198
199 ASSERT(length <= DMU_MAX_ACCESS);
200
201 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT;
202 if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz)
203 dbuf_flags |= DB_RF_NOPREFETCH;
204
205 rw_enter(&dn->dn_struct_rwlock, RW_READER);
206 if (dn->dn_datablkshift) {
207 int blkshift = dn->dn_datablkshift;
208 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) -
209 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift;
210 } else {
211 if (offset + length > dn->dn_datablksz) {
212 zfs_panic_recover("zfs: accessing past end of object "
213 "%llx/%llx (size=%u access=%llu+%llu)",
214 (longlong_t)dn->dn_objset->
215 os_dsl_dataset->ds_object,
216 (longlong_t)dn->dn_object, dn->dn_datablksz,
217 (longlong_t)offset, (longlong_t)length);
218 rw_exit(&dn->dn_struct_rwlock);
219 return (EIO);
220 }
221 nblks = 1;
222 }
223 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP);
224
225 if (dn->dn_objset->os_dsl_dataset)
226 dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool;
227 if (dp && dsl_pool_sync_context(dp))
228 start = gethrtime();
229 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL);
230 blkid = dbuf_whichblock(dn, offset);
231 for (i = 0; i < nblks; i++) {
232 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag);
233 if (db == NULL) {
234 rw_exit(&dn->dn_struct_rwlock);
235 dmu_buf_rele_array(dbp, nblks, tag);
236 zio_nowait(zio);
237 return (EIO);
238 }
239 /* initiate async i/o */
240 if (read) {
241 (void) dbuf_read(db, zio, dbuf_flags);
242 }
243 dbp[i] = &db->db;
244 }
245 rw_exit(&dn->dn_struct_rwlock);
246
247 /* wait for async i/o */
248 err = zio_wait(zio);
249 /* track read overhead when we are in sync context */
250 if (dp && dsl_pool_sync_context(dp))
251 dp->dp_read_overhead += gethrtime() - start;
252 if (err) {
253 dmu_buf_rele_array(dbp, nblks, tag);
254 return (err);
255 }
256
257 /* wait for other io to complete */
258 if (read) {
259 for (i = 0; i < nblks; i++) {
260 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i];
261 mutex_enter(&db->db_mtx);
262 while (db->db_state == DB_READ ||
263 db->db_state == DB_FILL)
264 cv_wait(&db->db_changed, &db->db_mtx);
265 if (db->db_state == DB_UNCACHED)
266 err = EIO;
267 mutex_exit(&db->db_mtx);
268 if (err) {
269 dmu_buf_rele_array(dbp, nblks, tag);
270 return (err);
271 }
272 }
273 }
274
275 *numbufsp = nblks;
276 *dbpp = dbp;
277 return (0);
278 }
279
280 static int
dmu_buf_hold_array(objset_t * os,uint64_t object,uint64_t offset,uint64_t length,int read,void * tag,int * numbufsp,dmu_buf_t *** dbpp)281 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset,
282 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
283 {
284 dnode_t *dn;
285 int err;
286
287 err = dnode_hold(os, object, FTAG, &dn);
288 if (err)
289 return (err);
290
291 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
292 numbufsp, dbpp, DMU_READ_PREFETCH);
293
294 dnode_rele(dn, FTAG);
295
296 return (err);
297 }
298
299 int
dmu_buf_hold_array_by_bonus(dmu_buf_t * db,uint64_t offset,uint64_t length,int read,void * tag,int * numbufsp,dmu_buf_t *** dbpp)300 dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset,
301 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp)
302 {
303 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode;
304 int err;
305
306 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag,
307 numbufsp, dbpp, DMU_READ_PREFETCH);
308
309 return (err);
310 }
311
312 void
dmu_buf_rele_array(dmu_buf_t ** dbp_fake,int numbufs,void * tag)313 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag)
314 {
315 int i;
316 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake;
317
318 if (numbufs == 0)
319 return;
320
321 for (i = 0; i < numbufs; i++) {
322 if (dbp[i])
323 dbuf_rele(dbp[i], tag);
324 }
325
326 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs);
327 }
328
329 void
dmu_prefetch(objset_t * os,uint64_t object,uint64_t offset,uint64_t len)330 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len)
331 {
332 dnode_t *dn;
333 uint64_t blkid;
334 int nblks, i, err;
335
336 if (zfs_prefetch_disable)
337 return;
338
339 if (len == 0) { /* they're interested in the bonus buffer */
340 dn = os->os_meta_dnode;
341
342 if (object == 0 || object >= DN_MAX_OBJECT)
343 return;
344
345 rw_enter(&dn->dn_struct_rwlock, RW_READER);
346 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t));
347 dbuf_prefetch(dn, blkid);
348 rw_exit(&dn->dn_struct_rwlock);
349 return;
350 }
351
352 /*
353 * XXX - Note, if the dnode for the requested object is not
354 * already cached, we will do a *synchronous* read in the
355 * dnode_hold() call. The same is true for any indirects.
356 */
357 err = dnode_hold(os, object, FTAG, &dn);
358 if (err != 0)
359 return;
360
361 rw_enter(&dn->dn_struct_rwlock, RW_READER);
362 if (dn->dn_datablkshift) {
363 int blkshift = dn->dn_datablkshift;
364 nblks = (P2ROUNDUP(offset+len, 1<<blkshift) -
365 P2ALIGN(offset, 1<<blkshift)) >> blkshift;
366 } else {
367 nblks = (offset < dn->dn_datablksz);
368 }
369
370 if (nblks != 0) {
371 blkid = dbuf_whichblock(dn, offset);
372 for (i = 0; i < nblks; i++)
373 dbuf_prefetch(dn, blkid+i);
374 }
375
376 rw_exit(&dn->dn_struct_rwlock);
377
378 dnode_rele(dn, FTAG);
379 }
380
381 /*
382 * Get the next "chunk" of file data to free. We traverse the file from
383 * the end so that the file gets shorter over time (if we crashes in the
384 * middle, this will leave us in a better state). We find allocated file
385 * data by simply searching the allocated level 1 indirects.
