dsl_dir.c (177698) | dsl_dir.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/dmu.h> | 26#include <sys/dmu.h> |
27#include <sys/dmu_objset.h> |
|
29#include <sys/dmu_tx.h> 30#include <sys/dsl_dataset.h> 31#include <sys/dsl_dir.h> 32#include <sys/dsl_prop.h> 33#include <sys/dsl_synctask.h> | 28#include <sys/dmu_tx.h> 29#include <sys/dsl_dataset.h> 30#include <sys/dsl_dir.h> 31#include <sys/dsl_prop.h> 32#include <sys/dsl_synctask.h> |
33#include <sys/dsl_deleg.h> |
|
34#include <sys/spa.h> 35#include <sys/zap.h> 36#include <sys/zio.h> 37#include <sys/arc.h> | 34#include <sys/spa.h> 35#include <sys/zap.h> 36#include <sys/zio.h> 37#include <sys/arc.h> |
38#include <sys/sunddi.h> |
|
38#include "zfs_namecheck.h" 39 | 39#include "zfs_namecheck.h" 40 |
40static uint64_t dsl_dir_estimated_space(dsl_dir_t *dd); 41static void dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx); | 41static uint64_t dsl_dir_space_towrite(dsl_dir_t *dd); 42static void dsl_dir_set_reservation_sync(void *arg1, void *arg2, 43 cred_t *cr, dmu_tx_t *tx); |
42 43 44/* ARGSUSED */ 45static void 46dsl_dir_evict(dmu_buf_t *db, void *arg) 47{ 48 dsl_dir_t *dd = arg; 49 dsl_pool_t *dp = dd->dd_pool; 50 int t; 51 52 for (t = 0; t < TXG_SIZE; t++) { 53 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t)); 54 ASSERT(dd->dd_tempreserved[t] == 0); 55 ASSERT(dd->dd_space_towrite[t] == 0); 56 } 57 | 44 45 46/* ARGSUSED */ 47static void 48dsl_dir_evict(dmu_buf_t *db, void *arg) 49{ 50 dsl_dir_t *dd = arg; 51 dsl_pool_t *dp = dd->dd_pool; 52 int t; 53 54 for (t = 0; t < TXG_SIZE; t++) { 55 ASSERT(!txg_list_member(&dp->dp_dirty_dirs, dd, t)); 56 ASSERT(dd->dd_tempreserved[t] == 0); 57 ASSERT(dd->dd_space_towrite[t] == 0); 58 } 59 |
58 ASSERT3U(dd->dd_used_bytes, ==, dd->dd_phys->dd_used_bytes); 59 | |
60 if (dd->dd_parent) 61 dsl_dir_close(dd->dd_parent, dd); 62 63 spa_close(dd->dd_pool->dp_spa, dd); 64 65 /* 66 * The props callback list should be empty since they hold the 67 * dir open. --- 18 unchanged lines hidden (view full) --- 86 if (err) 87 return (err); 88 dd = dmu_buf_get_user(dbuf); 89#ifdef ZFS_DEBUG 90 { 91 dmu_object_info_t doi; 92 dmu_object_info_from_db(dbuf, &doi); 93 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR); | 60 if (dd->dd_parent) 61 dsl_dir_close(dd->dd_parent, dd); 62 63 spa_close(dd->dd_pool->dp_spa, dd); 64 65 /* 66 * The props callback list should be empty since they hold the 67 * dir open. --- 18 unchanged lines hidden (view full) --- 86 if (err) 87 return (err); 88 dd = dmu_buf_get_user(dbuf); 89#ifdef ZFS_DEBUG 90 { 91 dmu_object_info_t doi; 92 dmu_object_info_from_db(dbuf, &doi); 93 ASSERT3U(doi.doi_type, ==, DMU_OT_DSL_DIR); |
94 ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t)); |
|
94 } 95#endif | 95 } 96#endif |
96 /* XXX assert bonus buffer size is correct */ | |
97 if (dd == NULL) { 98 dsl_dir_t *winner; 99 int err; 100 101 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); 102 dd->dd_object = ddobj; 103 dd->dd_dbuf = dbuf; 104 dd->dd_pool = dp; 105 dd->dd_phys = dbuf->db_data; | 97 if (dd == NULL) { 98 dsl_dir_t *winner; 99 int err; 100 101 dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP); 102 dd->dd_object = ddobj; 103 dd->dd_dbuf = dbuf; 104 dd->dd_pool = dp; 105 dd->dd_phys = dbuf->db_data; |
106 dd->dd_used_bytes = dd->dd_phys->dd_used_bytes; | |
107 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); 108 109 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t), 110 offsetof(dsl_prop_cb_record_t, cbr_node)); 111 112 if (dd->dd_phys->dd_parent_obj) { 113 err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj, 114 NULL, dd, &dd->dd_parent); | 106 mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL); 107 108 list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t), 109 offsetof(dsl_prop_cb_record_t, cbr_node)); 110 111 if (dd->dd_phys->dd_parent_obj) { 112 err = dsl_dir_open_obj(dp, dd->dd_phys->dd_parent_obj, 113 NULL, dd, &dd->dd_parent); |
115 if (err) { 116 mutex_destroy(&dd->dd_lock); 117 kmem_free(dd, sizeof (dsl_dir_t)); 118 dmu_buf_rele(dbuf, tag); 119 return (err); 120 } | 114 if (err) 115 goto errout; |
121 if (tail) { 122#ifdef ZFS_DEBUG 123 uint64_t foundobj; 124 125 err = zap_lookup(dp->dp_meta_objset, | 116 if (tail) { 117#ifdef ZFS_DEBUG 118 uint64_t foundobj; 119 120 err = zap_lookup(dp->dp_meta_objset, |
126 dd->dd_parent->dd_phys-> 127 dd_child_dir_zapobj, | 121 dd->dd_parent->dd_phys->dd_child_dir_zapobj, |
128 tail, sizeof (foundobj), 1, &foundobj); 129 ASSERT(err || foundobj == ddobj); 130#endif 131 (void) strcpy(dd->dd_myname, tail); 132 } else { 133 err = zap_value_search(dp->dp_meta_objset, | 122 tail, sizeof (foundobj), 1, &foundobj); 123 ASSERT(err || foundobj == ddobj); 124#endif 125 (void) strcpy(dd->dd_myname, tail); 126 } else { 127 err = zap_value_search(dp->dp_meta_objset, |
134 dd->dd_parent->dd_phys-> 135 dd_child_dir_zapobj, 136 ddobj, dd->dd_myname); | 128 dd->dd_parent->dd_phys->dd_child_dir_zapobj, 129 ddobj, 0, dd->dd_myname); |
137 } | 130 } |
138 if (err) { 139 dsl_dir_close(dd->dd_parent, dd); 140 mutex_destroy(&dd->dd_lock); 141 kmem_free(dd, sizeof (dsl_dir_t)); 142 dmu_buf_rele(dbuf, tag); 143 return (err); 144 } | 131 if (err) 132 goto errout; |
145 } else { 146 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); 147 } 148 149 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys, 150 dsl_dir_evict); 151 if (winner) { 152 if (dd->dd_parent) --- 16 unchanged lines hidden (view full) --- 169 * pointer to the dd_pool, which has a pointer to the spa_t. 170 */ 171 spa_open_ref(dp->dp_spa, tag); 172 ASSERT3P(dd->dd_pool, ==, dp); 173 ASSERT3U(dd->dd_object, ==, ddobj); 174 ASSERT3P(dd->dd_dbuf, ==, dbuf); 175 *ddp = dd; 176 return (0); | 133 } else { 134 (void) strcpy(dd->dd_myname, spa_name(dp->dp_spa)); 135 } 136 137 winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys, 138 dsl_dir_evict); 139 if (winner) { 140 if (dd->dd_parent) --- 16 unchanged lines hidden (view full) --- 157 * pointer to the dd_pool, which has a pointer to the spa_t. 158 */ 159 spa_open_ref(dp->dp_spa, tag); 160 ASSERT3P(dd->dd_pool, ==, dp); 161 ASSERT3U(dd->dd_object, ==, ddobj); 162 ASSERT3P(dd->dd_dbuf, ==, dbuf); 163 *ddp = dd; 164 return (0); |
165 166errout: 167 if (dd->dd_parent) 168 dsl_dir_close(dd->dd_parent, dd); 169 mutex_destroy(&dd->dd_lock); 170 kmem_free(dd, sizeof (dsl_dir_t)); 171 dmu_buf_rele(dbuf, tag); 172 return (err); 173 |
|
