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 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/dmu.h> 29 #include <sys/dmu_impl.h> 30 #include <sys/dmu_tx.h> 31 #include <sys/dbuf.h> 32 #include <sys/dnode.h> 33 #include <sys/zfs_context.h> 34 #include <sys/dmu_objset.h> 35 #include <sys/dmu_traverse.h> 36 #include <sys/dsl_dataset.h> 37 #include <sys/dsl_dir.h> 38 #include <sys/dsl_pool.h> 39 #include <sys/dsl_synctask.h> 40 #include <sys/dsl_prop.h> 41 #include <sys/dmu_zfetch.h> 42 #include <sys/zfs_ioctl.h> 43 #include <sys/zap.h> 44 #include <sys/zio_checksum.h> 45 #ifdef _KERNEL 46 #include <sys/vmsystm.h> 47 #endif 48 49 const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { 50 { byteswap_uint8_array, TRUE, "unallocated" }, 51 { zap_byteswap, TRUE, "object directory" }, 52 { byteswap_uint64_array, TRUE, "object array" }, 53 { byteswap_uint8_array, TRUE, "packed nvlist" }, 54 { byteswap_uint64_array, TRUE, "packed nvlist size" }, 55 { byteswap_uint64_array, TRUE, "bplist" }, 56 { byteswap_uint64_array, TRUE, "bplist header" }, 57 { byteswap_uint64_array, TRUE, "SPA space map header" }, 58 { byteswap_uint64_array, TRUE, "SPA space map" }, 59 { byteswap_uint64_array, TRUE, "ZIL intent log" }, 60 { dnode_buf_byteswap, TRUE, "DMU dnode" }, 61 { dmu_objset_byteswap, TRUE, "DMU objset" }, 62 { byteswap_uint64_array, TRUE, "DSL directory" }, 63 { zap_byteswap, TRUE, "DSL directory child map"}, 64 { zap_byteswap, TRUE, "DSL dataset snap map" }, 65 { zap_byteswap, TRUE, "DSL props" }, 66 { byteswap_uint64_array, TRUE, "DSL dataset" }, 67 { zfs_znode_byteswap, TRUE, "ZFS znode" }, 68 { zfs_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 }; 88 89 int 90 dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, 91 void *tag, dmu_buf_t **dbp) 92 { 93 dnode_t *dn; 94 uint64_t blkid; 95 dmu_buf_impl_t *db; 96 int err; 97 98 err = dnode_hold(os->os, object, FTAG, &dn); 99 if (err) 100 return (err); 101 blkid = dbuf_whichblock(dn, offset); 102 rw_enter(&dn->dn_struct_rwlock, RW_READER); 103 db = dbuf_hold(dn, blkid, tag); 104 rw_exit(&dn->dn_struct_rwlock); 105 if (db == NULL) { 106 err = EIO; 107 } else { 108 err = dbuf_read(db, NULL, DB_RF_CANFAIL); 109 if (err) { 110 dbuf_rele(db, tag); 111 db = NULL; 112 } 113 } 114 115 dnode_rele(dn, FTAG); 116 *dbp = &db->db; 117 return (err); 118 } 119 120 int 121 dmu_bonus_max(void) 122 { 123 return (DN_MAX_BONUSLEN); 124 } 125 126 int 127 dmu_set_bonus(dmu_buf_t *db, int newsize, dmu_tx_t *tx) 128 { 129 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 130 131 if (dn->dn_bonus != (dmu_buf_impl_t *)db) 132 return (EINVAL); 133 if (newsize < 0 || newsize > db->db_size) 134 return (EINVAL); 135 dnode_setbonuslen(dn, newsize, tx); 136 return (0); 137 } 138 139 /* 140 * returns ENOENT, EIO, or 0. 141 */ 142 int 143 dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) 144 { 145 dnode_t *dn; 146 dmu_buf_impl_t *db; 147 int error; 148 149 error = dnode_hold(os->os, object, FTAG, &dn); 150 if (error) 151 return (error); 152 153 rw_enter(&dn->dn_struct_rwlock, RW_READER); 154 if (dn->dn_bonus == NULL) { 155 rw_exit(&dn->dn_struct_rwlock); 156 rw_enter(&dn->dn_struct_rwlock, RW_WRITER); 157 if (dn->dn_bonus == NULL) 158 dbuf_create_bonus(dn); 159 } 160 db = dn->dn_bonus; 161 rw_exit(&dn->dn_struct_rwlock); 162 163 /* as long as the bonus buf is held, the dnode will be held */ 164 if (refcount_add(&db->db_holds, tag) == 1) 165 VERIFY(dnode_add_ref(dn, db)); 166 167 dnode_rele(dn, FTAG); 168 169 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED)); 170 171 *dbp = &db->db; 172 return (0); 173 } 174 175 /* 176 * Note: longer-term, we should modify all of the dmu_buf_*() interfaces 177 * to take a held dnode rather than <os, object> -- the lookup is wasteful, 178 * and can induce severe lock contention when writing to several files 179 * whose dnodes are in the same block. 