dmu.c revision 268657
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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2011, 2014 by Delphix. All rights reserved. 24 */ 25/* Copyright (c) 2013 by Saso Kiselkov. All rights reserved. */ 26/* Copyright (c) 2013, Joyent, Inc. All rights reserved. */ 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#include <sys/zio_compress.h> 46#include <sys/sa.h> 47#ifdef _KERNEL 48#include <sys/vm.h> 49#include <sys/zfs_znode.h> 50#endif 51 52/* 53 * Enable/disable nopwrite feature. 54 */ 55int zfs_nopwrite_enabled = 1; 56SYSCTL_DECL(_vfs_zfs); 57TUNABLE_INT("vfs.zfs.nopwrite_enabled", &zfs_nopwrite_enabled); 58SYSCTL_INT(_vfs_zfs, OID_AUTO, nopwrite_enabled, CTLFLAG_RDTUN, 59 &zfs_nopwrite_enabled, 0, "Enable nopwrite feature"); 60 61const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { 62 { DMU_BSWAP_UINT8, TRUE, "unallocated" }, 63 { DMU_BSWAP_ZAP, TRUE, "object directory" }, 64 { DMU_BSWAP_UINT64, TRUE, "object array" }, 65 { DMU_BSWAP_UINT8, TRUE, "packed nvlist" }, 66 { DMU_BSWAP_UINT64, TRUE, "packed nvlist size" }, 67 { DMU_BSWAP_UINT64, TRUE, "bpobj" }, 68 { DMU_BSWAP_UINT64, TRUE, "bpobj header" }, 69 { DMU_BSWAP_UINT64, TRUE, "SPA space map header" }, 70 { DMU_BSWAP_UINT64, TRUE, "SPA space map" }, 71 { DMU_BSWAP_UINT64, TRUE, "ZIL intent log" }, 72 { DMU_BSWAP_DNODE, TRUE, "DMU dnode" }, 73 { DMU_BSWAP_OBJSET, TRUE, "DMU objset" }, 74 { DMU_BSWAP_UINT64, TRUE, "DSL directory" }, 75 { DMU_BSWAP_ZAP, TRUE, "DSL directory child map"}, 76 { DMU_BSWAP_ZAP, TRUE, "DSL dataset snap map" }, 77 { DMU_BSWAP_ZAP, TRUE, "DSL props" }, 78 { DMU_BSWAP_UINT64, TRUE, "DSL dataset" }, 79 { DMU_BSWAP_ZNODE, TRUE, "ZFS znode" }, 80 { DMU_BSWAP_OLDACL, TRUE, "ZFS V0 ACL" }, 81 { DMU_BSWAP_UINT8, FALSE, "ZFS plain file" }, 82 { DMU_BSWAP_ZAP, TRUE, "ZFS directory" }, 83 { DMU_BSWAP_ZAP, TRUE, "ZFS master node" }, 84 { DMU_BSWAP_ZAP, TRUE, "ZFS delete queue" }, 85 { DMU_BSWAP_UINT8, FALSE, "zvol object" }, 86 { DMU_BSWAP_ZAP, TRUE, "zvol prop" }, 87 { DMU_BSWAP_UINT8, FALSE, "other uint8[]" }, 88 { DMU_BSWAP_UINT64, FALSE, "other uint64[]" }, 89 { DMU_BSWAP_ZAP, TRUE, "other ZAP" }, 90 { DMU_BSWAP_ZAP, TRUE, "persistent error log" }, 91 { DMU_BSWAP_UINT8, TRUE, "SPA history" }, 92 { DMU_BSWAP_UINT64, TRUE, "SPA history offsets" }, 93 { DMU_BSWAP_ZAP, TRUE, "Pool properties" }, 94 { DMU_BSWAP_ZAP, TRUE, "DSL permissions" }, 95 { DMU_BSWAP_ACL, TRUE, "ZFS ACL" }, 96 { DMU_BSWAP_UINT8, TRUE, "ZFS SYSACL" }, 97 { DMU_BSWAP_UINT8, TRUE, "FUID table" }, 98 { DMU_BSWAP_UINT64, TRUE, "FUID table size" }, 99 { DMU_BSWAP_ZAP, TRUE, "DSL dataset next clones"}, 100 { DMU_BSWAP_ZAP, TRUE, "scan work queue" }, 101 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group used" }, 102 { DMU_BSWAP_ZAP, TRUE, "ZFS user/group quota" }, 103 { DMU_BSWAP_ZAP, TRUE, "snapshot refcount tags"}, 104 { DMU_BSWAP_ZAP, TRUE, "DDT ZAP algorithm" }, 105 { DMU_BSWAP_ZAP, TRUE, "DDT statistics" }, 106 { DMU_BSWAP_UINT8, TRUE, "System attributes" }, 107 { DMU_BSWAP_ZAP, TRUE, "SA master node" }, 108 { DMU_BSWAP_ZAP, TRUE, "SA attr registration" }, 109 { DMU_BSWAP_ZAP, TRUE, "SA attr layouts" }, 110 { DMU_BSWAP_ZAP, TRUE, "scan translations" }, 111 { DMU_BSWAP_UINT8, FALSE, "deduplicated block" }, 112 { DMU_BSWAP_ZAP, TRUE, "DSL deadlist map" }, 113 { DMU_BSWAP_UINT64, TRUE, "DSL deadlist map hdr" }, 114 { DMU_BSWAP_ZAP, TRUE, "DSL dir clones" }, 115 { DMU_BSWAP_UINT64, TRUE, "bpobj subobj" } 116}; 117 118const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { 119 { byteswap_uint8_array, "uint8" }, 120 { byteswap_uint16_array, "uint16" }, 121 { byteswap_uint32_array, "uint32" }, 122 { byteswap_uint64_array, "uint64" }, 123 { zap_byteswap, "zap" }, 124 { dnode_buf_byteswap, "dnode" }, 125 { dmu_objset_byteswap, "objset" }, 126 { zfs_znode_byteswap, "znode" }, 127 { zfs_oldacl_byteswap, "oldacl" }, 128 { zfs_acl_byteswap, "acl" } 129}; 130 131int 132dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, 133 void *tag, dmu_buf_t **dbp) 134{ 135 dnode_t *dn; 136 uint64_t blkid; 137 dmu_buf_impl_t *db; 138 int err; 139 140 err = dnode_hold(os, object, FTAG, &dn); 141 if (err) 142 return (err); 143 blkid = dbuf_whichblock(dn, offset); 144 rw_enter(&dn->dn_struct_rwlock, RW_READER); 145 db = dbuf_hold(dn, blkid, tag); 146 rw_exit(&dn->dn_struct_rwlock); 147 dnode_rele(dn, FTAG); 148 149 if (db == NULL) { 150 *dbp = NULL; 151 return (SET_ERROR(EIO)); 152 } 153 154 *dbp = &db->db; 155 return (err); 156} 157 158int 159dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 160 void *tag, dmu_buf_t **dbp, int flags) 161{ 162 int err; 163 int db_flags = DB_RF_CANFAIL; 164 165 if (flags & DMU_READ_NO_PREFETCH) 166 db_flags |= DB_RF_NOPREFETCH; 167 168 err = dmu_buf_hold_noread(os, object, offset, tag, dbp); 169 if (err == 0) { 170 dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); 171 err = dbuf_read(db, NULL, db_flags); 172 if (err != 0) { 173 dbuf_rele(db, tag); 174 *dbp = NULL; 175 } 176 } 177 178 return (err); 179} 180 181int 182dmu_bonus_max(void) 183{ 184 return (DN_MAX_BONUSLEN); 185} 186 187int 188dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) 189{ 190 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 191 dnode_t *dn; 192 int error; 193 194 DB_DNODE_ENTER(db); 195 dn = DB_DNODE(db); 196 197 if (dn->dn_bonus != db) { 198 error = SET_ERROR(EINVAL); 199 } else if (newsize < 0 || newsize > db_fake->db_size) { 200 error = SET_ERROR(EINVAL); 201 } else { 202 dnode_setbonuslen(dn, newsize, tx); 203 error = 0; 204 } 205 206 DB_DNODE_EXIT(db); 207 return (error); 208} 209 210int 211dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) 212{ 213 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 214 dnode_t *dn; 215 int error; 216 217 DB_DNODE_ENTER(db); 218 dn = DB_DNODE(db); 219 220 if (!DMU_OT_IS_VALID(type)) { 221 error = SET_ERROR(EINVAL); 222 } else if (dn->dn_bonus != db) { 223 error = SET_ERROR(EINVAL); 224 } else { 225 dnode_setbonus_type(dn, type, tx); 226 error = 0; 227 } 228 229 DB_DNODE_EXIT(db); 230 return (error); 231} 232 233dmu_object_type_t 234dmu_get_bonustype(dmu_buf_t *db_fake) 235{ 236 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 237 dnode_t *dn; 238 dmu_object_type_t type; 239 240 DB_DNODE_ENTER(db); 241 dn = DB_DNODE(db); 242 type = dn->dn_bonustype; 243 DB_DNODE_EXIT(db); 244 245 return (type); 246} 247 248int 249dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) 250{ 251 dnode_t *dn; 252 int error; 253 254 error = dnode_hold(os, object, FTAG, &dn); 255 dbuf_rm_spill(dn, tx); 256 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 257 dnode_rm_spill(dn, tx); 258 rw_exit(&dn->dn_struct_rwlock); 259 dnode_rele(dn, FTAG); 260 return (error); 261} 262 263/* 264 * returns ENOENT, EIO, or 0. 