177} 178 179void 180dsl_dir_close(dsl_dir_t *dd, void *tag) 181{ 182 dprintf_dd(dd, "%s\n", ""); 183 spa_close(dd->dd_pool->dp_spa, tag); 184 dmu_buf_rele(dd->dd_dbuf, tag); --- 214 unchanged lines hidden (view full) --- 399 */ 400int 401dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp) 402{ 403 return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp)); 404} 405 406uint64_t | 174} 175 176void 177dsl_dir_close(dsl_dir_t *dd, void *tag) 178{ 179 dprintf_dd(dd, "%s\n", ""); 180 spa_close(dd->dd_pool->dp_spa, tag); 181 dmu_buf_rele(dd->dd_dbuf, tag); --- 214 unchanged lines hidden (view full) --- 396 */ 397int 398dsl_dir_open(const char *name, void *tag, dsl_dir_t **ddp, const char **tailp) 399{ 400 return (dsl_dir_open_spa(NULL, name, tag, ddp, tailp)); 401} 402 403uint64_t |
407dsl_dir_create_sync(dsl_dir_t *pds, const char *name, dmu_tx_t *tx) | 404dsl_dir_create_sync(dsl_pool_t *dp, dsl_dir_t *pds, const char *name, 405 dmu_tx_t *tx) |
408{ | 406{ |
409 objset_t *mos = pds->dd_pool->dp_meta_objset; | 407 objset_t *mos = dp->dp_meta_objset; |
410 uint64_t ddobj; 411 dsl_dir_phys_t *dsphys; 412 dmu_buf_t *dbuf; 413 414 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, 415 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); | 408 uint64_t ddobj; 409 dsl_dir_phys_t *dsphys; 410 dmu_buf_t *dbuf; 411 412 ddobj = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, 413 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); |
416 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj, 417 name, sizeof (uint64_t), 1, &ddobj, tx)); | 414 if (pds) { 415 VERIFY(0 == zap_add(mos, pds->dd_phys->dd_child_dir_zapobj, 416 name, sizeof (uint64_t), 1, &ddobj, tx)); 417 } else { 418 /* it's the root dir */ 419 VERIFY(0 == zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, 420 DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1, &ddobj, tx)); 421 } |
418 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf)); 419 dmu_buf_will_dirty(dbuf, tx); 420 dsphys = dbuf->db_data; 421 422 dsphys->dd_creation_time = gethrestime_sec(); | 422 VERIFY(0 == dmu_bonus_hold(mos, ddobj, FTAG, &dbuf)); 423 dmu_buf_will_dirty(dbuf, tx); 424 dsphys = dbuf->db_data; 425 426 dsphys->dd_creation_time = gethrestime_sec(); |
423 dsphys->dd_parent_obj = pds->dd_object; | 427 if (pds) 428 dsphys->dd_parent_obj = pds->dd_object; |
424 dsphys->dd_props_zapobj = zap_create(mos, 425 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); 426 dsphys->dd_child_dir_zapobj = zap_create(mos, 427 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); | 429 dsphys->dd_props_zapobj = zap_create(mos, 430 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); 431 dsphys->dd_child_dir_zapobj = zap_create(mos, 432 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); |
433 if (spa_version(dp->dp_spa) >= SPA_VERSION_USED_BREAKDOWN) 434 dsphys->dd_flags |= DD_FLAG_USED_BREAKDOWN; |
|
428 dmu_buf_rele(dbuf, FTAG); 429 430 return (ddobj); 431} 432 433/* ARGSUSED */ 434int 435dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) --- 20 unchanged lines hidden (view full) --- 456 return (err); 457 if (count != 0) 458 return (EEXIST); 459 460 return (0); 461} 462 463void | 435 dmu_buf_rele(dbuf, FTAG); 436 437 return (ddobj); 438} 439 440/* ARGSUSED */ 441int 442dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx) --- 20 unchanged lines hidden (view full) --- 463 return (err); 464 if (count != 0) 465 return (EEXIST); 466 467 return (0); 468} 469 470void |
464dsl_dir_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx) | 471dsl_dir_destroy_sync(void *arg1, void *tag, cred_t *cr, dmu_tx_t *tx) |
465{ 466 dsl_dir_t *dd = arg1; 467 objset_t *mos = dd->dd_pool->dp_meta_objset; 468 uint64_t val, obj; | 472{ 473 dsl_dir_t *dd = arg1; 474 objset_t *mos = dd->dd_pool->dp_meta_objset; 475 uint64_t val, obj; |
476 dd_used_t t; |
|
469 470 ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock)); 471 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 472 473 /* Remove our reservation. */ 474 val = 0; | 477 478 ASSERT(RW_WRITE_HELD(&dd->dd_pool->dp_config_rwlock)); 479 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0); 480 481 /* Remove our reservation. */ 482 val = 0; |
475 dsl_dir_set_reservation_sync(dd, &val, tx); 476 ASSERT3U(dd->dd_used_bytes, ==, 0); | 483 dsl_dir_set_reservation_sync(dd, &val, cr, tx); 484 ASSERT3U(dd->dd_phys->dd_used_bytes, ==, 0); |
477 ASSERT3U(dd->dd_phys->dd_reserved, ==, 0); | 485 ASSERT3U(dd->dd_phys->dd_reserved, ==, 0); |
486 for (t = 0; t < DD_USED_NUM; t++) 487 ASSERT3U(dd->dd_phys->dd_used_breakdown[t], ==, 0); |
|
478 479 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx)); 480 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx)); | 488 489 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_child_dir_zapobj, tx)); 490 VERIFY(0 == zap_destroy(mos, dd->dd_phys->dd_props_zapobj, tx)); |
491 VERIFY(0 == dsl_deleg_destroy(mos, dd->dd_phys->dd_deleg_zapobj, tx)); |
|
481 VERIFY(0 == zap_remove(mos, 482 dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx)); 483 484 obj = dd->dd_object; 485 dsl_dir_close(dd, tag); 486 VERIFY(0 == dmu_object_free(mos, obj, tx)); 487} 488 | 492 VERIFY(0 == zap_remove(mos, 493 dd->dd_parent->dd_phys->dd_child_dir_zapobj, dd->dd_myname, tx)); 494 495 obj = dd->dd_object; 496 dsl_dir_close(dd, tag); 497 VERIFY(0 == dmu_object_free(mos, obj, tx)); 498} 499 |
489void 490dsl_dir_create_root(objset_t *mos, uint64_t *ddobjp, dmu_tx_t *tx) | 500boolean_t 501dsl_dir_is_clone(dsl_dir_t *dd) |
491{ | 502{ |
492 dsl_dir_phys_t *dsp; 493 dmu_buf_t *dbuf; 494 int error; 495 496 *ddobjp = dmu_object_alloc(mos, DMU_OT_DSL_DIR, 0, 497 DMU_OT_DSL_DIR, sizeof (dsl_dir_phys_t), tx); 498 499 error = zap_add(mos, DMU_POOL_DIRECTORY_OBJECT, DMU_POOL_ROOT_DATASET, 500 sizeof (uint64_t), 1, ddobjp, tx); 501 ASSERT3U(error, ==, 0); 502 503 VERIFY(0 == dmu_bonus_hold(mos, *ddobjp, FTAG, &dbuf)); 504 dmu_buf_will_dirty(dbuf, tx); 505 dsp = dbuf->db_data; 506 507 dsp->dd_creation_time = gethrestime_sec(); 508 dsp->dd_props_zapobj = zap_create(mos, 509 DMU_OT_DSL_PROPS, DMU_OT_NONE, 0, tx); 510 dsp->dd_child_dir_zapobj = zap_create(mos, 511 DMU_OT_DSL_DIR_CHILD_MAP, DMU_OT_NONE, 0, tx); 512 513 dmu_buf_rele(dbuf, FTAG); | 503 return (dd->dd_phys->dd_origin_obj && 504 (dd->dd_pool->dp_origin_snap == NULL || 505 dd->dd_phys->dd_origin_obj != 506 dd->dd_pool->dp_origin_snap->ds_object)); |
514} 515 516void 517dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) 518{ | 507} 508 509void 510dsl_dir_stats(dsl_dir_t *dd, nvlist_t *nv) 511{ |
