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 #include <sys/sa.h> 44 #ifdef _KERNEL 45 #include <sys/vmsystm.h> 46 #include <sys/zfs_znode.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_oldacl_byteswap, TRUE, "ZFS V0 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 { zap_byteswap, TRUE, "Pool properties" }, 82 { zap_byteswap, TRUE, "DSL permissions" }, 83 { zfs_acl_byteswap, TRUE, "ZFS ACL" }, 84 { byteswap_uint8_array, TRUE, "ZFS SYSACL" }, 85 { byteswap_uint8_array, TRUE, "FUID table" }, 86 { byteswap_uint64_array, TRUE, "FUID table size" }, 87 { zap_byteswap, TRUE, "DSL dataset next clones"}, 88 { zap_byteswap, TRUE, "scrub work queue" }, 89 { zap_byteswap, TRUE, "ZFS user/group used" }, 90 { zap_byteswap, TRUE, "ZFS user/group quota" }, 91 { zap_byteswap, TRUE, "snapshot refcount tags"}, 92 { zap_byteswap, TRUE, "DDT ZAP algorithm" }, 93 { zap_byteswap, TRUE, "DDT statistics" }, 94 { byteswap_uint8_array, TRUE, "System attributes" }, 95 { zap_byteswap, TRUE, "SA master node" }, 96 { zap_byteswap, TRUE, "SA attr registration" }, 97 { zap_byteswap, TRUE, "SA attr layouts" }, }; 98 99 int 100 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 101 void *tag, dmu_buf_t **dbp) 102 { 103 dnode_t *dn; 104 uint64_t blkid; 105 dmu_buf_impl_t *db; 106 int err; 107 108 err = dnode_hold(os, object, FTAG, &dn); 109 if (err) 110 return (err); 111 blkid = dbuf_whichblock(dn, offset); 112 rw_enter(&dn->dn_struct_rwlock, RW_READER); 113 db = dbuf_hold(dn, blkid, tag); 114 rw_exit(&dn->dn_struct_rwlock); 115 if (db == NULL) { 116 err = EIO; 117 } else { 118 err = dbuf_read(db, NULL, DB_RF_CANFAIL); 119 if (err) { 120 dbuf_rele(db, tag); 121 db = NULL; 122 } 123 } 124 125 dnode_rele(dn, FTAG); 126 *dbp = &db->db; 127 return (err); 128 } 129 130 int 131 dmu_bonus_max(void) 132 { 133 return (DN_MAX_BONUSLEN); 134 } 135 136 int 137 dmu_set_bonus(dmu_buf_t *db, int newsize, dmu_tx_t *tx) 138 { 139 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 140 141 if (dn->dn_bonus != (dmu_buf_impl_t *)db) 142 return (EINVAL); 143 if (newsize < 0 || newsize > db->db_size) 144 return (EINVAL); 145 dnode_setbonuslen(dn, newsize, tx); 146 return (0); 147 } 148 149 int 150 dmu_set_bonustype(dmu_buf_t *db, dmu_object_type_t type, dmu_tx_t *tx) 151 { 152 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 153 154 if (type > DMU_OT_NUMTYPES) 155 return (EINVAL); 156 157 if (dn->dn_bonus != (dmu_buf_impl_t *)db) 158 return (EINVAL); 159 160 dnode_setbonus_type(dn, type, tx); 161 return (0); 162 } 163 164 int 165 dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) 166 { 167 dnode_t *dn; 168 int error; 169 170 error = dnode_hold(os, object, FTAG, &dn); 171 dbuf_rm_spill(dn, tx); 172 dnode_rele(dn, FTAG); 173 return (error); 174 } 175 176 /* 177 * returns ENOENT, EIO, or 0. 178 */ 179 int 180 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) 181 { 182 dnode_t *dn; 183 dmu_buf_impl_t *db; 184 int error; 185 186 error = dnode_hold(os, object, FTAG, &dn); 187 if (error) 188 return (error); 189 190 rw_enter(&dn->dn_struct_rwlock, RW_READER); 191 if (dn->dn_bonus == NULL) { 192 rw_exit(&dn->dn_struct_rwlock); 193 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 194 if (dn->dn_bonus == NULL) 195 dbuf_create_bonus(dn); 196 } 197 db = dn->dn_bonus; 198 rw_exit(&dn->dn_struct_rwlock); 199 200 /* as long as the bonus buf is held, the dnode will be held */ 201 if (refcount_add(&db->db_holds, tag) == 1) 202 VERIFY(dnode_add_ref(dn, db)); 203 204 dnode_rele(dn, FTAG); 205 206 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED)); 207 208 *dbp = &db->db; 209 return (0); 210 } 211 212 /* 213 * returns ENOENT, EIO, or 0. 214 * 215 * This interface will allocate a blank spill dbuf when a spill blk 216 * doesn't already exist on the dnode. 217 * 218 * if you only want to find an already existing spill db, then 219 * dmu_spill_hold_existing() should be used. 220 */ 221 int 222 dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) 223 { 224 dmu_buf_impl_t *db = NULL; 225 int err; 226 227 if ((flags & DB_RF_HAVESTRUCT) == 0) 228 rw_enter(&dn->dn_struct_rwlock, RW_READER); 229 230 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); 231 232 if ((flags & DB_RF_HAVESTRUCT) == 0) 233 rw_exit(&dn->dn_struct_rwlock); 234 235 ASSERT(db != NULL); 236 err = dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | flags); 237 *dbp = &db->db; 238 return (err); 239 } 240 241 int 242 dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) 243 { 244 dnode_t *dn = ((dmu_buf_impl_t *)bonus)->db_dnode; 245 int err; 246 247 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) 248 return (EINVAL); 249 rw_enter(&dn->dn_struct_rwlock, RW_READER); 250 251 if (!dn->dn_have_spill) { 252 rw_exit(&dn->dn_struct_rwlock); 253 return (ENOENT); 254 } 255 err = dmu_spill_hold_by_dnode(dn, DB_RF_HAVESTRUCT, tag, dbp); 256 rw_exit(&dn->dn_struct_rwlock); 257 return (err); 258 } 259 260 int 261 dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) 262 { 263 return (dmu_spill_hold_by_dnode(((dmu_buf_impl_t *)bonus)->db_dnode, 264 0, tag, dbp)); 265 } 266 267 /* 268 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces 269 * to take a held dnode rather than <os, object> -- the lookup is wasteful, 270 * and can induce severe lock contention when writing to several files 271 * whose dnodes are in the same block. 