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 2009 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 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 127 dmu_bonus_max(void) 128 { 129 return (DN_MAX_BONUSLEN); 130 } 131 132 int 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 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 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 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 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 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 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 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 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 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 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 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 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 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 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 #ifdef _KERNEL 665 int 666 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) 667 { 668 dmu_buf_t **dbp; 669 int numbufs, i, err; 670 671 /* 672 * NB: we could do this block-at-a-time, but it's nice 673 * to be reading in parallel. 674 */ 675 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG, 676 &numbufs, &dbp); 677 if (err) 678 return (err); 679 680 for (i = 0; i < numbufs; i++) { 681 int tocpy; 682 int bufoff; 683 dmu_buf_t *db = dbp[i]; 684 685 ASSERT(size > 0); 686 687 bufoff = uio->uio_loffset - db->db_offset; 688 tocpy = (int)MIN(db->db_size - bufoff, size); 689 690 err = uiomove((char *)db->db_data + bufoff, tocpy, 691 UIO_READ, uio); 692 if (err) 693 break; 694 695 size -= tocpy; 696 } 697 dmu_buf_rele_array(dbp, numbufs, FTAG); 698 699 return (err); 700 } 701 702 int 703 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, 704 dmu_tx_t *tx) 705 { 706 dmu_buf_t **dbp; 707 int numbufs, i; 708 int err = 0; 709 710 if (size == 0) 711 return (0); 712 713 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, 714 FALSE, FTAG, &numbufs, &dbp); 715 if (err) 716 return (err); 717 718 for (i = 0; i < numbufs; i++) { 719 int tocpy; 720 int bufoff; 721 dmu_buf_t *db = dbp[i]; 722 723 ASSERT(size > 0); 724 725 bufoff = uio->uio_loffset - db->db_offset; 726 tocpy = (int)MIN(db->db_size - bufoff, size); 727 728 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 729 730 if (tocpy == db->db_size) 731 dmu_buf_will_fill(db, tx); 732 else 733 dmu_buf_will_dirty(db, tx); 734 735 /* 736 * XXX uiomove could block forever (eg. nfs-backed 737 * pages). There needs to be a uiolockdown() function 738 * to lock the pages in memory, so that uiomove won't 739 * block. 740 */ 741 err = uiomove((char *)db->db_data + bufoff, tocpy, 742 UIO_WRITE, uio); 743 744 if (tocpy == db->db_size) 745 dmu_buf_fill_done(db, tx); 746 747 if (err) 748 break; 749 750 size -= tocpy; 751 } 752 dmu_buf_rele_array(dbp, numbufs, FTAG); 753 return (err); 754 } 755 756 int 757 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 758 page_t *pp, dmu_tx_t *tx) 759 { 760 dmu_buf_t **dbp; 761 int numbufs, i; 762 int err; 763 764 if (size == 0) 765 return (0); 766 767 err = dmu_buf_hold_array(os, object, offset, size, 768 FALSE, FTAG, &numbufs, &dbp); 769 if (err) 770 return (err); 771 772 for (i = 0; i < numbufs; i++) { 773 int tocpy, copied, thiscpy; 774 int bufoff; 775 dmu_buf_t *db = dbp[i]; 776 caddr_t va; 777 778 ASSERT(size > 0); 779 ASSERT3U(db->db_size, >=, PAGESIZE); 780 781 bufoff = offset - db->db_offset; 782 tocpy = (int)MIN(db->db_size - bufoff, size); 783 784 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 785 786 if (tocpy == db->db_size) 787 dmu_buf_will_fill(db, tx); 788 else 789 dmu_buf_will_dirty(db, tx); 790 791 for (copied = 0; copied < tocpy; copied += PAGESIZE) { 792 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); 793 thiscpy = MIN(PAGESIZE, tocpy - copied); 794 va = zfs_map_page(pp, S_READ); 795 bcopy(va, (char *)db->db_data + bufoff, thiscpy); 796 zfs_unmap_page(pp, va); 797 pp = pp->p_next; 798 bufoff += PAGESIZE; 799 } 800 801 if (tocpy == db->db_size) 802 dmu_buf_fill_done(db, tx); 803 804 offset += tocpy; 805 size -= tocpy; 806 } 807 dmu_buf_rele_array(dbp, numbufs, FTAG); 808 return (err); 809 } 810 #endif 811 812 /* 813 * Allocate a loaned anonymous arc buffer. 