519 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, 520 dsl_dir_space_available(dd, NULL, 0, TRUE)); 521 | |
522 mutex_enter(&dd->dd_lock); | 512 mutex_enter(&dd->dd_lock); |
523 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, dd->dd_used_bytes); 524 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, 525 dd->dd_phys->dd_quota); | 513 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED, 514 dd->dd_phys->dd_used_bytes); 515 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_QUOTA, dd->dd_phys->dd_quota); |
526 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, 527 dd->dd_phys->dd_reserved); 528 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 529 dd->dd_phys->dd_compressed_bytes == 0 ? 100 : 530 (dd->dd_phys->dd_uncompressed_bytes * 100 / 531 dd->dd_phys->dd_compressed_bytes)); | 516 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_RESERVATION, 517 dd->dd_phys->dd_reserved); 518 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, 519 dd->dd_phys->dd_compressed_bytes == 0 ? 100 : 520 (dd->dd_phys->dd_uncompressed_bytes * 100 / 521 dd->dd_phys->dd_compressed_bytes)); |
522 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 523 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDSNAP, 524 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP]); 525 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDDS, 526 dd->dd_phys->dd_used_breakdown[DD_USED_HEAD]); 527 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDREFRESERV, 528 dd->dd_phys->dd_used_breakdown[DD_USED_REFRSRV]); 529 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USEDCHILD, 530 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD] + 531 dd->dd_phys->dd_used_breakdown[DD_USED_CHILD_RSRV]); 532 } |
|
532 mutex_exit(&dd->dd_lock); 533 | 533 mutex_exit(&dd->dd_lock); 534 |
534 if (dd->dd_phys->dd_clone_parent_obj) { | 535 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); 536 if (dsl_dir_is_clone(dd)) { |
535 dsl_dataset_t *ds; 536 char buf[MAXNAMELEN]; 537 | 537 dsl_dataset_t *ds; 538 char buf[MAXNAMELEN]; 539 |
538 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER); 539 VERIFY(0 == dsl_dataset_open_obj(dd->dd_pool, 540 dd->dd_phys->dd_clone_parent_obj, 541 NULL, DS_MODE_NONE, FTAG, &ds)); | 540 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, 541 dd->dd_phys->dd_origin_obj, FTAG, &ds)); |
542 dsl_dataset_name(ds, buf); | 542 dsl_dataset_name(ds, buf); |
543 dsl_dataset_close(ds, DS_MODE_NONE, FTAG); 544 rw_exit(&dd->dd_pool->dp_config_rwlock); 545 | 543 dsl_dataset_rele(ds, FTAG); |
546 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); 547 } | 544 dsl_prop_nvlist_add_string(nv, ZFS_PROP_ORIGIN, buf); 545 } |
546 rw_exit(&dd->dd_pool->dp_config_rwlock); |
|
548} 549 550void 551dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) 552{ 553 dsl_pool_t *dp = dd->dd_pool; 554 555 ASSERT(dd->dd_phys); --- 19 unchanged lines hidden (view full) --- 575 576 dmu_buf_will_dirty(dd->dd_dbuf, tx); 577 578 mutex_enter(&dd->dd_lock); 579 ASSERT3U(dd->dd_tempreserved[tx->tx_txg&TXG_MASK], ==, 0); 580 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, 581 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024); 582 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0; | 547} 548 549void 550dsl_dir_dirty(dsl_dir_t *dd, dmu_tx_t *tx) 551{ 552 dsl_pool_t *dp = dd->dd_pool; 553 554 ASSERT(dd->dd_phys); --- 19 unchanged lines hidden (view full) --- 574 575 dmu_buf_will_dirty(dd->dd_dbuf, tx); 576 577 mutex_enter(&dd->dd_lock); 578 ASSERT3U(dd->dd_tempreserved[tx->tx_txg&TXG_MASK], ==, 0); 579 dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg, 580 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024); 581 dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0; |
583 dd->dd_phys->dd_used_bytes = dd->dd_used_bytes; | |
584 mutex_exit(&dd->dd_lock); 585 586 /* release the hold from dsl_dir_dirty */ 587 dmu_buf_rele(dd->dd_dbuf, dd); 588} 589 590static uint64_t | 582 mutex_exit(&dd->dd_lock); 583 584 /* release the hold from dsl_dir_dirty */ 585 dmu_buf_rele(dd->dd_dbuf, dd); 586} 587 588static uint64_t |
591dsl_dir_estimated_space(dsl_dir_t *dd) | 589dsl_dir_space_towrite(dsl_dir_t *dd) |
592{ | 590{ |
593 int64_t space; | 591 uint64_t space = 0; |
594 int i; 595 596 ASSERT(MUTEX_HELD(&dd->dd_lock)); 597 | 592 int i; 593 594 ASSERT(MUTEX_HELD(&dd->dd_lock)); 595 |
598 space = dd->dd_phys->dd_used_bytes; 599 ASSERT(space >= 0); | |
600 for (i = 0; i < TXG_SIZE; i++) { 601 space += dd->dd_space_towrite[i&TXG_MASK]; 602 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0); 603 } 604 return (space); 605} 606 607/* --- 17 unchanged lines hidden (view full) --- 625 if (dd->dd_parent != NULL) { 626 parentspace = dsl_dir_space_available(dd->dd_parent, 627 ancestor, delta, ondiskonly); 628 } 629 630 mutex_enter(&dd->dd_lock); 631 if (dd->dd_phys->dd_quota != 0) 632 quota = dd->dd_phys->dd_quota; | 596 for (i = 0; i < TXG_SIZE; i++) { 597 space += dd->dd_space_towrite[i&TXG_MASK]; 598 ASSERT3U(dd->dd_space_towrite[i&TXG_MASK], >=, 0); 599 } 600 return (space); 601} 602 603/* --- 17 unchanged lines hidden (view full) --- 621 if (dd->dd_parent != NULL) { 622 parentspace = dsl_dir_space_available(dd->dd_parent, 623 ancestor, delta, ondiskonly); 624 } 625 626 mutex_enter(&dd->dd_lock); 627 if (dd->dd_phys->dd_quota != 0) 628 quota = dd->dd_phys->dd_quota; |
633 if (ondiskonly) { 634 used = dd->dd_used_bytes; 635 } else { 636 used = dsl_dir_estimated_space(dd); 637 } 638 if (dd == ancestor) 639 used += delta; | 629 used = dd->dd_phys->dd_used_bytes; 630 if (!ondiskonly) 631 used += dsl_dir_space_towrite(dd); |
640 641 if (dd->dd_parent == NULL) { 642 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE); 643 quota = MIN(quota, poolsize); 644 } 645 646 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) { 647 /* 648 * We have some space reserved, in addition to what our 649 * parent gave us. 650 */ 651 parentspace += dd->dd_phys->dd_reserved - used; 652 } 653 | 632 633 if (dd->dd_parent == NULL) { 634 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, FALSE); 635 quota = MIN(quota, poolsize); 636 } 637 638 if (dd->dd_phys->dd_reserved > used && parentspace != UINT64_MAX) { 639 /* 640 * We have some space reserved, in addition to what our 641 * parent gave us. 642 */ 643 parentspace += dd->dd_phys->dd_reserved - used; 644 } 645 |
646 if (dd == ancestor) { 647 ASSERT(delta <= 0); 648 ASSERT(used >= -delta); 649 used += delta; 650 if (parentspace != UINT64_MAX) 651 parentspace -= delta; 652 } 653 |
|
