Deleted Added
sdiff udiff text old ( 177698 ) new ( 185029 )
full compact
dnode.c (177698) dnode.c (185029)
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

--- 5 unchanged lines hidden (view full) ---

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/*
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

--- 5 unchanged lines hidden (view full) ---

14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21/*
22 * Copyright 2007 Sun Microsystems, Inc. All rights reserved.
22 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
23 * Use is subject to license terms.
24 */
25
26#pragma ident "%Z%%M% %I% %E% SMI"
27
28#include <sys/zfs_context.h>
29#include <sys/dbuf.h>
30#include <sys/dnode.h>
31#include <sys/dmu.h>
32#include <sys/dmu_impl.h>
33#include <sys/dmu_tx.h>
34#include <sys/dmu_objset.h>
35#include <sys/dsl_dir.h>

--- 201 unchanged lines hidden (view full) ---

237
238 if (rp1->fr_blkid < rp2->fr_blkid)
239 return (-1);
240 else if (rp1->fr_blkid > rp2->fr_blkid)
241 return (1);
242 else return (0);
243}
244
26#include <sys/zfs_context.h>
27#include <sys/dbuf.h>
28#include <sys/dnode.h>
29#include <sys/dmu.h>
30#include <sys/dmu_impl.h>
31#include <sys/dmu_tx.h>
32#include <sys/dmu_objset.h>
33#include <sys/dsl_dir.h>

--- 201 unchanged lines hidden (view full) ---

235
236 if (rp1->fr_blkid < rp2->fr_blkid)
237 return (-1);
238 else if (rp1->fr_blkid > rp2->fr_blkid)
239 return (1);
240 else return (0);
241}
242
243void
244dnode_setbonuslen(dnode_t *dn, int newsize, dmu_tx_t *tx)
245{
246 ASSERT3U(refcount_count(&dn->dn_holds), >=, 1);
247
248 dnode_setdirty(dn, tx);
249 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
250 ASSERT3U(newsize, <=, DN_MAX_BONUSLEN -
251 (dn->dn_nblkptr-1) * sizeof (blkptr_t));
252 dn->dn_bonuslen = newsize;
253 if (newsize == 0)
254 dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = DN_ZERO_BONUSLEN;
255 else
256 dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen;
257 rw_exit(&dn->dn_struct_rwlock);
258}
259
245static void
246dnode_setdblksz(dnode_t *dn, int size)
247{
248 ASSERT3U(P2PHASE(size, SPA_MINBLOCKSIZE), ==, 0);
249 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE);
250 ASSERT3U(size, >=, SPA_MINBLOCKSIZE);
251 ASSERT3U(size >> SPA_MINBLOCKSHIFT, <,
252 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8));

--- 27 unchanged lines hidden (view full) ---

280
281 dmu_zfetch_init(&dn->dn_zfetch, dn);
282
283 ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES);
284 mutex_enter(&os->os_lock);
285 list_insert_head(&os->os_dnodes, dn);
286 mutex_exit(&os->os_lock);
287
260static void
261dnode_setdblksz(dnode_t *dn, int size)
262{
263 ASSERT3U(P2PHASE(size, SPA_MINBLOCKSIZE), ==, 0);
264 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE);
265 ASSERT3U(size, >=, SPA_MINBLOCKSIZE);
266 ASSERT3U(size >> SPA_MINBLOCKSHIFT, <,
267 1<<(sizeof (dn->dn_phys->dn_datablkszsec) * 8));

--- 27 unchanged lines hidden (view full) ---

295
296 dmu_zfetch_init(&dn->dn_zfetch, dn);
297
298 ASSERT(dn->dn_phys->dn_type < DMU_OT_NUMTYPES);
299 mutex_enter(&os->os_lock);
300 list_insert_head(&os->os_dnodes, dn);
301 mutex_exit(&os->os_lock);
302
303 arc_space_consume(sizeof (dnode_t));
288 return (dn);
289}
290
291static void
292dnode_destroy(dnode_t *dn)
293{
294 objset_impl_t *os = dn->dn_objset;
295

--- 18 unchanged lines hidden (view full) ---

314 }
315 dmu_zfetch_rele(&dn->dn_zfetch);
316 if (dn->dn_bonus) {
317 mutex_enter(&dn->dn_bonus->db_mtx);
318 dbuf_evict(dn->dn_bonus);
319 dn->dn_bonus = NULL;
320 }
321 kmem_cache_free(dnode_cache, dn);
304 return (dn);
305}
306
307static void
308dnode_destroy(dnode_t *dn)
309{
310 objset_impl_t *os = dn->dn_objset;
311

--- 18 unchanged lines hidden (view full) ---

330 }
331 dmu_zfetch_rele(&dn->dn_zfetch);
332 if (dn->dn_bonus) {
333 mutex_enter(&dn->dn_bonus->db_mtx);
334 dbuf_evict(dn->dn_bonus);
335 dn->dn_bonus = NULL;
336 }
337 kmem_cache_free(dnode_cache, dn);
338 arc_space_return(sizeof (dnode_t));
322}
323
324void
325dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
326 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
327{
328 int i;
329

--- 27 unchanged lines hidden (view full) ---

357 ASSERT3U(dn->dn_assigned_txg, ==, 0);
358 ASSERT(refcount_is_zero(&dn->dn_tx_holds));
359 ASSERT3U(refcount_count(&dn->dn_holds), <=, 1);
360 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL);
361
362 for (i = 0; i < TXG_SIZE; i++) {
363 ASSERT3U(dn->dn_next_nlevels[i], ==, 0);
364 ASSERT3U(dn->dn_next_indblkshift[i], ==, 0);
339}
340
341void
342dnode_allocate(dnode_t *dn, dmu_object_type_t ot, int blocksize, int ibs,
343 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
344{
345 int i;
346

--- 27 unchanged lines hidden (view full) ---

374 ASSERT3U(dn->dn_assigned_txg, ==, 0);
375 ASSERT(refcount_is_zero(&dn->dn_tx_holds));
376 ASSERT3U(refcount_count(&dn->dn_holds), <=, 1);
377 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL);
378
379 for (i = 0; i < TXG_SIZE; i++) {
380 ASSERT3U(dn->dn_next_nlevels[i], ==, 0);
381 ASSERT3U(dn->dn_next_indblkshift[i], ==, 0);
382 ASSERT3U(dn->dn_next_bonuslen[i], ==, 0);
365 ASSERT3U(dn->dn_next_blksz[i], ==, 0);
366 ASSERT(!list_link_active(&dn->dn_dirty_link[i]));
367 ASSERT3P(list_head(&dn->dn_dirty_records[i]), ==, NULL);
368 ASSERT3U(avl_numnodes(&dn->dn_ranges[i]), ==, 0);
369 }
370
371 dn->dn_type = ot;
372 dnode_setdblksz(dn, blocksize);

--- 11 unchanged lines hidden (view full) ---

384 kmem_free(dn->dn_dirtyctx_firstset, 1);
385 dn->dn_dirtyctx_firstset = NULL;
386 }
387
388 dn->dn_allocated_txg = tx->tx_txg;
389
390 dnode_setdirty(dn, tx);
391 dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs;
383 ASSERT3U(dn->dn_next_blksz[i], ==, 0);
384 ASSERT(!list_link_active(&dn->dn_dirty_link[i]));
385 ASSERT3P(list_head(&dn->dn_dirty_records[i]), ==, NULL);
386 ASSERT3U(avl_numnodes(&dn->dn_ranges[i]), ==, 0);
387 }
388
389 dn->dn_type = ot;
390 dnode_setdblksz(dn, blocksize);