265 */ 266int 267dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) 268{ 269 dnode_t *dn; 270 dmu_buf_impl_t *db; 271 int error; 272 273 error = dnode_hold(os, object, FTAG, &dn); 274 if (error) 275 return (error); 276 277 rw_enter(&dn->dn_struct_rwlock, RW_READER); 278 if (dn->dn_bonus == NULL) { 279 rw_exit(&dn->dn_struct_rwlock); 280 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 281 if (dn->dn_bonus == NULL) 282 dbuf_create_bonus(dn); 283 } 284 db = dn->dn_bonus; 285 286 /* as long as the bonus buf is held, the dnode will be held */ 287 if (refcount_add(&db->db_holds, tag) == 1) { 288 VERIFY(dnode_add_ref(dn, db)); 289 (void) atomic_inc_32_nv(&dn->dn_dbufs_count); 290 } 291 292 /* 293 * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's 294 * hold and incrementing the dbuf count to ensure that dnode_move() sees 295 * a dnode hold for every dbuf. 296 */ 297 rw_exit(&dn->dn_struct_rwlock); 298 299 dnode_rele(dn, FTAG); 300 301 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); 302 303 *dbp = &db->db; 304 return (0); 305} 306 307/* 308 * returns ENOENT, EIO, or 0. 309 * 310 * This interface will allocate a blank spill dbuf when a spill blk 311 * doesn't already exist on the dnode. 312 * 313 * if you only want to find an already existing spill db, then 314 * dmu_spill_hold_existing() should be used. 315 */ 316int 317dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) 318{ 319 dmu_buf_impl_t *db = NULL; 320 int err; 321 322 if ((flags & DB_RF_HAVESTRUCT) == 0) 323 rw_enter(&dn->dn_struct_rwlock, RW_READER); 324 325 db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); 326 327 if ((flags & DB_RF_HAVESTRUCT) == 0) 328 rw_exit(&dn->dn_struct_rwlock); 329 330 ASSERT(db != NULL); 331 err = dbuf_read(db, NULL, flags); 332 if (err == 0) 333 *dbp = &db->db; 334 else 335 dbuf_rele(db, tag); 336 return (err); 337} 338 339int 340dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) 341{ 342 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; 343 dnode_t *dn; 344 int err; 345 346 DB_DNODE_ENTER(db); 347 dn = DB_DNODE(db); 348 349 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { 350 err = SET_ERROR(EINVAL); 351 } else { 352 rw_enter(&dn->dn_struct_rwlock, RW_READER); 353 354 if (!dn->dn_have_spill) { 355 err = SET_ERROR(ENOENT); 356 } else { 357 err = dmu_spill_hold_by_dnode(dn, 358 DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); 359 } 360 361 rw_exit(&dn->dn_struct_rwlock); 362 } 363 364 DB_DNODE_EXIT(db); 365 return (err); 366} 367 368int 369dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) 370{ 371 dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; 372 dnode_t *dn; 373 int err; 374 375 DB_DNODE_ENTER(db); 376 dn = DB_DNODE(db); 377 err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); 378 DB_DNODE_EXIT(db); 379 380 return (err); 381} 382 383/* 384 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces 385 * to take a held dnode rather than <os, object> -- the lookup is wasteful, 386 * and can induce severe lock contention when writing to several files 387 * whose dnodes are in the same block. 388 */ 389static int 390dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, 391 int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) 392{ 393 dmu_buf_t **dbp; 394 uint64_t blkid, nblks, i; 395 uint32_t dbuf_flags; 396 int err; 397 zio_t *zio; 398 399 ASSERT(length <= DMU_MAX_ACCESS); 400 401 dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT; 402 if (flags & DMU_READ_NO_PREFETCH || length > zfetch_array_rd_sz) 403 dbuf_flags |= DB_RF_NOPREFETCH; 404 405 rw_enter(&dn->dn_struct_rwlock, RW_READER); 406 if (dn->dn_datablkshift) { 407 int blkshift = dn->dn_datablkshift; 408 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - 409 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; 410 } else { 411 if (offset + length > dn->dn_datablksz) { 412 zfs_panic_recover("zfs: accessing past end of object " 413 "%llx/%llx (size=%u access=%llu+%llu)", 414 (longlong_t)dn->dn_objset-> 415 os_dsl_dataset->ds_object, 416 (longlong_t)dn->dn_object, dn->dn_datablksz, 417 (longlong_t)offset, (longlong_t)length); 418 rw_exit(&dn->dn_struct_rwlock); 419 return (SET_ERROR(EIO)); 420 } 421 nblks = 1; 422 } 423 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); 424 425 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); 426 blkid = dbuf_whichblock(dn, offset); 427 for (i = 0; i < nblks; i++) { 428 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); 429 if (db == NULL) { 430 rw_exit(&dn->dn_struct_rwlock); 431 dmu_buf_rele_array(dbp, nblks, tag); 432 zio_nowait(zio); 433 return (SET_ERROR(EIO)); 434 } 435 /* initiate async i/o */ 436 if (read) 437 (void) dbuf_read(db, zio, dbuf_flags); 438#ifdef _KERNEL 439 else 440 curthread->td_ru.ru_oublock++; 441#endif 442 dbp[i] = &db->db; 443 } 444 rw_exit(&dn->dn_struct_rwlock); 445 446 /* wait for async i/o */ 447 err = zio_wait(zio); 448 if (err) { 449 dmu_buf_rele_array(dbp, nblks, tag); 450 return (err); 451 } 452 453 /* wait for other io to complete */ 454 if (read) { 455 for (i = 0; i < nblks; i++) { 456 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; 457 mutex_enter(&db->db_mtx); 458 while (db->db_state == DB_READ || 459 db->db_state == DB_FILL) 460 cv_wait(&db->db_changed, &db->db_mtx); 461 if (db->db_state == DB_UNCACHED) 462 err = SET_ERROR(EIO); 463 mutex_exit(&db->db_mtx); 464 if (err) { 465 dmu_buf_rele_array(dbp, nblks, tag); 466 return (err); 467 } 468 } 469 } 470 471 *numbufsp = nblks; 472 *dbpp = dbp; 473 return (0); 474} 475 476static int 477dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, 478 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 479{ 480 dnode_t *dn; 481 int err; 482 483 err = dnode_hold(os, object, FTAG, &dn); 484 if (err) 485 return (err); 486 487 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 488 numbufsp, dbpp, DMU_READ_PREFETCH); 489 490 dnode_rele(dn, FTAG); 491 492 return (err); 493} 494 495int 496dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, 497 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 498{ 499 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 500 dnode_t *dn; 501 int err; 502 503 DB_DNODE_ENTER(db); 504 dn = DB_DNODE(db); 505 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 506 numbufsp, dbpp, DMU_READ_PREFETCH); 507 DB_DNODE_EXIT(db); 508 509 return (err); 510} 511 512void 513dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) 514{ 515 int i; 516 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; 517 518 if (numbufs == 0) 519 return; 520 521 for (i = 0; i < numbufs; i++) { 522 if (dbp[i]) 523 dbuf_rele(dbp[i], tag); 524 } 525 526 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); 527} 528 529/* 530 * Issue prefetch i/os for the given blocks. 