654 if (used > quota) { 655 /* over quota */ 656 myspace = 0; 657 658 /* 659 * While it's OK to be a little over quota, if 660 * we think we are using more space than there 661 * is in the pool (which is already 1.6% more than --- 11 unchanged lines hidden (view full) --- 673 674 mutex_exit(&dd->dd_lock); 675 676 return (myspace); 677} 678 679struct tempreserve { 680 list_node_t tr_node; | 654 if (used > quota) { 655 /* over quota */ 656 myspace = 0; 657 658 /* 659 * While it's OK to be a little over quota, if 660 * we think we are using more space than there 661 * is in the pool (which is already 1.6% more than --- 11 unchanged lines hidden (view full) --- 673 674 mutex_exit(&dd->dd_lock); 675 676 return (myspace); 677} 678 679struct tempreserve { 680 list_node_t tr_node; |
681 dsl_pool_t *tr_dp; |
|
681 dsl_dir_t *tr_ds; 682 uint64_t tr_size; 683}; 684 | 682 dsl_dir_t *tr_ds; 683 uint64_t tr_size; 684}; 685 |
685/* 686 * Reserve space in this dsl_dir, to be used in this tx's txg. 687 * After the space has been dirtied (and thus 688 * dsl_dir_willuse_space() has been called), the reservation should 689 * be canceled, using dsl_dir_tempreserve_clear(). 690 */ | |
691static int | 686static int |
692dsl_dir_tempreserve_impl(dsl_dir_t *dd, 693 uint64_t asize, boolean_t netfree, list_t *tr_list, dmu_tx_t *tx) | 687dsl_dir_tempreserve_impl(dsl_dir_t *dd, uint64_t asize, boolean_t netfree, 688 boolean_t ignorequota, boolean_t checkrefquota, list_t *tr_list, 689 dmu_tx_t *tx, boolean_t first) |
694{ 695 uint64_t txg = tx->tx_txg; | 690{ 691 uint64_t txg = tx->tx_txg; |
696 uint64_t est_used, quota, parent_rsrv; 697 int edquot = EDQUOT; | 692 uint64_t est_inflight, used_on_disk, quota, parent_rsrv; 693 struct tempreserve *tr; 694 int enospc = EDQUOT; |
698 int txgidx = txg & TXG_MASK; 699 int i; | 695 int txgidx = txg & TXG_MASK; 696 int i; |
700 struct tempreserve *tr; | 697 uint64_t ref_rsrv = 0; |
701 702 ASSERT3U(txg, !=, 0); | 698 699 ASSERT3U(txg, !=, 0); |
703 ASSERT3S(asize, >=, 0); | 700 ASSERT3S(asize, >, 0); |
704 705 mutex_enter(&dd->dd_lock); | 701 702 mutex_enter(&dd->dd_lock); |
703 |
|
706 /* 707 * Check against the dsl_dir's quota. We don't add in the delta 708 * when checking for over-quota because they get one free hit. 709 */ | 704 /* 705 * Check against the dsl_dir's quota. We don't add in the delta 706 * when checking for over-quota because they get one free hit. 707 */ |
710 est_used = dsl_dir_estimated_space(dd); | 708 est_inflight = dsl_dir_space_towrite(dd); |
711 for (i = 0; i < TXG_SIZE; i++) | 709 for (i = 0; i < TXG_SIZE; i++) |
712 est_used += dd->dd_tempreserved[i]; | 710 est_inflight += dd->dd_tempreserved[i]; 711 used_on_disk = dd->dd_phys->dd_used_bytes; |
713 | 712 |
714 quota = UINT64_MAX; | 713 /* 714 * On the first iteration, fetch the dataset's used-on-disk and 715 * refreservation values. Also, if checkrefquota is set, test if 716 * allocating this space would exceed the dataset's refquota. 717 */ 718 if (first && tx->tx_objset) { 719 int error; 720 dsl_dataset_t *ds = tx->tx_objset->os->os_dsl_dataset; |
715 | 721 |
716 if (dd->dd_phys->dd_quota) | 722 error = dsl_dataset_check_quota(ds, checkrefquota, 723 asize, est_inflight, &used_on_disk, &ref_rsrv); 724 if (error) { 725 mutex_exit(&dd->dd_lock); 726 return (error); 727 } 728 } 729 730 /* 731 * If this transaction will result in a net free of space, 732 * we want to let it through. 733 */ 734 if (ignorequota || netfree || dd->dd_phys->dd_quota == 0) 735 quota = UINT64_MAX; 736 else |
717 quota = dd->dd_phys->dd_quota; 718 719 /* | 737 quota = dd->dd_phys->dd_quota; 738 739 /* |
720 * If this transaction will result in a net free of space, we want 721 * to let it through, but we have to be careful: the space that it 722 * frees won't become available until *after* this txg syncs. 723 * Therefore, to ensure that it's possible to remove files from 724 * a full pool without inducing transient overcommits, we throttle | 740 * Adjust the quota against the actual pool size at the root. 741 * To ensure that it's possible to remove files from a full 742 * pool without inducing transient overcommits, we throttle |
725 * netfree transactions against a quota that is slightly larger, 726 * but still within the pool's allocation slop. In cases where 727 * we're very close to full, this will allow a steady trickle of 728 * removes to get through. 729 */ 730 if (dd->dd_parent == NULL) { 731 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree); 732 if (poolsize < quota) { 733 quota = poolsize; | 743 * netfree transactions against a quota that is slightly larger, 744 * but still within the pool's allocation slop. In cases where 745 * we're very close to full, this will allow a steady trickle of 746 * removes to get through. 747 */ 748 if (dd->dd_parent == NULL) { 749 uint64_t poolsize = dsl_pool_adjustedsize(dd->dd_pool, netfree); 750 if (poolsize < quota) { 751 quota = poolsize; |
734 edquot = ENOSPC; | 752 enospc = ENOSPC; |
735 } | 753 } |
736 } else if (netfree) { 737 quota = UINT64_MAX; | |
738 } 739 740 /* 741 * If they are requesting more space, and our current estimate | 754 } 755 756 /* 757 * If they are requesting more space, and our current estimate |
742 * is over quota. They get to try again unless the actual | 758 * is over quota, they get to try again unless the actual |
743 * on-disk is over quota and there are no pending changes (which 744 * may free up space for us). 745 */ | 759 * on-disk is over quota and there are no pending changes (which 760 * may free up space for us). 761 */ |
746 if (asize > 0 && est_used > quota) { 747 if (dd->dd_space_towrite[txg & TXG_MASK] != 0 || 748 dd->dd_space_towrite[(txg-1) & TXG_MASK] != 0 || 749 dd->dd_space_towrite[(txg-2) & TXG_MASK] != 0 || 750 dd->dd_used_bytes < quota) 751 edquot = ERESTART; 752 dprintf_dd(dd, "failing: used=%lluK est_used = %lluK " | 762 if (used_on_disk + est_inflight > quota) { 763 if (est_inflight > 0 || used_on_disk < quota) 764 enospc = ERESTART; 765 dprintf_dd(dd, "failing: used=%lluK inflight = %lluK " |
753 "quota=%lluK tr=%lluK err=%d\n", | 766 "quota=%lluK tr=%lluK err=%d\n", |
754 dd->dd_used_bytes>>10, est_used>>10, 755 quota>>10, asize>>10, edquot); | 767 used_on_disk>>10, est_inflight>>10, 768 quota>>10, asize>>10, enospc); |
756 mutex_exit(&dd->dd_lock); | 769 mutex_exit(&dd->dd_lock); |
757 return (edquot); | 770 return (enospc); |