--- 11 unchanged lines hidden (view full) ---

402 kmem_free(dn->dn_dirtyctx_firstset, 1);
403 dn->dn_dirtyctx_firstset = NULL;
404 }
405
406 dn->dn_allocated_txg = tx->tx_txg;
407
408 dnode_setdirty(dn, tx);
409 dn->dn_next_indblkshift[tx->tx_txg & TXG_MASK] = ibs;
410 dn->dn_next_bonuslen[tx->tx_txg & TXG_MASK] = dn->dn_bonuslen;
392 dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz;
393}
394
395void
396dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
397 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
398{
411 dn->dn_next_blksz[tx->tx_txg & TXG_MASK] = dn->dn_datablksz;
412}
413
414void
415dnode_reallocate(dnode_t *dn, dmu_object_type_t ot, int blocksize,
416 dmu_object_type_t bonustype, int bonuslen, dmu_tx_t *tx)
417{
399 int i;
418 int i, old_nblkptr;
400 dmu_buf_impl_t *db = NULL;
401
402 ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE);
403 ASSERT3U(blocksize, <=, SPA_MAXBLOCKSIZE);
404 ASSERT3U(blocksize % SPA_MINBLOCKSIZE, ==, 0);
405 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx));
406 ASSERT(tx->tx_txg != 0);
407 ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) ||
408 (bonustype != DMU_OT_NONE && bonuslen != 0));
409 ASSERT3U(bonustype, <, DMU_OT_NUMTYPES);
410 ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN);
411
412 for (i = 0; i < TXG_SIZE; i++)
413 ASSERT(!list_link_active(&dn->dn_dirty_link[i]));
414
415 /* clean up any unreferenced dbufs */
419 dmu_buf_impl_t *db = NULL;
420
421 ASSERT3U(blocksize, >=, SPA_MINBLOCKSIZE);
422 ASSERT3U(blocksize, <=, SPA_MAXBLOCKSIZE);
423 ASSERT3U(blocksize % SPA_MINBLOCKSIZE, ==, 0);
424 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT || dmu_tx_private_ok(tx));
425 ASSERT(tx->tx_txg != 0);
426 ASSERT((bonustype == DMU_OT_NONE && bonuslen == 0) ||
427 (bonustype != DMU_OT_NONE && bonuslen != 0));
428 ASSERT3U(bonustype, <, DMU_OT_NUMTYPES);
429 ASSERT3U(bonuslen, <=, DN_MAX_BONUSLEN);
430
431 for (i = 0; i < TXG_SIZE; i++)
432 ASSERT(!list_link_active(&dn->dn_dirty_link[i]));
433
434 /* clean up any unreferenced dbufs */
416 (void) dnode_evict_dbufs(dn, 0);
435 dnode_evict_dbufs(dn);
417 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL);
418
419 /*
420 * XXX I should really have a generation number to tell if we
421 * need to do this...
422 */
423 if (blocksize != dn->dn_datablksz ||
424 dn->dn_bonustype != bonustype || dn->dn_bonuslen != bonuslen) {

--- 6 unchanged lines hidden (view full) ---

431 if (blocksize != dn->dn_datablksz &&
432 (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) ||
433 list_head(&dn->dn_dbufs) != NULL)) {
434 db = dbuf_hold(dn, 0, FTAG);
435 dbuf_new_size(db, blocksize, tx);
436 }
437 dnode_setdblksz(dn, blocksize);
438 dnode_setdirty(dn, tx);
436 ASSERT3P(list_head(&dn->dn_dbufs), ==, NULL);
437
438 /*
439 * XXX I should really have a generation number to tell if we
440 * need to do this...
441 */
442 if (blocksize != dn->dn_datablksz ||
443 dn->dn_bonustype != bonustype || dn->dn_bonuslen != bonuslen) {

--- 6 unchanged lines hidden (view full) ---

450 if (blocksize != dn->dn_datablksz &&
451 (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) ||
452 list_head(&dn->dn_dbufs) != NULL)) {
453 db = dbuf_hold(dn, 0, FTAG);
454 dbuf_new_size(db, blocksize, tx);
455 }
456 dnode_setdblksz(dn, blocksize);
457 dnode_setdirty(dn, tx);
458 dn->dn_next_bonuslen[tx->tx_txg&TXG_MASK] = bonuslen;
439 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize;
440 rw_exit(&dn->dn_struct_rwlock);
459 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = blocksize;
460 rw_exit(&dn->dn_struct_rwlock);
441 if (db) {
461 if (db)
442 dbuf_rele(db, FTAG);
462 dbuf_rele(db, FTAG);
443 db = NULL;
444 }
445
446 /* change type */
447 dn->dn_type = ot;
448
463
464 /* change type */
465 dn->dn_type = ot;
466
449 if (dn->dn_bonuslen != bonuslen) {
450 /* change bonus size */
451 if (bonuslen == 0)
452 bonuslen = 1; /* XXX */
453 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
454 if (dn->dn_bonus == NULL)
455 dn->dn_bonus = dbuf_create_bonus(dn);
456 db = dn->dn_bonus;
457 rw_exit(&dn->dn_struct_rwlock);
458 if (refcount_add(&db->db_holds, FTAG) == 1)
459 dnode_add_ref(dn, db);
460 VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED));
461 mutex_enter(&db->db_mtx);
462 ASSERT3U(db->db.db_size, ==, dn->dn_bonuslen);
463 ASSERT(db->db.db_data != NULL);
464 db->db.db_size = bonuslen;
465 mutex_exit(&db->db_mtx);
466 (void) dbuf_dirty(db, tx);
467 }
468
469 /* change bonus size and type */
470 mutex_enter(&dn->dn_mtx);
467 /* change bonus size and type */
468 mutex_enter(&dn->dn_mtx);
469 old_nblkptr = dn->dn_nblkptr;
471 dn->dn_bonustype = bonustype;
472 dn->dn_bonuslen = bonuslen;
473 dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT);
474 dn->dn_checksum = ZIO_CHECKSUM_INHERIT;
475 dn->dn_compress = ZIO_COMPRESS_INHERIT;
476 ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR);
477
470 dn->dn_bonustype = bonustype;
471 dn->dn_bonuslen = bonuslen;
472 dn->dn_nblkptr = 1 + ((DN_MAX_BONUSLEN - bonuslen) >> SPA_BLKPTRSHIFT);
473 dn->dn_checksum = ZIO_CHECKSUM_INHERIT;
474 dn->dn_compress = ZIO_COMPRESS_INHERIT;
475 ASSERT3U(dn->dn_nblkptr, <=, DN_MAX_NBLKPTR);
476
478 /*
479 * NB: we have to do the dbuf_rele after we've changed the
480 * dn_bonuslen, for the sake of dbuf_verify().
481 */
482 if (db)
483 dbuf_rele(db, FTAG);
477 /* XXX - for now, we can't make nblkptr smaller */
478 ASSERT3U(dn->dn_nblkptr, >=, old_nblkptr);
484
479
480 /* fix up the bonus db_size if dn_nblkptr has changed */
481 if (dn->dn_bonus && dn->dn_bonuslen != old_nblkptr) {
482 dn->dn_bonus->db.db_size =
483 DN_MAX_BONUSLEN - (dn->dn_nblkptr-1) * sizeof (blkptr_t);
484 ASSERT(dn->dn_bonuslen <= dn->dn_bonus->db.db_size);
485 }
486
485 dn->dn_allocated_txg = tx->tx_txg;
486 mutex_exit(&dn->dn_mtx);
487}
488
489void
490dnode_special_close(dnode_t *dn)
491{
492 /*