386 */
387 static int
get_next_chunk(dnode_t * dn,uint64_t * start,uint64_t limit)388 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit)
389 {
390 uint64_t len = *start - limit;
391 uint64_t blkcnt = 0;
392 uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1));
393 uint64_t iblkrange =
394 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT);
395
396 ASSERT(limit <= *start);
397
398 if (len <= iblkrange * maxblks) {
399 *start = limit;
400 return (0);
401 }
402 ASSERT(ISP2(iblkrange));
403
404 while (*start > limit && blkcnt < maxblks) {
405 int err;
406
407 /* find next allocated L1 indirect */
408 err = dnode_next_offset(dn,
409 DNODE_FIND_BACKWARDS, start, 2, 1, 0);
410
411 /* if there are no more, then we are done */
412 if (err == ESRCH) {
413 *start = limit;
414 return (0);
415 } else if (err) {
416 return (err);
417 }
418 blkcnt += 1;
419
420 /* reset offset to end of "next" block back */
421 *start = P2ALIGN(*start, iblkrange);
422 if (*start <= limit)
423 *start = limit;
424 else
425 *start -= 1;
426 }
427 return (0);
428 }
429
430 static int
dmu_free_long_range_impl(objset_t * os,dnode_t * dn,uint64_t offset,uint64_t length,boolean_t free_dnode)431 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset,
432 uint64_t length, boolean_t free_dnode)
433 {
434 dmu_tx_t *tx;
435 uint64_t object_size, start, end, len;
436 boolean_t trunc = (length == DMU_OBJECT_END);
437 int align, err;
438
439 align = 1 << dn->dn_datablkshift;
440 ASSERT(align > 0);
441 object_size = align == 1 ? dn->dn_datablksz :
442 (dn->dn_maxblkid + 1) << dn->dn_datablkshift;
443
444 end = offset + length;
445 if (trunc || end > object_size)
446 end = object_size;
447 if (end <= offset)
448 return (0);
449 length = end - offset;
450
451 while (length) {
452 start = end;
453 /* assert(offset <= start) */
454 err = get_next_chunk(dn, &start, offset);
455 if (err)
456 return (err);
457 len = trunc ? DMU_OBJECT_END : end - start;
458
459 tx = dmu_tx_create(os);
460 dmu_tx_hold_free(tx, dn->dn_object, start, len);
461 err = dmu_tx_assign(tx, TXG_WAIT);
462 if (err) {
463 dmu_tx_abort(tx);
464 return (err);
465 }
466
467 dnode_free_range(dn, start, trunc ? -1 : len, tx);
468
469 if (start == 0 && free_dnode) {
470 ASSERT(trunc);
471 dnode_free(dn, tx);
472 }
473
474 length -= end - start;
475
476 dmu_tx_commit(tx);
477 end = start;
478 }
479 return (0);
480 }
481
482 int
dmu_free_long_range(objset_t * os,uint64_t object,uint64_t offset,uint64_t length)483 dmu_free_long_range(objset_t *os, uint64_t object,
484 uint64_t offset, uint64_t length)
485 {
486 dnode_t *dn;
487 int err;
488
489 err = dnode_hold(os, object, FTAG, &dn);
490 if (err != 0)
491 return (err);
492 err = dmu_free_long_range_impl(os, dn, offset, length, FALSE);
493 dnode_rele(dn, FTAG);
494 return (err);
495 }
496
497 int
dmu_free_object(objset_t * os,uint64_t object)498 dmu_free_object(objset_t *os, uint64_t object)
499 {
500 dnode_t *dn;
501 dmu_tx_t *tx;
502 int err;
503
504 err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED,
505 FTAG, &dn);
506 if (err != 0)
507 return (err);
508 if (dn->dn_nlevels == 1) {
509 tx = dmu_tx_create(os);
510 dmu_tx_hold_bonus(tx, object);
511 dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END);
512 err = dmu_tx_assign(tx, TXG_WAIT);
513 if (err == 0) {
514 dnode_free_range(dn, 0, DMU_OBJECT_END, tx);
515 dnode_free(dn, tx);
516 dmu_tx_commit(tx);
517 } else {
518 dmu_tx_abort(tx);
519 }
520 } else {
521 err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE);
522 }
523 dnode_rele(dn, FTAG);
524 return (err);
525 }
526
527 int
dmu_free_range(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,dmu_tx_t * tx)528 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset,
529 uint64_t size, dmu_tx_t *tx)
530 {
531 dnode_t *dn;
532 int err = dnode_hold(os, object, FTAG, &dn);
533 if (err)
534 return (err);
535 ASSERT(offset < UINT64_MAX);
536 ASSERT(size == -1ULL || size <= UINT64_MAX - offset);
537 dnode_free_range(dn, offset, size, tx);
538 dnode_rele(dn, FTAG);
539 return (0);
540 }
541
542 int
dmu_read(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,void * buf,uint32_t flags)543 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
544 void *buf, uint32_t flags)
545 {
546 dnode_t *dn;
547 dmu_buf_t **dbp;
548 int numbufs, err;
549
550 err = dnode_hold(os, object, FTAG, &dn);
551 if (err)
552 return (err);
553
554 /*
555 * Deal with odd block sizes, where there can't be data past the first
556 * block. If we ever do the tail block optimization, we will need to
557 * handle that here as well.
