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 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/dmu.h> 29 #include <sys/dmu_impl.h> 30 #include <sys/dmu_tx.h> 31 #include <sys/dbuf.h> 32 #include <sys/dnode.h> 33 #include <sys/zfs_context.h> 34 #include <sys/dmu_objset.h> 35 #include <sys/dmu_traverse.h> 36 #include <sys/dsl_dataset.h> 37 #include <sys/dsl_dir.h> 38 #include <sys/dsl_pool.h> 39 #include <sys/dsl_synctask.h> 40 #include <sys/dsl_prop.h> 41 #include <sys/dmu_zfetch.h> 42 #include <sys/zfs_ioctl.h> 43 #include <sys/zap.h> 44 #include <sys/zio_checksum.h> 45 #ifdef _KERNEL 46 #include <sys/vmsystm.h> 47 #endif 48 49 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { 50 { byteswap_uint8_array, TRUE, "unallocated" }, 51 { zap_byteswap, TRUE, "object directory" }, 52 { byteswap_uint64_array, TRUE, "object array" }, 53 { byteswap_uint8_array, TRUE, "packed nvlist" }, 54 { byteswap_uint64_array, TRUE, "packed nvlist size" }, 55 { byteswap_uint64_array, TRUE, "bplist" }, 56 { byteswap_uint64_array, TRUE, "bplist header" }, 57 { byteswap_uint64_array, TRUE, "SPA space map header" }, 58 { byteswap_uint64_array, TRUE, "SPA space map" }, 59 { byteswap_uint64_array, TRUE, "ZIL intent log" }, 60 { dnode_buf_byteswap, TRUE, "DMU dnode" }, 61 { dmu_objset_byteswap, TRUE, "DMU objset" }, 62 { byteswap_uint64_array, TRUE, "DSL directory" }, 63 { zap_byteswap, TRUE, "DSL directory child map"}, 64 { zap_byteswap, TRUE, "DSL dataset snap map" }, 65 { zap_byteswap, TRUE, "DSL props" }, 66 { byteswap_uint64_array, TRUE, "DSL dataset" }, 67 { zfs_znode_byteswap, TRUE, "ZFS znode" }, 68 { zfs_acl_byteswap, TRUE, "ZFS ACL" }, 69 { byteswap_uint8_array, FALSE, "ZFS plain file" }, 70 { zap_byteswap, TRUE, "ZFS directory" }, 71 { zap_byteswap, TRUE, "ZFS master node" }, 72 { zap_byteswap, TRUE, "ZFS delete queue" }, 73 { byteswap_uint8_array, FALSE, "zvol object" }, 74 { zap_byteswap, TRUE, "zvol prop" }, 75 { byteswap_uint8_array, FALSE, "other uint8[]" }, 76 { byteswap_uint64_array, FALSE, "other uint64[]" }, 77 { zap_byteswap, TRUE, "other ZAP" }, 78 { zap_byteswap, TRUE, "persistent error log" }, 79 { byteswap_uint8_array, TRUE, "SPA history" }, 80 { byteswap_uint64_array, TRUE, "SPA history offsets" }, 81 }; 82 83 int 84 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 85 void *tag, dmu_buf_t **dbp) 86 { 87 dnode_t *dn; 88 uint64_t blkid; 89 dmu_buf_impl_t *db; 90 int err; 91 92 err = dnode_hold(os->os, object, FTAG, &dn); 93 if (err) 94 return (err); 95 blkid = dbuf_whichblock(dn, offset); 96 rw_enter(&dn->dn_struct_rwlock, RW_READER); 97 db = dbuf_hold(dn, blkid, tag); 98 rw_exit(&dn->dn_struct_rwlock); 99 if (db == NULL) { 100 err = EIO; 101 } else { 102 err = dbuf_read(db, NULL, DB_RF_CANFAIL); 103 if (err) { 104 dbuf_rele(db, tag); 105 db = NULL; 106 } 107 } 108 109 dnode_rele(dn, FTAG); 110 *dbp = &db->db; 111 return (err); 112 } 113 114 int 115 dmu_bonus_max(void) 116 { 117 return (DN_MAX_BONUSLEN); 118 } 119 120 /* 121 * returns ENOENT, EIO, or 0. 