--- 61 unchanged lines hidden (view full) ---

554 int epb, idx, err;
555 int drop_struct_lock = FALSE;
556 int type;
557 uint64_t blk;
558 dnode_t *mdn, *dn;
559 dmu_buf_impl_t *db;
560 dnode_t **children_dnodes;
561
487 dn->dn_allocated_txg = tx->tx_txg;
488 mutex_exit(&dn->dn_mtx);
489}
490
491void
492dnode_special_close(dnode_t *dn)
493{
494 /*

--- 61 unchanged lines hidden (view full) ---

556 int epb, idx, err;
557 int drop_struct_lock = FALSE;
558 int type;
559 uint64_t blk;
560 dnode_t *mdn, *dn;
561 dmu_buf_impl_t *db;
562 dnode_t **children_dnodes;
563
564 /*
565 * If you are holding the spa config lock as writer, you shouldn't
566 * be asking the DMU to do *anything*.
567 */
568 ASSERT(spa_config_held(os->os_spa, SCL_ALL, RW_WRITER) == 0);
569
562 if (object == 0 || object >= DN_MAX_OBJECT)
563 return (EINVAL);
564
565 mdn = os->os_meta_dnode;
566
567 DNODE_VERIFY(mdn);
568
569 if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) {

--- 27 unchanged lines hidden (view full) ---

597 if (winner = dmu_buf_set_user(&db->db, children_dnodes, NULL,
598 dnode_buf_pageout)) {
599 kmem_free(children_dnodes, epb * sizeof (dnode_t *));
600 children_dnodes = winner;
601 }
602 }
603
604 if ((dn = children_dnodes[idx]) == NULL) {
570 if (object == 0 || object >= DN_MAX_OBJECT)
571 return (EINVAL);
572
573 mdn = os->os_meta_dnode;
574
575 DNODE_VERIFY(mdn);
576
577 if (!RW_WRITE_HELD(&mdn->dn_struct_rwlock)) {

--- 27 unchanged lines hidden (view full) ---

605 if (winner = dmu_buf_set_user(&db->db, children_dnodes, NULL,
606 dnode_buf_pageout)) {
607 kmem_free(children_dnodes, epb * sizeof (dnode_t *));
608 children_dnodes = winner;
609 }
610 }
611
612 if ((dn = children_dnodes[idx]) == NULL) {
613 dnode_phys_t *dnp = (dnode_phys_t *)db->db.db_data+idx;
605 dnode_t *winner;
614 dnode_t *winner;
606 dn = dnode_create(os, (dnode_phys_t *)db->db.db_data+idx,
607 db, object);
615
616 dn = dnode_create(os, dnp, db, object);
608 winner = atomic_cas_ptr(&children_dnodes[idx], NULL, dn);
609 if (winner != NULL) {
610 dnode_destroy(dn);
611 dn = winner;
612 }
613 }
614
615 mutex_enter(&dn->dn_mtx);

--- 23 unchanged lines hidden (view full) ---

639 * Return held dnode if the object is allocated, NULL if not.
640 */
641int
642dnode_hold(objset_impl_t *os, uint64_t object, void *tag, dnode_t **dnp)
643{
644 return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp));
645}
646
617 winner = atomic_cas_ptr(&children_dnodes[idx], NULL, dn);
618 if (winner != NULL) {
619 dnode_destroy(dn);
620 dn = winner;
621 }
622 }
623
624 mutex_enter(&dn->dn_mtx);

--- 23 unchanged lines hidden (view full) ---

648 * Return held dnode if the object is allocated, NULL if not.
649 */
650int
651dnode_hold(objset_impl_t *os, uint64_t object, void *tag, dnode_t **dnp)
652{
653 return (dnode_hold_impl(os, object, DNODE_MUST_BE_ALLOCATED, tag, dnp));
654}
655
647void
656/*
657 * Can only add a reference if there is already at least one
658 * reference on the dnode. Returns FALSE if unable to add a
659 * new reference.
660 */
661boolean_t
648dnode_add_ref(dnode_t *dn, void *tag)
649{
662dnode_add_ref(dnode_t *dn, void *tag)
663{
650 ASSERT(refcount_count(&dn->dn_holds) > 0);
651 (void) refcount_add(&dn->dn_holds, tag);
664 mutex_enter(&dn->dn_mtx);
665 if (refcount_is_zero(&dn->dn_holds)) {
666 mutex_exit(&dn->dn_mtx);
667 return (FALSE);
668 }
669 VERIFY(1 < refcount_add(&dn->dn_holds, tag));
670 mutex_exit(&dn->dn_mtx);
671 return (TRUE);
652}
653
654void
655dnode_rele(dnode_t *dn, void *tag)
656{
657 uint64_t refs;
658
672}
673
674void
675dnode_rele(dnode_t *dn, void *tag)
676{
677 uint64_t refs;
678
679 mutex_enter(&dn->dn_mtx);
659 refs = refcount_remove(&dn->dn_holds, tag);
680 refs = refcount_remove(&dn->dn_holds, tag);
681 mutex_exit(&dn->dn_mtx);
660 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */
661 if (refs == 0 && dn->dn_dbuf)
662 dbuf_rele(dn->dn_dbuf, dn);
663}
664
665void
666dnode_setdirty(dnode_t *dn, dmu_tx_t *tx)
667{

--- 19 unchanged lines hidden (view full) ---

687 */
688 if (list_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) {
689 mutex_exit(&os->os_lock);
690 return;
691 }
692
693 ASSERT(!refcount_is_zero(&dn->dn_holds) || list_head(&dn->dn_dbufs));
694 ASSERT(dn->dn_datablksz != 0);
682 /* NOTE: the DNODE_DNODE does not have a dn_dbuf */
683 if (refs == 0 && dn->dn_dbuf)
684 dbuf_rele(dn->dn_dbuf, dn);
685}
686
687void
688dnode_setdirty(dnode_t *dn, dmu_tx_t *tx)
689{

--- 19 unchanged lines hidden (view full) ---

709 */
710 if (list_link_active(&dn->dn_dirty_link[txg & TXG_MASK])) {
711 mutex_exit(&os->os_lock);
712 return;
713 }
714
715 ASSERT(!refcount_is_zero(&dn->dn_holds) || list_head(&dn->dn_dbufs));
716 ASSERT(dn->dn_datablksz != 0);
717 ASSERT3U(dn->dn_next_bonuslen[txg&TXG_MASK], ==, 0);
695 ASSERT3U(dn->dn_next_blksz[txg&TXG_MASK], ==, 0);
696
697 dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n",
698 dn->dn_object, txg);
699
700 if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) {
701 list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn);
702 } else {