272 */ 273 static int 274 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, 275 int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) 276 { 277 dsl_pool_t *dp = NULL; 278 dmu_buf_t **dbp; 279 uint64_t blkid, nblks, i; 280 uint32_t dbuf_flags; 281 int err; 282 zio_t *zio; 283 hrtime_t start; 284 285 ASSERT(length <= DMU_MAX_ACCESS); 286 287 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT; 288 if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz) 289 dbuf_flags |= DB_RF_NOPREFETCH; 290 291 rw_enter(&dn->dn_struct_rwlock, RW_READER); 292 if (dn->dn_datablkshift) { 293 int blkshift = dn->dn_datablkshift; 294 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - 295 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; 296 } else { 297 if (offset + length > dn->dn_datablksz) { 298 zfs_panic_recover("zfs: accessing past end of object " 299 "%llx/%llx (size=%u access=%llu+%llu)", 300 (longlong_t)dn->dn_objset-> 301 os_dsl_dataset->ds_object, 302 (longlong_t)dn->dn_object, dn->dn_datablksz, 303 (longlong_t)offset, (longlong_t)length); 304 rw_exit(&dn->dn_struct_rwlock); 305 return (EIO); 306 } 307 nblks = 1; 308 } 309 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); 310 311 if (dn->dn_objset->os_dsl_dataset) 312 dp = dn->dn_objset->os_dsl_dataset->ds_dir->dd_pool; 313 if (dp && dsl_pool_sync_context(dp)) 314 start = gethrtime(); 315 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); 316 blkid = dbuf_whichblock(dn, offset); 317 for (i = 0; i < nblks; i++) { 318 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); 319 if (db == NULL) { 320 rw_exit(&dn->dn_struct_rwlock); 321 dmu_buf_rele_array(dbp, nblks, tag); 322 zio_nowait(zio); 323 return (EIO); 324 } 325 /* initiate async i/o */ 326 if (read) { 327 (void) dbuf_read(db, zio, dbuf_flags); 328 } 329 dbp[i] = &db->db; 330 } 331 rw_exit(&dn->dn_struct_rwlock); 332 333 /* wait for async i/o */ 334 err = zio_wait(zio); 335 /* track read overhead when we are in sync context */ 336 if (dp && dsl_pool_sync_context(dp)) 337 dp->dp_read_overhead += gethrtime() - start; 338 if (err) { 339 dmu_buf_rele_array(dbp, nblks, tag); 340 return (err); 341 } 342 343 /* wait for other io to complete */ 344 if (read) { 345 for (i = 0; i < nblks; i++) { 346 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; 347 mutex_enter(&db->db_mtx); 348 while (db->db_state == DB_READ || 349 db->db_state == DB_FILL) 350 cv_wait(&db->db_changed, &db->db_mtx); 351 if (db->db_state == DB_UNCACHED) 352 err = EIO; 353 mutex_exit(&db->db_mtx); 354 if (err) { 355 dmu_buf_rele_array(dbp, nblks, tag); 356 return (err); 357 } 358 } 359 } 360 361 *numbufsp = nblks; 362 *dbpp = dbp; 363 return (0); 364 } 365 366 static int 367 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, 368 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 369 { 370 dnode_t *dn; 371 int err; 372 373 err = dnode_hold(os, object, FTAG, &dn); 374 if (err) 375 return (err); 376 377 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 378 numbufsp, dbpp, DMU_READ_PREFETCH); 379 380 dnode_rele(dn, FTAG); 381 382 return (err); 383 } 384 385 int 386 dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, 387 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 388 { 389 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 390 int err; 391 392 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 393 numbufsp, dbpp, DMU_READ_PREFETCH); 394 395 return (err); 396 } 397 398 void 399 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) 400 { 401 int i; 402 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; 403 404 if (numbufs == 0) 405 return; 406 407 for (i = 0; i < numbufs; i++) { 408 if (dbp[i]) 409 dbuf_rele(dbp[i], tag); 410 } 411 412 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); 413 } 414 415 void 416 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) 417 { 418 dnode_t *dn; 419 uint64_t blkid; 420 int nblks, i, err; 421 422 if (zfs_prefetch_disable) 423 return; 424 425 if (len == 0) { /* they're interested in the bonus buffer */ 426 dn = os->os_meta_dnode; 427 428 if (object == 0 || object >= DN_MAX_OBJECT) 429 return; 430 431 rw_enter(&dn->dn_struct_rwlock, RW_READER); 432 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); 433 dbuf_prefetch(dn, blkid); 434 rw_exit(&dn->dn_struct_rwlock); 435 return; 436 } 437 438 /* 439 * XXX - Note, if the dnode for the requested object is not 440 * already cached, we will do a *synchronous* read in the 441 * dnode_hold() call. The same is true for any indirects. 