758 } 759 760 /* We need to up our estimated delta before dropping dd_lock */ 761 dd->dd_tempreserved[txgidx] += asize; 762 | 771 } 772 773 /* We need to up our estimated delta before dropping dd_lock */ 774 dd->dd_tempreserved[txgidx] += asize; 775 |
763 parent_rsrv = parent_delta(dd, est_used, asize); | 776 parent_rsrv = parent_delta(dd, used_on_disk + est_inflight, 777 asize - ref_rsrv); |
764 mutex_exit(&dd->dd_lock); 765 | 778 mutex_exit(&dd->dd_lock); 779 |
766 tr = kmem_alloc(sizeof (struct tempreserve), KM_SLEEP); | 780 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); |
767 tr->tr_ds = dd; 768 tr->tr_size = asize; 769 list_insert_tail(tr_list, tr); 770 771 /* see if it's OK with our parent */ 772 if (dd->dd_parent && parent_rsrv) { | 781 tr->tr_ds = dd; 782 tr->tr_size = asize; 783 list_insert_tail(tr_list, tr); 784 785 /* see if it's OK with our parent */ 786 if (dd->dd_parent && parent_rsrv) { |
787 boolean_t ismos = (dd->dd_phys->dd_head_dataset_obj == 0); 788 |
|
773 return (dsl_dir_tempreserve_impl(dd->dd_parent, | 789 return (dsl_dir_tempreserve_impl(dd->dd_parent, |
774 parent_rsrv, netfree, tr_list, tx)); | 790 parent_rsrv, netfree, ismos, TRUE, tr_list, tx, FALSE)); |
775 } else { 776 return (0); 777 } 778} 779 780/* 781 * Reserve space in this dsl_dir, to be used in this tx's txg. | 791 } else { 792 return (0); 793 } 794} 795 796/* 797 * Reserve space in this dsl_dir, to be used in this tx's txg. |
782 * After the space has been dirtied (and thus 783 * dsl_dir_willuse_space() has been called), the reservation should 784 * be canceled, using dsl_dir_tempreserve_clear(). | 798 * After the space has been dirtied (and dsl_dir_willuse_space() 799 * has been called), the reservation should be canceled, using 800 * dsl_dir_tempreserve_clear(). |
785 */ 786int | 801 */ 802int |
787dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, 788 uint64_t asize, uint64_t fsize, void **tr_cookiep, dmu_tx_t *tx) | 803dsl_dir_tempreserve_space(dsl_dir_t *dd, uint64_t lsize, uint64_t asize, 804 uint64_t fsize, uint64_t usize, void **tr_cookiep, dmu_tx_t *tx) |
789{ | 805{ |
790 int err = 0; | 806 int err; |
791 list_t *tr_list; 792 | 807 list_t *tr_list; 808 |
809 if (asize == 0) { 810 *tr_cookiep = NULL; 811 return (0); 812 } 813 |
|
793 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); 794 list_create(tr_list, sizeof (struct tempreserve), 795 offsetof(struct tempreserve, tr_node)); | 814 tr_list = kmem_alloc(sizeof (list_t), KM_SLEEP); 815 list_create(tr_list, sizeof (struct tempreserve), 816 offsetof(struct tempreserve, tr_node)); |
796 ASSERT3S(asize, >=, 0); | 817 ASSERT3S(asize, >, 0); |
797 ASSERT3S(fsize, >=, 0); 798 | 818 ASSERT3S(fsize, >=, 0); 819 |
799 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize, 800 tr_list, tx); 801 | 820 err = arc_tempreserve_space(lsize, tx->tx_txg); |
802 if (err == 0) { 803 struct tempreserve *tr; 804 | 821 if (err == 0) { 822 struct tempreserve *tr; 823 |
805 err = arc_tempreserve_space(lsize); 806 if (err == 0) { 807 tr = kmem_alloc(sizeof (struct tempreserve), KM_SLEEP); 808 tr->tr_ds = NULL; 809 tr->tr_size = lsize; 810 list_insert_tail(tr_list, tr); | 824 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 825 tr->tr_size = lsize; 826 list_insert_tail(tr_list, tr); 827 828 err = dsl_pool_tempreserve_space(dd->dd_pool, asize, tx); 829 } else { 830 if (err == EAGAIN) { 831 txg_delay(dd->dd_pool, tx->tx_txg, 1); 832 err = ERESTART; |
811 } | 833 } |
834 dsl_pool_memory_pressure(dd->dd_pool); |
|
812 } 813 | 835 } 836 |
837 if (err == 0) { 838 struct tempreserve *tr; 839 840 tr = kmem_zalloc(sizeof (struct tempreserve), KM_SLEEP); 841 tr->tr_dp = dd->dd_pool; 842 tr->tr_size = asize; 843 list_insert_tail(tr_list, tr); 844 845 err = dsl_dir_tempreserve_impl(dd, asize, fsize >= asize, 846 FALSE, asize > usize, tr_list, tx, TRUE); 847 } 848 |
|
814 if (err) 815 dsl_dir_tempreserve_clear(tr_list, tx); 816 else 817 *tr_cookiep = tr_list; | 849 if (err) 850 dsl_dir_tempreserve_clear(tr_list, tx); 851 else 852 *tr_cookiep = tr_list; |
853 |
|
818 return (err); 819} 820 821/* 822 * Clear a temporary reservation that we previously made with 823 * dsl_dir_tempreserve_space(). 824 */ 825void 826dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) 827{ 828 int txgidx = tx->tx_txg & TXG_MASK; 829 list_t *tr_list = tr_cookie; 830 struct tempreserve *tr; 831 832 ASSERT3U(tx->tx_txg, !=, 0); 833 | 854 return (err); 855} 856 857/* 858 * Clear a temporary reservation that we previously made with 859 * dsl_dir_tempreserve_space(). 860 */ 861void 862dsl_dir_tempreserve_clear(void *tr_cookie, dmu_tx_t *tx) 863{ 864 int txgidx = tx->tx_txg & TXG_MASK; 865 list_t *tr_list = tr_cookie; 866 struct tempreserve *tr; 867 868 ASSERT3U(tx->tx_txg, !=, 0); 869 |
870 if (tr_cookie == NULL) 871 return; 872 |
|
834 while (tr = list_head(tr_list)) { | 873 while (tr = list_head(tr_list)) { |
835 if (tr->tr_ds == NULL) { 836 arc_tempreserve_clear(tr->tr_size); 837 } else { | 874 if (tr->tr_dp) { 875 dsl_pool_tempreserve_clear(tr->tr_dp, tr->tr_size, tx); 876 } else if (tr->tr_ds) { |
838 mutex_enter(&tr->tr_ds->dd_lock); 839 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, 840 tr->tr_size); 841 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; 842 mutex_exit(&tr->tr_ds->dd_lock); | 877 mutex_enter(&tr->tr_ds->dd_lock); 878 ASSERT3U(tr->tr_ds->dd_tempreserved[txgidx], >=, 879 tr->tr_size); 880 tr->tr_ds->dd_tempreserved[txgidx] -= tr->tr_size; 881 mutex_exit(&tr->tr_ds->dd_lock); |
882 } else { 883 arc_tempreserve_clear(tr->tr_size); |
|
843 } 844 list_remove(tr_list, tr); 845 kmem_free(tr, sizeof (struct tempreserve)); 846 } 847 848 kmem_free(tr_list, sizeof (list_t)); 849} 850 | 884 } 885 list_remove(tr_list, tr); 886 kmem_free(tr, sizeof (struct tempreserve)); 887 } 888 889 kmem_free(tr_list, sizeof (list_t)); 890} 891 |
851/* 852 * Call in open context when we think we're going to write/free space, 853 * eg. when dirtying data. Be conservative (ie. OK to write less than 854 * this or free more than this, but don't write more or free less). 855 */ 856void 857dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) | 892static void 893dsl_dir_willuse_space_impl(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) |
858{ 859 int64_t parent_space; 860 uint64_t est_used; 861 862 mutex_enter(&dd->dd_lock); 863 if (space > 0) 864 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; 865 | 894{ 895 int64_t parent_space; 896 uint64_t est_used; 897 898 mutex_enter(&dd->dd_lock); 899 if (space > 0) 900 dd->dd_space_towrite[tx->tx_txg & TXG_MASK] += space; 901 |