531 * 532 * Note: The assumption is that we *know* these blocks will be needed 533 * almost immediately. Therefore, the prefetch i/os will be issued at 534 * ZIO_PRIORITY_SYNC_READ 535 * 536 * Note: indirect blocks and other metadata will be read synchronously, 537 * causing this function to block if they are not already cached. 538 */ 539void 540dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) 541{ 542 dnode_t *dn; 543 uint64_t blkid; 544 int nblks, err; 545 546 if (zfs_prefetch_disable) 547 return; 548 549 if (len == 0) { /* they're interested in the bonus buffer */ 550 dn = DMU_META_DNODE(os); 551 552 if (object == 0 || object >= DN_MAX_OBJECT) 553 return; 554 555 rw_enter(&dn->dn_struct_rwlock, RW_READER); 556 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); 557 dbuf_prefetch(dn, blkid, ZIO_PRIORITY_SYNC_READ); 558 rw_exit(&dn->dn_struct_rwlock); 559 return; 560 } 561 562 /* 563 * XXX - Note, if the dnode for the requested object is not 564 * already cached, we will do a *synchronous* read in the 565 * dnode_hold() call. The same is true for any indirects. 566 */ 567 err = dnode_hold(os, object, FTAG, &dn); 568 if (err != 0) 569 return; 570 571 rw_enter(&dn->dn_struct_rwlock, RW_READER); 572 if (dn->dn_datablkshift) { 573 int blkshift = dn->dn_datablkshift; 574 nblks = (P2ROUNDUP(offset + len, 1 << blkshift) - 575 P2ALIGN(offset, 1 << blkshift)) >> blkshift; 576 } else { 577 nblks = (offset < dn->dn_datablksz); 578 } 579 580 if (nblks != 0) { 581 blkid = dbuf_whichblock(dn, offset); 582 for (int i = 0; i < nblks; i++) 583 dbuf_prefetch(dn, blkid + i, ZIO_PRIORITY_SYNC_READ); 584 } 585 586 rw_exit(&dn->dn_struct_rwlock); 587 588 dnode_rele(dn, FTAG); 589} 590 591/* 592 * Get the next "chunk" of file data to free. We traverse the file from 593 * the end so that the file gets shorter over time (if we crashes in the 594 * middle, this will leave us in a better state). We find allocated file 595 * data by simply searching the allocated level 1 indirects. 596 * 597 * On input, *start should be the first offset that does not need to be 598 * freed (e.g. "offset + length"). On return, *start will be the first 599 * offset that should be freed. 600 */ 601static int 602get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum) 603{ 604 uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); 605 /* bytes of data covered by a level-1 indirect block */ 606 uint64_t iblkrange = 607 dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); 608 609 ASSERT3U(minimum, <=, *start); 610 611 if (*start - minimum <= iblkrange * maxblks) { 612 *start = minimum; 613 return (0); 614 } 615 ASSERT(ISP2(iblkrange)); 616 617 for (uint64_t blks = 0; *start > minimum && blks < maxblks; blks++) { 618 int err; 619 620 /* 621 * dnode_next_offset(BACKWARDS) will find an allocated L1 622 * indirect block at or before the input offset. We must 623 * decrement *start so that it is at the end of the region 624 * to search. 625 */ 626 (*start)--; 627 err = dnode_next_offset(dn, 628 DNODE_FIND_BACKWARDS, start, 2, 1, 0); 629 630 /* if there are no indirect blocks before start, we are done */ 631 if (err == ESRCH) { 632 *start = minimum; 633 break; 634 } else if (err != 0) { 635 return (err); 636 } 637 638 /* set start to the beginning of this L1 indirect */ 639 *start = P2ALIGN(*start, iblkrange); 640 } 641 if (*start < minimum) 642 *start = minimum; 643 return (0); 644} 645 646static int 647dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, 648 uint64_t length) 649{ 650 uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; 651 int err; 652 653 if (offset >= object_size) 654 return (0); 655 656 if (length == DMU_OBJECT_END || offset + length > object_size) 657 length = object_size - offset; 658 659 while (length != 0) { 660 uint64_t chunk_end, chunk_begin; 661 662 chunk_end = chunk_begin = offset + length; 663 664 /* move chunk_begin backwards to the beginning of this chunk */ 665 err = get_next_chunk(dn, &chunk_begin, offset); 666 if (err) 667 return (err); 668 ASSERT3U(chunk_begin, >=, offset); 669 ASSERT3U(chunk_begin, <=, chunk_end); 670 671 dmu_tx_t *tx = dmu_tx_create(os); 672 dmu_tx_hold_free(tx, dn->dn_object, 673 chunk_begin, chunk_end - chunk_begin); 674 err = dmu_tx_assign(tx, TXG_WAIT); 675 if (err) { 676 dmu_tx_abort(tx); 677 return (err); 678 } 679 dnode_free_range(dn, chunk_begin, chunk_end - chunk_begin, tx); 680 dmu_tx_commit(tx); 681 682 length -= chunk_end - chunk_begin; 683 } 684 return (0); 685} 686 687int 688dmu_free_long_range(objset_t *os, uint64_t object, 689 uint64_t offset, uint64_t length) 690{ 691 dnode_t *dn; 692 int err; 693 694 err = dnode_hold(os, object, FTAG, &dn); 695 if (err != 0) 696 return (err); 697 err = dmu_free_long_range_impl(os, dn, offset, length); 698 699 /* 700 * It is important to zero out the maxblkid when freeing the entire 701 * file, so that (a) subsequent calls to dmu_free_long_range_impl() 702 * will take the fast path, and (b) dnode_reallocate() can verify 703 * that the entire file has been freed. 704 */ 705 if (err == 0 && offset == 0 && length == DMU_OBJECT_END) 706 dn->dn_maxblkid = 0; 707 708 dnode_rele(dn, FTAG); 709 return (err); 710} 711 712int 713dmu_free_long_object(objset_t *os, uint64_t object) 714{ 715 dmu_tx_t *tx; 716 int err; 717 718 err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); 719 if (err != 0) 720 return (err); 721 722 tx = dmu_tx_create(os); 723 dmu_tx_hold_bonus(tx, object); 724 dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); 725 err = dmu_tx_assign(tx, TXG_WAIT); 726 if (err == 0) { 727 err = dmu_object_free(os, object, tx); 728 dmu_tx_commit(tx); 729 } else { 730 dmu_tx_abort(tx); 731 } 732 733 return (err); 734} 735 736int 737dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 738 uint64_t size, dmu_tx_t *tx) 739{ 740 dnode_t *dn; 741 int err = dnode_hold(os, object, FTAG, &dn); 742 if (err) 743 return (err); 744 ASSERT(offset < UINT64_MAX); 745 ASSERT(size == -1ULL || size <= UINT64_MAX - offset); 746 dnode_free_range(dn, offset, size, tx); 747 dnode_rele(dn, FTAG); 748 return (0); 749} 750 751int 752dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 753 void *buf, uint32_t flags) 754{ 755 dnode_t *dn; 756 dmu_buf_t **dbp; 757 int numbufs, err; 758 759 err = dnode_hold(os, object, FTAG, &dn); 760 if (err) 761 return (err); 762 763 /* 764 * Deal with odd block sizes, where there can't be data past the first 765 * block. If we ever do the tail block optimization, we will need to 766 * handle that here as well. 767 */ 768 if (dn->dn_maxblkid == 0) { 769 int newsz = offset > dn->dn_datablksz ? 