442 */ 443 err = dnode_hold(os, object, FTAG, &dn); 444 if (err != 0) 445 return; 446 447 rw_enter(&dn->dn_struct_rwlock, RW_READER); 448 if (dn->dn_datablkshift) { 449 int blkshift = dn->dn_datablkshift; 450 nblks = (P2ROUNDUP(offset+len, 1<<blkshift) - 451 P2ALIGN(offset, 1<<blkshift)) >> blkshift; 452 } else { 453 nblks = (offset < dn->dn_datablksz); 454 } 455 456 if (nblks != 0) { 457 blkid = dbuf_whichblock(dn, offset); 458 for (i = 0; i < nblks; i++) 459 dbuf_prefetch(dn, blkid+i); 460 } 461 462 rw_exit(&dn->dn_struct_rwlock); 463 464 dnode_rele(dn, FTAG); 465 } 466 467 /* 468 * Get the next "chunk" of file data to free. We traverse the file from 469 * the end so that the file gets shorter over time (if we crashes in the 470 * middle, this will leave us in a better state). We find allocated file 471 * data by simply searching the allocated level 1 indirects. 472 */ 473 static int 474 get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t limit) 475 { 476 uint64_t len = *start - limit; 477 uint64_t blkcnt = 0; 478 uint64_t maxblks = DMU_MAX_ACCESS / (1ULL << (dn->dn_indblkshift + 1)); 479 uint64_t iblkrange = 480 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); 481 482 ASSERT(limit <= *start); 483 484 if (len <= iblkrange * maxblks) { 485 *start = limit; 486 return (0); 487 } 488 ASSERT(ISP2(iblkrange)); 489 490 while (*start > limit && blkcnt < maxblks) { 491 int err; 492 493 /* find next allocated L1 indirect */ 494 err = dnode_next_offset(dn, 495 DNODE_FIND_BACKWARDS, start, 2, 1, 0); 496 497 /* if there are no more, then we are done */ 498 if (err == ESRCH) { 499 *start = limit; 500 return (0); 501 } else if (err) { 502 return (err); 503 } 504 blkcnt += 1; 505 506 /* reset offset to end of "next" block back */ 507 *start = P2ALIGN(*start, iblkrange); 508 if (*start <= limit) 509 *start = limit; 510 else 511 *start -= 1; 512 } 513 return (0); 514 } 515 516 static int 517 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, 518 uint64_t length, boolean_t free_dnode) 519 { 520 dmu_tx_t *tx; 521 uint64_t object_size, start, end, len; 522 boolean_t trunc = (length == DMU_OBJECT_END); 523 int align, err; 524 525 align = 1 << dn->dn_datablkshift; 526 ASSERT(align > 0); 527 object_size = align == 1 ? dn->dn_datablksz : 528 (dn->dn_maxblkid + 1) << dn->dn_datablkshift; 529 530 end = offset + length; 531 if (trunc || end > object_size) 532 end = object_size; 533 if (end <= offset) 534 return (0); 535 length = end - offset; 536 537 while (length) { 538 start = end; 539 /* assert(offset <= start) */ 540 err = get_next_chunk(dn, &start, offset); 541 if (err) 542 return (err); 543 len = trunc ? DMU_OBJECT_END : end - start; 544 545 tx = dmu_tx_create(os); 546 dmu_tx_hold_free(tx, dn->dn_object, start, len); 547 err = dmu_tx_assign(tx, TXG_WAIT); 548 if (err) { 549 dmu_tx_abort(tx); 550 return (err); 551 } 552 553 dnode_free_range(dn, start, trunc ? -1 : len, tx); 554 555 if (start == 0 && free_dnode) { 556 ASSERT(trunc); 557 dnode_free(dn, tx); 558 } 559 560 length -= end - start; 561 562 dmu_tx_commit(tx); 563 end = start; 564 } 565 return (0); 566 } 567 568 int 569 dmu_free_long_range(objset_t *os, uint64_t object, 570 uint64_t offset, uint64_t length) 571 { 572 dnode_t *dn; 573 int err; 574 575 err = dnode_hold(os, object, FTAG, &dn); 576 if (err != 0) 577 return (err); 578 err = dmu_free_long_range_impl(os, dn, offset, length, FALSE); 579 dnode_rele(dn, FTAG); 580 return (err); 581 } 582 583 int 584 dmu_free_object(objset_t *os, uint64_t object) 585 { 586 dnode_t *dn; 587 dmu_tx_t *tx; 588 int err; 589 590 err = dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, 591 FTAG, &dn); 592 if (err != 0) 593 return (err); 594 if (dn->dn_nlevels == 1) { 595 tx = dmu_tx_create(os); 596 dmu_tx_hold_bonus(tx, object); 597 dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END); 598 err = dmu_tx_assign(tx, TXG_WAIT); 599 if (err == 0) { 600 dnode_free_range(dn, 0, DMU_OBJECT_END, tx); 601 dnode_free(dn, tx); 602 dmu_tx_commit(tx); 603 } else { 604 dmu_tx_abort(tx); 605 } 606 } else { 607 err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE); 608 } 609 dnode_rele(dn, FTAG); 610 return (err); 611 } 612 613 int 614 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 615 uint64_t size, dmu_tx_t *tx) 616 { 617 dnode_t *dn; 618 int err = dnode_hold(os, object, FTAG, &dn); 619 if (err) 620 return (err); 621 ASSERT(offset < UINT64_MAX); 622 ASSERT(size == -1ULL || size <= UINT64_MAX - offset); 623 dnode_free_range(dn, offset, size, tx); 624 dnode_rele(dn, FTAG); 625 return (0); 626 } 627 628 int 629 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 630 void *buf, uint32_t flags) 631 { 632 dnode_t *dn; 633 dmu_buf_t **dbp; 634 int numbufs, err; 635 636 err = dnode_hold(os, object, FTAG, &dn); 637 if (err) 638 return (err); 639 640 /* 641 * Deal with odd block sizes, where there can't be data past the first 642 * block. If we ever do the tail block optimization, we will need to 643 * handle that here as well. 644 */ 645 if (dn->dn_maxblkid == 0) { 646 int newsz = offset > dn->dn_datablksz ? 