180 */ 181 static int 182 dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, 183 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 184 { 185 dmu_buf_t **dbp; 186 uint64_t blkid, nblks, i; 187 uint32_t flags; 188 int err; 189 zio_t *zio; 190 191 ASSERT(length <= DMU_MAX_ACCESS); 192 193 flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT; 194 if (length > zfetch_array_rd_sz) 195 flags |= DB_RF_NOPREFETCH; 196 197 rw_enter(&dn->dn_struct_rwlock, RW_READER); 198 if (dn->dn_datablkshift) { 199 int blkshift = dn->dn_datablkshift; 200 nblks = (P2ROUNDUP(offset+length, 1ULL<<blkshift) - 201 P2ALIGN(offset, 1ULL<<blkshift)) >> blkshift; 202 } else { 203 if (offset + length > dn->dn_datablksz) { 204 zfs_panic_recover("zfs: accessing past end of object " 205 "%llx/%llx (size=%u access=%llu+%llu)", 206 (longlong_t)dn->dn_objset-> 207 os_dsl_dataset->ds_object, 208 (longlong_t)dn->dn_object, dn->dn_datablksz, 209 (longlong_t)offset, (longlong_t)length); 210 return (EIO); 211 } 212 nblks = 1; 213 } 214 dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); 215 216 zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, TRUE); 217 blkid = dbuf_whichblock(dn, offset); 218 for (i = 0; i < nblks; i++) { 219 dmu_buf_impl_t *db = dbuf_hold(dn, blkid+i, tag); 220 if (db == NULL) { 221 rw_exit(&dn->dn_struct_rwlock); 222 dmu_buf_rele_array(dbp, nblks, tag); 223 zio_nowait(zio); 224 return (EIO); 225 } 226 /* initiate async i/o */ 227 if (read) { 228 rw_exit(&dn->dn_struct_rwlock); 229 (void) dbuf_read(db, zio, flags); 230 rw_enter(&dn->dn_struct_rwlock, RW_READER); 231 } 232 dbp[i] = &db->db; 233 } 234 rw_exit(&dn->dn_struct_rwlock); 235 236 /* wait for async i/o */ 237 err = zio_wait(zio); 238 if (err) { 239 dmu_buf_rele_array(dbp, nblks, tag); 240 return (err); 241 } 242 243 /* wait for other io to complete */ 244 if (read) { 245 for (i = 0; i < nblks; i++) { 246 dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; 247 mutex_enter(&db->db_mtx); 248 while (db->db_state == DB_READ || 249 db->db_state == DB_FILL) 250 cv_wait(&db->db_changed, &db->db_mtx); 251 if (db->db_state == DB_UNCACHED) 252 err = EIO; 253 mutex_exit(&db->db_mtx); 254 if (err) { 255 dmu_buf_rele_array(dbp, nblks, tag); 256 return (err); 257 } 258 } 259 } 260 261 *numbufsp = nblks; 262 *dbpp = dbp; 263 return (0); 264 } 265 266 static int 267 dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, 268 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 269 { 270 dnode_t *dn; 271 int err; 272 273 err = dnode_hold(os->os, object, FTAG, &dn); 274 if (err) 275 return (err); 276 277 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 278 numbufsp, dbpp); 279 280 dnode_rele(dn, FTAG); 281 282 return (err); 283 } 284 285 int 286 dmu_buf_hold_array_by_bonus(dmu_buf_t *db, uint64_t offset, 287 uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) 288 { 289 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 290 int err; 291 292 err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, 293 numbufsp, dbpp); 294 295 return (err); 296 } 297 298 void 299 dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) 300 { 301 int i; 302 dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; 303 304 if (numbufs == 0) 305 return; 306 307 for (i = 0; i < numbufs; i++) { 308 if (dbp[i]) 309 dbuf_rele(dbp[i], tag); 310 } 311 312 kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); 313 } 314 315 void 316 dmu_prefetch(objset_t *os, uint64_t object, uint64_t offset, uint64_t len) 317 { 318 dnode_t *dn; 319 uint64_t blkid; 320 int nblks, i, err; 321 322 if (zfs_prefetch_disable) 323 return; 324 325 if (len == 0) { /* they're interested in the bonus buffer */ 326 dn = os->os->os_meta_dnode; 327 328 if (object == 0 || object >= DN_MAX_OBJECT) 329 return; 330 331 rw_enter(&dn->dn_struct_rwlock, RW_READER); 332 blkid = dbuf_whichblock(dn, object * sizeof (dnode_phys_t)); 333 dbuf_prefetch(dn, blkid); 334 rw_exit(&dn->dn_struct_rwlock); 335 return; 336 } 337 338 /* 339 * XXX - Note, if the dnode for the requested object is not 340 * already cached, we will do a *synchronous* read in the 341 * dnode_hold() call. The same is true for any indirects. 342 */ 343 err = dnode_hold(os->os, object, FTAG, &dn); 344 if (err != 0) 345 return; 346 347 rw_enter(&dn->dn_struct_rwlock, RW_READER); 348 if (dn->dn_datablkshift) { 349 int blkshift = dn->dn_datablkshift; 350 nblks = (P2ROUNDUP(offset+len, 1<<blkshift) - 351 P2ALIGN(offset, 1<<blkshift)) >> blkshift; 352 } else { 353 nblks = (offset < dn->dn_datablksz); 354 } 355 356 if (nblks != 0) { 357 blkid = dbuf_whichblock(dn, offset); 358 for (i = 0; i < nblks; i++) 359 dbuf_prefetch(dn, blkid+i); 360 } 361 362 rw_exit(&dn->dn_struct_rwlock); 363 364 dnode_rele(dn, FTAG); 365 } 366 367 static int 368 get_next_chunk(dnode_t *dn, uint64_t *offset, uint64_t limit) 369 { 370 uint64_t len = limit - *offset; 371 uint64_t chunk_len = dn->dn_datablksz * DMU_MAX_DELETEBLKCNT; 372 uint64_t dn_used; 373 int err; 374 375 ASSERT(limit <= *offset); 376 377 dn_used = dn->dn_phys->dn_used << 378 (dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES ? 0 : DEV_BSHIFT); 379 if (len <= chunk_len || dn_used <= chunk_len) { 380 *offset = limit; 381 return (0); 382 } 383 384 while (*offset > limit) { 385 uint64_t initial_offset = *offset; 386 uint64_t delta; 387 388 /* skip over allocated data */ 389 err = dnode_next_offset(dn, 390 DNODE_FIND_HOLE|DNODE_FIND_BACKWARDS, offset, 1, 1, 0); 391 if (err == ESRCH) 392 *offset = limit; 393 else if (err) 394 return (err); 395 396 ASSERT3U(*offset, <=, initial_offset); 397 delta = initial_offset - *offset; 398 if (delta >= chunk_len) { 399 *offset += delta - chunk_len; 400 return (0); 401 } 402 chunk_len -= delta; 403 404 /* skip over unallocated data */ 405 err = dnode_next_offset(dn, 406 DNODE_FIND_BACKWARDS, offset, 1, 1, 0); 407 if (err == ESRCH) 408 *offset = limit; 409 else if (err) 410 return (err); 411 412 if (*offset < limit) 413 *offset = limit; 414 ASSERT3U(*offset, <, initial_offset); 415 } 416 return (0); 417 } 418 419 static int 420 dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, 421 uint64_t length, boolean_t free_dnode) 422 { 423 dmu_tx_t *tx; 424 uint64_t object_size, start, end, len; 425 boolean_t trunc = (length == DMU_OBJECT_END); 426 int align, err; 427 428 align = 1 << dn->dn_datablkshift; 429 ASSERT(align > 0); 430 object_size = align == 1 ? dn->dn_datablksz : 431 (dn->dn_maxblkid + 1) << dn->dn_datablkshift; 432 433 if (trunc || (end = offset + length) > object_size) 434 end = object_size; 435 if (end <= offset) 436 return (0); 437 length = end - offset; 438 439 while (length) { 440 start = end; 441 err = get_next_chunk(dn, &start, offset); 442 if (err) 443 return (err); 444 len = trunc ? DMU_OBJECT_END : end - start; 445 446 tx = dmu_tx_create(os); 447 dmu_tx_hold_free(tx, dn->dn_object, start, len); 448 err = dmu_tx_assign(tx, TXG_WAIT); 449 if (err) { 450 dmu_tx_abort(tx); 451 return (err); 452 } 453 454 dnode_free_range(dn, start, trunc ? -1 : len, tx); 455 456 if (start == 0 && trunc && free_dnode) 457 dnode_free(dn, tx); 458 