558 */
559 if (dn->dn_maxblkid == 0) {
560 int newsz = offset > dn->dn_datablksz ? 0 :
561 MIN(size, dn->dn_datablksz - offset);
562 bzero((char *)buf + newsz, size - newsz);
563 size = newsz;
564 }
565
566 while (size > 0) {
567 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2);
568 int i;
569
570 /*
571 * NB: we could do this block-at-a-time, but it's nice
572 * to be reading in parallel.
573 */
574 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen,
575 TRUE, FTAG, &numbufs, &dbp, flags);
576 if (err)
577 break;
578
579 for (i = 0; i < numbufs; i++) {
580 int tocpy;
581 int bufoff;
582 dmu_buf_t *db = dbp[i];
583
584 ASSERT(size > 0);
585
586 bufoff = offset - db->db_offset;
587 tocpy = (int)MIN(db->db_size - bufoff, size);
588
589 bcopy((char *)db->db_data + bufoff, buf, tocpy);
590
591 offset += tocpy;
592 size -= tocpy;
593 buf = (char *)buf + tocpy;
594 }
595 dmu_buf_rele_array(dbp, numbufs, FTAG);
596 }
597 dnode_rele(dn, FTAG);
598 return (err);
599 }
600
601 void
dmu_write(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,const void * buf,dmu_tx_t * tx)602 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
603 const void *buf, dmu_tx_t *tx)
604 {
605 dmu_buf_t **dbp;
606 int numbufs, i;
607
608 if (size == 0)
609 return;
610
611 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
612 FALSE, FTAG, &numbufs, &dbp));
613
614 for (i = 0; i < numbufs; i++) {
615 int tocpy;
616 int bufoff;
617 dmu_buf_t *db = dbp[i];
618
619 ASSERT(size > 0);
620
621 bufoff = offset - db->db_offset;
622 tocpy = (int)MIN(db->db_size - bufoff, size);
623
624 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
625
626 if (tocpy == db->db_size)
627 dmu_buf_will_fill(db, tx);
628 else
629 dmu_buf_will_dirty(db, tx);
630
631 bcopy(buf, (char *)db->db_data + bufoff, tocpy);
632
633 if (tocpy == db->db_size)
634 dmu_buf_fill_done(db, tx);
635
636 offset += tocpy;
637 size -= tocpy;
638 buf = (char *)buf + tocpy;
639 }
640 dmu_buf_rele_array(dbp, numbufs, FTAG);
641 }
642
643 void
dmu_prealloc(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,dmu_tx_t * tx)644 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
645 dmu_tx_t *tx)
646 {
647 dmu_buf_t **dbp;
648 int numbufs, i;
649
650 if (size == 0)
651 return;
652
653 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size,
654 FALSE, FTAG, &numbufs, &dbp));
655
656 for (i = 0; i < numbufs; i++) {
657 dmu_buf_t *db = dbp[i];
658
659 dmu_buf_will_not_fill(db, tx);
660 }
661 dmu_buf_rele_array(dbp, numbufs, FTAG);
662 }
663
664 /*
665 * DMU support for xuio
666 */
667 kstat_t *xuio_ksp = NULL;
668
669 int
dmu_xuio_init(xuio_t * xuio,int nblk)670 dmu_xuio_init(xuio_t *xuio, int nblk)
671 {
672 dmu_xuio_t *priv;
673 uio_t *uio = &xuio->xu_uio;
674
675 uio->uio_iovcnt = nblk;
676 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP);
677
678 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP);
679 priv->cnt = nblk;
680 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP);
681 priv->iovp = uio->uio_iov;
682 XUIO_XUZC_PRIV(xuio) = priv;
683
684 #ifdef PORT_SOLARIS
685 if (XUIO_XUZC_RW(xuio) == UIO_READ)
686 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk);
687 else
688 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk);
689 #endif
690
691 return (0);
692 }
693
694 void
dmu_xuio_fini(xuio_t * xuio)695 dmu_xuio_fini(xuio_t *xuio)
696 {
697 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
698 int nblk = priv->cnt;
699
700 kmem_free(priv->iovp, nblk * sizeof (iovec_t));
701 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *));
702 kmem_free(priv, sizeof (dmu_xuio_t));
703
704 #ifdef PORT_SOLARIS
705 if (XUIO_XUZC_RW(xuio) == UIO_READ)
706 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk);
707 else
708 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk);
709 #endif
710 }
711
712 /*
713 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf }
714 * and increase priv->next by 1.