0 : 647 MIN(size, dn->dn_datablksz - offset); 648 bzero((char *)buf + newsz, size - newsz); 649 size = newsz; 650 } 651 652 while (size > 0) { 653 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); 654 int i; 655 656 /* 657 * NB: we could do this block-at-a-time, but it's nice 658 * to be reading in parallel. 659 */ 660 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, 661 TRUE, FTAG, &numbufs, &dbp, flags); 662 if (err) 663 break; 664 665 for (i = 0; i < numbufs; i++) { 666 int tocpy; 667 int bufoff; 668 dmu_buf_t *db = dbp[i]; 669 670 ASSERT(size > 0); 671 672 bufoff = offset - db->db_offset; 673 tocpy = (int)MIN(db->db_size - bufoff, size); 674 675 bcopy((char *)db->db_data + bufoff, buf, tocpy); 676 677 offset += tocpy; 678 size -= tocpy; 679 buf = (char *)buf + tocpy; 680 } 681 dmu_buf_rele_array(dbp, numbufs, FTAG); 682 } 683 dnode_rele(dn, FTAG); 684 return (err); 685 } 686 687 void 688 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 689 const void *buf, dmu_tx_t *tx) 690 { 691 dmu_buf_t **dbp; 692 int numbufs, i; 693 694 if (size == 0) 695 return; 696 697 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 698 FALSE, FTAG, &numbufs, &dbp)); 699 700 for (i = 0; i < numbufs; i++) { 701 int tocpy; 702 int bufoff; 703 dmu_buf_t *db = dbp[i]; 704 705 ASSERT(size > 0); 706 707 bufoff = offset - db->db_offset; 708 tocpy = (int)MIN(db->db_size - bufoff, size); 709 710 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 711 712 if (tocpy == db->db_size) 713 dmu_buf_will_fill(db, tx); 714 else 715 dmu_buf_will_dirty(db, tx); 716 717 bcopy(buf, (char *)db->db_data + bufoff, tocpy); 718 719 if (tocpy == db->db_size) 720 dmu_buf_fill_done(db, tx); 721 722 offset += tocpy; 723 size -= tocpy; 724 buf = (char *)buf + tocpy; 725 } 726 dmu_buf_rele_array(dbp, numbufs, FTAG); 727 } 728 729 void 730 dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 731 dmu_tx_t *tx) 732 { 733 dmu_buf_t **dbp; 734 int numbufs, i; 735 736 if (size == 0) 737 return; 738 739 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 740 FALSE, FTAG, &numbufs, &dbp)); 741 742 for (i = 0; i < numbufs; i++) { 743 dmu_buf_t *db = dbp[i]; 744 745 dmu_buf_will_not_fill(db, tx); 746 } 747 dmu_buf_rele_array(dbp, numbufs, FTAG); 748 } 749 750 /* 751 * DMU support for xuio 752 */ 753 kstat_t *xuio_ksp = NULL; 754 755 int 756 dmu_xuio_init(xuio_t *xuio, int nblk) 757 { 758 dmu_xuio_t *priv; 759 uio_t *uio = &xuio->xu_uio; 760 761 uio->uio_iovcnt = nblk; 762 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); 763 764 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); 765 priv->cnt = nblk; 766 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); 767 priv->iovp = uio->uio_iov; 768 XUIO_XUZC_PRIV(xuio) = priv; 769 770 if (XUIO_XUZC_RW(xuio) == UIO_READ) 771 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); 772 else 773 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); 774 775 return (0); 776 } 777 778 void 779 dmu_xuio_fini(xuio_t *xuio) 780 { 781 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 782 int nblk = priv->cnt; 783 784 kmem_free(priv->iovp, nblk * sizeof (iovec_t)); 785 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); 786 kmem_free(priv, sizeof (dmu_xuio_t)); 787 788 if (XUIO_XUZC_RW(xuio) == UIO_READ) 789 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); 790 else 791 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); 792 } 793 794 /* 795 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } 796 * and increase priv->next by 1. 797 */ 798 int 799 dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) 800 { 801 struct iovec *iov; 802 uio_t *uio = &xuio->xu_uio; 803 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 804 int i = priv->next++; 805 806 ASSERT(i < priv->cnt); 807 ASSERT(off + n <= arc_buf_size(abuf)); 808 iov = uio->uio_iov + i; 809 iov->iov_base = (char *)abuf->b_data + off; 810 iov->iov_len = n; 811 priv->bufs[i] = abuf; 812 return (0); 813 } 814 815 int 816 dmu_xuio_cnt(xuio_t *xuio) 817 { 818 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 819 return (priv->cnt); 820 } 821 822 arc_buf_t * 823 dmu_xuio_arcbuf(xuio_t *xuio, int i) 824 { 825 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 826 827 ASSERT(i < priv->cnt); 828 return (priv->bufs[i]); 829 } 830 831 void 832 dmu_xuio_clear(xuio_t *xuio, int i) 833 { 834 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 835 836 ASSERT(i < priv->cnt); 837 priv->bufs[i] = NULL; 838 } 839 840 static void 841 xuio_stat_init(void) 842 { 843 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", 844 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), 845 KSTAT_FLAG_VIRTUAL); 846 if (xuio_ksp != NULL) { 847 xuio_ksp->ks_data = &xuio_stats; 848 kstat_install(xuio_ksp); 849 } 850 } 851 852 static void 853 xuio_stat_fini(void) 854 { 855 if (xuio_ksp != NULL) { 856 kstat_delete(xuio_ksp); 857 xuio_ksp = NULL; 858 } 859 } 860 861 void 862 xuio_stat_wbuf_copied() 863 { 864 XUIOSTAT_BUMP(xuiostat_wbuf_copied); 865 } 866 867 void 868 xuio_stat_wbuf_nocopy() 869 { 870 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); 871 } 872 873 #ifdef _KERNEL 874 int 875 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) 876 { 877 dmu_buf_t **dbp; 878 int numbufs, i, err; 879 xuio_t *xuio = NULL; 880 881 /* 882 * NB: we could do this block-at-a-time, but it's nice 883 * to be reading in parallel. 