0 : 770 MIN(size, dn->dn_datablksz - offset); 771 bzero((char *)buf + newsz, size - newsz); 772 size = newsz; 773 } 774 775 while (size > 0) { 776 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); 777 int i; 778 779 /* 780 * NB: we could do this block-at-a-time, but it's nice 781 * to be reading in parallel. 782 */ 783 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, 784 TRUE, FTAG, &numbufs, &dbp, flags); 785 if (err) 786 break; 787 788 for (i = 0; i < numbufs; i++) { 789 int tocpy; 790 int bufoff; 791 dmu_buf_t *db = dbp[i]; 792 793 ASSERT(size > 0); 794 795 bufoff = offset - db->db_offset; 796 tocpy = (int)MIN(db->db_size - bufoff, size); 797 798 bcopy((char *)db->db_data + bufoff, buf, tocpy); 799 800 offset += tocpy; 801 size -= tocpy; 802 buf = (char *)buf + tocpy; 803 } 804 dmu_buf_rele_array(dbp, numbufs, FTAG); 805 } 806 dnode_rele(dn, FTAG); 807 return (err); 808} 809 810void 811dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 812 const void *buf, dmu_tx_t *tx) 813{ 814 dmu_buf_t **dbp; 815 int numbufs, i; 816 817 if (size == 0) 818 return; 819 820 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 821 FALSE, FTAG, &numbufs, &dbp)); 822 823 for (i = 0; i < numbufs; i++) { 824 int tocpy; 825 int bufoff; 826 dmu_buf_t *db = dbp[i]; 827 828 ASSERT(size > 0); 829 830 bufoff = offset - db->db_offset; 831 tocpy = (int)MIN(db->db_size - bufoff, size); 832 833 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 834 835 if (tocpy == db->db_size) 836 dmu_buf_will_fill(db, tx); 837 else 838 dmu_buf_will_dirty(db, tx); 839 840 bcopy(buf, (char *)db->db_data + bufoff, tocpy); 841 842 if (tocpy == db->db_size) 843 dmu_buf_fill_done(db, tx); 844 845 offset += tocpy; 846 size -= tocpy; 847 buf = (char *)buf + tocpy; 848 } 849 dmu_buf_rele_array(dbp, numbufs, FTAG); 850} 851 852void 853dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 854 dmu_tx_t *tx) 855{ 856 dmu_buf_t **dbp; 857 int numbufs, i; 858 859 if (size == 0) 860 return; 861 862 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 863 FALSE, FTAG, &numbufs, &dbp)); 864 865 for (i = 0; i < numbufs; i++) { 866 dmu_buf_t *db = dbp[i]; 867 868 dmu_buf_will_not_fill(db, tx); 869 } 870 dmu_buf_rele_array(dbp, numbufs, FTAG); 871} 872 873void 874dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, 875 void *data, uint8_t etype, uint8_t comp, int uncompressed_size, 876 int compressed_size, int byteorder, dmu_tx_t *tx) 877{ 878 dmu_buf_t *db; 879 880 ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); 881 ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); 882 VERIFY0(dmu_buf_hold_noread(os, object, offset, 883 FTAG, &db)); 884 885 dmu_buf_write_embedded(db, 886 data, (bp_embedded_type_t)etype, (enum zio_compress)comp, 887 uncompressed_size, compressed_size, byteorder, tx); 888 889 dmu_buf_rele(db, FTAG); 890} 891 892/* 893 * DMU support for xuio 894 */ 895kstat_t *xuio_ksp = NULL; 896 897int 898dmu_xuio_init(xuio_t *xuio, int nblk) 899{ 900 dmu_xuio_t *priv; 901 uio_t *uio = &xuio->xu_uio; 902 903 uio->uio_iovcnt = nblk; 904 uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); 905 906 priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); 907 priv->cnt = nblk; 908 priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); 909 priv->iovp = uio->uio_iov; 910 XUIO_XUZC_PRIV(xuio) = priv; 911 912 if (XUIO_XUZC_RW(xuio) == UIO_READ) 913 XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); 914 else 915 XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); 916 917 return (0); 918} 919 920void 921dmu_xuio_fini(xuio_t *xuio) 922{ 923 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 924 int nblk = priv->cnt; 925 926 kmem_free(priv->iovp, nblk * sizeof (iovec_t)); 927 kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); 928 kmem_free(priv, sizeof (dmu_xuio_t)); 929 930 if (XUIO_XUZC_RW(xuio) == UIO_READ) 931 XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); 932 else 933 XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); 934} 935 936/* 937 * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } 938 * and increase priv->next by 1. 939 */ 940int 941dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) 942{ 943 struct iovec *iov; 944 uio_t *uio = &xuio->xu_uio; 945 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 946 int i = priv->next++; 947 948 ASSERT(i < priv->cnt); 949 ASSERT(off + n <= arc_buf_size(abuf)); 950 iov = uio->uio_iov + i; 951 iov->iov_base = (char *)abuf->b_data + off; 952 iov->iov_len = n; 953 priv->bufs[i] = abuf; 954 return (0); 955} 956 957int 958dmu_xuio_cnt(xuio_t *xuio) 959{ 960 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 961 return (priv->cnt); 962} 963 964arc_buf_t * 965dmu_xuio_arcbuf(xuio_t *xuio, int i) 966{ 967 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 968 969 ASSERT(i < priv->cnt); 970 return (priv->bufs[i]); 971} 972 973void 974dmu_xuio_clear(xuio_t *xuio, int i) 975{ 976 dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); 977 978 ASSERT(i < priv->cnt); 979 priv->bufs[i] = NULL; 980} 981 982static void 983xuio_stat_init(void) 984{ 985 xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", 986 KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), 987 KSTAT_FLAG_VIRTUAL); 988 if (xuio_ksp != NULL) { 989 xuio_ksp->ks_data = &xuio_stats; 990 kstat_install(xuio_ksp); 991 } 992} 993 994static void 995xuio_stat_fini(void) 996{ 997 if (xuio_ksp != NULL) { 998 kstat_delete(xuio_ksp); 999 xuio_ksp = NULL; 1000 } 1001} 1002 1003void 1004xuio_stat_wbuf_copied() 1005{ 1006 XUIOSTAT_BUMP(xuiostat_wbuf_copied); 1007} 1008 1009void 1010xuio_stat_wbuf_nocopy() 1011{ 1012 XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); 1013} 1014 1015#ifdef _KERNEL 1016int 1017dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) 1018{ 1019 dmu_buf_t **dbp; 1020 int numbufs, i, err; 1021 xuio_t *xuio = NULL; 1022 1023 /* 1024 * NB: we could do this block-at-a-time, but it's nice 1025 * to be reading in parallel. 