1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2017 by Delphix. All rights reserved. 24 */ 25 26 #include <sys/dmu.h> 27 #include <sys/dmu_tx.h> 28 #include <sys/dsl_pool.h> 29 #include <sys/dsl_dir.h> 30 #include <sys/dsl_synctask.h> 31 #include <sys/metaslab.h> 32 33 #define DST_AVG_BLKSHIFT 14 34 35 /* ARGSUSED */ 36 static int 37 dsl_null_checkfunc(void *arg, dmu_tx_t *tx) 38 { 39 return (0); 40 } 41 42 static int 43 dsl_sync_task_common(const char *pool, dsl_checkfunc_t *checkfunc, 44 dsl_syncfunc_t *syncfunc, void *arg, 45 int blocks_modified, zfs_space_check_t space_check, boolean_t early) 46 { 47 spa_t *spa; 48 dmu_tx_t *tx; 49 int err; 50 dsl_sync_task_t dst = { 0 }; 51 dsl_pool_t *dp; 52 53 err = spa_open(pool, &spa, FTAG); 54 if (err != 0) 55 return (err); 56 dp = spa_get_dsl(spa); 57 58 top: 59 tx = dmu_tx_create_dd(dp->dp_mos_dir); 60 VERIFY0(dmu_tx_assign(tx, TXG_WAIT)); 61 62 dst.dst_pool = dp; 63 dst.dst_txg = dmu_tx_get_txg(tx); 64 dst.dst_space = blocks_modified << DST_AVG_BLKSHIFT; 65 dst.dst_space_check = space_check; 66 dst.dst_checkfunc = checkfunc != NULL ? checkfunc : dsl_null_checkfunc; 67 dst.dst_syncfunc = syncfunc; 68 dst.dst_arg = arg; 69 dst.dst_error = 0; 70 dst.dst_nowaiter = B_FALSE; 71 72 dsl_pool_config_enter(dp, FTAG); 73 err = dst.dst_checkfunc(arg, tx); 74 dsl_pool_config_exit(dp, FTAG); 75 76 if (err != 0) { 77 dmu_tx_commit(tx); 78 spa_close(spa, FTAG); 79 return (err); 80 } 81 82 txg_list_t *task_list = (early) ? 83 &dp->dp_early_sync_tasks : &dp->dp_sync_tasks; 84 VERIFY(txg_list_add_tail(task_list, &dst, dst.dst_txg)); 85 86 dmu_tx_commit(tx); 87 88 txg_wait_synced(dp, dst.dst_txg); 89 90 if (dst.dst_error == EAGAIN) { 91 txg_wait_synced(dp, dst.dst_txg + TXG_DEFER_SIZE); 92 goto top; 93 } 94 95 spa_close(spa, FTAG); 96 return (dst.dst_error); 97 } 98 99 /* 100 * Called from open context to perform a callback in syncing context. Waits 101 * for the operation to complete. 102 * 103 * The checkfunc will be called from open context as a preliminary check 104 * which can quickly fail. If it succeeds, it will be called again from 105 * syncing context. The checkfunc should generally be designed to work 106 * properly in either context, but if necessary it can check 107 * dmu_tx_is_syncing(tx). 108 * 109 * The synctask infrastructure enforces proper locking strategy with respect 110 * to the dp_config_rwlock -- the lock will always be held when the callbacks 111 * are called. It will be held for read during the open-context (preliminary) 112 * call to the checkfunc, and then held for write from syncing context during 113 * the calls to the check and sync funcs. 114 * 115 * A dataset or pool name can be passed as the first argument. Typically, 116 * the check func will hold, check the return value of the hold, and then 117 * release the dataset. The sync func will VERIFYO(hold()) the dataset. 118 * This is safe because no changes can be made between the check and sync funcs, 119 * and the sync func will only be called if the check func successfully opened 120 * the dataset. 121 */ 122 int 123 dsl_sync_task(const char *pool, dsl_checkfunc_t *checkfunc, 124 dsl_syncfunc_t *syncfunc, void *arg, 125 int blocks_modified, zfs_space_check_t space_check) 126 { 127 return (dsl_sync_task_common(pool, checkfunc, syncfunc, arg, 128 blocks_modified, space_check, B_FALSE)); 129 } 130 131 /* 132 * An early synctask works exactly as a standard synctask with one important 133 * difference on the way it is handled during syncing context. Standard 134 * synctasks run after we've written out all the dirty blocks of dirty 135 * datasets. Early synctasks are executed before writing out any dirty data, 136 * and thus before standard synctasks. 