884 */ 885 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG, 886 &numbufs, &dbp); 887 if (err) 888 return (err); 889 890 if (uio->uio_extflg == UIO_XUIO) 891 xuio = (xuio_t *)uio; 892 893 for (i = 0; i < numbufs; i++) { 894 int tocpy; 895 int bufoff; 896 dmu_buf_t *db = dbp[i]; 897 898 ASSERT(size > 0); 899 900 bufoff = uio->uio_loffset - db->db_offset; 901 tocpy = (int)MIN(db->db_size - bufoff, size); 902 903 if (xuio) { 904 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; 905 arc_buf_t *dbuf_abuf = dbi->db_buf; 906 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); 907 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); 908 if (!err) { 909 uio->uio_resid -= tocpy; 910 uio->uio_loffset += tocpy; 911 } 912 913 if (abuf == dbuf_abuf) 914 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); 915 else 916 XUIOSTAT_BUMP(xuiostat_rbuf_copied); 917 } else { 918 err = uiomove((char *)db->db_data + bufoff, tocpy, 919 UIO_READ, uio); 920 } 921 if (err) 922 break; 923 924 size -= tocpy; 925 } 926 dmu_buf_rele_array(dbp, numbufs, FTAG); 927 928 return (err); 929 } 930 931 int 932 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, 933 dmu_tx_t *tx) 934 { 935 dmu_buf_t **dbp; 936 int numbufs, i; 937 int err = 0; 938 939 if (size == 0) 940 return (0); 941 942 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, 943 FALSE, FTAG, &numbufs, &dbp); 944 if (err) 945 return (err); 946 947 for (i = 0; i < numbufs; i++) { 948 int tocpy; 949 int bufoff; 950 dmu_buf_t *db = dbp[i]; 951 952 ASSERT(size > 0); 953 954 bufoff = uio->uio_loffset - db->db_offset; 955 tocpy = (int)MIN(db->db_size - bufoff, size); 956 957 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 958 959 if (tocpy == db->db_size) 960 dmu_buf_will_fill(db, tx); 961 else 962 dmu_buf_will_dirty(db, tx); 963 964 /* 965 * XXX uiomove could block forever (eg. nfs-backed 966 * pages). There needs to be a uiolockdown() function 967 * to lock the pages in memory, so that uiomove won't 968 * block. 969 */ 970 err = uiomove((char *)db->db_data + bufoff, tocpy, 971 UIO_WRITE, uio); 972 973 if (tocpy == db->db_size) 974 dmu_buf_fill_done(db, tx); 975 976 if (err) 977 break; 978 979 size -= tocpy; 980 } 981 dmu_buf_rele_array(dbp, numbufs, FTAG); 982 return (err); 983 } 984 985 int 986 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 987 page_t *pp, dmu_tx_t *tx) 988 { 989 dmu_buf_t **dbp; 990 int numbufs, i; 991 int err; 992 993 if (size == 0) 994 return (0); 995 996 err = dmu_buf_hold_array(os, object, offset, size, 997 FALSE, FTAG, &numbufs, &dbp); 998 if (err) 999 return (err); 1000 1001 for (i = 0; i < numbufs; i++) { 1002 int tocpy, copied, thiscpy; 1003 int bufoff; 1004 dmu_buf_t *db = dbp[i]; 1005 caddr_t va; 1006 1007 ASSERT(size > 0); 1008 ASSERT3U(db->db_size, >=, PAGESIZE); 1009 1010 bufoff = offset - db->db_offset; 1011 tocpy = (int)MIN(db->db_size - bufoff, size); 1012 1013 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 1014 1015 if (tocpy == db->db_size) 1016 dmu_buf_will_fill(db, tx); 1017 else 1018 dmu_buf_will_dirty(db, tx); 1019 1020 for (copied = 0; copied < tocpy; copied += PAGESIZE) { 1021 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); 1022 thiscpy = MIN(PAGESIZE, tocpy - copied); 1023 va = zfs_map_page(pp, S_READ); 1024 bcopy(va, (char *)db->db_data + bufoff, thiscpy); 1025 zfs_unmap_page(pp, va); 1026 pp = pp->p_next; 1027 bufoff += PAGESIZE; 1028 } 1029 1030 if (tocpy == db->db_size) 1031 dmu_buf_fill_done(db, tx); 1032 1033 offset += tocpy; 1034 size -= tocpy; 1035 } 1036 dmu_buf_rele_array(dbp, numbufs, FTAG); 1037 return (err); 1038 } 1039 #endif 1040 1041 /* 1042 * Allocate a loaned anonymous arc buffer. 1043 */ 1044 arc_buf_t * 1045 dmu_request_arcbuf(dmu_buf_t *handle, int size) 1046 { 1047 dnode_t *dn = ((dmu_buf_impl_t *)handle)->db_dnode; 1048 1049 return (arc_loan_buf(dn->dn_objset->os_spa, size)); 1050 } 1051 1052 /* 1053 * Free a loaned arc buffer. 1054 */ 1055 void 1056 dmu_return_arcbuf(arc_buf_t *buf) 1057 { 1058 arc_return_buf(buf, FTAG); 1059 VERIFY(arc_buf_remove_ref(buf, FTAG) == 1); 1060 } 1061 1062 /* 1063 * When possible directly assign passed loaned arc buffer to a dbuf. 1064 * If this is not possible copy the contents of passed arc buf via 1065 * dmu_write(). 