814 */ 815 arc_buf_t * 816 dmu_request_arcbuf(dmu_buf_t *handle, int size) 817 { 818 dnode_t *dn = ((dmu_buf_impl_t *)handle)->db_dnode; 819 820 return (arc_loan_buf(dn->dn_objset->os_spa, size)); 821 } 822 823 /* 824 * Free a loaned arc buffer. 825 */ 826 void 827 dmu_return_arcbuf(arc_buf_t *buf) 828 { 829 arc_return_buf(buf, FTAG); 830 VERIFY(arc_buf_remove_ref(buf, FTAG) == 1); 831 } 832 833 /* 834 * When possible directly assign passed loaned arc buffer to a dbuf. 835 * If this is not possible copy the contents of passed arc buf via 836 * dmu_write(). 837 */ 838 void 839 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, 840 dmu_tx_t *tx) 841 { 842 dnode_t *dn = ((dmu_buf_impl_t *)handle)->db_dnode; 843 dmu_buf_impl_t *db; 844 uint32_t blksz = (uint32_t)arc_buf_size(buf); 845 uint64_t blkid; 846 847 rw_enter(&dn->dn_struct_rwlock, RW_READER); 848 blkid = dbuf_whichblock(dn, offset); 849 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); 850 rw_exit(&dn->dn_struct_rwlock); 851 852 if (offset == db->db.db_offset && blksz == db->db.db_size) { 853 dbuf_assign_arcbuf(db, buf, tx); 854 dbuf_rele(db, FTAG); 855 } else { 856 dbuf_rele(db, FTAG); 857 dmu_write(dn->dn_objset, dn->dn_object, offset, blksz, 858 buf->b_data, tx); 859 dmu_return_arcbuf(buf); 860 } 861 } 862 863 typedef struct { 864 dbuf_dirty_record_t *dsa_dr; 865 dmu_sync_cb_t *dsa_done; 866 zgd_t *dsa_zgd; 867 dmu_tx_t *dsa_tx; 868 } dmu_sync_arg_t; 869 870 /* ARGSUSED */ 871 static void 872 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) 873 { 874 dmu_sync_arg_t *dsa = varg; 875 dmu_buf_t *db = dsa->dsa_zgd->zgd_db; 876 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 877 blkptr_t *bp = zio->io_bp; 878 879 if (zio->io_error == 0) { 880 if (BP_IS_HOLE(bp)) { 881 /* 882 * A block of zeros may compress to a hole, but the 883 * block size still needs to be known for replay. 884 */ 885 BP_SET_LSIZE(bp, db->db_size); 886 } else { 887 ASSERT(BP_GET_TYPE(bp) == dn->dn_type); 888 ASSERT(BP_GET_LEVEL(bp) == 0); 889 bp->blk_fill = 1; 890 } 891 } 892 } 893 894 static void 895 dmu_sync_late_arrival_ready(zio_t *zio) 896 { 897 dmu_sync_ready(zio, NULL, zio->io_private); 898 } 899 900 /* ARGSUSED */ 901 static void 902 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) 903 { 904 dmu_sync_arg_t *dsa = varg; 905 dbuf_dirty_record_t *dr = dsa->dsa_dr; 906 dmu_buf_impl_t *db = dr->dr_dbuf; 907 908 mutex_enter(&db->db_mtx); 909 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); 910 if (zio->io_error == 0) { 911 dr->dt.dl.dr_overridden_by = *zio->io_bp; 912 dr->dt.dl.dr_override_state = DR_OVERRIDDEN; 913 dr->dt.dl.dr_copies = zio->io_prop.zp_copies; 914 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by)) 915 BP_ZERO(&dr->dt.dl.dr_overridden_by); 916 } else { 917 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; 918 } 919 cv_broadcast(&db->db_changed); 920 mutex_exit(&db->db_mtx); 921 922 dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 923 924 kmem_free(dsa, sizeof (*dsa)); 925 } 926 927 static void 928 dmu_sync_late_arrival_done(zio_t *zio) 929 { 930 blkptr_t *bp = zio->io_bp; 931 dmu_sync_arg_t *dsa = zio->io_private; 932 933 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { 934 ASSERT(zio->io_bp->blk_birth == zio->io_txg); 935 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); 936 zio_free(zio->io_spa, zio->io_txg, zio->io_bp); 937 } 938 939 dmu_tx_commit(dsa->dsa_tx); 940 941 dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 942 943 kmem_free(dsa, sizeof (*dsa)); 944 } 945 946 static int 947 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, 948 zio_prop_t *zp, zbookmark_t *zb) 949 { 950 dmu_sync_arg_t *dsa; 951 dmu_tx_t *tx; 952 953 tx = dmu_tx_create(os); 954 dmu_tx_hold_space(tx, zgd->zgd_db->db_size); 955 if (dmu_tx_assign(tx, TXG_NOWAIT) != 0) { 956 dmu_tx_abort(tx); 957 return (EIO); /* Make zl_get_data do txg_waited_synced() */ 958 } 959 960 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 961 dsa->dsa_dr = NULL; 962 dsa->dsa_done = done; 963 dsa->dsa_zgd = zgd; 964 dsa->dsa_tx = tx; 965 966 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, 967 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, 968 dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa, 969 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); 970 971 return (0); 972 } 973 974 /* 975 * Intent log support: sync the block associated with db to disk. 976 * N.B. and XXX: the caller is responsible for making sure that the 977 * data isn't changing while dmu_sync() is writing it. 978 * 979 * Return values: 980 * 981 * EEXIST: this txg has already been synced, so there's nothing to to. 982 * The caller should not log the write. 983 * 984 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. 985 * The caller should not log the write. 986 * 987 * EALREADY: this block is already in the process of being synced. 988 * The caller should track its progress (somehow). 989 * 990 * EIO: could not do the I/O. 991 * The caller should do a txg_wait_synced(). 992 * 993 * 0: the I/O has been initiated. 994 * The caller should log this blkptr in the done callback. 995 * It is possible that the I/O will fail, in which case 996 * the error will be reported to the done callback and 997 * propagated to pio from zio_done(). 998 */ 999 int 1000 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) 1001 { 1002 blkptr_t *bp = zgd->zgd_bp; 1003 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; 1004 objset_t *os = db->db_objset; 1005 dsl_dataset_t *ds = os->os_dsl_dataset; 1006 dbuf_dirty_record_t *dr; 1007 dmu_sync_arg_t *dsa; 1008 zbookmark_t zb; 1009 zio_prop_t zp; 1010 1011 ASSERT(pio != NULL); 1012 ASSERT(BP_IS_HOLE(bp)); 1013 ASSERT(txg != 0); 1014 1015 SET_BOOKMARK(&zb, ds->ds_object, 1016 db->db.db_object, db->db_level, db->db_blkid); 1017 1018 dmu_write_policy(os, db->db_dnode, db->db_level, WP_DMU_SYNC, &zp); 1019 1020 /* 1021 * If we're frozen (running ziltest), we always need to generate a bp. 1022 */ 1023 if (txg > spa_freeze_txg(os->os_spa)) 1024 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1025 1026 /* 1027 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() 1028 * and us. If we determine that this txg is not yet syncing, 1029 * but it begins to sync a moment later, that's OK because the 1030 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. 1031 */ 1032 mutex_enter(&db->db_mtx); 1033 1034 if (txg <= spa_last_synced_txg(os->os_spa)) { 1035 /* 1036 * This txg has already synced. There's nothing to do. 1037 */ 1038 mutex_exit(&db->db_mtx); 1039 return (EEXIST); 1040 } 1041 1042 if (txg <= spa_syncing_txg(os->os_spa)) { 1043 /* 1044 * This txg is currently syncing, so we can't mess with 1045 * the dirty record anymore; just write a new log block. 1046 */ 1047 mutex_exit(&db->db_mtx); 1048 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1049 } 1050 1051 dr = db->db_last_dirty; 1052 while (dr && dr->dr_txg != txg) 1053 dr = dr->dr_next; 1054 1055 if (dr == NULL) { 1056 /* 1057 * There's no dr for this dbuf, so it must have been freed. 1058 * There's no need to log writes to freed blocks, so we're done. 1059 */ 1060 mutex_exit(&db->db_mtx); 1061 return (ENOENT); 1062 } 1063 1064 ASSERT(dr->dr_txg == txg); 1065 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || 1066 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { 1067 /* 1068 * We have already issued a sync write for this buffer, 1069 * or this buffer has already been synced. It could not 1070 * have been dirtied since, or we would have cleared the state. 