137 * 138 * For that reason, early synctasks can affect the process of writing dirty 139 * changes to disk for the txg that they run and should be used with caution. 140 * In addition, early synctasks should not dirty any metaslabs as this would 141 * invalidate the precodition/invariant for subsequent early synctasks. 142 * [see dsl_pool_sync() and dsl_early_sync_task_verify()] 143 */ 144 int 145 dsl_early_sync_task(const char *pool, dsl_checkfunc_t *checkfunc, 146 dsl_syncfunc_t *syncfunc, void *arg, 147 int blocks_modified, zfs_space_check_t space_check) 148 { 149 return (dsl_sync_task_common(pool, checkfunc, syncfunc, arg, 150 blocks_modified, space_check, B_TRUE)); 151 } 152 153 static void 154 dsl_sync_task_nowait_common(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg, 155 int blocks_modified, zfs_space_check_t space_check, dmu_tx_t *tx, 156 boolean_t early) 157 { 158 dsl_sync_task_t *dst = kmem_zalloc(sizeof (*dst), KM_SLEEP); 159 160 dst->dst_pool = dp; 161 dst->dst_txg = dmu_tx_get_txg(tx); 162 dst->dst_space = blocks_modified << DST_AVG_BLKSHIFT; 163 dst->dst_space_check = space_check; 164 dst->dst_checkfunc = dsl_null_checkfunc; 165 dst->dst_syncfunc = syncfunc; 166 dst->dst_arg = arg; 167 dst->dst_error = 0; 168 dst->dst_nowaiter = B_TRUE; 169 170 txg_list_t *task_list = (early) ? 171 &dp->dp_early_sync_tasks : &dp->dp_sync_tasks; 172 VERIFY(txg_list_add_tail(task_list, dst, dst->dst_txg)); 173 } 174 175 void 176 dsl_sync_task_nowait(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg, 177 int blocks_modified, zfs_space_check_t space_check, dmu_tx_t *tx) 178 { 179 dsl_sync_task_nowait_common(dp, syncfunc, arg, 180 blocks_modified, space_check, tx, B_FALSE); 181 } 182 183 void 184 dsl_early_sync_task_nowait(dsl_pool_t *dp, dsl_syncfunc_t *syncfunc, void *arg, 185 int blocks_modified, zfs_space_check_t space_check, dmu_tx_t *tx) 186 { 187 dsl_sync_task_nowait_common(dp, syncfunc, arg, 188 blocks_modified, space_check, tx, B_TRUE); 189 } 190 191 /* 192 * Called in syncing context to execute the synctask. 193 */ 194 void 195 dsl_sync_task_sync(dsl_sync_task_t *dst, dmu_tx_t *tx) 196 { 197 dsl_pool_t *dp = dst->dst_pool; 198 199 ASSERT0(dst->dst_error); 200 201 /* 202 * Check for sufficient space. 203 * 204 * When the sync task was created, the caller specified the 205 * type of space checking required. See the comment in 206 * zfs_space_check_t for details on the semantics of each 207 * type of space checking. 208 * 209 * We just check against what's on-disk; we don't want any 210 * in-flight accounting to get in our way, because open context 211 * may have already used up various in-core limits 212 * (arc_tempreserve, dsl_pool_tempreserve). 213 */ 214 if (dst->dst_space_check != ZFS_SPACE_CHECK_NONE) { 215 uint64_t quota = dsl_pool_unreserved_space(dp, 216 dst->dst_space_check); 217 uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes; 218 219 /* MOS space is triple-dittoed, so we multiply by 3. */ 220 if (used + dst->dst_space * 3 > quota) { 221 dst->dst_error = SET_ERROR(ENOSPC); 222 if (dst->dst_nowaiter) 223 kmem_free(dst, sizeof (*dst)); 224 return; 225 } 226 } 227 228 /* 229 * Check for errors by calling checkfunc. 230 */ 231 rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG); 232 dst->dst_error = dst->dst_checkfunc(dst->dst_arg, tx); 233 if (dst->dst_error == 0) 234 dst->dst_syncfunc(dst->dst_arg, tx); 235 rrw_exit(&dp->dp_config_rwlock, FTAG); 236 if (dst->dst_nowaiter) 237 kmem_free(dst, sizeof (*dst)); 238 } 239