122 */ 123 int 124 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) 125 { 126 dnode_t *dn; 127 int err, count; 128 dmu_buf_impl_t *db; 129 130 err = dnode_hold(os->os, object, FTAG, &dn); 131 if (err) 132 return (err); 133 134 rw_enter(&dn->dn_struct_rwlock, RW_READER); 135 if (dn->dn_bonus == NULL) { 136 rw_exit(&dn->dn_struct_rwlock); 137 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 138 if (dn->dn_bonus == NULL) 139 dn->dn_bonus = dbuf_create_bonus(dn); 140 } 141 db = dn->dn_bonus; 142 rw_exit(&dn->dn_struct_rwlock); 143 mutex_enter(&db->db_mtx); 144 count = refcount_add(&db->db_holds, tag); 145 mutex_exit(&db->db_mtx); 146 if (count == 1) 147 dnode_add_ref(dn, db); 148 dnode_rele(dn, FTAG); 149 150 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED)); 151 152 *dbp = &db->db; 153 return (0); 154 } 155 156 /* 157 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces 158 * to take a held dnode rather than <os, object> -- the lookup is wasteful, 159 * and can induce severe lock contention when writing to several files 160 * whose dnodes are in the same block. 161 */ 162 static int 163 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, 164 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 165 { 166 dmu_buf_t **dbp; 167 uint64_t blkid, nblks, i; 168 uint32_t flags; 169 int err; 170 zio_t *zio; 171 172 ASSERT(length <= DMU_MAX_ACCESS); 173 174 flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT; 175 if (length > zfetch_array_rd_sz) 176 flags |= DB_RF_NOPREFETCH; 177 178 rw_enter(&dn->dn_struct_rwlock, RW_READER); 179 if (dn->dn_datablkshift) { 180 int blkshift = dn->dn_datablkshift; 181 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - 182 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; 183 } else { 184 ASSERT3U(offset + length, <=, dn->dn_datablksz); 185 nblks = 1; 186 } 187 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); 188 189 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, TRUE); 190 blkid = dbuf_whichblock(dn, offset); 191 for (i = 0; i < nblks; i++) { 192 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); 193 if (db == NULL) { 194 rw_exit(&dn->dn_struct_rwlock); 195 dmu_buf_rele_array(dbp, nblks, tag); 196 zio_nowait(zio); 197 return (EIO); 198 } 199 /* initiate async i/o */ 200 if (read) { 201 rw_exit(&dn->dn_struct_rwlock); 202 (void) dbuf_read(db, zio, flags); 203 rw_enter(&dn->dn_struct_rwlock, RW_READER); 204 } 205 dbp[i] = &db->db; 206 } 207 rw_exit(&dn->dn_struct_rwlock); 208 209 /* wait for async i/o */ 210 err = zio_wait(zio); 211 if (err) { 212 dmu_buf_rele_array(dbp, nblks, tag); 213 return (err); 214 } 215 216 /* wait for other io to complete */ 217 if (read) { 218 for (i = 0; i < nblks; i++) { 219 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; 220 mutex_enter(&db->db_mtx); 221 while (db->db_state == DB_READ || 222 db->db_state == DB_FILL) 223 cv_wait(&db->db_changed, &db->db_mtx); 224 if (db->db_state == DB_UNCACHED) 225 err = EIO; 226 mutex_exit(&db->db_mtx); 227 if (err) { 228 dmu_buf_rele_array(dbp, nblks, tag); 229 return (err); 230 } 231 } 232 } 233 234 *numbufsp = nblks; 235 *dbpp = dbp; 236 return (0); 237 } 238 239 static int 240 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, 241 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 242 { 243 dnode_t *dn; 244 int err; 245 246 err = dnode_hold(os->os, object, FTAG, &dn); 247 if (err) 248 return (err); 249 250 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 251 numbufsp, dbpp); 252 253 dnode_rele(dn, FTAG); 254 255 return (err); 256 } 257 258 int 259 dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, 260 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 261 { 262 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 263 int err; 264 265 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 266 numbufsp, dbpp); 267 268 return (err); 269 } 270 271 void 272 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) 273 { 274 int i; 275 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; 276 277 if (numbufs == 0) 278 