1066 */ 1067 void 1068 dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, 1069 dmu_tx_t *tx) 1070 { 1071 dnode_t *dn = ((dmu_buf_impl_t *)handle)->db_dnode; 1072 dmu_buf_impl_t *db; 1073 uint32_t blksz = (uint32_t)arc_buf_size(buf); 1074 uint64_t blkid; 1075 1076 rw_enter(&dn->dn_struct_rwlock, RW_READER); 1077 blkid = dbuf_whichblock(dn, offset); 1078 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); 1079 rw_exit(&dn->dn_struct_rwlock); 1080 1081 if (offset == db->db.db_offset && blksz == db->db.db_size) { 1082 dbuf_assign_arcbuf(db, buf, tx); 1083 dbuf_rele(db, FTAG); 1084 } else { 1085 dbuf_rele(db, FTAG); 1086 dmu_write(dn->dn_objset, dn->dn_object, offset, blksz, 1087 buf->b_data, tx); 1088 dmu_return_arcbuf(buf); 1089 XUIOSTAT_BUMP(xuiostat_wbuf_copied); 1090 } 1091 } 1092 1093 typedef struct { 1094 dbuf_dirty_record_t *dsa_dr; 1095 dmu_sync_cb_t *dsa_done; 1096 zgd_t *dsa_zgd; 1097 dmu_tx_t *dsa_tx; 1098 } dmu_sync_arg_t; 1099 1100 /* ARGSUSED */ 1101 static void 1102 dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) 1103 { 1104 dmu_sync_arg_t *dsa = varg; 1105 dmu_buf_t *db = dsa->dsa_zgd->zgd_db; 1106 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 1107 blkptr_t *bp = zio->io_bp; 1108 1109 if (zio->io_error == 0) { 1110 if (BP_IS_HOLE(bp)) { 1111 /* 1112 * A block of zeros may compress to a hole, but the 1113 * block size still needs to be known for replay. 1114 */ 1115 BP_SET_LSIZE(bp, db->db_size); 1116 } else { 1117 ASSERT(BP_GET_TYPE(bp) == dn->dn_type); 1118 ASSERT(BP_GET_LEVEL(bp) == 0); 1119 bp->blk_fill = 1; 1120 } 1121 } 1122 } 1123 1124 static void 1125 dmu_sync_late_arrival_ready(zio_t *zio) 1126 { 1127 dmu_sync_ready(zio, NULL, zio->io_private); 1128 } 1129 1130 /* ARGSUSED */ 1131 static void 1132 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) 1133 { 1134 dmu_sync_arg_t *dsa = varg; 1135 dbuf_dirty_record_t *dr = dsa->dsa_dr; 1136 dmu_buf_impl_t *db = dr->dr_dbuf; 1137 1138 mutex_enter(&db->db_mtx); 1139 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); 1140 if (zio->io_error == 0) { 1141 dr->dt.dl.dr_overridden_by = *zio->io_bp; 1142 dr->dt.dl.dr_override_state = DR_OVERRIDDEN; 1143 dr->dt.dl.dr_copies = zio->io_prop.zp_copies; 1144 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by)) 1145 BP_ZERO(&dr->dt.dl.dr_overridden_by); 1146 } else { 1147 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; 1148 } 1149 cv_broadcast(&db->db_changed); 1150 mutex_exit(&db->db_mtx); 1151 1152 dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 1153 1154 kmem_free(dsa, sizeof (*dsa)); 1155 } 1156 1157 static void 1158 dmu_sync_late_arrival_done(zio_t *zio) 1159 { 1160 blkptr_t *bp = zio->io_bp; 1161 dmu_sync_arg_t *dsa = zio->io_private; 1162 1163 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { 1164 ASSERT(zio->io_bp->blk_birth == zio->io_txg); 1165 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); 1166 zio_free(zio->io_spa, zio->io_txg, zio->io_bp); 1167 } 1168 1169 dmu_tx_commit(dsa->dsa_tx); 1170 1171 dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 1172 1173 kmem_free(dsa, sizeof (*dsa)); 1174 } 1175 1176 static int 1177 dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, 1178 zio_prop_t *zp, zbookmark_t *zb) 1179 { 1180 dmu_sync_arg_t *dsa; 1181 dmu_tx_t *tx; 1182 1183 tx = dmu_tx_create(os); 1184 dmu_tx_hold_space(tx, zgd->zgd_db->db_size); 1185 if (dmu_tx_assign(tx, TXG_WAIT) != 0) { 1186 dmu_tx_abort(tx); 1187 return (EIO); /* Make zl_get_data do txg_waited_synced() */ 1188 } 1189 1190 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1191 dsa->dsa_dr = NULL; 1192 dsa->dsa_done = done; 1193 dsa->dsa_zgd = zgd; 1194 dsa->dsa_tx = tx; 1195 1196 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, 1197 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, 1198 dmu_sync_late_arrival_ready, dmu_sync_late_arrival_done, dsa, 1199 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); 1200 1201 return (0); 1202 } 1203 1204 /* 1205 * Intent log support: sync the block associated with db to disk. 1206 * N.B. and XXX: the caller is responsible for making sure that the 1207 * data isn't changing while dmu_sync() is writing it. 1208 * 1209 * Return values: 1210 * 1211 * EEXIST: this txg has already been synced, so there's nothing to to. 1212 * The caller should not log the write. 1213 * 1214 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. 1215 * The caller should not log the write. 1216 * 1217 * EALREADY: this block is already in the process of being synced. 1218 * The caller should track its progress (somehow). 1219 * 1220 * EIO: could not do the I/O. 1221 * The caller should do a txg_wait_synced(). 1222 * 1223 * 0: the I/O has been initiated. 1224 * The caller should log this blkptr in the done callback. 1225 * It is possible that the I/O will fail, in which case 1226 * the error will be reported to the done callback and 1227 * propagated to pio from zio_done(). 