1026 */ 1027 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG, 1028 &numbufs, &dbp); 1029 if (err) 1030 return (err); 1031 1032#ifdef UIO_XUIO 1033 if (uio->uio_extflg == UIO_XUIO) 1034 xuio = (xuio_t *)uio; 1035#endif 1036 1037 for (i = 0; i < numbufs; i++) { 1038 int tocpy; 1039 int bufoff; 1040 dmu_buf_t *db = dbp[i]; 1041 1042 ASSERT(size > 0); 1043 1044 bufoff = uio->uio_loffset - db->db_offset; 1045 tocpy = (int)MIN(db->db_size - bufoff, size); 1046 1047 if (xuio) { 1048 dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; 1049 arc_buf_t *dbuf_abuf = dbi->db_buf; 1050 arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); 1051 err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); 1052 if (!err) { 1053 uio->uio_resid -= tocpy; 1054 uio->uio_loffset += tocpy; 1055 } 1056 1057 if (abuf == dbuf_abuf) 1058 XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); 1059 else 1060 XUIOSTAT_BUMP(xuiostat_rbuf_copied); 1061 } else { 1062 err = uiomove((char *)db->db_data + bufoff, tocpy, 1063 UIO_READ, uio); 1064 } 1065 if (err) 1066 break; 1067 1068 size -= tocpy; 1069 } 1070 dmu_buf_rele_array(dbp, numbufs, FTAG); 1071 1072 return (err); 1073} 1074 1075static int 1076dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) 1077{ 1078 dmu_buf_t **dbp; 1079 int numbufs; 1080 int err = 0; 1081 int i; 1082 1083 err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, 1084 FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); 1085 if (err) 1086 return (err); 1087 1088 for (i = 0; i < numbufs; i++) { 1089 int tocpy; 1090 int bufoff; 1091 dmu_buf_t *db = dbp[i]; 1092 1093 ASSERT(size > 0); 1094 1095 bufoff = uio->uio_loffset - db->db_offset; 1096 tocpy = (int)MIN(db->db_size - bufoff, size); 1097 1098 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 1099 1100 if (tocpy == db->db_size) 1101 dmu_buf_will_fill(db, tx); 1102 else 1103 dmu_buf_will_dirty(db, tx); 1104 1105 /* 1106 * XXX uiomove could block forever (eg. nfs-backed 1107 * pages). There needs to be a uiolockdown() function 1108 * to lock the pages in memory, so that uiomove won't 1109 * block. 1110 */ 1111 err = uiomove((char *)db->db_data + bufoff, tocpy, 1112 UIO_WRITE, uio); 1113 1114 if (tocpy == db->db_size) 1115 dmu_buf_fill_done(db, tx); 1116 1117 if (err) 1118 break; 1119 1120 size -= tocpy; 1121 } 1122 1123 dmu_buf_rele_array(dbp, numbufs, FTAG); 1124 return (err); 1125} 1126 1127int 1128dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, 1129 dmu_tx_t *tx) 1130{ 1131 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; 1132 dnode_t *dn; 1133 int err; 1134 1135 if (size == 0) 1136 return (0); 1137 1138 DB_DNODE_ENTER(db); 1139 dn = DB_DNODE(db); 1140 err = dmu_write_uio_dnode(dn, uio, size, tx); 1141 DB_DNODE_EXIT(db); 1142 1143 return (err); 1144} 1145 1146int 1147dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, 1148 dmu_tx_t *tx) 1149{ 1150 dnode_t *dn; 1151 int err; 1152 1153 if (size == 0) 1154 return (0); 1155 1156 err = dnode_hold(os, object, FTAG, &dn); 1157 if (err) 1158 return (err); 1159 1160 err = dmu_write_uio_dnode(dn, uio, size, tx); 1161 1162 dnode_rele(dn, FTAG); 1163 1164 return (err); 1165} 1166 1167#ifdef sun 1168int 1169dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 1170 page_t *pp, dmu_tx_t *tx) 1171{ 1172 dmu_buf_t **dbp; 1173 int numbufs, i; 1174 int err; 1175 1176 if (size == 0) 1177 return (0); 1178 1179 err = dmu_buf_hold_array(os, object, offset, size, 1180 FALSE, FTAG, &numbufs, &dbp); 1181 if (err) 1182 return (err); 1183 1184 for (i = 0; i < numbufs; i++) { 1185 int tocpy, copied, thiscpy; 1186 int bufoff; 1187 dmu_buf_t *db = dbp[i]; 1188 caddr_t va; 1189 1190 ASSERT(size > 0); 1191 ASSERT3U(db->db_size, >=, PAGESIZE); 1192 1193 bufoff = offset - db->db_offset; 1194 tocpy = (int)MIN(db->db_size - bufoff, size); 1195 1196 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 1197 1198 if (tocpy == db->db_size) 1199 dmu_buf_will_fill(db, tx); 1200 else 1201 dmu_buf_will_dirty(db, tx); 1202 1203 for (copied = 0; copied < tocpy; copied += PAGESIZE) { 1204 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); 1205 thiscpy = MIN(PAGESIZE, tocpy - copied); 1206 va = zfs_map_page(pp, S_READ); 1207 bcopy(va, (char *)db->db_data + bufoff, thiscpy); 1208 zfs_unmap_page(pp, va); 1209 pp = pp->p_next; 1210 bufoff += PAGESIZE; 1211 } 1212 1213 if (tocpy == db->db_size) 1214 dmu_buf_fill_done(db, tx); 1215 1216 offset += tocpy; 1217 size -= tocpy; 1218 } 1219 dmu_buf_rele_array(dbp, numbufs, FTAG); 1220 return (err); 1221} 1222 1223#else 1224 1225int 1226dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 1227 vm_page_t *ma, dmu_tx_t *tx) 1228{ 1229 dmu_buf_t **dbp; 1230 struct sf_buf *sf; 1231 int numbufs, i; 1232 int err; 1233 1234 if (size == 0) 1235 return (0); 1236 1237 err = dmu_buf_hold_array(os, object, offset, size, 1238 FALSE, FTAG, &numbufs, &dbp); 1239 if (err) 1240 return (err); 1241 1242 for (i = 0; i < numbufs; i++) { 1243 int tocpy, copied, thiscpy; 1244 int bufoff; 1245 dmu_buf_t *db = dbp[i]; 1246 caddr_t va; 1247 1248 ASSERT(size > 0); 1249 ASSERT3U(db->db_size, >=, PAGESIZE); 1250 1251 bufoff = offset - db->db_offset; 1252 tocpy = (int)MIN(db->db_size - bufoff, size); 1253 1254 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 1255 1256 if (tocpy == db->db_size) 1257 dmu_buf_will_fill(db, tx); 1258 else 1259 dmu_buf_will_dirty(db, tx); 1260 1261 for (copied = 0; copied < tocpy; copied += PAGESIZE) { 1262 ASSERT3U(ptoa((*ma)->pindex), ==, db->db_offset + bufoff); 1263 thiscpy = MIN(PAGESIZE, tocpy - copied); 1264 va = zfs_map_page(*ma, &sf); 1265 bcopy(va, (char *)db->db_data + bufoff, thiscpy); 1266 zfs_unmap_page(sf); 1267 ma += 1; 1268 bufoff += PAGESIZE; 1269 } 1270 1271 if (tocpy == db->db_size) 1272 dmu_buf_fill_done(db, tx); 1273 1274 offset += tocpy; 1275 size -= tocpy; 1276 } 1277 dmu_buf_rele_array(dbp, numbufs, FTAG); 1278 return (err); 1279} 1280#endif /* sun */ 1281#endif 1282 1283/* 1284 * Allocate a loaned anonymous arc buffer. 1285 */ 1286arc_buf_t * 1287dmu_request_arcbuf(dmu_buf_t *handle, int size) 1288{ 1289 dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; 1290 1291 return (arc_loan_buf(db->db_objset->os_spa, size)); 1292} 1293 1294/* 1295 * Free a loaned arc buffer. 1296 */ 1297void 1298dmu_return_arcbuf(arc_buf_t *buf) 1299{ 1300 arc_return_buf(buf, FTAG); 1301 VERIFY(arc_buf_remove_ref(buf, FTAG)); 1302} 1303 1304/* 1305 * When possible directly assign passed loaned arc buffer to a dbuf. 1306 * If this is not possible copy the contents of passed arc buf via 1307 * dmu_write(). 