459 length -= end - start; 460 461 dmu_tx_commit(tx); 462 end = start; 463 trunc = FALSE; 464 } 465 return (0); 466 } 467 468 int 469 dmu_free_long_range(objset_t *os, uint64_t object, 470 uint64_t offset, uint64_t length) 471 { 472 dnode_t *dn; 473 int err; 474 475 err = dnode_hold(os->os, object, FTAG, &dn); 476 if (err != 0) 477 return (err); 478 err = dmu_free_long_range_impl(os, dn, offset, length, FALSE); 479 dnode_rele(dn, FTAG); 480 return (err); 481 } 482 483 int 484 dmu_free_object(objset_t *os, uint64_t object) 485 { 486 dnode_t *dn; 487 dmu_tx_t *tx; 488 int err; 489 490 err = dnode_hold_impl(os->os, object, DNODE_MUST_BE_ALLOCATED, 491 FTAG, &dn); 492 if (err != 0) 493 return (err); 494 if (dn->dn_nlevels == 1) { 495 tx = dmu_tx_create(os); 496 dmu_tx_hold_bonus(tx, object); 497 dmu_tx_hold_free(tx, dn->dn_object, 0, DMU_OBJECT_END); 498 err = dmu_tx_assign(tx, TXG_WAIT); 499 if (err == 0) { 500 dnode_free_range(dn, 0, DMU_OBJECT_END, tx); 501 dnode_free(dn, tx); 502 dmu_tx_commit(tx); 503 } else { 504 dmu_tx_abort(tx); 505 } 506 } else { 507 err = dmu_free_long_range_impl(os, dn, 0, DMU_OBJECT_END, TRUE); 508 } 509 dnode_rele(dn, FTAG); 510 return (err); 511 } 512 513 int 514 dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, 515 uint64_t size, dmu_tx_t *tx) 516 { 517 dnode_t *dn; 518 int err = dnode_hold(os->os, object, FTAG, &dn); 519 if (err) 520 return (err); 521 ASSERT(offset < UINT64_MAX); 522 ASSERT(size == -1ULL || size <= UINT64_MAX - offset); 523 dnode_free_range(dn, offset, size, tx); 524 dnode_rele(dn, FTAG); 525 return (0); 526 } 527 528 int 529 dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 530 void *buf) 531 { 532 dnode_t *dn; 533 dmu_buf_t **dbp; 534 int numbufs, i, err; 535 536 err = dnode_hold(os->os, object, FTAG, &dn); 537 if (err) 538 return (err); 539 540 /* 541 * Deal with odd block sizes, where there can't be data past the first 542 * block. If we ever do the tail block optimization, we will need to 543 * handle that here as well. 544 */ 545 if (dn->dn_datablkshift == 0) { 546 int newsz = offset > dn->dn_datablksz ? 0 : 547 MIN(size, dn->dn_datablksz - offset); 548 bzero((char *)buf + newsz, size - newsz); 549 size = newsz; 550 } 551 552 while (size > 0) { 553 uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); 554 555 /* 556 * NB: we could do this block-at-a-time, but it's nice 557 * to be reading in parallel. 558 */ 559 err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, 560 TRUE, FTAG, &numbufs, &dbp); 561 if (err) 562 break; 563 564 for (i = 0; i < numbufs; i++) { 565 int tocpy; 566 int bufoff; 567 dmu_buf_t *db = dbp[i]; 568 569 ASSERT(size > 0); 570 571 bufoff = offset - db->db_offset; 572 tocpy = (int)MIN(db->db_size - bufoff, size); 573 574 bcopy((char *)db->db_data + bufoff, buf, tocpy); 575 576 offset += tocpy; 577 size -= tocpy; 578 buf = (char *)buf + tocpy; 579 } 580 dmu_buf_rele_array(dbp, numbufs, FTAG); 581 } 582 dnode_rele(dn, FTAG); 583 return (err); 584 } 585 586 void 587 dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 588 const void *buf, dmu_tx_t *tx) 589 { 590 dmu_buf_t **dbp; 591 int numbufs, i; 592 593 if (size == 0) 594 return; 595 596 VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, 597 FALSE, FTAG, &numbufs, &dbp)); 598 599 for (i = 0; i < numbufs; i++) { 600 int tocpy; 601 int bufoff; 602 dmu_buf_t *db = dbp[i]; 603 604 ASSERT(size > 0); 605 606 bufoff = offset - db->db_offset; 607 tocpy = (int)MIN(db->db_size - bufoff, size); 608 609 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 610 611 if (tocpy == db->db_size) 612 dmu_buf_will_fill(db, tx); 613 else 614 dmu_buf_will_dirty(db, tx); 615 616 bcopy(buf, (char *)db->db_data + bufoff, tocpy); 617 618 if (tocpy == db->db_size) 619 dmu_buf_fill_done(db, tx); 620 621 offset += tocpy; 622 size -= tocpy; 623 buf = (char *)buf + tocpy; 624 } 625 dmu_buf_rele_array(dbp, numbufs, FTAG); 626 } 627 628 #ifdef _KERNEL 629 int 630 dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) 631 { 632 dmu_buf_t **dbp; 633 int numbufs, i, err; 634 635 /* 636 * NB: we could do this block-at-a-time, but it's nice 637 * to be reading in parallel. 