1071 */ 1072 mutex_exit(&db->db_mtx); 1073 return (EALREADY); 1074 } 1075 1076 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); 1077 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; 1078 mutex_exit(&db->db_mtx); 1079 1080 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1081 dsa->dsa_dr = dr; 1082 dsa->dsa_done = done; 1083 dsa->dsa_zgd = zgd; 1084 dsa->dsa_tx = NULL; 1085 1086 zio_nowait(arc_write(pio, os->os_spa, txg, 1087 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp, 1088 dmu_sync_ready, dmu_sync_done, dsa, 1089 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); 1090 1091 return (0); 1092 } 1093 1094 int 1095 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, 1096 dmu_tx_t *tx) 1097 { 1098 dnode_t *dn; 1099 int err; 1100 1101 err = dnode_hold(os, object, FTAG, &dn); 1102 if (err) 1103 return (err); 1104 err = dnode_set_blksz(dn, size, ibs, tx); 1105 dnode_rele(dn, FTAG); 1106 return (err); 1107 } 1108 1109 void 1110 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 1111 dmu_tx_t *tx) 1112 { 1113 dnode_t *dn; 1114 1115 /* XXX assumes dnode_hold will not get an i/o error */ 1116 (void) dnode_hold(os, object, FTAG, &dn); 1117 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS); 1118 dn->dn_checksum = checksum; 1119 dnode_setdirty(dn, tx); 1120 dnode_rele(dn, FTAG); 1121 } 1122 1123 void 1124 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 1125 dmu_tx_t *tx) 1126 { 1127 dnode_t *dn; 1128 1129 /* XXX assumes dnode_hold will not get an i/o error */ 1130 (void) dnode_hold(os, object, FTAG, &dn); 1131 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS); 1132 dn->dn_compress = compress; 1133 dnode_setdirty(dn, tx); 1134 dnode_rele(dn, FTAG); 1135 } 1136 1137 int zfs_mdcomp_disable = 0; 1138 1139 void 1140 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) 1141 { 1142 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; 1143 boolean_t ismd = (level > 0 || dmu_ot[type].ot_metadata); 1144 enum zio_checksum checksum = os->os_checksum; 1145 enum zio_compress compress = os->os_compress; 1146 enum zio_checksum dedup_checksum = os->os_dedup_checksum; 1147 boolean_t dedup; 1148 boolean_t dedup_verify = os->os_dedup_verify; 1149 int copies = os->os_copies; 1150 1151 /* 1152 * Determine checksum setting. 1153 */ 1154 if (ismd) { 1155 /* 1156 * Metadata always gets checksummed. If the data 1157 * checksum is multi-bit correctable, and it's not a 1158 * ZBT-style checksum, then it's suitable for metadata 1159 * as well. Otherwise, the metadata checksum defaults 1160 * to fletcher4. 1161 */ 1162 if (zio_checksum_table[checksum].ci_correctable < 1 || 1163 zio_checksum_table[checksum].ci_zbt) 1164 checksum = ZIO_CHECKSUM_FLETCHER_4; 1165 } else { 1166 checksum = zio_checksum_select(dn->dn_checksum, checksum); 1167 } 1168 1169 /* 1170 * Determine compression setting. 1171 */ 1172 if (ismd) { 1173 /* 1174 * XXX -- we should design a compression algorithm 1175 * that specializes in arrays of bps. 1176 */ 1177 compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY : 1178 ZIO_COMPRESS_LZJB; 1179 } else { 1180 compress = zio_compress_select(dn->dn_compress, compress); 1181 } 1182 1183 /* 1184 * Determine dedup setting. If we are in dmu_sync(), we won't 1185 * actually dedup now because that's all done in syncing context; 1186 * but we do want to use the dedup checkum. If the checksum is not 1187 * strong enough to ensure unique signatures, force dedup_verify. 1188 */ 1189 dedup = (!ismd && dedup_checksum != ZIO_CHECKSUM_OFF); 1190 if (dedup) { 1191 checksum = dedup_checksum; 1192 if (!zio_checksum_table[checksum].ci_dedup) 1193 dedup_verify = 1; 1194 } 1195 1196 if (wp & WP_DMU_SYNC) 1197 dedup = 0; 1198 1199 if (wp & WP_NOFILL) { 1200 ASSERT(!ismd && level == 0); 1201 checksum = ZIO_CHECKSUM_OFF; 1202 compress = ZIO_COMPRESS_OFF; 1203 dedup = B_FALSE; 1204 } 1205 1206 zp->zp_checksum = checksum; 1207 zp->zp_compress = compress; 1208 zp->zp_type = type; 1209 zp->zp_level = level; 1210 zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa)); 1211 zp->zp_dedup = dedup; 1212 zp->zp_dedup_verify = dedup && dedup_verify; 1213 } 1214 1215 int 1216 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) 1217 { 1218 dnode_t *dn; 1219 int i, err; 1220 1221 err = dnode_hold(os, object, FTAG, &dn); 1222 if (err) 1223 return (err); 1224 /* 1225 * Sync any current changes before 1226 * we go trundling through the block pointers. 