1308 */ 1309void 1310dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, 1311 dmu_tx_t *tx) 1312{ 1313 dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; 1314 dnode_t *dn; 1315 dmu_buf_impl_t *db; 1316 uint32_t blksz = (uint32_t)arc_buf_size(buf); 1317 uint64_t blkid; 1318 1319 DB_DNODE_ENTER(dbuf); 1320 dn = DB_DNODE(dbuf); 1321 rw_enter(&dn->dn_struct_rwlock, RW_READER); 1322 blkid = dbuf_whichblock(dn, offset); 1323 VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); 1324 rw_exit(&dn->dn_struct_rwlock); 1325 DB_DNODE_EXIT(dbuf); 1326 1327 if (offset == db->db.db_offset && blksz == db->db.db_size) { 1328 dbuf_assign_arcbuf(db, buf, tx); 1329 dbuf_rele(db, FTAG); 1330 } else { 1331 objset_t *os; 1332 uint64_t object; 1333 1334 DB_DNODE_ENTER(dbuf); 1335 dn = DB_DNODE(dbuf); 1336 os = dn->dn_objset; 1337 object = dn->dn_object; 1338 DB_DNODE_EXIT(dbuf); 1339 1340 dbuf_rele(db, FTAG); 1341 dmu_write(os, object, offset, blksz, buf->b_data, tx); 1342 dmu_return_arcbuf(buf); 1343 XUIOSTAT_BUMP(xuiostat_wbuf_copied); 1344 } 1345} 1346 1347typedef struct { 1348 dbuf_dirty_record_t *dsa_dr; 1349 dmu_sync_cb_t *dsa_done; 1350 zgd_t *dsa_zgd; 1351 dmu_tx_t *dsa_tx; 1352} dmu_sync_arg_t; 1353 1354/* ARGSUSED */ 1355static void 1356dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) 1357{ 1358 dmu_sync_arg_t *dsa = varg; 1359 dmu_buf_t *db = dsa->dsa_zgd->zgd_db; 1360 blkptr_t *bp = zio->io_bp; 1361 1362 if (zio->io_error == 0) { 1363 if (BP_IS_HOLE(bp)) { 1364 /* 1365 * A block of zeros may compress to a hole, but the 1366 * block size still needs to be known for replay. 1367 */ 1368 BP_SET_LSIZE(bp, db->db_size); 1369 } else if (!BP_IS_EMBEDDED(bp)) { 1370 ASSERT(BP_GET_LEVEL(bp) == 0); 1371 bp->blk_fill = 1; 1372 } 1373 } 1374} 1375 1376static void 1377dmu_sync_late_arrival_ready(zio_t *zio) 1378{ 1379 dmu_sync_ready(zio, NULL, zio->io_private); 1380} 1381 1382/* ARGSUSED */ 1383static void 1384dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) 1385{ 1386 dmu_sync_arg_t *dsa = varg; 1387 dbuf_dirty_record_t *dr = dsa->dsa_dr; 1388 dmu_buf_impl_t *db = dr->dr_dbuf; 1389 1390 mutex_enter(&db->db_mtx); 1391 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); 1392 if (zio->io_error == 0) { 1393 dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); 1394 if (dr->dt.dl.dr_nopwrite) { 1395 blkptr_t *bp = zio->io_bp; 1396 blkptr_t *bp_orig = &zio->io_bp_orig; 1397 uint8_t chksum = BP_GET_CHECKSUM(bp_orig); 1398 1399 ASSERT(BP_EQUAL(bp, bp_orig)); 1400 ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); 1401 ASSERT(zio_checksum_table[chksum].ci_dedup); 1402 } 1403 dr->dt.dl.dr_overridden_by = *zio->io_bp; 1404 dr->dt.dl.dr_override_state = DR_OVERRIDDEN; 1405 dr->dt.dl.dr_copies = zio->io_prop.zp_copies; 1406 if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by)) 1407 BP_ZERO(&dr->dt.dl.dr_overridden_by); 1408 } else { 1409 dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; 1410 } 1411 cv_broadcast(&db->db_changed); 1412 mutex_exit(&db->db_mtx); 1413 1414 dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 1415 1416 kmem_free(dsa, sizeof (*dsa)); 1417} 1418 1419static void 1420dmu_sync_late_arrival_done(zio_t *zio) 1421{ 1422 blkptr_t *bp = zio->io_bp; 1423 dmu_sync_arg_t *dsa = zio->io_private; 1424 blkptr_t *bp_orig = &zio->io_bp_orig; 1425 1426 if (zio->io_error == 0 && !BP_IS_HOLE(bp)) { 1427 /* 1428 * If we didn't allocate a new block (i.e. ZIO_FLAG_NOPWRITE) 1429 * then there is nothing to do here. Otherwise, free the 1430 * newly allocated block in this txg. 1431 */ 1432 if (zio->io_flags & ZIO_FLAG_NOPWRITE) { 1433 ASSERT(BP_EQUAL(bp, bp_orig)); 1434 } else { 1435 ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); 1436 ASSERT(zio->io_bp->blk_birth == zio->io_txg); 1437 ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); 1438 zio_free(zio->io_spa, zio->io_txg, zio->io_bp); 1439 } 1440 } 1441 1442 dmu_tx_commit(dsa->dsa_tx); 1443 1444 dsa->dsa_done(dsa->dsa_zgd, zio->io_error); 1445 1446 kmem_free(dsa, sizeof (*dsa)); 1447} 1448 1449static int 1450dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, 1451 zio_prop_t *zp, zbookmark_phys_t *zb) 1452{ 1453 dmu_sync_arg_t *dsa; 1454 dmu_tx_t *tx; 1455 1456 tx = dmu_tx_create(os); 1457 dmu_tx_hold_space(tx, zgd->zgd_db->db_size); 1458 if (dmu_tx_assign(tx, TXG_WAIT) != 0) { 1459 dmu_tx_abort(tx); 1460 /* Make zl_get_data do txg_waited_synced() */ 1461 return (SET_ERROR(EIO)); 1462 } 1463 1464 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1465 dsa->dsa_dr = NULL; 1466 dsa->dsa_done = done; 1467 dsa->dsa_zgd = zgd; 1468 dsa->dsa_tx = tx; 1469 1470 zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, 1471 zgd->zgd_db->db_data, zgd->zgd_db->db_size, zp, 1472 dmu_sync_late_arrival_ready, NULL, dmu_sync_late_arrival_done, dsa, 1473 ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); 1474 1475 return (0); 1476} 1477 1478/* 1479 * Intent log support: sync the block associated with db to disk. 1480 * N.B. and XXX: the caller is responsible for making sure that the 1481 * data isn't changing while dmu_sync() is writing it. 1482 * 1483 * Return values: 1484 * 1485 * EEXIST: this txg has already been synced, so there's nothing to do. 1486 * The caller should not log the write. 1487 * 1488 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. 1489 * The caller should not log the write. 1490 * 1491 * EALREADY: this block is already in the process of being synced. 1492 * The caller should track its progress (somehow). 1493 * 1494 * EIO: could not do the I/O. 1495 * The caller should do a txg_wait_synced(). 1496 * 1497 * 0: the I/O has been initiated. 1498 * The caller should log this blkptr in the done callback. 1499 * It is possible that the I/O will fail, in which case 1500 * the error will be reported to the done callback and 1501 * propagated to pio from zio_done(). 1502 */ 1503int 1504dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) 1505{ 1506 blkptr_t *bp = zgd->zgd_bp; 1507 dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; 1508 objset_t *os = db->db_objset; 1509 dsl_dataset_t *ds = os->os_dsl_dataset; 1510 dbuf_dirty_record_t *dr; 1511 dmu_sync_arg_t *dsa; 1512 zbookmark_phys_t zb; 1513 zio_prop_t zp; 1514 dnode_t *dn; 1515 1516 ASSERT(pio != NULL); 1517 ASSERT(txg != 0); 1518 1519 SET_BOOKMARK(&zb, ds->ds_object, 1520 db->db.db_object, db->db_level, db->db_blkid); 1521 1522 DB_DNODE_ENTER(db); 1523 dn = DB_DNODE(db); 1524 dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); 1525 DB_DNODE_EXIT(db); 1526 1527 /* 1528 * If we're frozen (running ziltest), we always need to generate a bp. 1529 */ 1530 if (txg > spa_freeze_txg(os->os_spa)) 1531 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1532 1533 /* 1534 * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() 1535 * and us. If we determine that this txg is not yet syncing, 1536 * but it begins to sync a moment later, that's OK because the 1537 * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. 