--- 6 unchanged lines hidden (view full) ---

709 * The dnode maintains a hold on its containing dbuf as
710 * long as there are holds on it. Each instantiated child
711 * dbuf maintaines a hold on the dnode. When the last child
712 * drops its hold, the dnode will drop its hold on the
713 * containing dbuf. We add a "dirty hold" here so that the
714 * dnode will hang around after we finish processing its
715 * children.
716 */
718 ASSERT3U(dn->dn_next_blksz[txg&TXG_MASK], ==, 0);
719
720 dprintf_ds(os->os_dsl_dataset, "obj=%llu txg=%llu\n",
721 dn->dn_object, txg);
722
723 if (dn->dn_free_txg > 0 && dn->dn_free_txg <= txg) {
724 list_insert_tail(&os->os_free_dnodes[txg&TXG_MASK], dn);
725 } else {

--- 6 unchanged lines hidden (view full) ---

732 * The dnode maintains a hold on its containing dbuf as
733 * long as there are holds on it. Each instantiated child
734 * dbuf maintaines a hold on the dnode. When the last child
735 * drops its hold, the dnode will drop its hold on the
736 * containing dbuf. We add a "dirty hold" here so that the
737 * dnode will hang around after we finish processing its
738 * children.
739 */
717 dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg);
740 VERIFY(dnode_add_ref(dn, (void *)(uintptr_t)tx->tx_txg));
718
719 (void) dbuf_dirty(dn->dn_dbuf, tx);
720
721 dsl_dataset_dirty(os->os_dsl_dataset, tx);
722}
723
724void
725dnode_free(dnode_t *dn, dmu_tx_t *tx)

--- 31 unchanged lines hidden (view full) ---

757/*
758 * Try to change the block size for the indicated dnode. This can only
759 * succeed if there are no blocks allocated or dirty beyond first block
760 */
761int
762dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx)
763{
764 dmu_buf_impl_t *db, *db_next;
741
742 (void) dbuf_dirty(dn->dn_dbuf, tx);
743
744 dsl_dataset_dirty(os->os_dsl_dataset, tx);
745}
746
747void
748dnode_free(dnode_t *dn, dmu_tx_t *tx)

--- 31 unchanged lines hidden (view full) ---

780/*
781 * Try to change the block size for the indicated dnode. This can only
782 * succeed if there are no blocks allocated or dirty beyond first block
783 */
784int
785dnode_set_blksz(dnode_t *dn, uint64_t size, int ibs, dmu_tx_t *tx)
786{
787 dmu_buf_impl_t *db, *db_next;
765 int have_db0 = FALSE;
788 int err;
766
767 if (size == 0)
768 size = SPA_MINBLOCKSIZE;
769 if (size > SPA_MAXBLOCKSIZE)
770 size = SPA_MAXBLOCKSIZE;
771 else
772 size = P2ROUNDUP(size, SPA_MINBLOCKSIZE);
773

--- 8 unchanged lines hidden (view full) ---

782 /* Check for any allocated blocks beyond the first */
783 if (dn->dn_phys->dn_maxblkid != 0)
784 goto fail;
785
786 mutex_enter(&dn->dn_dbufs_mtx);
787 for (db = list_head(&dn->dn_dbufs); db; db = db_next) {
788 db_next = list_next(&dn->dn_dbufs, db);
789
789
790 if (size == 0)
791 size = SPA_MINBLOCKSIZE;
792 if (size > SPA_MAXBLOCKSIZE)
793 size = SPA_MAXBLOCKSIZE;
794 else
795 size = P2ROUNDUP(size, SPA_MINBLOCKSIZE);
796

--- 8 unchanged lines hidden (view full) ---

805 /* Check for any allocated blocks beyond the first */
806 if (dn->dn_phys->dn_maxblkid != 0)
807 goto fail;
808
809 mutex_enter(&dn->dn_dbufs_mtx);
810 for (db = list_head(&dn->dn_dbufs); db; db = db_next) {
811 db_next = list_next(&dn->dn_dbufs, db);
812
790 if (db->db_blkid == 0) {
791 have_db0 = TRUE;
792 } else if (db->db_blkid != DB_BONUS_BLKID) {
813 if (db->db_blkid != 0 && db->db_blkid != DB_BONUS_BLKID) {
793 mutex_exit(&dn->dn_dbufs_mtx);
794 goto fail;
795 }
796 }
797 mutex_exit(&dn->dn_dbufs_mtx);
798
799 if (ibs && dn->dn_nlevels != 1)
800 goto fail;
801
814 mutex_exit(&dn->dn_dbufs_mtx);
815 goto fail;
816 }
817 }
818 mutex_exit(&dn->dn_dbufs_mtx);
819
820 if (ibs && dn->dn_nlevels != 1)
821 goto fail;
822
802 db = NULL;
803 if (!BP_IS_HOLE(&dn->dn_phys->dn_blkptr[0]) || have_db0) {
804 /* obtain the old block */
805 db = dbuf_hold(dn, 0, FTAG);
823 /* resize the old block */
824 err = dbuf_hold_impl(dn, 0, 0, TRUE, FTAG, &db);
825 if (err == 0)
806 dbuf_new_size(db, size, tx);
826 dbuf_new_size(db, size, tx);
807 }
827 else if (err != ENOENT)
828 goto fail;
808
809 dnode_setdblksz(dn, size);
810 dnode_setdirty(dn, tx);
811 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size;
812 if (ibs) {
813 dn->dn_indblkshift = ibs;
814 dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs;
815 }
829
830 dnode_setdblksz(dn, size);
831 dnode_setdirty(dn, tx);
832 dn->dn_next_blksz[tx->tx_txg&TXG_MASK] = size;
833 if (ibs) {
834 dn->dn_indblkshift = ibs;
835 dn->dn_next_indblkshift[tx->tx_txg&TXG_MASK] = ibs;
836 }
816
837 /* rele after we have fixed the blocksize in the dnode */
817 if (db)
818 dbuf_rele(db, FTAG);
819
820 rw_exit(&dn->dn_struct_rwlock);
821 return (0);
822
823fail:
824 rw_exit(&dn->dn_struct_rwlock);
825 return (ENOTSUP);
826}
827
838 if (db)
839 dbuf_rele(db, FTAG);
840
841 rw_exit(&dn->dn_struct_rwlock);
842 return (0);
843
844fail:
845 rw_exit(&dn->dn_struct_rwlock);
846 return (ENOTSUP);
847}
848
849/* read-holding callers must not rely on the lock being continuously held */
828void
850void
829dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx)
851dnode_new_blkid(dnode_t *dn, uint64_t blkid, dmu_tx_t *tx, boolean_t have_read)
830{
831 uint64_t txgoff = tx->tx_txg & TXG_MASK;
852{
853 uint64_t txgoff = tx->tx_txg & TXG_MASK;
832 int drop_struct_lock = FALSE;
833 int epbs, new_nlevels;
834 uint64_t sz;
835
836 ASSERT(blkid != DB_BONUS_BLKID);
837
854 int epbs, new_nlevels;
855 uint64_t sz;
856
857 ASSERT(blkid != DB_BONUS_BLKID);
858
838 if (!RW_WRITE_HELD(&dn->dn_struct_rwlock)) {
839 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
840 drop_struct_lock = TRUE;
859 ASSERT(have_read ?
860 RW_READ_HELD(&dn->dn_struct_rwlock) :
861 RW_WRITE_HELD(&dn->dn_struct_rwlock));
862
863 /*
864 * if we have a read-lock, check to see if we need to do any work
865 * before upgrading to a write-lock.
866 */
867 if (have_read) {
868 if (blkid <= dn->dn_maxblkid)
869 return;
870
871 if (!rw_tryupgrade(&dn->dn_struct_rwlock)) {
872 rw_exit(&dn->dn_struct_rwlock);
873 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
874 }
841 }
842
843 if (blkid <= dn->dn_maxblkid)
844 goto out;
845
846 dn->dn_maxblkid = blkid;
847
848 /*

