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 https://opensource.org/licenses/CDDL-1.0. 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, 2018 by Delphix. All rights reserved. 24 */ 25 26 /* Portions Copyright 2010 Robert Milkowski */ 27 28 #ifndef _SYS_ZIL_IMPL_H 29 #define _SYS_ZIL_IMPL_H 30 31 #include <sys/zil.h> 32 #include <sys/dmu_objset.h> 33 34 #ifdef __cplusplus 35 extern "C" { 36 #endif 37 38 /* 39 * Possible states for a given lwb structure. 40 * 41 * An lwb will start out in the "closed" state, and then transition to 42 * the "opened" state via a call to zil_lwb_write_open(). When 43 * transitioning from "closed" to "opened" the zilog's "zl_issuer_lock" 44 * must be held. 45 * 46 * After the lwb is "opened", it can transition into the "issued" state 47 * via zil_lwb_write_issue(). Again, the zilog's "zl_issuer_lock" must 48 * be held when making this transition. 49 * 50 * After the lwb's write zio completes, it transitions into the "write 51 * done" state via zil_lwb_write_done(); and then into the "flush done" 52 * state via zil_lwb_flush_vdevs_done(). When transitioning from 53 * "issued" to "write done", and then from "write done" to "flush done", 54 * the zilog's "zl_lock" must be held, *not* the "zl_issuer_lock". 55 * 56 * The zilog's "zl_issuer_lock" can become heavily contended in certain 57 * workloads, so we specifically avoid acquiring that lock when 58 * transitioning an lwb from "issued" to "done". This allows us to avoid 59 * having to acquire the "zl_issuer_lock" for each lwb ZIO completion, 60 * which would have added more lock contention on an already heavily 61 * contended lock. 62 * 63 * Additionally, correctness when reading an lwb's state is often 64 * achieved by exploiting the fact that these state transitions occur in 65 * this specific order; i.e. "closed" to "opened" to "issued" to "done". 66 * 67 * Thus, if an lwb is in the "closed" or "opened" state, holding the 68 * "zl_issuer_lock" will prevent a concurrent thread from transitioning 69 * that lwb to the "issued" state. Likewise, if an lwb is already in the 70 * "issued" state, holding the "zl_lock" will prevent a concurrent 71 * thread from transitioning that lwb to the "write done" state. 72 */ 73 typedef enum { 74 LWB_STATE_CLOSED, 75 LWB_STATE_OPENED, 76 LWB_STATE_ISSUED, 77 LWB_STATE_WRITE_DONE, 78 LWB_STATE_FLUSH_DONE, 79 LWB_NUM_STATES 80 } lwb_state_t; 81 82 /* 83 * Log write block (lwb) 84 * 85 * Prior to an lwb being issued to disk via zil_lwb_write_issue(), it 86 * will be protected by the zilog's "zl_issuer_lock". Basically, prior 87 * to it being issued, it will only be accessed by the thread that's 88 * holding the "zl_issuer_lock". After the lwb is issued, the zilog's 89 * "zl_lock" is used to protect the lwb against concurrent access. 90 */ 91 typedef struct lwb { 92 zilog_t *lwb_zilog; /* back pointer to log struct */ 93 blkptr_t lwb_blk; /* on disk address of this log blk */ 94 boolean_t lwb_fastwrite; /* is blk marked for fastwrite? */ 95 boolean_t lwb_slog; /* lwb_blk is on SLOG device */ 96 int lwb_nused; /* # used bytes in buffer */ 97 int lwb_sz; /* size of block and buffer */ 98 lwb_state_t lwb_state; /* the state of this lwb */ 99 char *lwb_buf; /* log write buffer */ 100 zio_t *lwb_write_zio; /* zio for the lwb buffer */ 101 zio_t *lwb_root_zio; /* root zio for lwb write and flushes */ 102 uint64_t lwb_issued_txg; /* the txg when the write is issued */ 103 uint64_t lwb_max_txg; /* highest txg in this lwb */ 104 list_node_t lwb_node; /* zilog->zl_lwb_list linkage */ 105 list_t lwb_itxs; /* list of itx's */ 106 list_t lwb_waiters; /* list of zil_commit_waiter's */ 107 avl_tree_t lwb_vdev_tree; /* vdevs to flush after lwb write */ 108 kmutex_t lwb_vdev_lock; /* protects lwb_vdev_tree */ 109 hrtime_t lwb_issued_timestamp; /* when was the lwb issued? */ 110 } lwb_t; 111 112 /* 113 * ZIL commit waiter. 114 * 115 * This structure is allocated each time zil_commit() is called, and is 116 * used by zil_commit() to communicate with other parts of the ZIL, such 117 * that zil_commit() can know when it safe for it return. For more 118 * details, see the comment above zil_commit(). 119 * 120 * The "zcw_lock" field is used to protect the commit waiter against 121 * concurrent access. This lock is often acquired while already holding 122 * the zilog's "zl_issuer_lock" or "zl_lock"; see the functions 123 * zil_process_commit_list() and zil_lwb_flush_vdevs_done() as examples 124 * of this. Thus, one must be careful not to acquire the 125 * "zl_issuer_lock" or "zl_lock" when already holding the "zcw_lock"; 126 * e.g. see the zil_commit_waiter_timeout() function. 