638 */ 639 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, TRUE, FTAG, 640 &numbufs, &dbp); 641 if (err) 642 return (err); 643 644 for (i = 0; i < numbufs; i++) { 645 int tocpy; 646 int bufoff; 647 dmu_buf_t *db = dbp[i]; 648 649 ASSERT(size > 0); 650 651 bufoff = uio->uio_loffset - db->db_offset; 652 tocpy = (int)MIN(db->db_size - bufoff, size); 653 654 err = uiomove((char *)db->db_data + bufoff, tocpy, 655 UIO_READ, uio); 656 if (err) 657 break; 658 659 size -= tocpy; 660 } 661 dmu_buf_rele_array(dbp, numbufs, FTAG); 662 663 return (err); 664 } 665 666 int 667 dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, 668 dmu_tx_t *tx) 669 { 670 dmu_buf_t **dbp; 671 int numbufs, i; 672 int err = 0; 673 674 if (size == 0) 675 return (0); 676 677 err = dmu_buf_hold_array(os, object, uio->uio_loffset, size, 678 FALSE, FTAG, &numbufs, &dbp); 679 if (err) 680 return (err); 681 682 for (i = 0; i < numbufs; i++) { 683 int tocpy; 684 int bufoff; 685 dmu_buf_t *db = dbp[i]; 686 687 ASSERT(size > 0); 688 689 bufoff = uio->uio_loffset - db->db_offset; 690 tocpy = (int)MIN(db->db_size - bufoff, size); 691 692 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 693 694 if (tocpy == db->db_size) 695 dmu_buf_will_fill(db, tx); 696 else 697 dmu_buf_will_dirty(db, tx); 698 699 /* 700 * XXX uiomove could block forever (eg. nfs-backed 701 * pages). There needs to be a uiolockdown() function 702 * to lock the pages in memory, so that uiomove won't 703 * block. 704 */ 705 err = uiomove((char *)db->db_data + bufoff, tocpy, 706 UIO_WRITE, uio); 707 708 if (tocpy == db->db_size) 709 dmu_buf_fill_done(db, tx); 710 711 if (err) 712 break; 713 714 size -= tocpy; 715 } 716 dmu_buf_rele_array(dbp, numbufs, FTAG); 717 return (err); 718 } 719 720 int 721 dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, 722 page_t *pp, dmu_tx_t *tx) 723 { 724 dmu_buf_t **dbp; 725 int numbufs, i; 726 int err; 727 728 if (size == 0) 729 return (0); 730 731 err = dmu_buf_hold_array(os, object, offset, size, 732 FALSE, FTAG, &numbufs, &dbp); 733 if (err) 734 return (err); 735 736 for (i = 0; i < numbufs; i++) { 737 int tocpy, copied, thiscpy; 738 int bufoff; 739 dmu_buf_t *db = dbp[i]; 740 caddr_t va; 741 742 ASSERT(size > 0); 743 ASSERT3U(db->db_size, >=, PAGESIZE); 744 745 bufoff = offset - db->db_offset; 746 tocpy = (int)MIN(db->db_size - bufoff, size); 747 748 ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); 749 750 if (tocpy == db->db_size) 751 dmu_buf_will_fill(db, tx); 752 else 753 dmu_buf_will_dirty(db, tx); 754 755 for (copied = 0; copied < tocpy; copied += PAGESIZE) { 756 ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); 757 thiscpy = MIN(PAGESIZE, tocpy - copied); 758 va = ppmapin(pp, PROT_READ, (caddr_t)-1); 759 bcopy(va, (char *)db->db_data + bufoff, thiscpy); 760 ppmapout(va); 761 pp = pp->p_next; 762 bufoff += PAGESIZE; 763 } 764 765 if (tocpy == db->db_size) 766 dmu_buf_fill_done(db, tx); 767 768 if (err) 769 break; 770 771 offset += tocpy; 772 size -= tocpy; 773 } 774 dmu_buf_rele_array(dbp, numbufs, FTAG); 775 return (err); 776 } 777 #endif 778 779 typedef struct { 780 dbuf_dirty_record_t *dr; 781 dmu_sync_cb_t *done; 