1538 */ 1539 mutex_enter(&db->db_mtx); 1540 1541 if (txg <= spa_last_synced_txg(os->os_spa)) { 1542 /* 1543 * This txg has already synced. There's nothing to do. 1544 */ 1545 mutex_exit(&db->db_mtx); 1546 return (SET_ERROR(EEXIST)); 1547 } 1548 1549 if (txg <= spa_syncing_txg(os->os_spa)) { 1550 /* 1551 * This txg is currently syncing, so we can't mess with 1552 * the dirty record anymore; just write a new log block. 1553 */ 1554 mutex_exit(&db->db_mtx); 1555 return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); 1556 } 1557 1558 dr = db->db_last_dirty; 1559 while (dr && dr->dr_txg != txg) 1560 dr = dr->dr_next; 1561 1562 if (dr == NULL) { 1563 /* 1564 * There's no dr for this dbuf, so it must have been freed. 1565 * There's no need to log writes to freed blocks, so we're done. 1566 */ 1567 mutex_exit(&db->db_mtx); 1568 return (SET_ERROR(ENOENT)); 1569 } 1570 1571 ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); 1572 1573 /* 1574 * Assume the on-disk data is X, the current syncing data is Y, 1575 * and the current in-memory data is Z (currently in dmu_sync). 1576 * X and Z are identical but Y is has been modified. Normally, 1577 * when X and Z are the same we will perform a nopwrite but if Y 1578 * is different we must disable nopwrite since the resulting write 1579 * of Y to disk can free the block containing X. If we allowed a 1580 * nopwrite to occur the block pointing to Z would reference a freed 1581 * block. Since this is a rare case we simplify this by disabling 1582 * nopwrite if the current dmu_sync-ing dbuf has been modified in 1583 * a previous transaction. 1584 */ 1585 if (dr->dr_next) 1586 zp.zp_nopwrite = B_FALSE; 1587 1588 ASSERT(dr->dr_txg == txg); 1589 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || 1590 dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { 1591 /* 1592 * We have already issued a sync write for this buffer, 1593 * or this buffer has already been synced. It could not 1594 * have been dirtied since, or we would have cleared the state. 1595 */ 1596 mutex_exit(&db->db_mtx); 1597 return (SET_ERROR(EALREADY)); 1598 } 1599 1600 ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); 1601 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; 1602 mutex_exit(&db->db_mtx); 1603 1604 dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 1605 dsa->dsa_dr = dr; 1606 dsa->dsa_done = done; 1607 dsa->dsa_zgd = zgd; 1608 dsa->dsa_tx = NULL; 1609 1610 zio_nowait(arc_write(pio, os->os_spa, txg, 1611 bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), 1612 DBUF_IS_L2COMPRESSIBLE(db), &zp, dmu_sync_ready, 1613 NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE, 1614 ZIO_FLAG_CANFAIL, &zb)); 1615 1616 return (0); 1617} 1618 1619int 1620dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, 1621 dmu_tx_t *tx) 1622{ 1623 dnode_t *dn; 1624 int err; 1625 1626 err = dnode_hold(os, object, FTAG, &dn); 1627 if (err) 1628 return (err); 1629 err = dnode_set_blksz(dn, size, ibs, tx); 1630 dnode_rele(dn, FTAG); 1631 return (err); 1632} 1633 1634void 1635dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 1636 dmu_tx_t *tx) 1637{ 1638 dnode_t *dn; 1639 1640 /* 1641 * Send streams include each object's checksum function. This 1642 * check ensures that the receiving system can understand the 1643 * checksum function transmitted. 1644 */ 1645 ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); 1646 1647 VERIFY0(dnode_hold(os, object, FTAG, &dn)); 1648 ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); 1649 dn->dn_checksum = checksum; 1650 dnode_setdirty(dn, tx); 1651 dnode_rele(dn, FTAG); 1652} 1653 1654void 1655dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 1656 dmu_tx_t *tx) 1657{ 1658 dnode_t *dn; 1659 1660 /* 1661 * Send streams include each object's compression function. This 1662 * check ensures that the receiving system can understand the 1663 * compression function transmitted. 1664 */ 1665 ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); 1666 1667 VERIFY0(dnode_hold(os, object, FTAG, &dn)); 1668 dn->dn_compress = compress; 1669 dnode_setdirty(dn, tx); 1670 dnode_rele(dn, FTAG); 1671} 1672 1673int zfs_mdcomp_disable = 0; 1674TUNABLE_INT("vfs.zfs.mdcomp_disable", &zfs_mdcomp_disable); 1675SYSCTL_INT(_vfs_zfs, OID_AUTO, mdcomp_disable, CTLFLAG_RW, 1676 &zfs_mdcomp_disable, 0, "Disable metadata compression"); 1677 1678/* 1679 * When the "redundant_metadata" property is set to "most", only indirect 1680 * blocks of this level and higher will have an additional ditto block. 1681 */ 1682int zfs_redundant_metadata_most_ditto_level = 2; 1683 1684void 1685dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) 1686{ 1687 dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; 1688 boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || 1689 (wp & WP_SPILL)); 1690 enum zio_checksum checksum = os->os_checksum; 1691 enum zio_compress compress = os->os_compress; 1692 enum zio_checksum dedup_checksum = os->os_dedup_checksum; 1693 boolean_t dedup = B_FALSE; 1694 boolean_t nopwrite = B_FALSE; 1695 boolean_t dedup_verify = os->os_dedup_verify; 1696 int copies = os->os_copies; 1697 1698 /* 1699 * We maintain different write policies for each of the following 1700 * types of data: 1701 * 1. metadata 1702 * 2. preallocated blocks (i.e. level-0 blocks of a dump device) 1703 * 3. all other level 0 blocks 1704 */ 1705 if (ismd) { 1706 /* 1707 * XXX -- we should design a compression algorithm 1708 * that specializes in arrays of bps. 1709 */ 1710 compress = zfs_mdcomp_disable ? ZIO_COMPRESS_EMPTY : 1711 ZIO_COMPRESS_LZJB; 1712 1713 /* 1714 * Metadata always gets checksummed. If the data 1715 * checksum is multi-bit correctable, and it's not a 1716 * ZBT-style checksum, then it's suitable for metadata 1717 * as well. Otherwise, the metadata checksum defaults 1718 * to fletcher4. 