return; 279 280 for (i = 0; i < numbufs; i++) { 281 if (dbp[i]) 282 dbuf_rele(dbp[i], tag); 283 } 284 285 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); 286 } 287 288 void 289 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) 290 { 291 dnode_t *dn; 292 uint64_t blkid; 293 int nblks, i, err; 294 295 if (zfs_prefetch_disable) 296 return; 297 298 if (len == 0) { /* they're interested in the bonus buffer */ 299 dn = os->os->os_meta_dnode; 300 301 if (object == 0 || object >= DN_MAX_OBJECT) 302 return; 303 304 rw_enter(&dn->dn_struct_rwlock, RW_READER); 305 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); 306 dbuf_prefetch(dn, blkid); 307 rw_exit(&dn->dn_struct_rwlock); 308 return; 309 } 310 311 /* 312 * XXX - Note, if the dnode for the requested object is not 313 * already cached, we will do a *synchronous* read in the 314 * dnode_hold() call. The same is true for any indirects. 315 */ 316 err = dnode_hold(os->os, object, FTAG, &dn); 317 if (err != 0) 318 return; 319 320 rw_enter(&dn->dn_struct_rwlock, RW_READER); 321 if (dn->dn_datablkshift) { 322 int blkshift = dn->dn_datablkshift; 323 nblks = (P2ROUNDUP(offset+len, 1<<blkshift) - 324 P2ALIGN(offset, 1<<blkshift)) >> blkshift; 325 } else { 326 nblks = (offset < dn->dn_datablksz); 327 } 328 329 if (nblks != 0) { 330 blkid = dbuf_whichblock(dn, offset); 331 for (i = 0; i < nblks; i++) 332 dbuf_prefetch(dn, blkid+i); 333 } 334 335 rw_exit(&dn->dn_struct_rwlock); 336 337 dnode_rele(dn, FTAG); 338 } 339 340 int 341 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 342 uint64_t size, dmu_tx_t *tx) 343 { 344 dnode_t *dn; 345 int err = dnode_hold(os->os, object, FTAG, &dn); 346 if (err) 347 return (err); 348 ASSERT(offset < UINT64_MAX); 349 ASSERT(size == -1ULL || size <= UINT64_MAX - offset); 350 dnode_free_range(dn, offset, size, tx); 351 dnode_rele(dn, FTAG); 352 return (0); 353 } 354 355 int 356 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 357 void *buf) 358 { 359 dnode_t *dn; 360 dmu_buf_t **dbp; 361 int numbufs, i, err; 362 363 /* 364 * Deal with odd block sizes, where there can't be data past the 365 * first block. 366 */ 367 err = dnode_hold(os->os, object, FTAG, &dn); 368 if (err) 369 return (err); 370 if (dn->dn_datablkshift == 0) { 371 int newsz = offset > dn->dn_datablksz ? 0 : 372 MIN(size, dn->dn_datablksz - offset); 373 bzero((char *)buf + newsz, size - newsz); 374 size = newsz; 375 } 376 377 while (size > 0) { 378 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); 379 int err; 380 381 /* 382 * NB: we could do this block-at-a-time, but it's nice 383 * to be reading in parallel. 384 */ 385 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, 386 TRUE, FTAG, &numbufs, &dbp); 387 if (err) 388 return (err); 389 390 for (i = 0; i < numbufs; i++) { 391 int tocpy; 392 int bufoff; 393 dmu_buf_t *db = dbp[i]; 394 395 ASSERT(size > 0); 396 397 bufoff = offset - db->db_offset; 398 tocpy = (int)MIN(db->db_size - bufoff, size); 399 400 bcopy((char *)db->db_data + bufoff, buf, tocpy); 401 402 offset += tocpy; 403 size -= tocpy; 404 buf = (char *)buf + tocpy; 405 } 406 dmu_buf_rele_array(dbp, numbufs, FTAG); 407 } 408 dnode_rele(dn, FTAG); 409 return (0); 410 } 411 412 void 413 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 414 const void *buf, dmu_tx_t *tx) 415 { 416 dmu_buf_t **dbp; 417 int numbufs, i; 418 419 if (size == 0) 420 return; 421 422 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 423 FALSE, FTAG, &numbufs, &dbp)); 424 425 for (i = 0; i < numbufs; i++) { 426 int tocpy; 