782 void *arg; 783 } dmu_sync_arg_t; 784 785 /* ARGSUSED */ 786 static void 787 dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) 788 { 789 dmu_sync_arg_t *in = varg; 790 dbuf_dirty_record_t *dr = in->dr; 791 dmu_buf_impl_t *db = dr->dr_dbuf; 792 dmu_sync_cb_t *done = in->done; 793 794 if (!BP_IS_HOLE(zio->io_bp)) { 795 zio->io_bp->blk_fill = 1; 796 BP_SET_TYPE(zio->io_bp, db->db_dnode->dn_type); 797 BP_SET_LEVEL(zio->io_bp, 0); 798 } 799 800 mutex_enter(&db->db_mtx); 801 ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); 802 dr->dt.dl.dr_overridden_by = *zio->io_bp; /* structure assignment */ 803 dr->dt.dl.dr_override_state = DR_OVERRIDDEN; 804 cv_broadcast(&db->db_changed); 805 mutex_exit(&db->db_mtx); 806 807 if (done) 808 done(&(db->db), in->arg); 809 810 kmem_free(in, sizeof (dmu_sync_arg_t)); 811 } 812 813 /* 814 * Intent log support: sync the block associated with db to disk. 815 * N.B. and XXX: the caller is responsible for making sure that the 816 * data isn't changing while dmu_sync() is writing it. 817 * 818 * Return values: 819 * 820 * EEXIST: this txg has already been synced, so there's nothing to to. 821 * The caller should not log the write. 822 * 823 * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. 824 * The caller should not log the write. 825 * 826 * EALREADY: this block is already in the process of being synced. 827 * The caller should track its progress (somehow). 828 * 829 * EINPROGRESS: the IO has been initiated. 830 * The caller should log this blkptr in the callback. 831 * 832 * 0: completed. Sets *bp to the blkptr just written. 833 * The caller should log this blkptr immediately. 834 */ 835 int 836 dmu_sync(zio_t *pio, dmu_buf_t *db_fake, 837 blkptr_t *bp, uint64_t txg, dmu_sync_cb_t *done, void *arg) 838 { 839 dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; 840 objset_impl_t *os = db->db_objset; 841 dsl_pool_t *dp = os->os_dsl_dataset->ds_dir->dd_pool; 842 tx_state_t *tx = &dp->dp_tx; 843 dbuf_dirty_record_t *dr; 844 dmu_sync_arg_t *in; 845 zbookmark_t zb; 846 zio_t *zio; 847 int zio_flags; 848 int err; 849 850 ASSERT(BP_IS_HOLE(bp)); 851 ASSERT(txg != 0); 852 853 854 dprintf("dmu_sync txg=%llu, s,o,q %llu %llu %llu\n", 855 txg, tx->tx_synced_txg, tx->tx_open_txg, tx->tx_quiesced_txg); 856 857 /* 858 * XXX - would be nice if we could do this without suspending... 859 */ 860 txg_suspend(dp); 861 862 /* 863 * If this txg already synced, there's nothing to do. 864 */ 865 if (txg <= tx->tx_synced_txg) { 866 txg_resume(dp); 867 /* 868 * If we're running ziltest, we need the blkptr regardless. 869 */ 870 if (txg > spa_freeze_txg(dp->dp_spa)) { 871 /* if db_blkptr == NULL, this was an empty write */ 872 if (db->db_blkptr) 873 *bp = *db->db_blkptr; /* structure assignment */ 874 return (0); 875 } 876 return (EEXIST); 877 } 878 879 mutex_enter(&db->db_mtx); 880 881 if (txg == tx->tx_syncing_txg) { 882 while (db->db_data_pending) { 883 /* 884 * IO is in-progress. Wait for it to finish. 885 * XXX - would be nice to be able to somehow "attach" 886 * this zio to the parent zio passed in. 887 */ 888 cv_wait(&db->db_changed, &db->db_mtx); 889 if (!db->db_data_pending && 890 db->db_blkptr && BP_IS_HOLE(db->db_blkptr)) { 891 /* 892 * IO was compressed away 893 */ 894 *bp = *db->db_blkptr; /* structure assignment */ 895 mutex_exit(&db->db_mtx); 896 txg_resume(dp); 897 return (0); 898 } 899 ASSERT(db->db_data_pending || 900 (db->db_blkptr && db->db_blkptr->blk_birth == txg)); 901 } 902 903 if (db->db_blkptr && db->db_blkptr->blk_birth == txg) { 904 /* 905 * IO is already completed. 