866 est_used = dsl_dir_estimated_space(dd); | 902 est_used = dsl_dir_space_towrite(dd) + dd->dd_phys->dd_used_bytes; |
867 parent_space = parent_delta(dd, est_used, space); 868 mutex_exit(&dd->dd_lock); 869 870 /* Make sure that we clean up dd_space_to* */ 871 dsl_dir_dirty(dd, tx); 872 873 /* XXX this is potentially expensive and unnecessary... */ 874 if (parent_space && dd->dd_parent) | 903 parent_space = parent_delta(dd, est_used, space); 904 mutex_exit(&dd->dd_lock); 905 906 /* Make sure that we clean up dd_space_to* */ 907 dsl_dir_dirty(dd, tx); 908 909 /* XXX this is potentially expensive and unnecessary... */ 910 if (parent_space && dd->dd_parent) |
875 dsl_dir_willuse_space(dd->dd_parent, parent_space, tx); | 911 dsl_dir_willuse_space_impl(dd->dd_parent, parent_space, tx); |
876} 877 | 912} 913 |
914/* 915 * Call in open context when we think we're going to write/free space, 916 * eg. when dirtying data. Be conservative (ie. OK to write less than 917 * this or free more than this, but don't write more or free less). 918 */ 919void 920dsl_dir_willuse_space(dsl_dir_t *dd, int64_t space, dmu_tx_t *tx) 921{ 922 dsl_pool_willuse_space(dd->dd_pool, space, tx); 923 dsl_dir_willuse_space_impl(dd, space, tx); 924} 925 |
|
878/* call from syncing context when we actually write/free space for this dd */ 879void | 926/* call from syncing context when we actually write/free space for this dd */ 927void |
880dsl_dir_diduse_space(dsl_dir_t *dd, | 928dsl_dir_diduse_space(dsl_dir_t *dd, dd_used_t type, |
881 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) 882{ 883 int64_t accounted_delta; | 929 int64_t used, int64_t compressed, int64_t uncompressed, dmu_tx_t *tx) 930{ 931 int64_t accounted_delta; |
932 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); |
|
884 885 ASSERT(dmu_tx_is_syncing(tx)); | 933 934 ASSERT(dmu_tx_is_syncing(tx)); |
935 ASSERT(type < DD_USED_NUM); |
|
886 887 dsl_dir_dirty(dd, tx); 888 | 936 937 dsl_dir_dirty(dd, tx); 938 |
889 mutex_enter(&dd->dd_lock); 890 accounted_delta = parent_delta(dd, dd->dd_used_bytes, used); 891 ASSERT(used >= 0 || dd->dd_used_bytes >= -used); | 939 if (needlock) 940 mutex_enter(&dd->dd_lock); 941 accounted_delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, used); 942 ASSERT(used >= 0 || dd->dd_phys->dd_used_bytes >= -used); |
892 ASSERT(compressed >= 0 || 893 dd->dd_phys->dd_compressed_bytes >= -compressed); 894 ASSERT(uncompressed >= 0 || 895 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed); | 943 ASSERT(compressed >= 0 || 944 dd->dd_phys->dd_compressed_bytes >= -compressed); 945 ASSERT(uncompressed >= 0 || 946 dd->dd_phys->dd_uncompressed_bytes >= -uncompressed); |
896 dd->dd_used_bytes += used; | 947 dd->dd_phys->dd_used_bytes += used; |
897 dd->dd_phys->dd_uncompressed_bytes += uncompressed; 898 dd->dd_phys->dd_compressed_bytes += compressed; | 948 dd->dd_phys->dd_uncompressed_bytes += uncompressed; 949 dd->dd_phys->dd_compressed_bytes += compressed; |
899 mutex_exit(&dd->dd_lock); | |
900 | 950 |
951 if (dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) { 952 ASSERT(used > 0 || 953 dd->dd_phys->dd_used_breakdown[type] >= -used); 954 dd->dd_phys->dd_used_breakdown[type] += used; 955#ifdef DEBUG 956 dd_used_t t; 957 uint64_t u = 0; 958 for (t = 0; t < DD_USED_NUM; t++) 959 u += dd->dd_phys->dd_used_breakdown[t]; 960 ASSERT3U(u, ==, dd->dd_phys->dd_used_bytes); 961#endif 962 } 963 if (needlock) 964 mutex_exit(&dd->dd_lock); 965 |
|
901 if (dd->dd_parent != NULL) { | 966 if (dd->dd_parent != NULL) { |
902 dsl_dir_diduse_space(dd->dd_parent, | 967 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, |
903 accounted_delta, compressed, uncompressed, tx); | 968 accounted_delta, compressed, uncompressed, tx); |
969 dsl_dir_transfer_space(dd->dd_parent, 970 used - accounted_delta, 971 DD_USED_CHILD_RSRV, DD_USED_CHILD, tx); |
|
904 } 905} 906 | 972 } 973} 974 |
907/* ARGSUSED */ | 975void 976dsl_dir_transfer_space(dsl_dir_t *dd, int64_t delta, 977 dd_used_t oldtype, dd_used_t newtype, dmu_tx_t *tx) 978{ 979 boolean_t needlock = !MUTEX_HELD(&dd->dd_lock); 980 981 ASSERT(dmu_tx_is_syncing(tx)); 982 ASSERT(oldtype < DD_USED_NUM); 983 ASSERT(newtype < DD_USED_NUM); 984 985 if (delta == 0 || !(dd->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN)) 986 return; 987 988 dsl_dir_dirty(dd, tx); 989 if (needlock) 990 mutex_enter(&dd->dd_lock); 991 ASSERT(delta > 0 ? 992 dd->dd_phys->dd_used_breakdown[oldtype] >= delta : 993 dd->dd_phys->dd_used_breakdown[newtype] >= -delta); 994 ASSERT(dd->dd_phys->dd_used_bytes >= ABS(delta)); 995 dd->dd_phys->dd_used_breakdown[oldtype] -= delta; 996 dd->dd_phys->dd_used_breakdown[newtype] += delta; 997 if (needlock) 998 mutex_exit(&dd->dd_lock); 999} 1000 |
908static int 909dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) 910{ 911 dsl_dir_t *dd = arg1; 912 uint64_t *quotap = arg2; 913 uint64_t new_quota = *quotap; 914 int err = 0; 915 uint64_t towrite; 916 917 if (new_quota == 0) 918 return (0); 919 920 mutex_enter(&dd->dd_lock); 921 /* 922 * If we are doing the preliminary check in open context, and 923 * there are pending changes, then don't fail it, since the | 1001static int 1002dsl_dir_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx) 1003{ 1004 dsl_dir_t *dd = arg1; 1005 uint64_t *quotap = arg2; 1006 uint64_t new_quota = *quotap; 1007 int err = 0; 1008 uint64_t towrite; 1009 1010 if (new_quota == 0) 1011 return (0); 1012 1013 mutex_enter(&dd->dd_lock); 1014 /* 1015 * If we are doing the preliminary check in open context, and 1016 * there are pending changes, then don't fail it, since the |
924 * pending changes could under-estimat the amount of space to be | 1017 * pending changes could under-estimate the amount of space to be |
925 * freed up. 926 */ | 1018 * freed up. 1019 */ |
927 towrite = dd->dd_space_towrite[0] + dd->dd_space_towrite[1] + 928 dd->dd_space_towrite[2] + dd->dd_space_towrite[3]; | 1020 towrite = dsl_dir_space_towrite(dd); |
929 if ((dmu_tx_is_syncing(tx) || towrite == 0) && 930 (new_quota < dd->dd_phys->dd_reserved || | 1021 if ((dmu_tx_is_syncing(tx) || towrite == 0) && 1022 (new_quota < dd->dd_phys->dd_reserved || |
931 new_quota < dsl_dir_estimated_space(dd))) { | 1023 new_quota < dd->dd_phys->dd_used_bytes + towrite)) { |
932 err = ENOSPC; 933 } 934 mutex_exit(&dd->dd_lock); 935 return (err); 936} 937 | 1024 err = ENOSPC; 1025 } 1026 mutex_exit(&dd->dd_lock); 1027 return (err); 1028} 1029 |
1030/* ARGSUSED */ |
|
938static void | 1031static void |