--- 35 unchanged lines hidden (view full) ---

884 dr->dr_parent = new;
885 }
886 }
887 mutex_exit(&new->dt.di.dr_mtx);
888 mutex_exit(&dn->dn_mtx);
889 }
890
891out:
875 }
876
877 if (blkid <= dn->dn_maxblkid)
878 goto out;
879
880 dn->dn_maxblkid = blkid;
881
882 /*

--- 35 unchanged lines hidden (view full) ---

918 dr->dr_parent = new;
919 }
920 }
921 mutex_exit(&new->dt.di.dr_mtx);
922 mutex_exit(&dn->dn_mtx);
923 }
924
925out:
892 if (drop_struct_lock)
893 rw_exit(&dn->dn_struct_rwlock);
926 if (have_read)
927 rw_downgrade(&dn->dn_struct_rwlock);
894}
895
896void
897dnode_clear_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx)
898{
899 avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK];
900 avl_index_t where;
901 free_range_t *rp;

--- 44 unchanged lines hidden (view full) ---

946 }
947}
948
949void
950dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx)
951{
952 dmu_buf_impl_t *db;
953 uint64_t blkoff, blkid, nblks;
928}
929
930void
931dnode_clear_range(dnode_t *dn, uint64_t blkid, uint64_t nblks, dmu_tx_t *tx)
932{
933 avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK];
934 avl_index_t where;
935 free_range_t *rp;

--- 44 unchanged lines hidden (view full) ---

980 }
981}
982
983void
984dnode_free_range(dnode_t *dn, uint64_t off, uint64_t len, dmu_tx_t *tx)
985{
986 dmu_buf_impl_t *db;
987 uint64_t blkoff, blkid, nblks;
954 int blksz, head;
988 int blksz, blkshift, head, tail;
955 int trunc = FALSE;
989 int trunc = FALSE;
990 int epbs;
956
957 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
958 blksz = dn->dn_datablksz;
991
992 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
993 blksz = dn->dn_datablksz;
994 blkshift = dn->dn_datablkshift;
995 epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
959
996
960 /* If the range is past the end of the file, this is a no-op */
961 if (off >= blksz * (dn->dn_maxblkid+1))
962 goto out;
963 if (len == -1ULL) {
964 len = UINT64_MAX - off;
965 trunc = TRUE;
966 }
967
968 /*
969 * First, block align the region to free:
970 */
971 if (ISP2(blksz)) {
972 head = P2NPHASE(off, blksz);
973 blkoff = P2PHASE(off, blksz);
997 if (len == -1ULL) {
998 len = UINT64_MAX - off;
999 trunc = TRUE;
1000 }
1001
1002 /*
1003 * First, block align the region to free:
1004 */
1005 if (ISP2(blksz)) {
1006 head = P2NPHASE(off, blksz);
1007 blkoff = P2PHASE(off, blksz);
1008 if ((off >> blkshift) > dn->dn_maxblkid)
1009 goto out;
974 } else {
975 ASSERT(dn->dn_maxblkid == 0);
976 if (off == 0 && len >= blksz) {
1010 } else {
1011 ASSERT(dn->dn_maxblkid == 0);
1012 if (off == 0 && len >= blksz) {
977 /* Freeing the whole block; don't do any head. */
978 head = 0;
1013 /* Freeing the whole block; fast-track this request */
1014 blkid = 0;
1015 nblks = 1;
1016 goto done;
1017 } else if (off >= blksz) {
1018 /* Freeing past end-of-data */
1019 goto out;
979 } else {
980 /* Freeing part of the block. */
981 head = blksz - off;
982 ASSERT3U(head, >, 0);
983 }
984 blkoff = off;
985 }
986 /* zero out any partial block data at the start of the range */

--- 16 unchanged lines hidden (view full) ---

1003 }
1004 dbuf_rele(db, FTAG);
1005 }
1006 off += head;
1007 len -= head;
1008 }
1009
1010 /* If the range was less than one block, we're done */
1020 } else {
1021 /* Freeing part of the block. */
1022 head = blksz - off;
1023 ASSERT3U(head, >, 0);
1024 }
1025 blkoff = off;
1026 }
1027 /* zero out any partial block data at the start of the range */