1227 */ 1228 for (i = 0; i < TXG_SIZE; i++) { 1229 if (list_link_active(&dn->dn_dirty_link[i])) 1230 break; 1231 } 1232 if (i != TXG_SIZE) { 1233 dnode_rele(dn, FTAG); 1234 txg_wait_synced(dmu_objset_pool(os), 0); 1235 err = dnode_hold(os, object, FTAG, &dn); 1236 if (err) 1237 return (err); 1238 } 1239 1240 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); 1241 dnode_rele(dn, FTAG); 1242 1243 return (err); 1244 } 1245 1246 void 1247 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) 1248 { 1249 dnode_phys_t *dnp; 1250 1251 rw_enter(&dn->dn_struct_rwlock, RW_READER); 1252 mutex_enter(&dn->dn_mtx); 1253 1254 dnp = dn->dn_phys; 1255 1256 doi->doi_data_block_size = dn->dn_datablksz; 1257 doi->doi_metadata_block_size = dn->dn_indblkshift ? 1258 1ULL << dn->dn_indblkshift : 0; 1259 doi->doi_type = dn->dn_type; 1260 doi->doi_bonus_type = dn->dn_bonustype; 1261 doi->doi_bonus_size = dn->dn_bonuslen; 1262 doi->doi_indirection = dn->dn_nlevels; 1263 doi->doi_checksum = dn->dn_checksum; 1264 doi->doi_compress = dn->dn_compress; 1265 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; 1266 doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz; 1267 doi->doi_fill_count = 0; 1268 for (int i = 0; i < dnp->dn_nblkptr; i++) 1269 doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill; 1270 1271 mutex_exit(&dn->dn_mtx); 1272 rw_exit(&dn->dn_struct_rwlock); 1273 } 1274 1275 /* 1276 * Get information on a DMU object. 1277 * If doi is NULL, just indicates whether the object exists. 1278 */ 1279 int 1280 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) 1281 { 1282 dnode_t *dn; 1283 int err = dnode_hold(os, object, FTAG, &dn); 1284 1285 if (err) 1286 return (err); 1287 1288 if (doi != NULL) 1289 dmu_object_info_from_dnode(dn, doi); 1290 1291 dnode_rele(dn, FTAG); 1292 return (0); 1293 } 1294 1295 /* 1296 * As above, but faster; can be used when you have a held dbuf in hand. 1297 */ 1298 void 1299 dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi) 1300 { 1301 dmu_object_info_from_dnode(((dmu_buf_impl_t *)db)->db_dnode, doi); 1302 } 1303 1304 /* 1305 * Faster still when you only care about the size. 1306 * This is specifically optimized for zfs_getattr(). 1307 */ 1308 void 1309 dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, u_longlong_t *nblk512) 1310 { 1311 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 1312 1313 *blksize = dn->dn_datablksz; 1314 /* add 1 for dnode space */ 1315 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> 1316 SPA_MINBLOCKSHIFT) + 1; 1317 } 1318 1319 void 1320 byteswap_uint64_array(void *vbuf, size_t size) 1321 { 1322 uint64_t *buf = vbuf; 1323 size_t count = size >> 3; 1324 int i; 1325 1326 ASSERT((size & 7) == 0); 1327 1328 for (i = 0; i < count; i++) 1329 buf[i] = BSWAP_64(buf[i]); 1330 } 1331 1332 void 1333 byteswap_uint32_array(void *vbuf, size_t size) 1334 { 1335 uint32_t *buf = vbuf; 1336 size_t count = size >> 2; 1337 int i; 1338 1339 ASSERT((size & 3) == 0); 1340 1341 for (i = 0; i < count; i++) 1342 buf[i] = BSWAP_32(buf[i]); 1343 } 1344 1345 void 1346 byteswap_uint16_array(void *vbuf, size_t size) 1347 { 1348 uint16_t *buf = vbuf; 1349 size_t count = size >> 1; 1350 int i; 1351 1352 ASSERT((size & 1) == 0); 1353 1354 for (i = 0; i < count; i++) 1355 buf[i] = BSWAP_16(buf[i]); 1356 } 1357 1358 /* ARGSUSED */ 1359 void 1360 byteswap_uint8_array(void *vbuf, size_t size) 1361 { 1362 } 1363 1364 void 1365 dmu_init(void) 1366 { 1367 dbuf_init(); 1368 dnode_init(); 1369 zfetch_init(); 1370 arc_init(); 1371 l2arc_init(); 1372 } 1373 1374 void 1375 dmu_fini(void) 1376 { 1377 arc_fini(); 1378 zfetch_fini(); 1379 dnode_fini(); 1380 dbuf_fini(); 1381 l2arc_fini(); 1382 } 1383