715 */
716 int
dmu_xuio_add(xuio_t * xuio,arc_buf_t * abuf,offset_t off,size_t n)717 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n)
718 {
719 struct iovec *iov;
720 uio_t *uio = &xuio->xu_uio;
721 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
722 int i = priv->next++;
723
724 ASSERT(i < priv->cnt);
725 ASSERT(off + n <= arc_buf_size(abuf));
726 iov = uio->uio_iov + i;
727 iov->iov_base = (char *)abuf->b_data + off;
728 iov->iov_len = n;
729 priv->bufs[i] = abuf;
730 return (0);
731 }
732
733 int
dmu_xuio_cnt(xuio_t * xuio)734 dmu_xuio_cnt(xuio_t *xuio)
735 {
736 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
737 return (priv->cnt);
738 }
739
740 arc_buf_t *
dmu_xuio_arcbuf(xuio_t * xuio,int i)741 dmu_xuio_arcbuf(xuio_t *xuio, int i)
742 {
743 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
744
745 ASSERT(i < priv->cnt);
746 return (priv->bufs[i]);
747 }
748
749 void
dmu_xuio_clear(xuio_t * xuio,int i)750 dmu_xuio_clear(xuio_t *xuio, int i)
751 {
752 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio);
753
754 ASSERT(i < priv->cnt);
755 priv->bufs[i] = NULL;
756 }
757
758 #ifdef PORT_SOLARIS
759 static void
xuio_stat_init(void)760 xuio_stat_init(void)
761 {
762 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc",
763 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t),
764 KSTAT_FLAG_VIRTUAL);
765 if (xuio_ksp != NULL) {
766 xuio_ksp->ks_data = &xuio_stats;
767 kstat_install(xuio_ksp);
768 }
769 }
770
771 static void
xuio_stat_fini(void)772 xuio_stat_fini(void)
773 {
774 if (xuio_ksp != NULL) {
775 kstat_delete(xuio_ksp);
776 xuio_ksp = NULL;
777 }
778 }
779 #endif
780
781 void
xuio_stat_wbuf_copied()782 xuio_stat_wbuf_copied()
783 {
784 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
785 }
786
787 void
xuio_stat_wbuf_nocopy()788 xuio_stat_wbuf_nocopy()
789 {
790 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy);
791 }
792
793 #ifdef _KERNEL
794 int
dmu_read_uio(objset_t * os,uint64_t object,uio_t * uio,uint64_t size)795 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size)
796 {
797 dmu_buf_t **dbp;
798 int numbufs, i, err;
799 xuio_t *xuio = NULL;
800
801 /*
802 * NB: we could do this block-at-a-time, but it's nice
803 * to be reading in parallel.
804 */
805 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG,
806 &numbufs, &dbp);
807 if (err)
808 return (err);
809
810 #ifndef __NetBSD__ /* XXX xuio */
811 if (uio->uio_extflg == UIO_XUIO)
812 xuio = (xuio_t *)uio;
813 #endif
814
815 for (i = 0; i < numbufs; i++) {
816 int tocpy;
817 int bufoff;
818 dmu_buf_t *db = dbp[i];
819
820 ASSERT(size > 0);
821
822 bufoff = uio->uio_loffset - db->db_offset;
823 tocpy = (int)MIN(db->db_size - bufoff, size);
824
825 if (xuio) {
826 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db;
827 arc_buf_t *dbuf_abuf = dbi->db_buf;
828 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi);
829 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy);
830 if (!err) {
831 uio->uio_resid -= tocpy;
832 uio->uio_loffset += tocpy;
833 }
834
835 if (abuf == dbuf_abuf)
836 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy);
837 else
838 XUIOSTAT_BUMP(xuiostat_rbuf_copied);
839 } else {
840 err = uiomove((char *)db->db_data + bufoff, tocpy,
841 UIO_READ, uio);
842 }
843 if (err)
844 break;
845
846 size -= tocpy;
847 }
848 dmu_buf_rele_array(dbp, numbufs, FTAG);
849
850 return (err);
851 }
852
853 int
dmu_write_uio(objset_t * os,uint64_t object,uio_t * uio,uint64_t size,dmu_tx_t * tx)854 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size,
855 dmu_tx_t *tx)
856 {
857 dmu_buf_t **dbp;
858 int numbufs, i;
859 int err = 0;
860
861 if (size == 0)
862 return (0);
863
864 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size,
865 FALSE, FTAG, &numbufs, &dbp);
866 if (err)
867 return (err);
868
869 for (i = 0; i < numbufs; i++) {
870 int tocpy;
871 int bufoff;
872 dmu_buf_t *db = dbp[i];
873
874 ASSERT(size > 0);
875
876 bufoff = uio->uio_loffset - db->db_offset;
877 tocpy = (int)MIN(db->db_size - bufoff, size);
878
879 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
880
881 if (tocpy == db->db_size)
882 dmu_buf_will_fill(db, tx);
883 else
884 dmu_buf_will_dirty(db, tx);
885
886 /*
887 * XXX uiomove could block forever (eg. nfs-backed
888 * pages). There needs to be a uiolockdown() function
889 * to lock the pages in memory, so that uiomove won't
890 * block.