--- 16 unchanged lines hidden (view full) ---

1044 }
1045 dbuf_rele(db, FTAG);
1046 }
1047 off += head;
1048 len -= head;
1049 }
1050
1051 /* If the range was less than one block, we're done */
1011 if (len == 0 || off >= blksz * (dn->dn_maxblkid+1))
1052 if (len == 0)
1012 goto out;
1013
1053 goto out;
1054
1014 if (!ISP2(blksz)) {
1015 /*
1016 * They are freeing the whole block of a
1017 * non-power-of-two blocksize file. Skip all the messy
1018 * math.
1019 */
1020 ASSERT3U(off, ==, 0);
1021 ASSERT3U(len, >=, blksz);
1022 blkid = 0;
1023 nblks = 1;
1024 } else {
1025 int tail;
1026 int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
1027 int blkshift = dn->dn_datablkshift;
1055 /* If the remaining range is past end of file, we're done */
1056 if ((off >> blkshift) > dn->dn_maxblkid)
1057 goto out;
1028
1058
1029 /* If the remaining range is past end of file, we're done */
1030 if (off > dn->dn_maxblkid << blkshift)
1031 goto out;
1059 ASSERT(ISP2(blksz));
1060 if (trunc)
1061 tail = 0;
1062 else
1063 tail = P2PHASE(len, blksz);
1032
1064
1033 if (off + len == UINT64_MAX)
1034 tail = 0;
1035 else
1036 tail = P2PHASE(len, blksz);
1037
1038 ASSERT3U(P2PHASE(off, blksz), ==, 0);
1039 /* zero out any partial block data at the end of the range */
1040 if (tail) {
1041 if (len < tail)
1042 tail = len;
1043 if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off+len),
1044 TRUE, FTAG, &db) == 0) {
1045 /* don't dirty if not on disk and not dirty */
1046 if (db->db_last_dirty ||
1047 (db->db_blkptr &&
1048 !BP_IS_HOLE(db->db_blkptr))) {
1049 rw_exit(&dn->dn_struct_rwlock);
1050 dbuf_will_dirty(db, tx);
1051 rw_enter(&dn->dn_struct_rwlock,
1052 RW_WRITER);
1053 bzero(db->db.db_data, tail);
1054 }
1055 dbuf_rele(db, FTAG);
1065 ASSERT3U(P2PHASE(off, blksz), ==, 0);
1066 /* zero out any partial block data at the end of the range */
1067 if (tail) {
1068 if (len < tail)
1069 tail = len;
1070 if (dbuf_hold_impl(dn, 0, dbuf_whichblock(dn, off+len),
1071 TRUE, FTAG, &db) == 0) {
1072 /* don't dirty if not on disk and not dirty */
1073 if (db->db_last_dirty ||
1074 (db->db_blkptr && !BP_IS_HOLE(db->db_blkptr))) {
1075 rw_exit(&dn->dn_struct_rwlock);
1076 dbuf_will_dirty(db, tx);
1077 rw_enter(&dn->dn_struct_rwlock, RW_WRITER);
1078 bzero(db->db.db_data, tail);
1056 }
1079 }
1057 len -= tail;
1080 dbuf_rele(db, FTAG);
1058 }
1081 }
1059 /* If the range did not include a full block, we are done */
1060 if (len == 0)
1061 goto out;
1082 len -= tail;
1083 }
1062
1084
1063 /* dirty the left indirects */
1064 if (dn->dn_nlevels > 1 && off != 0) {
1065 db = dbuf_hold_level(dn, 1,
1066 (off - head) >> (blkshift + epbs), FTAG);
1085 /* If the range did not include a full block, we are done */
1086 if (len == 0)
1087 goto out;
1088
1089 ASSERT(IS_P2ALIGNED(off, blksz));
1090 ASSERT(trunc || IS_P2ALIGNED(len, blksz));
1091 blkid = off >> blkshift;
1092 nblks = len >> blkshift;
1093 if (trunc)
1094 nblks += 1;
1095
1096 /*
1097 * Read in and mark all the level-1 indirects dirty,
1098 * so that they will stay in memory until syncing phase.
1099 * Always dirty the first and last indirect to make sure
1100 * we dirty all the partial indirects.
1101 */
1102 if (dn->dn_nlevels > 1) {
1103 uint64_t i, first, last;
1104 int shift = epbs + dn->dn_datablkshift;
1105
1106 first = blkid >> epbs;
1107 if (db = dbuf_hold_level(dn, 1, first, FTAG)) {
1067 dbuf_will_dirty(db, tx);
1068 dbuf_rele(db, FTAG);
1069 }
1108 dbuf_will_dirty(db, tx);
1109 dbuf_rele(db, FTAG);
1110 }
1070
1071 /* dirty the right indirects */
1072 if (dn->dn_nlevels > 1 && !trunc) {
1073 db = dbuf_hold_level(dn, 1,
1074 (off + len + tail - 1) >> (blkshift + epbs), FTAG);
1111 if (trunc)
1112 last = dn->dn_maxblkid >> epbs;
1113 else
1114 last = (blkid + nblks - 1) >> epbs;
1115 if (last > first && (db = dbuf_hold_level(dn, 1, last, FTAG))) {
1075 dbuf_will_dirty(db, tx);
1076 dbuf_rele(db, FTAG);
1077 }
1116 dbuf_will_dirty(db, tx);
1117 dbuf_rele(db, FTAG);
1118 }
1119 for (i = first + 1; i < last; i++) {
1120 uint64_t ibyte = i << shift;
1121 int err;
1078
1122
1079 /*
1080 * Finally, add this range to the dnode range list, we
1081 * will finish up this free operation in the syncing phase.
1082 */
1083 ASSERT(IS_P2ALIGNED(off, 1<<blkshift));
1084 ASSERT(off + len == UINT64_MAX ||
1085 IS_P2ALIGNED(len, 1<<blkshift));
1086 blkid = off >> blkshift;
1087 nblks = len >> blkshift;
1088
1089 if (trunc)
1090 dn->dn_maxblkid = (blkid ? blkid - 1 : 0);
1123 err = dnode_next_offset(dn,
1124 DNODE_FIND_HAVELOCK, &ibyte, 1, 1, 0);
1125 i = ibyte >> shift;
1126 if (err == ESRCH || i >= last)
1127 break;
1128 ASSERT(err == 0);
1129 db = dbuf_hold_level(dn, 1, i, FTAG);
1130 if (db) {
1131 dbuf_will_dirty(db, tx);
1132 dbuf_rele(db, FTAG);
1133 }
1134 }
1091 }
1135 }
1092
1136done:
1137 /*
1138 * Add this range to the dnode range list.
1139 * We will finish up this free operation in the syncing phase.
1140 */
1093 mutex_enter(&dn->dn_mtx);
1094 dnode_clear_range(dn, blkid, nblks, tx);
1095 {
1096 free_range_t *rp, *found;
1097 avl_index_t where;
1098 avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK];
1099
1100 /* Add new range to dn_ranges */
1101 rp = kmem_alloc(sizeof (free_range_t), KM_SLEEP);
1102 rp->fr_blkid = blkid;
1103 rp->fr_nblks = nblks;
1104 found = avl_find(tree, rp, &where);
1105 ASSERT(found == NULL);
1106 avl_insert(tree, rp, where);
1107 dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n",
1108 blkid, nblks, tx->tx_txg);
1109 }
1110 mutex_exit(&dn->dn_mtx);
1111
1141 mutex_enter(&dn->dn_mtx);
1142 dnode_clear_range(dn, blkid, nblks, tx);
1143 {
1144 free_range_t *rp, *found;
1145 avl_index_t where;
1146 avl_tree_t *tree = &dn->dn_ranges[tx->tx_txg&TXG_MASK];
1147
1148 /* Add new range to dn_ranges */
1149 rp = kmem_alloc(sizeof (free_range_t), KM_SLEEP);
1150 rp->fr_blkid = blkid;
1151 rp->fr_nblks = nblks;
1152 found = avl_find(tree, rp, &where);
1153 ASSERT(found == NULL);
1154 avl_insert(tree, rp, where);
1155 dprintf_dnode(dn, "blkid=%llu nblks=%llu txg=%llu\n",
1156 blkid, nblks, tx->tx_txg);
1157 }
1158 mutex_exit(&dn->dn_mtx);
1159
1112 dbuf_free_range(dn, blkid, nblks, tx);
1160 dbuf_free_range(dn, blkid, blkid + nblks - 1, tx);
1113 dnode_setdirty(dn, tx);
1114out:
1161 dnode_setdirty(dn, tx);
1162out:
1163 if (trunc && dn->dn_maxblkid >= (off >> blkshift))
1164 dn->dn_maxblkid = (off >> blkshift ? (off >> blkshift) - 1 : 0);
1165
1115 rw_exit(&dn->dn_struct_rwlock);
1116}
1117
1118/* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */
1119uint64_t
1120dnode_block_freed(dnode_t *dn, uint64_t blkid)
1121{
1122 free_range_t range_tofind;

--- 51 unchanged lines hidden (view full) ---

1174 mutex_enter(&dn->dn_mtx);
1175 space = DN_USED_BYTES(dn->dn_phys);
1176 if (delta > 0) {
1177 ASSERT3U(space + delta, >=, space); /* no overflow */
1178 } else {
1179 ASSERT3U(space, >=, -delta); /* no underflow */
1180 }
1181 space += delta;
1166 rw_exit(&dn->dn_struct_rwlock);
1167}
1168
1169/* return TRUE if this blkid was freed in a recent txg, or FALSE if it wasn't */
1170uint64_t
1171dnode_block_freed(dnode_t *dn, uint64_t blkid)
1172{
1173 free_range_t range_tofind;