1228 */ 1229 int 1230 dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) 1231 { 1232 blkptr_t *bp = zgd->zgd_bp; 1233 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; 1234 objset_t *os = db->db_objset; 1235 dsl_dataset_t *ds = os->os_dsl_dataset; 1236 dbuf_dirty_record_t *dr; 1237 dmu_sync_arg_t *dsa; 1238 zbookmark_t zb; 1239 zio_prop_t zp; 1240 1241 ASSERT(pio != NULL); 1242 ASSERT(BP_IS_HOLE(bp)); 1243 ASSERT(txg != 0); 1244 1245 SET_BOOKMARK(&zb, ds->ds_object, 1246 db->db.db_object, db->db_level, db->db_blkid); 1247 1248 dmu_write_policy(os, db->db_dnode, db->db_level, WP_DMU_SYNC, &zp); 1249 1250 /* 1251 * If we're frozen (running ziltest), we always need to generate a bp. 1252 */ 1253 if (txg > spa_freeze_txg(os->os_spa)) 1254 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1255 1256 /* 1257 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() 1258 * and us. If we determine that this txg is not yet syncing, 1259 * but it begins to sync a moment later, that's OK because the 1260 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. 1261 */ 1262 mutex_enter(&db->db_mtx); 1263 1264 if (txg <= spa_last_synced_txg(os->os_spa)) { 1265 /* 1266 * This txg has already synced. There's nothing to do. 1267 */ 1268 mutex_exit(&db->db_mtx); 1269 return (EEXIST); 1270 } 1271 1272 if (txg <= spa_syncing_txg(os->os_spa)) { 1273 /* 1274 * This txg is currently syncing, so we can't mess with 1275 * the dirty record anymore; just write a new log block. 1276 */ 1277 mutex_exit(&db->db_mtx); 1278 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1279 } 1280 1281 dr = db->db_last_dirty; 1282 while (dr && dr->dr_txg != txg) 1283 dr = dr->dr_next; 1284 1285 if (dr == NULL) { 1286 /* 1287 * There's no dr for this dbuf, so it must have been freed. 1288 * There's no need to log writes to freed blocks, so we're done. 1289 */ 1290 mutex_exit(&db->db_mtx); 1291 return (ENOENT); 1292 } 1293 1294 ASSERT(dr->dr_txg == txg); 1295 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || 1296 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { 1297 /* 1298 * We have already issued a sync write for this buffer, 1299 * or this buffer has already been synced. It could not 1300 * have been dirtied since, or we would have cleared the state. 1301 */ 1302 mutex_exit(&db->db_mtx); 1303 return (EALREADY); 1304 } 1305 1306 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); 1307 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; 1308 mutex_exit(&db->db_mtx); 1309 1310 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1311 dsa->dsa_dr = dr; 1312 dsa->dsa_done = done; 1313 dsa->dsa_zgd = zgd; 1314 dsa->dsa_tx = NULL; 1315 1316 zio_nowait(arc_write(pio, os->os_spa, txg, 1317 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp, 1318 dmu_sync_ready, dmu_sync_done, dsa, 1319 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); 1320 1321 return (0); 1322 } 1323 1324 int 1325 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, 1326 dmu_tx_t *tx) 1327 { 1328 dnode_t *dn; 1329 int err; 1330 1331 err = dnode_hold(os, object, FTAG, &dn); 1332 if (err) 1333 return (err); 1334 err = dnode_set_blksz(dn, size, ibs, tx); 1335 dnode_rele(dn, FTAG); 1336 return (err); 1337 } 1338 1339 void 1340 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 1341 dmu_tx_t *tx) 1342 { 1343 dnode_t *dn; 1344 1345 /* XXX assumes dnode_hold will not get an i/o error */ 1346 (void) dnode_hold(os, object, FTAG, &dn); 1347 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS); 1348 dn->dn_checksum = checksum; 1349 dnode_setdirty(dn, tx); 1350 dnode_rele(dn, FTAG); 1351 } 1352 1353 void 1354 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 1355 dmu_tx_t *tx) 1356 { 1357 dnode_t *dn; 1358 1359 /* XXX assumes dnode_hold will not get an i/o error */ 1360 (void) dnode_hold(os, object, FTAG, &dn); 1361 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS); 1362 dn->dn_compress = compress; 1363 dnode_setdirty(dn, tx); 1364 dnode_rele(dn, FTAG); 1365 } 1366 1367 int zfs_mdcomp_disable = 0; 1368 1369 void 1370 dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) 1371 { 1372 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; 1373 boolean_t ismd = (level > 0 || dmu_ot[type].ot_metadata); 1374 enum zio_checksum checksum = os->os_checksum; 1375 enum zio_compress compress = os->os_compress; 1376 enum zio_checksum dedup_checksum = os->os_dedup_checksum; 1377 boolean_t dedup; 1378 boolean_t dedup_verify = os->os_dedup_verify; 1379 int copies = os->os_copies; 1380 1381 /* 1382 * Determine checksum setting. 1383 */ 1384 if (ismd) { 1385 /* 1386 * Metadata always gets checksummed. If the data 1387 * checksum is multi-bit correctable, and it's not a 1388 * ZBT-style checksum, then it's suitable for metadata 1389 * as well. Otherwise, the metadata checksum defaults 1390 * to fletcher4. 1391 */ 1392 if (zio_checksum_table[checksum].ci_correctable < 1 || 1393 zio_checksum_table[checksum].ci_eck) 1394 checksum = ZIO_CHECKSUM_FLETCHER_4; 1395 } else { 1396 checksum = zio_checksum_select(dn->dn_checksum, checksum); 1397 } 1398 1399 /* 1400 * Determine compression setting. 1401 */ 1402 if (ismd) { 1403 /* 1404 * XXX -- we should design a compression algorithm 1405 * that specializes in arrays of bps. 1406 */ 1407 compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY : 1408 ZIO_COMPRESS_LZJB; 1409 } else { 1410 compress = zio_compress_select(dn->dn_compress, compress); 1411 } 1412 1413 /* 1414 * Determine dedup setting. If we are in dmu_sync(), we won't 1415 * actually dedup now because that's all done in syncing context; 1416 * but we do want to use the dedup checkum. If the checksum is not 1417 * strong enough to ensure unique signatures, force dedup_verify. 