891 */
892 err = uiomove((char *)db->db_data + bufoff, tocpy,
893 UIO_WRITE, uio);
894
895 if (tocpy == db->db_size)
896 dmu_buf_fill_done(db, tx);
897
898 if (err)
899 break;
900
901 size -= tocpy;
902 }
903 dmu_buf_rele_array(dbp, numbufs, FTAG);
904 return (err);
905 }
906
907 #ifndef __NetBSD__
908 int
dmu_write_pages(objset_t * os,uint64_t object,uint64_t offset,uint64_t size,page_t * pp,dmu_tx_t * tx)909 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size,
910 page_t *pp, dmu_tx_t *tx)
911 {
912 dmu_buf_t **dbp;
913 int numbufs, i;
914 int err;
915
916 if (size == 0)
917 return (0);
918
919 err = dmu_buf_hold_array(os, object, offset, size,
920 FALSE, FTAG, &numbufs, &dbp);
921 if (err)
922 return (err);
923
924 for (i = 0; i < numbufs; i++) {
925 int tocpy, copied, thiscpy;
926 int bufoff;
927 dmu_buf_t *db = dbp[i];
928 caddr_t va;
929
930 ASSERT(size > 0);
931 ASSERT3U(db->db_size, >=, PAGESIZE);
932
933 bufoff = offset - db->db_offset;
934 tocpy = (int)MIN(db->db_size - bufoff, size);
935
936 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size);
937
938 if (tocpy == db->db_size)
939 dmu_buf_will_fill(db, tx);
940 else
941 dmu_buf_will_dirty(db, tx);
942
943 for (copied = 0; copied < tocpy; copied += PAGESIZE) {
944 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff);
945 thiscpy = MIN(PAGESIZE, tocpy - copied);
946 va = zfs_map_page(pp, S_READ);
947 bcopy(va, (char *)db->db_data + bufoff, thiscpy);
948 zfs_unmap_page(pp, va);
949 pp = pp->p_next;
950 bufoff += PAGESIZE;
951 }
952
953 if (tocpy == db->db_size)
954 dmu_buf_fill_done(db, tx);
955
956 offset += tocpy;
957 size -= tocpy;
958 }
959 dmu_buf_rele_array(dbp, numbufs, FTAG);
960 return (err);
961 }
962 #endif /* __NetBSD__ */
963 #endif
964
965 /*
966 * Allocate a loaned anonymous arc buffer.
967 */
968 arc_buf_t *
dmu_request_arcbuf(dmu_buf_t * handle,int size)969 dmu_request_arcbuf(dmu_buf_t *handle, int size)
970 {
971 dnode_t *dn = ((dmu_buf_impl_t *)handle)->db_dnode;
972
973 return (arc_loan_buf(dn->dn_objset->os_spa, size));
974 }
975
976 /*
977 * Free a loaned arc buffer.
978 */
979 void
dmu_return_arcbuf(arc_buf_t * buf)980 dmu_return_arcbuf(arc_buf_t *buf)
981 {
982 arc_return_buf(buf, FTAG);
983 VERIFY(arc_buf_remove_ref(buf, FTAG) == 1);
984 }
985
986 /*
987 * When possible directly assign passed loaned arc buffer to a dbuf.
988 * If this is not possible copy the contents of passed arc buf via
989 * dmu_write().
990 */
991 void
dmu_assign_arcbuf(dmu_buf_t * handle,uint64_t offset,arc_buf_t * buf,dmu_tx_t * tx)992 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf,
993 dmu_tx_t *tx)
994 {
995 dnode_t *dn = ((dmu_buf_impl_t *)handle)->db_dnode;
996 dmu_buf_impl_t *db;
997 uint32_t blksz = (uint32_t)arc_buf_size(buf);
998 uint64_t blkid;
999
1000 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1001 blkid = dbuf_whichblock(dn, offset);
1002 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL);
1003 rw_exit(&dn->dn_struct_rwlock);
1004
1005 if (offset == db->db.db_offset && blksz == db->db.db_size) {
1006 dbuf_assign_arcbuf(db, buf, tx);
1007 dbuf_rele(db, FTAG);
1008 } else {
1009 dbuf_rele(db, FTAG);
1010 dmu_write(dn->dn_objset, dn->dn_object, offset, blksz,
1011 buf->b_data, tx);
1012 dmu_return_arcbuf(buf);
1013 XUIOSTAT_BUMP(xuiostat_wbuf_copied);
1014 }
1015 }
1016
1017 typedef struct {
1018 dbuf_dirty_record_t *dsa_dr;
1019 dmu_sync_cb_t *dsa_done;
1020 zgd_t *dsa_zgd;
1021 dmu_tx_t *dsa_tx;
1022 } dmu_sync_arg_t;
1023
1024 /* ARGSUSED */
1025 static void
dmu_sync_ready(zio_t * zio,arc_buf_t * buf,void * varg)1026 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg)
1027 {
1028 dmu_sync_arg_t *dsa = varg;
1029 dmu_buf_t *db = dsa->dsa_zgd->zgd_db;
1030 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode;
1031 blkptr_t *bp = zio->io_bp;
1032
1033 if (zio->io_error == 0) {
1034 if (BP_IS_HOLE(bp)) {
1035 /*
1036 * A block of zeros may compress to a hole, but the
1037 * block size still needs to be known for replay.