906 */ 907 *bp = *db->db_blkptr; /* structure assignment */ 908 mutex_exit(&db->db_mtx); 909 txg_resume(dp); 910 return (0); 911 } 912 } 913 914 dr = db->db_last_dirty; 915 while (dr && dr->dr_txg > txg) 916 dr = dr->dr_next; 917 if (dr == NULL || dr->dr_txg < txg) { 918 /* 919 * This dbuf isn't dirty, must have been free_range'd. 920 * There's no need to log writes to freed blocks, so we're done. 921 */ 922 mutex_exit(&db->db_mtx); 923 txg_resume(dp); 924 return (ENOENT); 925 } 926 927 ASSERT(dr->dr_txg == txg); 928 if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC) { 929 /* 930 * We have already issued a sync write for this buffer. 931 */ 932 mutex_exit(&db->db_mtx); 933 txg_resume(dp); 934 return (EALREADY); 935 } else if (dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { 936 /* 937 * This buffer has already been synced. It could not 938 * have been dirtied since, or we would have cleared the state. 939 */ 940 *bp = dr->dt.dl.dr_overridden_by; /* structure assignment */ 941 mutex_exit(&db->db_mtx); 942 txg_resume(dp); 943 return (0); 944 } 945 946 dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; 947 in = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); 948 in->dr = dr; 949 in->done = done; 950 in->arg = arg; 951 mutex_exit(&db->db_mtx); 952 txg_resume(dp); 953 954 zb.zb_objset = os->os_dsl_dataset->ds_object; 955 zb.zb_object = db->db.db_object; 956 zb.zb_level = db->db_level; 957 zb.zb_blkid = db->db_blkid; 958 zio_flags = ZIO_FLAG_MUSTSUCCEED; 959 if (dmu_ot[db->db_dnode->dn_type].ot_metadata || zb.zb_level != 0) 960 zio_flags |= ZIO_FLAG_METADATA; 961 zio = arc_write(pio, os->os_spa, 962 zio_checksum_select(db->db_dnode->dn_checksum, os->os_checksum), 963 zio_compress_select(db->db_dnode->dn_compress, os->os_compress), 964 dmu_get_replication_level(os, &zb, db->db_dnode->dn_type), 965 txg, bp, dr->dt.dl.dr_data, NULL, dmu_sync_done, in, 966 ZIO_PRIORITY_SYNC_WRITE, zio_flags, &zb); 967 968 if (pio) { 969 zio_nowait(zio); 970 err = EINPROGRESS; 971 } else { 972 err = zio_wait(zio); 973 ASSERT(err == 0); 974 } 975 return (err); 976 } 977 978 int 979 dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, 980 dmu_tx_t *tx) 981 { 982 dnode_t *dn; 983 int err; 984 985 err = dnode_hold(os->os, object, FTAG, &dn); 986 if (err) 987 return (err); 988 err = dnode_set_blksz(dn, size, ibs, tx); 989 dnode_rele(dn, FTAG); 990 return (err); 991 } 992 993 void 994 dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, 995 dmu_tx_t *tx) 996 { 997 dnode_t *dn; 998 999 /* XXX assumes dnode_hold will not get an i/o error */ 1000 (void) dnode_hold(os->os, object, FTAG, &dn); 1001 ASSERT(checksum < ZIO_CHECKSUM_FUNCTIONS); 1002 dn->dn_checksum = checksum; 1003 dnode_setdirty(dn, tx); 1004 dnode_rele(dn, FTAG); 1005 } 1006 1007 void 1008 dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, 1009 dmu_tx_t *tx) 1010 { 1011 dnode_t *dn; 1012 1013 /* XXX assumes dnode_hold will not get an i/o error */ 1014 (void) dnode_hold(os->os, object, FTAG, &dn); 1015 ASSERT(compress < ZIO_COMPRESS_FUNCTIONS); 1016 dn->dn_compress = compress; 1017 dnode_setdirty(dn, tx); 1018 dnode_rele(dn, FTAG); 1019 } 1020 1021 int 1022 dmu_get_replication_level(objset_impl_t *os, 1023 zbookmark_t *zb, dmu_object_type_t ot) 1024 { 1025 int ncopies = os->os_copies; 1026 1027 /* If it's the mos, it should have max copies set. */ 1028 ASSERT(zb->zb_objset != 0 || 1029 ncopies == spa_max_replication(os->os_spa)); 1030 1031 if (dmu_ot[ot].ot_metadata || zb->zb_level != 0) 1032 ncopies++; 1033 return (MIN(ncopies, spa_max_replication(os->os_spa))); 1034 } 1035 1036 int 1037 dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) 1038 { 1039 dnode_t *dn; 1040 int i, err; 1041 1042 err = dnode_hold(os->os, object, FTAG, &dn); 1043 if (err) 1044 return (err); 1045 /* 1046 * Sync any current changes before 1047 * we go trundling through the block pointers. 