939dsl_dir_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx) | 1032dsl_dir_set_quota_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) |
940{ 941 dsl_dir_t *dd = arg1; 942 uint64_t *quotap = arg2; 943 uint64_t new_quota = *quotap; 944 945 dmu_buf_will_dirty(dd->dd_dbuf, tx); 946 947 mutex_enter(&dd->dd_lock); 948 dd->dd_phys->dd_quota = new_quota; 949 mutex_exit(&dd->dd_lock); | 1033{ 1034 dsl_dir_t *dd = arg1; 1035 uint64_t *quotap = arg2; 1036 uint64_t new_quota = *quotap; 1037 1038 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1039 1040 mutex_enter(&dd->dd_lock); 1041 dd->dd_phys->dd_quota = new_quota; 1042 mutex_exit(&dd->dd_lock); |
1043 1044 spa_history_internal_log(LOG_DS_QUOTA, dd->dd_pool->dp_spa, 1045 tx, cr, "%lld dataset = %llu ", 1046 (longlong_t)new_quota, dd->dd_phys->dd_head_dataset_obj); |
|
950} 951 952int 953dsl_dir_set_quota(const char *ddname, uint64_t quota) 954{ 955 dsl_dir_t *dd; 956 int err; 957 958 err = dsl_dir_open(ddname, FTAG, &dd, NULL); 959 if (err) 960 return (err); | 1047} 1048 1049int 1050dsl_dir_set_quota(const char *ddname, uint64_t quota) 1051{ 1052 dsl_dir_t *dd; 1053 int err; 1054 1055 err = dsl_dir_open(ddname, FTAG, &dd, NULL); 1056 if (err) 1057 return (err); |
961 /* 962 * If someone removes a file, then tries to set the quota, we 963 * want to make sure the file freeing takes effect. 964 */ 965 txg_wait_open(dd->dd_pool, 0); | |
966 | 1058 |
967 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check, 968 dsl_dir_set_quota_sync, dd, "a, 0); | 1059 if (quota != dd->dd_phys->dd_quota) { 1060 /* 1061 * If someone removes a file, then tries to set the quota, we 1062 * want to make sure the file freeing takes effect. 1063 */ 1064 txg_wait_open(dd->dd_pool, 0); 1065 1066 err = dsl_sync_task_do(dd->dd_pool, dsl_dir_set_quota_check, 1067 dsl_dir_set_quota_sync, dd, "a, 0); 1068 } |
969 dsl_dir_close(dd, FTAG); 970 return (err); 971} 972 | 1069 dsl_dir_close(dd, FTAG); 1070 return (err); 1071} 1072 |
973/* ARGSUSED */ 974static int | 1073int |
975dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) 976{ 977 dsl_dir_t *dd = arg1; 978 uint64_t *reservationp = arg2; 979 uint64_t new_reservation = *reservationp; 980 uint64_t used, avail; 981 int64_t delta; 982 983 if (new_reservation > INT64_MAX) 984 return (EOVERFLOW); 985 986 /* 987 * If we are doing the preliminary check in open context, the 988 * space estimates may be inaccurate. 989 */ 990 if (!dmu_tx_is_syncing(tx)) 991 return (0); 992 993 mutex_enter(&dd->dd_lock); | 1074dsl_dir_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx) 1075{ 1076 dsl_dir_t *dd = arg1; 1077 uint64_t *reservationp = arg2; 1078 uint64_t new_reservation = *reservationp; 1079 uint64_t used, avail; 1080 int64_t delta; 1081 1082 if (new_reservation > INT64_MAX) 1083 return (EOVERFLOW); 1084 1085 /* 1086 * If we are doing the preliminary check in open context, the 1087 * space estimates may be inaccurate. 1088 */ 1089 if (!dmu_tx_is_syncing(tx)) 1090 return (0); 1091 1092 mutex_enter(&dd->dd_lock); |
994 used = dd->dd_used_bytes; | 1093 used = dd->dd_phys->dd_used_bytes; |
995 delta = MAX(used, new_reservation) - 996 MAX(used, dd->dd_phys->dd_reserved); 997 mutex_exit(&dd->dd_lock); 998 999 if (dd->dd_parent) { 1000 avail = dsl_dir_space_available(dd->dd_parent, 1001 NULL, 0, FALSE); 1002 } else { 1003 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used; 1004 } 1005 1006 if (delta > 0 && delta > avail) 1007 return (ENOSPC); 1008 if (delta > 0 && dd->dd_phys->dd_quota > 0 && 1009 new_reservation > dd->dd_phys->dd_quota) 1010 return (ENOSPC); 1011 return (0); 1012} 1013 | 1094 delta = MAX(used, new_reservation) - 1095 MAX(used, dd->dd_phys->dd_reserved); 1096 mutex_exit(&dd->dd_lock); 1097 1098 if (dd->dd_parent) { 1099 avail = dsl_dir_space_available(dd->dd_parent, 1100 NULL, 0, FALSE); 1101 } else { 1102 avail = dsl_pool_adjustedsize(dd->dd_pool, B_FALSE) - used; 1103 } 1104 1105 if (delta > 0 && delta > avail) 1106 return (ENOSPC); 1107 if (delta > 0 && dd->dd_phys->dd_quota > 0 && 1108 new_reservation > dd->dd_phys->dd_quota) 1109 return (ENOSPC); 1110 return (0); 1111} 1112 |
1113/* ARGSUSED */ |
|
1014static void | 1114static void |
1015dsl_dir_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx) | 1115dsl_dir_set_reservation_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) |
1016{ 1017 dsl_dir_t *dd = arg1; 1018 uint64_t *reservationp = arg2; 1019 uint64_t new_reservation = *reservationp; 1020 uint64_t used; 1021 int64_t delta; 1022 | 1116{ 1117 dsl_dir_t *dd = arg1; 1118 uint64_t *reservationp = arg2; 1119 uint64_t new_reservation = *reservationp; 1120 uint64_t used; 1121 int64_t delta; 1122 |
1123 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1124 |
|
1023 mutex_enter(&dd->dd_lock); | 1125 mutex_enter(&dd->dd_lock); |
1024 used = dd->dd_used_bytes; | 1126 used = dd->dd_phys->dd_used_bytes; |
1025 delta = MAX(used, new_reservation) - 1026 MAX(used, dd->dd_phys->dd_reserved); | 1127 delta = MAX(used, new_reservation) - 1128 MAX(used, dd->dd_phys->dd_reserved); |
1027 mutex_exit(&dd->dd_lock); 1028 1029 dmu_buf_will_dirty(dd->dd_dbuf, tx); | |
1030 dd->dd_phys->dd_reserved = new_reservation; 1031 1032 if (dd->dd_parent != NULL) { 1033 /* Roll up this additional usage into our ancestors */ | 1129 dd->dd_phys->dd_reserved = new_reservation; 1130 1131 if (dd->dd_parent != NULL) { 1132 /* Roll up this additional usage into our ancestors */ |
1034 dsl_dir_diduse_space(dd->dd_parent, delta, 0, 0, tx); | 1133 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 1134 delta, 0, 0, tx); |
1035 } | 1135 } |
1136 mutex_exit(&dd->dd_lock); 1137 1138 spa_history_internal_log(LOG_DS_RESERVATION, dd->dd_pool->dp_spa, 1139 tx, cr, "%lld dataset = %llu", 1140 (longlong_t)new_reservation, dd->dd_phys->dd_head_dataset_obj); |
|
1036} 1037 1038int 1039dsl_dir_set_reservation(const char *ddname, uint64_t reservation) 1040{ 1041 dsl_dir_t *dd; 1042 int err; 1043 --- 25 unchanged lines hidden (view full) --- 1069 */ 1070static int64_t 1071would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) 1072{ 1073 if (dd == ancestor) 1074 return (delta); 1075 1076 mutex_enter(&dd->dd_lock); | 1141} 1142 1143int 1144dsl_dir_set_reservation(const char *ddname, uint64_t reservation) 1145{ 1146 dsl_dir_t *dd; 1147 int err; 1148 --- 25 unchanged lines hidden (view full) --- 1174 */ 1175static int64_t 1176would_change(dsl_dir_t *dd, int64_t delta, dsl_dir_t *ancestor) 1177{ 1178 if (dd == ancestor) 1179 return (delta); 1180 1181 mutex_enter(&dd->dd_lock); |