1038 */
1039 BP_SET_LSIZE(bp, db->db_size);
1040 } else {
1041 ASSERT(BP_GET_TYPE(bp) == dn->dn_type);
1042 ASSERT(BP_GET_LEVEL(bp) == 0);
1043 bp->blk_fill = 1;
1044 }
1045 }
1046 }
1047
1048 static void
dmu_sync_late_arrival_ready(zio_t * zio)1049 dmu_sync_late_arrival_ready(zio_t *zio)
1050 {
1051 dmu_sync_ready(zio, NULL, zio->io_private);
1052 }
1053
1054 /* ARGSUSED */
1055 static void
dmu_sync_done(zio_t * zio,arc_buf_t * buf,void * varg)1056 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg)
1057 {
1058 dmu_sync_arg_t *dsa = varg;
1059 dbuf_dirty_record_t *dr = dsa->dsa_dr;
1060 dmu_buf_impl_t *db = dr->dr_dbuf;
1061
1062 mutex_enter(&db->db_mtx);
1063 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC);
1064 if (zio->io_error == 0) {
1065 dr->dt.dl.dr_overridden_by = *zio->io_bp;
1066 dr->dt.dl.dr_override_state = DR_OVERRIDDEN;
1067 dr->dt.dl.dr_copies = zio->io_prop.zp_copies;
1068 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by))
1069 BP_ZERO(&dr->dt.dl.dr_overridden_by);
1070 } else {
1071 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN;
1072 }
1073 cv_broadcast(&db->db_changed);
1074 mutex_exit(&db->db_mtx);
1075
1076 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1077
1078 kmem_free(dsa, sizeof (*dsa));
1079 }
1080
1081 static void
dmu_sync_late_arrival_done(zio_t * zio)1082 dmu_sync_late_arrival_done(zio_t *zio)
1083 {
1084 blkptr_t *bp = zio->io_bp;
1085 dmu_sync_arg_t *dsa = zio->io_private;
1086
1087 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) {
1088 ASSERT(zio->io_bp->blk_birth == zio->io_txg);
1089 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa));
1090 zio_free(zio->io_spa, zio->io_txg, zio->io_bp);
1091 }
1092
1093 dmu_tx_commit(dsa->dsa_tx);
1094
1095 dsa->dsa_done(dsa->dsa_zgd, zio->io_error);
1096
1097 kmem_free(dsa, sizeof (*dsa));
1098 }
1099
1100 static int
dmu_sync_late_arrival(zio_t * pio,objset_t * os,dmu_sync_cb_t * done,zgd_t * zgd,zio_prop_t * zp,zbookmark_t * zb)1101 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd,
1102 zio_prop_t *zp, zbookmark_t *zb)
1103 {
1104 dmu_sync_arg_t *dsa;
1105 dmu_tx_t *tx;
1106
1107 tx = dmu_tx_create(os);
1108 dmu_tx_hold_space(tx, zgd->zgd_db->db_size);
1109 if (dmu_tx_assign(tx, TXG_WAIT) != 0) {
1110 dmu_tx_abort(tx);
1111 return (EIO); /* Make zl_get_data do txg_waited_synced() */
1112 }
1113
1114 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1115 dsa->dsa_dr = NULL;
1116 dsa->dsa_done = done;
1117 dsa->dsa_zgd = zgd;
1118 dsa->dsa_tx = tx;
1119
1120 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp,
1121 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp,
1122 dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa,
1123 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb));
1124
1125 return (0);
1126 }
1127
1128 /*
1129 * Intent log support: sync the block associated with db to disk.
1130 * N.B. and XXX: the caller is responsible for making sure that the
1131 * data isn't changing while dmu_sync() is writing it.
1132 *
1133 * Return values:
1134 *
1135 * EEXIST: this txg has already been synced, so there's nothing to to.
1136 * The caller should not log the write.
1137 *
1138 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do.
1139 * The caller should not log the write.
1140 *
1141 * EALREADY: this block is already in the process of being synced.
1142 * The caller should track its progress (somehow).
1143 *
1144 * EIO: could not do the I/O.
1145 * The caller should do a txg_wait_synced().
1146 *
1147 * 0: the I/O has been initiated.
1148 * The caller should log this blkptr in the done callback.
1149 * It is possible that the I/O will fail, in which case
1150 * the error will be reported to the done callback and
1151 * propagated to pio from zio_done().
1152 */
1153 int
dmu_sync(zio_t * pio,uint64_t txg,dmu_sync_cb_t * done,zgd_t * zgd)1154 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd)
1155 {
1156 blkptr_t *bp = zgd->zgd_bp;
1157 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db;
1158 objset_t *os = db->db_objset;
1159 dsl_dataset_t *ds = os->os_dsl_dataset;
1160 dbuf_dirty_record_t *dr;
1161 dmu_sync_arg_t *dsa;
1162 zbookmark_t zb;
1163 zio_prop_t zp;
1164
1165 ASSERT(pio != NULL);
1166 ASSERT(BP_IS_HOLE(bp));
1167 ASSERT(txg != 0);
1168
1169 SET_BOOKMARK(&zb, ds->ds_object,
1170 db->db.db_object, db->db_level, db->db_blkid);
1171
1172 dmu_write_policy(os, db->db_dnode, db->db_level, WP_DMU_SYNC, &zp);
1173
1174 /*
1175 * If we're frozen (running ziltest), we always need to generate a bp.
1176 */
1177 if (txg > spa_freeze_txg(os->os_spa))
1178 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1179
1180 /*
1181 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf()
1182 * and us. If we determine that this txg is not yet syncing,
1183 * but it begins to sync a moment later, that's OK because the
1184 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx.
1185 */
1186 mutex_enter(&db->db_mtx);
1187
1188 if (txg <= spa_last_synced_txg(os->os_spa)) {
1189 /*
1190 * This txg has already synced. There's nothing to do.
1191 */
1192 mutex_exit(&db->db_mtx);
1193 return (EEXIST);
1194 }
1195
1196 if (txg <= spa_syncing_txg(os->os_spa)) {
1197 /*
1198 * This txg is currently syncing, so we can't mess with
1199 * the dirty record anymore; just write a new log block.
1200 */
1201 mutex_exit(&db->db_mtx);
1202 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb));
1203 }
1204
1205 dr = db->db_last_dirty;
1206 while (dr && dr->dr_txg != txg)
1207 dr = dr->dr_next;
1208
1209 if (dr == NULL) {
1210 /*
1211 * There's no dr for this dbuf, so it must have been freed.
1212 * There's no need to log writes to freed blocks, so we're done.
1213 */
1214 mutex_exit(&db->db_mtx);
1215 return (ENOENT);
1216 }
1217
1218 ASSERT(dr->dr_txg == txg);
1219 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC ||
1220 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) {
1221 /*
1222 * We have already issued a sync write for this buffer,
1223 * or this buffer has already been synced. It could not
1224 * have been dirtied since, or we would have cleared the state.