427 int bufoff; 428 dmu_buf_t *db = dbp[i]; 429 430 ASSERT(size > 0); 431 432 bufoff = offset - db->db_offset; 433 tocpy = (int)MIN(db->db_size - bufoff, size); 434 435 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 436 437 if (tocpy == db->db_size) 438 dmu_buf_will_fill(db, tx); 439 else 440 dmu_buf_will_dirty(db, tx); 441 442 bcopy(buf, (char *)db->db_data + bufoff, tocpy); 443 444 if (tocpy == db->db_size) 445 dmu_buf_fill_done(db, tx); 446 447 offset += tocpy; 448 size -= tocpy; 449 buf = (char *)buf + tocpy; 450 } 451 dmu_buf_rele_array(dbp, numbufs, FTAG); 452 } 453 454 #ifdef _KERNEL 455 int 456 dmu_write_uio(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 457 uio_t *uio, dmu_tx_t *tx) 458 { 459 dmu_buf_t **dbp; 460 int numbufs, i; 461 int err = 0; 462 463 if (size == 0) 464 return (0); 465 466 err = dmu_buf_hold_array(os, object, offset, size, 467 FALSE, FTAG, &numbufs, &dbp); 468 if (err) 469 return (err); 470 471 for (i = 0; i < numbufs; i++) { 472 int tocpy; 473 int bufoff; 474 dmu_buf_t *db = dbp[i]; 475 476 ASSERT(size > 0); 477 478 bufoff = offset - db->db_offset; 479 tocpy = (int)MIN(db->db_size - bufoff, size); 480 481 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 482 483 if (tocpy == db->db_size) 484 dmu_buf_will_fill(db, tx); 485 else 486 dmu_buf_will_dirty(db, tx); 487 488 /* 489 * XXX uiomove could block forever (eg. nfs-backed 490 * pages). There needs to be a uiolockdown() function 491 * to lock the pages in memory, so that uiomove won't 492 * block. 493 */ 494 err = uiomove((char *)db->db_data + bufoff, tocpy, 495 UIO_WRITE, uio); 496 497 if (tocpy == db->db_size) 498 dmu_buf_fill_done(db, tx); 499 500 if (err) 501 break; 502 503 offset += tocpy; 504 size -= tocpy; 505 } 506 dmu_buf_rele_array(dbp, numbufs, FTAG); 507 return (err); 508 } 509 510 int 511 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 512 page_t *pp, dmu_tx_t *tx) 513 { 514 dmu_buf_t **dbp; 515 int numbufs, i; 516 int err; 517 518 if (size == 0) 519 return (0); 520 521 err = dmu_buf_hold_array(os, object, offset, size, 522 FALSE, FTAG, &numbufs, &dbp); 523 if (err) 524 return (err); 525 526 for (i = 0; i < numbufs; i++) { 527 int tocpy, copied, thiscpy; 528 int bufoff; 529 dmu_buf_t *db = dbp[i]; 530 caddr_t va; 531 532 ASSERT(size > 0); 533 ASSERT3U(db->db_size, >=, PAGESIZE); 534 535 bufoff = offset - db->db_offset; 536 tocpy = (int)MIN(db->db_size - bufoff, size); 537 538 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 539 540 if (tocpy == db->db_size) 541 dmu_buf_will_fill(db, tx); 542 else 543 dmu_buf_will_dirty(db, tx); 544 545 for (copied = 0; copied < tocpy; copied += PAGESIZE) { 546 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); 547 thiscpy = MIN(PAGESIZE, tocpy - copied); 548 va = ppmapin(pp, PROT_READ, (caddr_t)-1); 549 bcopy(va, (char *)db->db_data + bufoff, thiscpy); 550 ppmapout(va); 551 pp = pp->p_next; 552 bufoff += PAGESIZE; 553 } 554 555 if (tocpy == db->db_size) 556 dmu_buf_fill_done(db, tx); 557 558 if (err) 559 break; 560 561 offset += tocpy; 562 size -= tocpy; 563 } 564 dmu_buf_rele_array(dbp, numbufs, FTAG); 565 return (err); 566 } 567 #endif 568 569 typedef struct { 570 uint64_t txg; 571 dmu_buf_impl_t *db; 572 dmu_sync_cb_t *done; 573 void *arg; 574 } dmu_sync_cbin_t; 575 576 typedef union { 577 dmu_sync_cbin_t data; 578 blkptr_t blk; 579 } dmu_sync_cbarg_t; 580 581 /* ARGSUSED */ 582 static void 583 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) 584 { 585 dmu_sync_cbin_t *in = (dmu_sync_cbin_t *)varg; 586 