--- 51 unchanged lines hidden (view full) ---

1225 mutex_enter(&dn->dn_mtx);
1226 space = DN_USED_BYTES(dn->dn_phys);
1227 if (delta > 0) {
1228 ASSERT3U(space + delta, >=, space); /* no overflow */
1229 } else {
1230 ASSERT3U(space, >=, -delta); /* no underflow */
1231 }
1232 space += delta;
1182 if (spa_version(dn->dn_objset->os_spa) < ZFS_VERSION_DNODE_BYTES) {
1233 if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_DNODE_BYTES) {
1183 ASSERT((dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) == 0);
1184 ASSERT3U(P2PHASE(space, 1<<DEV_BSHIFT), ==, 0);
1185 dn->dn_phys->dn_used = space >> DEV_BSHIFT;
1186 } else {
1187 dn->dn_phys->dn_used = space;
1188 dn->dn_phys->dn_flags |= DNODE_FLAG_USED_BYTES;
1189 }
1190 mutex_exit(&dn->dn_mtx);

--- 15 unchanged lines hidden (view full) ---

1206
1207 if (ds)
1208 dsl_dir_willuse_space(ds->ds_dir, space, tx);
1209
1210 dmu_tx_willuse_space(tx, space);
1211}
1212
1213static int
1234 ASSERT((dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) == 0);
1235 ASSERT3U(P2PHASE(space, 1<<DEV_BSHIFT), ==, 0);
1236 dn->dn_phys->dn_used = space >> DEV_BSHIFT;
1237 } else {
1238 dn->dn_phys->dn_used = space;
1239 dn->dn_phys->dn_flags |= DNODE_FLAG_USED_BYTES;
1240 }
1241 mutex_exit(&dn->dn_mtx);