1719 */ 1720 if (zio_checksum_table[checksum].ci_correctable < 1 || 1721 zio_checksum_table[checksum].ci_eck) 1722 checksum = ZIO_CHECKSUM_FLETCHER_4; 1723 1724 if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || 1725 (os->os_redundant_metadata == 1726 ZFS_REDUNDANT_METADATA_MOST && 1727 (level >= zfs_redundant_metadata_most_ditto_level || 1728 DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) 1729 copies++; 1730 } else if (wp & WP_NOFILL) { 1731 ASSERT(level == 0); 1732 1733 /* 1734 * If we're writing preallocated blocks, we aren't actually 1735 * writing them so don't set any policy properties. These 1736 * blocks are currently only used by an external subsystem 1737 * outside of zfs (i.e. dump) and not written by the zio 1738 * pipeline. 1739 */ 1740 compress = ZIO_COMPRESS_OFF; 1741 checksum = ZIO_CHECKSUM_NOPARITY; 1742 } else { 1743 compress = zio_compress_select(dn->dn_compress, compress); 1744 1745 checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? 1746 zio_checksum_select(dn->dn_checksum, checksum) : 1747 dedup_checksum; 1748 1749 /* 1750 * Determine dedup setting. If we are in dmu_sync(), 1751 * we won't actually dedup now because that's all 1752 * done in syncing context; but we do want to use the 1753 * dedup checkum. If the checksum is not strong 1754 * enough to ensure unique signatures, force 1755 * dedup_verify. 1756 */ 1757 if (dedup_checksum != ZIO_CHECKSUM_OFF) { 1758 dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; 1759 if (!zio_checksum_table[checksum].ci_dedup) 1760 dedup_verify = B_TRUE; 1761 } 1762 1763 /* 1764 * Enable nopwrite if we have a cryptographically secure 1765 * checksum that has no known collisions (i.e. SHA-256) 1766 * and compression is enabled. We don't enable nopwrite if 1767 * dedup is enabled as the two features are mutually exclusive. 1768 */ 1769 nopwrite = (!dedup && zio_checksum_table[checksum].ci_dedup && 1770 compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); 1771 } 1772 1773 zp->zp_checksum = checksum; 1774 zp->zp_compress = compress; 1775 zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; 1776 zp->zp_level = level; 1777 zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); 1778 zp->zp_dedup = dedup; 1779 zp->zp_dedup_verify = dedup && dedup_verify; 1780 zp->zp_nopwrite = nopwrite; 1781} 1782 1783int 1784dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) 1785{ 1786 dnode_t *dn; 1787 int i, err; 1788 1789 err = dnode_hold(os, object, FTAG, &dn); 1790 if (err) 1791 return (err); 1792 /* 1793 * Sync any current changes before 1794 * we go trundling through the block pointers. 1795 */ 1796 for (i = 0; i < TXG_SIZE; i++) { 1797 if (list_link_active(&dn->dn_dirty_link[i])) 1798 break; 1799 } 1800 if (i != TXG_SIZE) { 1801 dnode_rele(dn, FTAG); 1802 txg_wait_synced(dmu_objset_pool(os), 0); 1803 err = dnode_hold(os, object, FTAG, &dn); 1804 if (err) 1805 return (err); 1806 } 1807 1808 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); 1809 dnode_rele(dn, FTAG); 1810 1811 return (err); 1812} 1813 1814void 1815dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) 1816{ 1817 dnode_phys_t *dnp; 1818 1819 rw_enter(&dn->dn_struct_rwlock, RW_READER); 1820 mutex_enter(&dn->dn_mtx); 1821 1822 dnp = dn->dn_phys; 1823 1824 doi->doi_data_block_size = dn->dn_datablksz; 1825 doi->doi_metadata_block_size = dn->dn_indblkshift ? 1826 1ULL << dn->dn_indblkshift : 0; 1827 doi->doi_type = dn->dn_type; 1828 doi->doi_bonus_type = dn->dn_bonustype; 1829 doi->doi_bonus_size = dn->dn_bonuslen; 1830 doi->doi_indirection = dn->dn_nlevels; 1831 doi->doi_checksum = dn->dn_checksum; 1832 doi->doi_compress = dn->dn_compress; 1833 doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; 1834 doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; 1835 doi->doi_fill_count = 0; 1836 for (int i = 0; i < dnp->dn_nblkptr; i++) 1837 doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); 1838 1839 mutex_exit(&dn->dn_mtx); 1840 rw_exit(&dn->dn_struct_rwlock); 1841} 1842 1843/* 1844 * Get information on a DMU object. 1845 * If doi is NULL, just indicates whether the object exists. 1846 */ 1847int 1848dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) 1849{ 1850 dnode_t *dn; 1851 int err = dnode_hold(os, object, FTAG, &dn); 1852 1853 if (err) 1854 return (err); 1855 1856 if (doi != NULL) 1857 dmu_object_info_from_dnode(dn, doi); 1858 1859 dnode_rele(dn, FTAG); 1860 return (0); 1861} 1862 1863/* 1864 * As above, but faster; can be used when you have a held dbuf in hand. 1865 */ 1866void 1867dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) 1868{ 1869 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 1870 1871 DB_DNODE_ENTER(db); 1872 dmu_object_info_from_dnode(DB_DNODE(db), doi); 1873 DB_DNODE_EXIT(db); 1874} 1875 1876/* 1877 * Faster still when you only care about the size. 1878 * This is specifically optimized for zfs_getattr(). 1879 */ 1880void 1881dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, 1882 u_longlong_t *nblk512) 1883{ 1884 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 1885 dnode_t *dn; 1886 1887 DB_DNODE_ENTER(db); 1888 dn = DB_DNODE(db); 1889 1890 *blksize = dn->dn_datablksz; 1891 /* add 1 for dnode space */ 1892 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> 1893 SPA_MINBLOCKSHIFT) + 1; 1894 DB_DNODE_EXIT(db); 1895} 1896 1897void 1898byteswap_uint64_array(void *vbuf, size_t size) 1899{ 1900 uint64_t *buf = vbuf; 1901 size_t count = size >> 3; 1902 int i; 1903 1904 ASSERT((size & 7) == 0); 1905 1906 for (i = 0; i < count; i++) 1907 buf[i] = BSWAP_64(buf[i]); 1908} 1909 1910void 1911byteswap_uint32_array(void *vbuf, size_t size) 1912{ 1913 uint32_t *buf = vbuf; 1914 size_t count = size >> 2; 1915 int i; 1916 1917 ASSERT((size & 3) == 0); 1918 1919 for (i = 0; i < count; i++) 1920 buf[i] = BSWAP_32(buf[i]); 1921} 1922 1923void 1924byteswap_uint16_array(void *vbuf, size_t size) 1925{ 1926 uint16_t *buf = vbuf; 1927 size_t count = size >> 1; 1928 int i; 1929 1930 ASSERT((size & 1) == 0); 1931 1932 for (i = 0; i < count; i++) 1933 buf[i] = BSWAP_16(buf[i]); 1934} 1935 1936/* ARGSUSED */ 1937void 1938byteswap_uint8_array(void *vbuf, size_t size) 1939{ 1940} 1941 1942void 1943dmu_init(void) 1944{ 1945 zfs_dbgmsg_init(); 1946 sa_cache_init(); 1947 xuio_stat_init(); 1948 dmu_objset_init(); 1949 dnode_init(); 1950 dbuf_init(); 1951 zfetch_init(); 1952 zio_compress_init(); 1953 l2arc_init(); 1954 arc_init(); 1955} 1956 1957void 1958dmu_fini(void) 1959{ 1960 arc_fini(); /* arc depends on l2arc, so arc must go first */ 1961 l2arc_fini(); 1962 zfetch_fini(); 1963 zio_compress_fini(); 1964 dbuf_fini(); 1965 dnode_fini(); 1966 dmu_objset_fini(); 1967 xuio_stat_fini(); 1968 sa_cache_fini(); 1969 zfs_dbgmsg_fini(); 1970} 1971