dmu_buf_impl_t *db = in->db; 587 uint64_t txg = in->txg; 588 dmu_sync_cb_t *done = in->done; 589 void *arg = in->arg; 590 blkptr_t *blk = (blkptr_t *)varg; 591 592 if (!BP_IS_HOLE(zio->io_bp)) { 593 zio->io_bp->blk_fill = 1; 594 BP_SET_TYPE(zio->io_bp, db->db_dnode->dn_type); 595 BP_SET_LEVEL(zio->io_bp, 0); 596 } 597 598 *blk = *zio->io_bp; /* structure assignment */ 599 600 mutex_enter(&db->db_mtx); 601 ASSERT(db->db_d.db_overridden_by[txg&TXG_MASK] == IN_DMU_SYNC); 602 db->db_d.db_overridden_by[txg&TXG_MASK] = blk; 603 cv_broadcast(&db->db_changed); 604 mutex_exit(&db->db_mtx); 605 606 if (done) 607 done(&(db->db), arg); 608 } 609 610 /* 611 * Intent log support: sync the block associated with db to disk. 612 * N.B. and XXX: the caller is responsible for making sure that the 613 * data isn't changing while dmu_sync() is writing it. 614 * 615 * Return values: 616 * 617 * EEXIST: this txg has already been synced, so there's nothing to to. 618 * The caller should not log the write. 619 * 620 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. 621 * The caller should not log the write. 622 * 623 * EALREADY: this block is already in the process of being synced. 624 * The caller should track its progress (somehow). 625 * 626 * EINPROGRESS: the IO has been initiated. 627 * The caller should log this blkptr in the callback. 628 * 629 * 0: completed. Sets *bp to the blkptr just written. 630 * The caller should log this blkptr immediately. 631 */ 632 int 633 dmu_sync(zio_t *pio, dmu_buf_t *db_fake, 634 blkptr_t *bp, uint64_t txg, dmu_sync_cb_t *done, void *arg) 635 { 636 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 637 objset_impl_t *os = db->db_objset; 638 dsl_pool_t *dp = os->os_dsl_dataset->ds_dir->dd_pool; 639 tx_state_t *tx = &dp->dp_tx; 640 dmu_sync_cbin_t *in; 641 blkptr_t *blk; 642 zbookmark_t zb; 643 uint32_t arc_flag; 644 int err; 645 646 ASSERT(BP_IS_HOLE(bp)); 647 ASSERT(txg != 0); 648 649 650 dprintf("dmu_sync txg=%llu, s,o,q %llu %llu %llu\n", 651 txg, tx->tx_synced_txg, tx->tx_open_txg, tx->tx_quiesced_txg); 652 653 /* 654 * XXX - would be nice if we could do this without suspending... 655 */ 656 txg_suspend(dp); 657 658 /* 659 * If this txg already synced, there's nothing to do. 660 */ 661 if (txg <= tx->tx_synced_txg) { 662 txg_resume(dp); 663 /* 664 * If we're running ziltest, we need the blkptr regardless. 665 */ 666 if (txg > spa_freeze_txg(dp->dp_spa)) { 667 /* if db_blkptr == NULL, this was an empty write */ 668 if (db->db_blkptr) 669 *bp = *db->db_blkptr; /* structure assignment */ 670 return (0); 671 } 672 return (EEXIST); 673 } 674 675 mutex_enter(&db->db_mtx); 676 677 blk = db->db_d.db_overridden_by[txg&TXG_MASK]; 678 if (blk == IN_DMU_SYNC) { 679 /* 680 * We have already issued a sync write for this buffer. 681 */ 682 mutex_exit(&db->db_mtx); 683 txg_resume(dp); 684 return (EALREADY); 685 } else if (blk != NULL) { 686 /* 687 * This buffer had already been synced. It could not 688 * have been dirtied since, or we would have cleared blk. 689 */ 690 *bp = *blk; /* structure assignment */ 691 mutex_exit(&db->db_mtx); 692 txg_resume(dp); 693 return (0); 694 } 695 696 if (txg == tx->tx_syncing_txg) { 697 while (db->db_data_pending) { 698 /* 699 * IO is in-progress. Wait for it to finish. 