1418 */ 1419 dedup = (!ismd && dedup_checksum != ZIO_CHECKSUM_OFF); 1420 if (dedup) { 1421 checksum = dedup_checksum; 1422 if (!zio_checksum_table[checksum].ci_dedup) 1423 dedup_verify = 1; 1424 } 1425 1426 if (wp & WP_DMU_SYNC) 1427 dedup = 0; 1428 1429 if (wp & WP_NOFILL) { 1430 ASSERT(!ismd && level == 0); 1431 checksum = ZIO_CHECKSUM_OFF; 1432 compress = ZIO_COMPRESS_OFF; 1433 dedup = B_FALSE; 1434 } 1435 1436 zp->zp_checksum = checksum; 1437 zp->zp_compress = compress; 1438 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; 1439 zp->zp_level = level; 1440 zp->zp_copies = MIN(copies + ismd, spa_max_replication(os->os_spa)); 1441 zp->zp_dedup = dedup; 1442 zp->zp_dedup_verify = dedup && dedup_verify; 1443 } 1444 1445 int 1446 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) 1447 { 1448 dnode_t *dn; 1449 int i, err; 1450 1451 err = dnode_hold(os, object, FTAG, &dn); 1452 if (err) 1453 return (err); 1454 /* 1455 * Sync any current changes before 1456 * we go trundling through the block pointers. 1457 */ 1458 for (i = 0; i < TXG_SIZE; i++) { 1459 if (list_link_active(&dn->dn_dirty_link[i])) 1460 break; 1461 } 1462 if (i != TXG_SIZE) { 1463 dnode_rele(dn, FTAG); 1464 txg_wait_synced(dmu_objset_pool(os), 0); 1465 err = dnode_hold(os, object, FTAG, &dn); 1466 if (err) 1467 return (err); 1468 } 1469 1470 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); 1471 dnode_rele(dn, FTAG); 1472 1473 return (err); 1474 } 1475 1476 void 1477 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) 1478 { 1479 dnode_phys_t *dnp; 1480 1481 rw_enter(&dn->dn_struct_rwlock, RW_READER); 1482 mutex_enter(&dn->dn_mtx); 1483 1484 dnp = dn->dn_phys; 1485 1486 doi->doi_data_block_size = dn->dn_datablksz; 1487 doi->doi_metadata_block_size = dn->dn_indblkshift ? 1488 1ULL << dn->dn_indblkshift : 0; 1489 doi->doi_type = dn->dn_type; 1490 doi->doi_bonus_type = dn->dn_bonustype; 1491 doi->doi_bonus_size = dn->dn_bonuslen; 1492 doi->doi_indirection = dn->dn_nlevels; 1493 doi->doi_checksum = dn->dn_checksum; 1494 doi->doi_compress = dn->dn_compress; 1495 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; 1496 doi->doi_max_offset = (dnp->dn_maxblkid + 1) * dn->dn_datablksz; 1497 doi->doi_fill_count = 0; 1498 for (int i = 0; i < dnp->dn_nblkptr; i++) 1499 doi->doi_fill_count += dnp->dn_blkptr[i].blk_fill; 1500 1501 mutex_exit(&dn->dn_mtx); 1502 rw_exit(&dn->dn_struct_rwlock); 1503 } 1504 1505 /* 1506 * Get information on a DMU object. 1507 * If doi is NULL, just indicates whether the object exists. 1508 */ 1509 int 1510 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) 1511 { 1512 dnode_t *dn; 1513 int err = dnode_hold(os, object, FTAG, &dn); 1514 1515 if (err) 1516 return (err); 1517 1518 if (doi != NULL) 1519 dmu_object_info_from_dnode(dn, doi); 1520 1521 dnode_rele(dn, FTAG); 1522 return (0); 1523 } 1524 1525 /* 1526 * As above, but faster; can be used when you have a held dbuf in hand. 1527 */ 1528 void 1529 dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi) 1530 { 1531 dmu_object_info_from_dnode(((dmu_buf_impl_t *)db)->db_dnode, doi); 1532 } 1533 1534 /* 1535 * Faster still when you only care about the size. 1536 * This is specifically optimized for zfs_getattr(). 1537 */ 1538 void 1539 dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, u_longlong_t *nblk512) 1540 { 1541 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 1542 1543 *blksize = dn->dn_datablksz; 1544 /* add 1 for dnode space */ 1545 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> 1546 SPA_MINBLOCKSHIFT) + 1; 1547 } 1548 1549 void 1550 byteswap_uint64_array(void *vbuf, size_t size) 1551 { 1552 uint64_t *buf = vbuf; 1553 size_t count = size >> 3; 1554 int i; 1555 1556 ASSERT((size & 7) == 0); 1557 1558 for (i = 0; i < count; i++) 1559 buf[i] = BSWAP_64(buf[i]); 1560 } 1561 1562 void 1563 byteswap_uint32_array(void *vbuf, size_t size) 1564 { 1565 uint32_t *buf = vbuf; 1566 size_t count = size >> 2; 1567 int i; 1568 1569 ASSERT((size & 3) == 0); 1570 1571 for (i = 0; i < count; i++) 1572 buf[i] = BSWAP_32(buf[i]); 1573 } 1574 1575 void 1576 byteswap_uint16_array(void *vbuf, size_t size) 1577 { 1578 uint16_t *buf = vbuf; 1579 size_t count = size >> 1; 1580 int i; 1581 1582 ASSERT((size & 1) == 0); 1583 1584 for (i = 0; i < count; i++) 1585 buf[i] = BSWAP_16(buf[i]); 1586 } 1587 1588 /* ARGSUSED */ 1589 void 1590 byteswap_uint8_array(void *vbuf, size_t size) 1591 { 1592 } 1593 1594 void 1595 dmu_init(void) 1596 { 1597 dbuf_init(); 1598 dnode_init(); 1599 zfetch_init(); 1600 arc_init(); 1601 l2arc_init(); 1602 xuio_stat_init(); 1603 sa_cache_init(); 1604 } 1605 1606 void 1607 dmu_fini(void) 1608 { 1609 arc_fini(); 1610 zfetch_fini(); 1611 dnode_fini(); 1612 dbuf_fini(); 1613 l2arc_fini(); 1614 xuio_stat_fini(); 1615 sa_cache_fini(); 1616 } 1617