1077 delta = parent_delta(dd, dd->dd_used_bytes, delta); | 1182 delta = parent_delta(dd, dd->dd_phys->dd_used_bytes, delta); |
1078 mutex_exit(&dd->dd_lock); 1079 return (would_change(dd->dd_parent, delta, ancestor)); 1080} 1081 1082struct renamearg { 1083 dsl_dir_t *newparent; 1084 const char *mynewname; 1085}; 1086 | 1183 mutex_exit(&dd->dd_lock); 1184 return (would_change(dd->dd_parent, delta, ancestor)); 1185} 1186 1187struct renamearg { 1188 dsl_dir_t *newparent; 1189 const char *mynewname; 1190}; 1191 |
1087/* ARGSUSED */ | 1192/*ARGSUSED*/ |
1088static int 1089dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) 1090{ 1091 dsl_dir_t *dd = arg1; 1092 struct renamearg *ra = arg2; 1093 dsl_pool_t *dp = dd->dd_pool; 1094 objset_t *mos = dp->dp_meta_objset; 1095 int err; --- 9 unchanged lines hidden (view full) --- 1105 if (err == 0) 1106 return (EEXIST); 1107 if (err != ENOENT) 1108 return (err); 1109 1110 if (ra->newparent != dd->dd_parent) { 1111 /* is there enough space? */ 1112 uint64_t myspace = | 1193static int 1194dsl_dir_rename_check(void *arg1, void *arg2, dmu_tx_t *tx) 1195{ 1196 dsl_dir_t *dd = arg1; 1197 struct renamearg *ra = arg2; 1198 dsl_pool_t *dp = dd->dd_pool; 1199 objset_t *mos = dp->dp_meta_objset; 1200 int err; --- 9 unchanged lines hidden (view full) --- 1210 if (err == 0) 1211 return (EEXIST); 1212 if (err != ENOENT) 1213 return (err); 1214 1215 if (ra->newparent != dd->dd_parent) { 1216 /* is there enough space? */ 1217 uint64_t myspace = |
1113 MAX(dd->dd_used_bytes, dd->dd_phys->dd_reserved); | 1218 MAX(dd->dd_phys->dd_used_bytes, dd->dd_phys->dd_reserved); |
1114 1115 /* no rename into our descendant */ 1116 if (closest_common_ancestor(dd, ra->newparent) == dd) 1117 return (EINVAL); 1118 1119 if (err = dsl_dir_transfer_possible(dd->dd_parent, 1120 ra->newparent, myspace)) 1121 return (err); 1122 } 1123 1124 return (0); 1125} 1126 1127static void | 1219 1220 /* no rename into our descendant */ 1221 if (closest_common_ancestor(dd, ra->newparent) == dd) 1222 return (EINVAL); 1223 1224 if (err = dsl_dir_transfer_possible(dd->dd_parent, 1225 ra->newparent, myspace)) 1226 return (err); 1227 } 1228 1229 return (0); 1230} 1231 1232static void |
1128dsl_dir_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx) | 1233dsl_dir_rename_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx) |
1129{ 1130 dsl_dir_t *dd = arg1; 1131 struct renamearg *ra = arg2; 1132 dsl_pool_t *dp = dd->dd_pool; 1133 objset_t *mos = dp->dp_meta_objset; 1134 int err; 1135 1136 ASSERT(dmu_buf_refcount(dd->dd_dbuf) <= 2); 1137 1138 if (ra->newparent != dd->dd_parent) { | 1234{ 1235 dsl_dir_t *dd = arg1; 1236 struct renamearg *ra = arg2; 1237 dsl_pool_t *dp = dd->dd_pool; 1238 objset_t *mos = dp->dp_meta_objset; 1239 int err; 1240 1241 ASSERT(dmu_buf_refcount(dd->dd_dbuf) <= 2); 1242 1243 if (ra->newparent != dd->dd_parent) { |
1139 uint64_t myspace = 1140 MAX(dd->dd_used_bytes, dd->dd_phys->dd_reserved); 1141 1142 dsl_dir_diduse_space(dd->dd_parent, -myspace, | 1244 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD, 1245 -dd->dd_phys->dd_used_bytes, |
1143 -dd->dd_phys->dd_compressed_bytes, 1144 -dd->dd_phys->dd_uncompressed_bytes, tx); | 1246 -dd->dd_phys->dd_compressed_bytes, 1247 -dd->dd_phys->dd_uncompressed_bytes, tx); |
1145 dsl_dir_diduse_space(ra->newparent, myspace, | 1248 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD, 1249 dd->dd_phys->dd_used_bytes, |
1146 dd->dd_phys->dd_compressed_bytes, 1147 dd->dd_phys->dd_uncompressed_bytes, tx); | 1250 dd->dd_phys->dd_compressed_bytes, 1251 dd->dd_phys->dd_uncompressed_bytes, tx); |
1252 1253 if (dd->dd_phys->dd_reserved > dd->dd_phys->dd_used_bytes) { 1254 uint64_t unused_rsrv = dd->dd_phys->dd_reserved - 1255 dd->dd_phys->dd_used_bytes; 1256 1257 dsl_dir_diduse_space(dd->dd_parent, DD_USED_CHILD_RSRV, 1258 -unused_rsrv, 0, 0, tx); 1259 dsl_dir_diduse_space(ra->newparent, DD_USED_CHILD_RSRV, 1260 unused_rsrv, 0, 0, tx); 1261 } |
|
1148 } 1149 1150 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1151 1152 /* remove from old parent zapobj */ 1153 err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj, 1154 dd->dd_myname, tx); 1155 ASSERT3U(err, ==, 0); 1156 1157 (void) strcpy(dd->dd_myname, ra->mynewname); 1158 dsl_dir_close(dd->dd_parent, dd); 1159 dd->dd_phys->dd_parent_obj = ra->newparent->dd_object; 1160 VERIFY(0 == dsl_dir_open_obj(dd->dd_pool, 1161 ra->newparent->dd_object, NULL, dd, &dd->dd_parent)); 1162 1163 /* add to new parent zapobj */ 1164 err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj, 1165 dd->dd_myname, 8, 1, &dd->dd_object, tx); 1166 ASSERT3U(err, ==, 0); | 1262 } 1263 1264 dmu_buf_will_dirty(dd->dd_dbuf, tx); 1265 1266 /* remove from old parent zapobj */ 1267 err = zap_remove(mos, dd->dd_parent->dd_phys->dd_child_dir_zapobj, 1268 dd->dd_myname, tx); 1269 ASSERT3U(err, ==, 0); 1270 1271 (void) strcpy(dd->dd_myname, ra->mynewname); 1272 dsl_dir_close(dd->dd_parent, dd); 1273 dd->dd_phys->dd_parent_obj = ra->newparent->dd_object; 1274 VERIFY(0 == dsl_dir_open_obj(dd->dd_pool, 1275 ra->newparent->dd_object, NULL, dd, &dd->dd_parent)); 1276 1277 /* add to new parent zapobj */ 1278 err = zap_add(mos, ra->newparent->dd_phys->dd_child_dir_zapobj, 1279 dd->dd_myname, 8, 1, &dd->dd_object, tx); 1280 ASSERT3U(err, ==, 0); |
1281 1282 spa_history_internal_log(LOG_DS_RENAME, dd->dd_pool->dp_spa, 1283 tx, cr, "dataset = %llu", dd->dd_phys->dd_head_dataset_obj); |
|
1167} 1168 1169int 1170dsl_dir_rename(dsl_dir_t *dd, const char *newname) 1171{ 1172 struct renamearg ra; 1173 int err; 1174 --- 9 unchanged lines hidden (view full) --- 1184 } 1185 1186 /* new name should not already exist */ 1187 if (ra.mynewname == NULL) { 1188 err = EEXIST; 1189 goto out; 1190 } 1191 | 1284} 1285 1286int 1287dsl_dir_rename(dsl_dir_t *dd, const char *newname) 1288{ 1289 struct renamearg ra; 1290 int err; 1291 --- 9 unchanged lines hidden (view full) --- 1301 } 1302 1303 /* new name should not already exist */ 1304 if (ra.mynewname == NULL) { 1305 err = EEXIST; 1306 goto out; 1307 } 1308 |
1192 | |
1193 err = dsl_sync_task_do(dd->dd_pool, 1194 dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3); 1195 1196out: 1197 dsl_dir_close(ra.newparent, FTAG); 1198 return (err); 1199} 1200 --- 15 unchanged lines hidden --- | 1309 err = dsl_sync_task_do(dd->dd_pool, 1310 dsl_dir_rename_check, dsl_dir_rename_sync, dd, &ra, 3); 1311 1312out: 1313 dsl_dir_close(ra.newparent, FTAG); 1314 return (err); 1315} 1316 --- 15 unchanged lines hidden --- |