127 */ 128 typedef struct zil_commit_waiter { 129 kcondvar_t zcw_cv; /* signalled when "done" */ 130 kmutex_t zcw_lock; /* protects fields of this struct */ 131 list_node_t zcw_node; /* linkage in lwb_t:lwb_waiter list */ 132 lwb_t *zcw_lwb; /* back pointer to lwb when linked */ 133 boolean_t zcw_done; /* B_TRUE when "done", else B_FALSE */ 134 int zcw_zio_error; /* contains the zio io_error value */ 135 } zil_commit_waiter_t; 136 137 /* 138 * Intent log transaction lists 139 */ 140 typedef struct itxs { 141 list_t i_sync_list; /* list of synchronous itxs */ 142 avl_tree_t i_async_tree; /* tree of foids for async itxs */ 143 } itxs_t; 144 145 typedef struct itxg { 146 kmutex_t itxg_lock; /* lock for this structure */ 147 uint64_t itxg_txg; /* txg for this chain */ 148 itxs_t *itxg_itxs; /* sync and async itxs */ 149 } itxg_t; 150 151 /* for async nodes we build up an AVL tree of lists of async itxs per file */ 152 typedef struct itx_async_node { 153 uint64_t ia_foid; /* file object id */ 154 list_t ia_list; /* list of async itxs for this foid */ 155 avl_node_t ia_node; /* AVL tree linkage */ 156 } itx_async_node_t; 157 158 /* 159 * Vdev flushing: during a zil_commit(), we build up an AVL tree of the vdevs 160 * we've touched so we know which ones need a write cache flush at the end. 161 */ 162 typedef struct zil_vdev_node { 163 uint64_t zv_vdev; /* vdev to be flushed */ 164 avl_node_t zv_node; /* AVL tree linkage */ 165 } zil_vdev_node_t; 166 167 #define ZIL_PREV_BLKS 16 168 169 /* 170 * Stable storage intent log management structure. One per dataset. 171 */ 172 struct zilog { 173 kmutex_t zl_lock; /* protects most zilog_t fields */ 174 struct dsl_pool *zl_dmu_pool; /* DSL pool */ 175 spa_t *zl_spa; /* handle for read/write log */ 176 const zil_header_t *zl_header; /* log header buffer */ 177 objset_t *zl_os; /* object set we're logging */ 178 zil_get_data_t *zl_get_data; /* callback to get object content */ 179 lwb_t *zl_last_lwb_opened; /* most recent lwb opened */ 180 hrtime_t zl_last_lwb_latency; /* zio latency of last lwb done */ 181 uint64_t zl_lr_seq; /* on-disk log record sequence number */ 182 uint64_t zl_commit_lr_seq; /* last committed on-disk lr seq */ 183 uint64_t zl_destroy_txg; /* txg of last zil_destroy() */ 184 uint64_t zl_replayed_seq[TXG_SIZE]; /* last replayed rec seq */ 185 uint64_t zl_replaying_seq; /* current replay seq number */ 186 uint32_t zl_suspend; /* log suspend count */ 187 kcondvar_t zl_cv_suspend; /* log suspend completion */ 188 uint8_t zl_suspending; /* log is currently suspending */ 189 uint8_t zl_keep_first; /* keep first log block in destroy */ 190 uint8_t zl_replay; /* replaying records while set */ 191 uint8_t zl_stop_sync; /* for debugging */ 192 kmutex_t zl_issuer_lock; /* single writer, per ZIL, at a time */ 193 uint8_t zl_logbias; /* latency or throughput */ 194 uint8_t zl_sync; /* synchronous or asynchronous */ 195 int zl_parse_error; /* last zil_parse() error */ 196 uint64_t zl_parse_blk_seq; /* highest blk seq on last parse */ 197 uint64_t zl_parse_lr_seq; /* highest lr seq on last parse */ 198 uint64_t zl_parse_blk_count; /* number of blocks parsed */ 199 uint64_t zl_parse_lr_count; /* number of log records parsed */ 200 itxg_t zl_itxg[TXG_SIZE]; /* intent log txg chains */ 201 list_t zl_itx_commit_list; /* itx list to be committed */ 202 uint64_t zl_cur_used; /* current commit log size used */ 203 list_t zl_lwb_list; /* in-flight log write list */ 204 avl_tree_t zl_bp_tree; /* track bps during log parse */ 205 clock_t zl_replay_time; /* lbolt of when replay started */ 206 uint64_t zl_replay_blks; /* number of log blocks replayed */ 207 zil_header_t zl_old_header; /* debugging aid */ 208 uint_t zl_prev_blks[ZIL_PREV_BLKS]; /* size - sector rounded */ 209 uint_t zl_prev_rotor; /* rotor for zl_prev[] */ 210 txg_node_t zl_dirty_link; /* protected by dp_dirty_zilogs list */ 211 uint64_t zl_dirty_max_txg; /* highest txg used to dirty zilog */ 212 213 kmutex_t zl_lwb_io_lock; /* protect following members */ 214 uint64_t zl_lwb_inflight[TXG_SIZE]; /* io issued, but not done */ 215 kcondvar_t zl_lwb_io_cv; /* signal when the flush is done */ 216 uint64_t zl_lwb_max_issued_txg; /* max txg when lwb io issued */ 217 218 /* 219 * Max block size for this ZIL. Note that this can not be changed 220 * while the ZIL is in use because consumers (ZPL/zvol) need to take 221 * this into account when deciding between WR_COPIED and WR_NEED_COPY 222 * (see zil_max_copied_data()). 223 */ 224 uint64_t zl_max_block_size; 225 226 /* Pointer for per dataset zil sums */ 227 zil_sums_t *zl_sums; 228 }; 229 230 typedef struct zil_bp_node { 231 dva_t zn_dva; 232 avl_node_t zn_node; 233 } zil_bp_node_t; 234 235 #ifdef __cplusplus 236 } 237 #endif 238 239 #endif /* _SYS_ZIL_IMPL_H */ 240