1225 */
1226 mutex_exit(&db->db_mtx);
1227 return (EALREADY);
1228 }
1229
1230 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN);
1231 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC;
1232 mutex_exit(&db->db_mtx);
1233
1234 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP);
1235 dsa->dsa_dr = dr;
1236 dsa->dsa_done = done;
1237 dsa->dsa_zgd = zgd;
1238 dsa->dsa_tx = NULL;
1239
1240 zio_nowait(arc_write(pio, os->os_spa, txg,
1241 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp,
1242 dmu_sync_ready, dmu_sync_done, dsa,
1243 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb));
1244
1245 return (0);
1246 }
1247
1248 int
dmu_object_set_blocksize(objset_t * os,uint64_t object,uint64_t size,int ibs,dmu_tx_t * tx)1249 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs,
1250 dmu_tx_t *tx)
1251 {
1252 dnode_t *dn;
1253 int err;
1254
1255 err = dnode_hold(os, object, FTAG, &dn);
1256 if (err)
1257 return (err);
1258 err = dnode_set_blksz(dn, size, ibs, tx);
1259 dnode_rele(dn, FTAG);
1260 return (err);
1261 }
1262
1263 void
dmu_object_set_checksum(objset_t * os,uint64_t object,uint8_t checksum,dmu_tx_t * tx)1264 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum,
1265 dmu_tx_t *tx)
1266 {
1267 dnode_t *dn;
1268
1269 /* XXX assumes dnode_hold will not get an i/o error */
1270 (void) dnode_hold(os, object, FTAG, &dn);
1271 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS);
1272 dn->dn_checksum = checksum;
1273 dnode_setdirty(dn, tx);
1274 dnode_rele(dn, FTAG);
1275 }
1276
1277 void
dmu_object_set_compress(objset_t * os,uint64_t object,uint8_t compress,dmu_tx_t * tx)1278 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress,
1279 dmu_tx_t *tx)
1280 {
1281 dnode_t *dn;
1282
1283 /* XXX assumes dnode_hold will not get an i/o error */
1284 (void) dnode_hold(os, object, FTAG, &dn);
1285 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS);
1286 dn->dn_compress = compress;
1287 dnode_setdirty(dn, tx);
1288 dnode_rele(dn, FTAG);
1289 }
1290
1291 int zfs_mdcomp_disable = 0;
1292
1293 void
dmu_write_policy(objset_t * os,dnode_t * dn,int level,int wp,zio_prop_t * zp)1294 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp)
1295 {
1296 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET;
1297 boolean_t ismd = (level > 0 || dmu_ot[type].ot_metadata);
1298 enum zio_checksum checksum = os->os_checksum;
1299 enum zio_compress compress = os->os_compress;
1300 enum zio_checksum dedup_checksum = os->os_dedup_checksum;
1301 boolean_t dedup;
1302 boolean_t dedup_verify = os->os_dedup_verify;
1303 int copies = os->os_copies;
1304
1305 /*
1306 * Determine checksum setting.
1307 */
1308 if (ismd) {
1309 /*
1310 * Metadata always gets checksummed. If the data
1311 * checksum is multi-bit correctable, and it's not a
1312 * ZBT-style checksum, then it's suitable for metadata
1313 * as well. Otherwise, the metadata checksum defaults
1314 * to fletcher4.
1315 */
1316 if (zio_checksum_table[checksum].ci_correctable < 1 ||
1317 zio_checksum_table[checksum].ci_eck)
1318 checksum = ZIO_CHECKSUM_FLETCHER_4;
1319 } else {
1320 checksum = zio_checksum_select(dn->dn_checksum, checksum);
1321 }
1322
1323 /*
1324 * Determine compression setting.
1325 */
1326 if (ismd) {
1327 /*
1328 * XXX -- we should design a compression algorithm
1329 * that specializes in arrays of bps.
1330 */
1331 compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY :
1332 ZIO_COMPRESS_LZJB;
1333 } else {
1334 compress = zio_compress_select(dn->dn_compress, compress);
1335 }
1336
1337 /*
1338 * Determine dedup setting. If we are in dmu_sync(), we won't
1339 * actually dedup now because that's all done in syncing context;
1340 * but we do want to use the dedup checkum. If the checksum is not
1341 * strong enough to ensure unique signatures, force dedup_verify.
1342 */
1343 dedup = (!ismd && dedup_checksum != ZIO_CHECKSUM_OFF);
1344 if (dedup) {
1345 checksum = dedup_checksum;
1346 if (!zio_checksum_table[checksum].ci_dedup)
1347 dedup_verify = 1;
1348 }
1349
1350 if (wp & WP_DMU_SYNC)
1351 dedup = 0;
1352
1353 if (wp & WP_NOFILL) {
1354 ASSERT(!ismd && level == 0);
1355 checksum = ZIO_CHECKSUM_OFF;
1356 compress = ZIO_COMPRESS_OFF;
1357 dedup = B_FALSE;
1358 }
1359
1360 zp->zp_checksum = checksum;
1361 zp->zp_compress = compress;
1362 zp->zp_type = type;
1363 zp->zp_level = level;
1364 zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa));
1365 zp->zp_dedup = dedup;
1366 zp->zp_dedup_verify = dedup && dedup_verify;
1367 }
1368
1369 int
dmu_offset_next(objset_t * os,uint64_t object,boolean_t hole,uint64_t * off)1370 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off)
1371 {
1372 dnode_t *dn;
1373 int i, err;
1374
1375 err = dnode_hold(os, object, FTAG, &dn);
1376 if (err)
1377 return (err);
1378 /*
1379 * Sync any current changes before
1380 * we go trundling through the block pointers.