700 * XXX - would be nice to be able to somehow "attach" 701 * this zio to the parent zio passed in. 702 */ 703 cv_wait(&db->db_changed, &db->db_mtx); 704 if (!db->db_data_pending && 705 db->db_blkptr && BP_IS_HOLE(db->db_blkptr)) { 706 /* 707 * IO was compressed away 708 */ 709 *bp = *db->db_blkptr; /* structure assignment */ 710 mutex_exit(&db->db_mtx); 711 txg_resume(dp); 712 return (0); 713 } 714 ASSERT(db->db_data_pending || 715 (db->db_blkptr && db->db_blkptr->blk_birth == txg)); 716 } 717 718 if (db->db_blkptr && db->db_blkptr->blk_birth == txg) { 719 /* 720 * IO is already completed. 721 */ 722 *bp = *db->db_blkptr; /* structure assignment */ 723 mutex_exit(&db->db_mtx); 724 txg_resume(dp); 725 return (0); 726 } 727 } 728 729 if (db->db_d.db_data_old[txg&TXG_MASK] == NULL) { 730 /* 731 * This dbuf isn't dirty, must have been free_range'd. 732 * There's no need to log writes to freed blocks, so we're done. 733 */ 734 mutex_exit(&db->db_mtx); 735 txg_resume(dp); 736 return (ENOENT); 737 } 738 739 ASSERT(db->db_d.db_overridden_by[txg&TXG_MASK] == NULL); 740 db->db_d.db_overridden_by[txg&TXG_MASK] = IN_DMU_SYNC; 741 /* 742 * XXX - a little ugly to stash the blkptr in the callback 743 * buffer. We always need to make sure the following is true: 744 * ASSERT(sizeof(blkptr_t) >= sizeof(dmu_sync_cbin_t)); 745 */ 746 in = kmem_alloc(sizeof (blkptr_t), KM_SLEEP); 747 in->db = db; 748 in->txg = txg; 749 in->done = done; 750 in->arg = arg; 751 mutex_exit(&db->db_mtx); 752 txg_resume(dp); 753 754 arc_flag = pio == NULL ? ARC_WAIT : ARC_NOWAIT; 755 zb.zb_objset = os->os_dsl_dataset->ds_object; 756 zb.zb_object = db->db.db_object; 757 zb.zb_level = db->db_level; 758 zb.zb_blkid = db->db_blkid; 759 err = arc_write(pio, os->os_spa, 760 zio_checksum_select(db->db_dnode->dn_checksum, os->os_checksum), 761 zio_compress_select(db->db_dnode->dn_compress, os->os_compress), 762 dmu_get_replication_level(os->os_spa, &zb, db->db_dnode->dn_type), 763 txg, bp, db->db_d.db_data_old[txg&TXG_MASK], dmu_sync_done, in, 764 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, arc_flag, &zb); 765 ASSERT(err == 0); 766 767 return (arc_flag == ARC_NOWAIT ? EINPROGRESS : 0); 768 } 769 770 int 771 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, 772 dmu_tx_t *tx) 773 { 774 dnode_t *dn; 775 int err; 776 777 err = dnode_hold(os->os, object, FTAG, &dn); 778 if (err) 779 return (err); 780 err = dnode_set_blksz(dn, size, ibs, tx); 781 dnode_rele(dn, FTAG); 782 return (err); 783 } 784 785 void 786 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 787 dmu_tx_t *tx) 788 { 789 dnode_t *dn; 790 791 /* XXX assumes dnode_hold will not get an i/o error */ 792 (void) dnode_hold(os->os, object, FTAG, &dn); 793 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS); 794 dn->dn_checksum = checksum; 795 dnode_setdirty(dn, tx); 796 dnode_rele(dn, FTAG); 797 } 798 799 void 800 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 801 dmu_tx_t *tx) 802 { 803 dnode_t *dn; 804 805 /* XXX assumes dnode_hold will not get an i/o error */ 806 (void) dnode_hold(os->os, object, FTAG, &dn); 807 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS); 808 dn->dn_compress = compress; 809 dnode_setdirty(dn, tx); 810 dnode_rele(dn, FTAG); 811 } 812 813 /* 814 * XXX - eventually, this should take into account per-dataset (or 815 * even per-object?) user requests for higher levels of replication. 816 */ 817 int 818 dmu_get_replication_level(spa_t *spa, zbookmark_t *zb, dmu_object_type_t ot) 819 { 820 int ncopies = 1; 821 822 if (dmu_ot[ot].ot_metadata) 823 ncopies++; 824 if (zb->zb_level != 0) 825 ncopies++; 826 if (zb->zb_objset == 0 && zb->zb_object == 0) 827 ncopies++; 828 return (MIN(ncopies, spa_max_replication(spa))); 829 } 830 831 int 832 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) 833 { 834 dnode_t *dn; 835 int i, err; 836 837 err = dnode_hold(os->os, object, FTAG, &dn); 838 if (err) 839 return (err); 840 /* 841 * Sync any current changes before 842 * we go trundling through the block pointers. 