1048 */ 1049 for (i = 0; i < TXG_SIZE; i++) { 1050 if (list_link_active(&dn->dn_dirty_link[i])) 1051 break; 1052 } 1053 if (i != TXG_SIZE) { 1054 dnode_rele(dn, FTAG); 1055 txg_wait_synced(dmu_objset_pool(os), 0); 1056 err = dnode_hold(os->os, object, FTAG, &dn); 1057 if (err) 1058 return (err); 1059 } 1060 1061 err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); 1062 dnode_rele(dn, FTAG); 1063 1064 return (err); 1065 } 1066 1067 void 1068 dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) 1069 { 1070 rw_enter(&dn->dn_struct_rwlock, RW_READER); 1071 mutex_enter(&dn->dn_mtx); 1072 1073 doi->doi_data_block_size = dn->dn_datablksz; 1074 doi->doi_metadata_block_size = dn->dn_indblkshift ? 1075 1ULL << dn->dn_indblkshift : 0; 1076 doi->doi_indirection = dn->dn_nlevels; 1077 doi->doi_checksum = dn->dn_checksum; 1078 doi->doi_compress = dn->dn_compress; 1079 doi->doi_physical_blks = (DN_USED_BYTES(dn->dn_phys) + 1080 SPA_MINBLOCKSIZE/2) >> SPA_MINBLOCKSHIFT; 1081 doi->doi_max_block_offset = dn->dn_phys->dn_maxblkid; 1082 doi->doi_type = dn->dn_type; 1083 doi->doi_bonus_size = dn->dn_bonuslen; 1084 doi->doi_bonus_type = dn->dn_bonustype; 1085 1086 mutex_exit(&dn->dn_mtx); 1087 rw_exit(&dn->dn_struct_rwlock); 1088 } 1089 1090 /* 1091 * Get information on a DMU object. 1092 * If doi is NULL, just indicates whether the object exists. 1093 */ 1094 int 1095 dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) 1096 { 1097 dnode_t *dn; 1098 int err = dnode_hold(os->os, object, FTAG, &dn); 1099 1100 if (err) 1101 return (err); 1102 1103 if (doi != NULL) 1104 dmu_object_info_from_dnode(dn, doi); 1105 1106 dnode_rele(dn, FTAG); 1107 return (0); 1108 } 1109 1110 /* 1111 * As above, but faster; can be used when you have a held dbuf in hand. 1112 */ 1113 void 1114 dmu_object_info_from_db(dmu_buf_t *db, dmu_object_info_t *doi) 1115 { 1116 dmu_object_info_from_dnode(((dmu_buf_impl_t *)db)->db_dnode, doi); 1117 } 1118 1119 /* 1120 * Faster still when you only care about the size. 1121 * This is specifically optimized for zfs_getattr(). 1122 */ 1123 void 1124 dmu_object_size_from_db(dmu_buf_t *db, uint32_t *blksize, u_longlong_t *nblk512) 1125 { 1126 dnode_t *dn = ((dmu_buf_impl_t *)db)->db_dnode; 1127 1128 *blksize = dn->dn_datablksz; 1129 /* add 1 for dnode space */ 1130 *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> 1131 SPA_MINBLOCKSHIFT) + 1; 1132 } 1133 1134 void 1135 byteswap_uint64_array(void *vbuf, size_t size) 1136 { 1137 uint64_t *buf = vbuf; 1138 size_t count = size >> 3; 1139 int i; 1140 1141 ASSERT((size & 7) == 0); 1142 1143 for (i = 0; i < count; i++) 1144 buf[i] = BSWAP_64(buf[i]); 1145 } 1146 1147 void 1148 byteswap_uint32_array(void *vbuf, size_t size) 1149 { 1150 uint32_t *buf = vbuf; 1151 size_t count = size >> 2; 1152 int i; 1153 1154 ASSERT((size & 3) == 0); 1155 1156 for (i = 0; i < count; i++) 1157 buf[i] = BSWAP_32(buf[i]); 1158 } 1159 1160 void 1161 byteswap_uint16_array(void *vbuf, size_t size) 1162 { 1163 uint16_t *buf = vbuf; 1164 size_t count = size >> 1; 1165 int i; 1166 1167 ASSERT((size & 1) == 0); 1168 1169 for (i = 0; i < count; i++) 1170 buf[i] = BSWAP_16(buf[i]); 1171 } 1172 1173 /* ARGSUSED */ 1174 void 1175 byteswap_uint8_array(void *vbuf, size_t size) 1176 { 1177 } 1178 1179 void 1180 dmu_init(void) 1181 { 1182 dbuf_init(); 1183 dnode_init(); 1184 arc_init(); 1185 l2arc_init(); 1186 } 1187 1188 void 1189 dmu_fini(void) 1190 { 1191 arc_fini(); 1192 dnode_fini(); 1193 dbuf_fini(); 1194 l2arc_fini(); 1195 } 1196