1381 */
1382 for (i = 0; i < TXG_SIZE; i++) {
1383 if (list_link_active(&dn->dn_dirty_link[i]))
1384 break;
1385 }
1386 if (i != TXG_SIZE) {
1387 dnode_rele(dn, FTAG);
1388 txg_wait_synced(dmu_objset_pool(os), 0);
1389 err = dnode_hold(os, object, FTAG, &dn);
1390 if (err)
1391 return (err);
1392 }
1393
1394 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0);
1395 dnode_rele(dn, FTAG);
1396
1397 return (err);
1398 }
1399
1400 void
dmu_object_info_from_dnode(dnode_t * dn,dmu_object_info_t * doi)1401 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi)
1402 {
1403 dnode_phys_t *dnp;
1404
1405 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1406 mutex_enter(&dn->dn_mtx);
1407
1408 dnp = dn->dn_phys;
1409
1410 doi->doi_data_block_size = dn->dn_datablksz;
1411 doi->doi_metadata_block_size = dn->dn_indblkshift ?
1412 1ULL << dn->dn_indblkshift : 0;
1413 doi->doi_type = dn->dn_type;
1414 doi->doi_bonus_type = dn->dn_bonustype;
1415 doi->doi_bonus_size = dn->dn_bonuslen;
1416 doi->doi_indirection = dn->dn_nlevels;
1417 doi->doi_checksum = dn->dn_checksum;
1418 doi->doi_compress = dn->dn_compress;
1419 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9;
1420 doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz;
1421 doi->doi_fill_count = 0;
1422 for (int i = 0; i < dnp->dn_nblkptr; i++)
1423 doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill;
1424
1425 mutex_exit(&dn->dn_mtx);
1426 rw_exit(&dn->dn_struct_rwlock);
1427 }
1428
1429 /*
1430 * Get information on a DMU object.
1431 * If doi is NULL, just indicates whether the object exists.
1432 */
1433 int
dmu_object_info(objset_t * os,uint64_t object,dmu_object_info_t * doi)1434 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi)
1435 {
1436 dnode_t *dn;
1437 int err = dnode_hold(os, object, FTAG, &dn);
1438
1439 if (err)
1440 return (err);
1441
1442 if (doi != NULL)
1443 dmu_object_info_from_dnode(dn, doi);
1444
1445 dnode_rele(dn, FTAG);
1446 return (0);
1447 }
1448
1449 /*
1450 * As above, but faster; can be used when you have a held dbuf in hand.
1451 */
1452 void
dmu_object_info_from_db(dmu_buf_t * db,dmu_object_info_t * doi)1453 dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi)
1454 {
1455 dmu_object_info_from_dnode(((dmu_buf_impl_t *)db)->db_dnode, doi);
1456 }
1457
1458 /*
1459 * Faster still when you only care about the size.
1460 * This is specifically optimized for zfs_getattr().
1461 */
1462 void
dmu_object_size_from_db(dmu_buf_t * db,uint32_t * blksize,u_longlong_t * nblk512)1463 dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, u_longlong_t *nblk512)
1464 {
1465 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode;
1466
1467 *blksize = dn->dn_datablksz;
1468 /* add 1 for dnode space */
1469 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >>
1470 SPA_MINBLOCKSHIFT) + 1;
1471 }
1472
1473 void
byteswap_uint64_array(void * vbuf,size_t size)1474 byteswap_uint64_array(void *vbuf, size_t size)
1475 {
1476 uint64_t *buf = vbuf;
1477 size_t count = size >> 3;
1478 int i;
1479
1480 ASSERT((size & 7) == 0);
1481
1482 for (i = 0; i < count; i++)
1483 buf[i] = BSWAP_64(buf[i]);
1484 }
1485
1486 void
byteswap_uint32_array(void * vbuf,size_t size)1487 byteswap_uint32_array(void *vbuf, size_t size)
1488 {
1489 uint32_t *buf = vbuf;
1490 size_t count = size >> 2;
1491 int i;
1492
1493 ASSERT((size & 3) == 0);
1494
1495 for (i = 0; i < count; i++)
1496 buf[i] = BSWAP_32(buf[i]);
1497 }
1498
1499 void
byteswap_uint16_array(void * vbuf,size_t size)1500 byteswap_uint16_array(void *vbuf, size_t size)
1501 {
1502 uint16_t *buf = vbuf;
1503 size_t count = size >> 1;
1504 int i;
1505
1506 ASSERT((size & 1) == 0);
1507
1508 for (i = 0; i < count; i++)
1509 buf[i] = BSWAP_16(buf[i]);
1510 }
1511
1512 /* ARGSUSED */
1513 void
byteswap_uint8_array(void * vbuf,size_t size)1514 byteswap_uint8_array(void *vbuf, size_t size)
1515 {
1516 }
1517
1518 void
dmu_init(void)1519 dmu_init(void)
1520 {
1521 dbuf_init();
1522 dnode_init();
1523 zfetch_init();
1524 arc_init();
1525 l2arc_init();
1526 #ifdef PORT_SOLARIS
1527 xuio_stat_init();
1528 #endif
1529 }
1530
1531 void
dmu_fini(void)1532 dmu_fini(void)
1533 {
1534 arc_fini();
1535 zfetch_fini();
1536 dnode_fini();
1537 dbuf_fini();
1538 l2arc_fini();
1539 #ifdef PORT_SOLARIS
1540 xuio_stat_fini();
1541 #endif
1542 }
1543