843 */ 844 for (i = 0; i < TXG_SIZE; i++) { 845 if (list_link_active(&dn->dn_dirty_link[i])) 846 break; 847 } 848 if (i != TXG_SIZE) { 849 dnode_rele(dn, FTAG); 850 txg_wait_synced(dmu_objset_pool(os), 0); 851 err = dnode_hold(os->os, object, FTAG, &dn); 852 if (err) 853 return (err); 854 } 855 856 err = dnode_next_offset(dn, hole, off, 1, 1, 0); 857 dnode_rele(dn, FTAG); 858 859 return (err); 860 } 861 862 void 863 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) 864 { 865 rw_enter(&dn->dn_struct_rwlock, RW_READER); 866 mutex_enter(&dn->dn_mtx); 867 868 doi->doi_data_block_size = dn->dn_datablksz; 869 doi->doi_metadata_block_size = dn->dn_indblkshift ? 870 1ULL << dn->dn_indblkshift : 0; 871 doi->doi_indirection = dn->dn_nlevels; 872 doi->doi_checksum = dn->dn_checksum; 873 doi->doi_compress = dn->dn_compress; 874 doi->doi_physical_blks = (DN_USED_BYTES(dn->dn_phys) + 875 SPA_MINBLOCKSIZE/2) >> SPA_MINBLOCKSHIFT; 876 doi->doi_max_block_offset = dn->dn_phys->dn_maxblkid; 877 doi->doi_type = dn->dn_type; 878 doi->doi_bonus_size = dn->dn_bonuslen; 879 doi->doi_bonus_type = dn->dn_bonustype; 880 881 mutex_exit(&dn->dn_mtx); 882 rw_exit(&dn->dn_struct_rwlock); 883 } 884 885 /* 886 * Get information on a DMU object. 887 * If doi is NULL, just indicates whether the object exists. 888 */ 889 int 890 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) 891 { 892 dnode_t *dn; 893 int err = dnode_hold(os->os, object, FTAG, &dn); 894 895 if (err) 896 return (err); 897 898 if (doi != NULL) 899 dmu_object_info_from_dnode(dn, doi); 900 901 dnode_rele(dn, FTAG); 902 return (0); 903 } 904 905 /* 906 * As above, but faster; can be used when you have a held dbuf in hand. 907 */ 908 void 909 dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi) 910 { 911 dmu_object_info_from_dnode(((dmu_buf_impl_t *)db)->db_dnode, doi); 912 } 913 914 /* 915 * Faster still when you only care about the size. 916 * This is specifically optimized for zfs_getattr(). 917 */ 918 void 919 dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, u_longlong_t *nblk512) 920 { 921 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 922 923 *blksize = dn->dn_datablksz; 924 /* add 1 for dnode space */ 925 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> 926 SPA_MINBLOCKSHIFT) + 1; 927 } 928 929 void 930 byteswap_uint64_array(void *vbuf, size_t size) 931 { 932 uint64_t *buf = vbuf; 933 size_t count = size >> 3; 934 int i; 935 936 ASSERT((size & 7) == 0); 937 938 for (i = 0; i < count; i++) 939 buf[i] = BSWAP_64(buf[i]); 940 } 941 942 void 943 byteswap_uint32_array(void *vbuf, size_t size) 944 { 945 uint32_t *buf = vbuf; 946 size_t count = size >> 2; 947 int i; 948 949 ASSERT((size & 3) == 0); 950 951 for (i = 0; i < count; i++) 952 buf[i] = BSWAP_32(buf[i]); 953 } 954 955 void 956 byteswap_uint16_array(void *vbuf, size_t size) 957 { 958 uint16_t *buf = vbuf; 959 size_t count = size >> 1; 960 int i; 961 962 ASSERT((size & 1) == 0); 963 964 for (i = 0; i < count; i++) 965 buf[i] = BSWAP_16(buf[i]); 966 } 967 968 /* ARGSUSED */ 969 void 970 byteswap_uint8_array(void *vbuf, size_t size) 971 { 972 } 973 974 void 975 dmu_init(void) 976 { 977 dbuf_init(); 978 dnode_init(); 979 arc_init(); 980 } 981 982 void 983 dmu_fini(void) 984 { 985 arc_fini(); 986 dnode_fini(); 987 dbuf_fini(); 988 } 989