--- 15 unchanged lines hidden (view full) ---

1257
1258 if (ds)
1259 dsl_dir_willuse_space(ds->ds_dir, space, tx);
1260
1261 dmu_tx_willuse_space(tx, space);
1262}
1263
1264static int
1214dnode_next_offset_level(dnode_t *dn, boolean_t hole, uint64_t *offset,
1265dnode_next_offset_level(dnode_t *dn, int flags, uint64_t *offset,
1215 int lvl, uint64_t blkfill, uint64_t txg)
1216{
1217 dmu_buf_impl_t *db = NULL;
1218 void *data = NULL;
1219 uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT;
1220 uint64_t epb = 1ULL << epbs;
1221 uint64_t minfill, maxfill;
1266 int lvl, uint64_t blkfill, uint64_t txg)
1267{
1268 dmu_buf_impl_t *db = NULL;
1269 void *data = NULL;
1270 uint64_t epbs = dn->dn_phys->dn_indblkshift - SPA_BLKPTRSHIFT;
1271 uint64_t epb = 1ULL << epbs;
1272 uint64_t minfill, maxfill;
1222 int i, error, span;
1273 boolean_t hole;
1274 int i, inc, error, span;
1223
1224 dprintf("probing object %llu offset %llx level %d of %u\n",
1225 dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels);
1226
1275
1276 dprintf("probing object %llu offset %llx level %d of %u\n",
1277 dn->dn_object, *offset, lvl, dn->dn_phys->dn_nlevels);
1278
1279 hole = flags & DNODE_FIND_HOLE;
1280 inc = (flags & DNODE_FIND_BACKWARDS) ? -1 : 1;
1281 ASSERT(txg == 0 || !hole);
1282
1227 if (lvl == dn->dn_phys->dn_nlevels) {
1228 error = 0;
1229 epb = dn->dn_phys->dn_nblkptr;
1230 data = dn->dn_phys->dn_blkptr;
1231 } else {
1232 uint64_t blkid = dbuf_whichblock(dn, *offset) >> (epbs * lvl);
1233 error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FTAG, &db);
1234 if (error) {
1283 if (lvl == dn->dn_phys->dn_nlevels) {
1284 error = 0;
1285 epb = dn->dn_phys->dn_nblkptr;
1286 data = dn->dn_phys->dn_blkptr;
1287 } else {
1288 uint64_t blkid = dbuf_whichblock(dn, *offset) >> (epbs * lvl);
1289 error = dbuf_hold_impl(dn, lvl, blkid, TRUE, FTAG, &db);
1290 if (error) {
1235 if (error == ENOENT)
1236 return (hole ? 0 : ESRCH);
1237 return (error);
1291 if (error != ENOENT)
1292 return (error);
1293 if (hole)
1294 return (0);
1295 /*
1296 * This can only happen when we are searching up
1297 * the block tree for data. We don't really need to
1298 * adjust the offset, as we will just end up looking
1299 * at the pointer to this block in its parent, and its
1300 * going to be unallocated, so we will skip over it.
1301 */
1302 return (ESRCH);
1238 }
1239 error = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_HAVESTRUCT);
1240 if (error) {
1241 dbuf_rele(db, FTAG);
1242 return (error);
1243 }
1244 data = db->db.db_data;
1245 }
1246
1247 if (db && txg &&
1248 (db->db_blkptr == NULL || db->db_blkptr->blk_birth <= txg)) {
1303 }
1304 error = dbuf_read(db, NULL, DB_RF_CANFAIL | DB_RF_HAVESTRUCT);
1305 if (error) {
1306 dbuf_rele(db, FTAG);
1307 return (error);
1308 }
1309 data = db->db.db_data;
1310 }
1311
1312 if (db && txg &&
1313 (db->db_blkptr == NULL || db->db_blkptr->blk_birth <= txg)) {
1314 /*
1315 * This can only happen when we are searching up the tree
1316 * and these conditions mean that we need to keep climbing.
1317 */
1249 error = ESRCH;
1250 } else if (lvl == 0) {
1251 dnode_phys_t *dnp = data;
1252 span = DNODE_SHIFT;
1253 ASSERT(dn->dn_type == DMU_OT_DNODE);
1254
1318 error = ESRCH;
1319 } else if (lvl == 0) {
1320 dnode_phys_t *dnp = data;
1321 span = DNODE_SHIFT;
1322 ASSERT(dn->dn_type == DMU_OT_DNODE);
1323
1255 for (i = (*offset >> span) & (blkfill - 1); i < blkfill; i++) {
1324 for (i = (*offset >> span) & (blkfill - 1);
1325 i >= 0 && i < blkfill; i += inc) {
1256 boolean_t newcontents = B_TRUE;
1257 if (txg) {
1258 int j;
1259 newcontents = B_FALSE;
1260 for (j = 0; j < dnp[i].dn_nblkptr; j++) {
1261 if (dnp[i].dn_blkptr[j].blk_birth > txg)
1262 newcontents = B_TRUE;
1263 }
1264 }
1265 if (!dnp[i].dn_type == hole && newcontents)
1266 break;
1326 boolean_t newcontents = B_TRUE;
1327 if (txg) {
1328 int j;
1329 newcontents = B_FALSE;
1330 for (j = 0; j < dnp[i].dn_nblkptr; j++) {
1331 if (dnp[i].dn_blkptr[j].blk_birth > txg)
1332 newcontents = B_TRUE;
1333 }
1334 }
1335 if (!dnp[i].dn_type == hole && newcontents)
1336 break;
1267 *offset += 1ULL << span;
1337 *offset += (1ULL << span) * inc;
1268 }
1338 }
1269 if (i == blkfill)
1339 if (i < 0 || i == blkfill)
1270 error = ESRCH;
1271 } else {
1272 blkptr_t *bp = data;
1273 span = (lvl - 1) * epbs + dn->dn_datablkshift;
1274 minfill = 0;
1275 maxfill = blkfill << ((lvl - 1) * epbs);
1276
1277 if (hole)
1278 maxfill--;
1279 else
1280 minfill++;
1281
1282 for (i = (*offset >> span) & ((1ULL << epbs) - 1);
1340 error = ESRCH;
1341 } else {
1342 blkptr_t *bp = data;
1343 span = (lvl - 1) * epbs + dn->dn_datablkshift;
1344 minfill = 0;
1345 maxfill = blkfill << ((lvl - 1) * epbs);
1346
1347 if (hole)
1348 maxfill--;
1349 else
1350 minfill++;
1351
1352 for (i = (*offset >> span) & ((1ULL << epbs) - 1);
1283 i < epb; i++) {
1353 i >= 0 && i < epb; i += inc) {
1284 if (bp[i].blk_fill >= minfill &&
1285 bp[i].blk_fill <= maxfill &&
1354 if (bp[i].blk_fill >= minfill &&
1355 bp[i].blk_fill <= maxfill &&
1286 bp[i].blk_birth > txg)
1356 (hole || bp[i].blk_birth > txg))
1287 break;
1357 break;
1288 *offset += 1ULL << span;
1358 if (inc < 0 && *offset < (1ULL << span))
1359 *offset = 0;
1360 else
1361 *offset += (1ULL << span) * inc;
1289 }
1362 }
1290 if (i >= epb)
1363 if (i < 0 || i == epb)
1291 error = ESRCH;
1292 }
1293
1294 if (db)
1295 dbuf_rele(db, FTAG);
1296
1297 return (error);
1298}
1299
1300/*
1301 * Find the next hole, data, or sparse region at or after *offset.
1302 * The value 'blkfill' tells us how many items we expect to find
1303 * in an L0 data block; this value is 1 for normal objects,
1304 * DNODES_PER_BLOCK for the meta dnode, and some fraction of
1305 * DNODES_PER_BLOCK when searching for sparse regions thereof.
1306 *
1307 * Examples:
1308 *
1364 error = ESRCH;
1365 }
1366
1367 if (db)
1368 dbuf_rele(db, FTAG);
1369
1370 return (error);
1371}
1372
1373/*
1374 * Find the next hole, data, or sparse region at or after *offset.
1375 * The value 'blkfill' tells us how many items we expect to find
1376 * in an L0 data block; this value is 1 for normal objects,
1377 * DNODES_PER_BLOCK for the meta dnode, and some fraction of
1378 * DNODES_PER_BLOCK when searching for sparse regions thereof.
1379 *
1380 * Examples:
1381 *
1309 * dnode_next_offset(dn, hole, offset, 1, 1, 0);
1310 * Finds the next hole/data in a file.
1382 * dnode_next_offset(dn, flags, offset, 1, 1, 0);
1383 * Finds the next/previous hole/data in a file.
1311 * Used in dmu_offset_next().
1312 *
1384 * Used in dmu_offset_next().
1385 *
1313 * dnode_next_offset(mdn, hole, offset, 0, DNODES_PER_BLOCK, txg);
1386 * dnode_next_offset(mdn, flags, offset, 0, DNODES_PER_BLOCK, txg);
1314 * Finds the next free/allocated dnode an objset's meta-dnode.
1315 * Only finds objects that have new contents since txg (ie.
1316 * bonus buffer changes and content removal are ignored).
1317 * Used in dmu_object_next().
1318 *
1387 * Finds the next free/allocated dnode an objset's meta-dnode.
1388 * Only finds objects that have new contents since txg (ie.
1389 * bonus buffer changes and content removal are ignored).
1390 * Used in dmu_object_next().
1391 *
1319 * dnode_next_offset(mdn, TRUE, offset, 2, DNODES_PER_BLOCK >> 2, 0);
1392 * dnode_next_offset(mdn, DNODE_FIND_HOLE, offset, 2, DNODES_PER_BLOCK >> 2, 0);
1320 * Finds the next L2 meta-dnode bp that's at most 1/4 full.
1321 * Used in dmu_object_alloc().
1322 */
1323int
1393 * Finds the next L2 meta-dnode bp that's at most 1/4 full.
1394 * Used in dmu_object_alloc().
1395 */
1396int
1324dnode_next_offset(dnode_t *dn, boolean_t hole, uint64_t *offset,
1397dnode_next_offset(dnode_t *dn, int flags, uint64_t *offset,
1325 int minlvl, uint64_t blkfill, uint64_t txg)
1326{
1398 int minlvl, uint64_t blkfill, uint64_t txg)
1399{
1400 uint64_t initial_offset = *offset;
1327 int lvl, maxlvl;
1328 int error = 0;
1401 int lvl, maxlvl;
1402 int error = 0;
1329 uint64_t initial_offset = *offset;
1330
1403
1331 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1404 if (!(flags & DNODE_FIND_HAVELOCK))
1405 rw_enter(&dn->dn_struct_rwlock, RW_READER);
1332
1333 if (dn->dn_phys->dn_nlevels == 0) {
1406
1407 if (dn->dn_phys->dn_nlevels == 0) {
1334 rw_exit(&dn->dn_struct_rwlock);
1335 return (ESRCH);
1408 error = ESRCH;
1409 goto out;
1336 }
1337
1338 if (dn->dn_datablkshift == 0) {
1339 if (*offset < dn->dn_datablksz) {
1410 }
1411
1412 if (dn->dn_datablkshift == 0) {
1413 if (*offset < dn->dn_datablksz) {
1340 if (hole)
1414 if (flags & DNODE_FIND_HOLE)
1341 *offset = dn->dn_datablksz;
1342 } else {
1343 error = ESRCH;
1344 }
1415 *offset = dn->dn_datablksz;
1416 } else {
1417 error = ESRCH;
1418 }
1345 rw_exit(&dn->dn_struct_rwlock);
1346 return (error);
1419 goto out;
1347 }
1348
1349 maxlvl = dn->dn_phys->dn_nlevels;
1350
1351 for (lvl = minlvl; lvl <= maxlvl; lvl++) {
1352 error = dnode_next_offset_level(dn,
1420 }
1421
1422 maxlvl = dn->dn_phys->dn_nlevels;
1423
1424 for (lvl = minlvl; lvl <= maxlvl; lvl++) {
1425 error = dnode_next_offset_level(dn,
1353 hole, offset, lvl, blkfill, txg);
1426 flags, offset, lvl, blkfill, txg);
1354 if (error != ESRCH)
1355 break;
1356 }
1357
1427 if (error != ESRCH)
1428 break;
1429 }
1430
1358 while (--lvl >= minlvl && error == 0) {
1431 while (error == 0 && --lvl >= minlvl) {
1359 error = dnode_next_offset_level(dn,
1432 error = dnode_next_offset_level(dn,
1360 hole, offset, lvl, blkfill, txg);
1433 flags, offset, lvl, blkfill, txg);
1361 }
1362
1434 }
1435
1363 rw_exit(&dn->dn_struct_rwlock);
1364
1365 if (error == 0 && initial_offset > *offset)
1436 if (error == 0 && (flags & DNODE_FIND_BACKWARDS ?
1437 initial_offset < *offset : initial_offset > *offset))
1366 error = ESRCH;
1438 error = ESRCH;
1439out:
1440 if (!(flags & DNODE_FIND_HAVELOCK))
1441 rw_exit(&dn->dn_struct_rwlock);
1367
1368 return (error);
1369}
1442
1443 return (error);
1444}