include/sys/zil_impl.h

   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 "new" state, and transition to the "opened"
  42  * state via a call to zil_lwb_write_open() on first itx assignment.  When
  43  * transitioning from "new" to "opened" the zilog's "zl_issuer_lock" must be
  44  * held.
  45  *
  46  * After the lwb is "opened", it can be assigned number of itxs and transition
  47  * into the "closed" state via zil_lwb_write_close() when full or on timeout.
  48  * When transitioning from "opened" to "closed" the zilog's "zl_issuer_lock"
  49  * must be held.  New lwb allocation also takes "zl_lock" to protect the list.
  50  *
  51  * After the lwb is "closed", it can transition into the "ready" state via
  52  * zil_lwb_write_issue().  "zl_lock" must be held when making this transition.
  53  * Since it is done by the same thread, "zl_issuer_lock" is not needed.
  54  *
  55  * When lwb in "ready" state receives its block pointer, it can transition to
  56  * "issued". "zl_lock" must be held when making this transition.
  57  *
  58  * After the lwb's write zio completes, it transitions into the "write
  59  * done" state via zil_lwb_write_done(); and then into the "flush done"
  60  * state via zil_lwb_flush_vdevs_done(). When transitioning from
  61  * "issued" to "write done", and then from "write done" to "flush done",
  62  * the zilog's "zl_lock" must be held, *not* the "zl_issuer_lock".
  63  *
  64  * The zilog's "zl_issuer_lock" can become heavily contended in certain
  65  * workloads, so we specifically avoid acquiring that lock when
  66  * transitioning an lwb from "issued" to "done". This allows us to avoid
  67  * having to acquire the "zl_issuer_lock" for each lwb ZIO completion,
  68  * which would have added more lock contention on an already heavily
  69  * contended lock.
  70  *
  71  * Additionally, correctness when reading an lwb's state is often
  72  * achieved by exploiting the fact that these state transitions occur in
  73  * this specific order; i.e. "new" to "opened" to "closed" to "ready" to
  74  * "issued" to "write_done" and finally "flush_done".
  75  *
  76  * Thus, if an lwb is in the "new" or "opened" state, holding the
  77  * "zl_issuer_lock" will prevent a concurrent thread from transitioning
  78  * that lwb to the "closed" state. Likewise, if an lwb is already in the
  79  * "ready" state, holding the "zl_lock" will prevent a concurrent thread
  80  * from transitioning that lwb to the "issued" state.
  81  */
  82 typedef enum {
  83     LWB_STATE_NEW,
  84     LWB_STATE_OPENED,
  85     LWB_STATE_CLOSED,
  86     LWB_STATE_READY,
  87     LWB_STATE_ISSUED,
  88     LWB_STATE_WRITE_DONE,
  89     LWB_STATE_FLUSH_DONE,
  90     LWB_NUM_STATES
  91 } lwb_state_t;
  92
  93 /*
  94  * Log write block (lwb)
  95  *
  96  * Prior to an lwb being issued to disk via zil_lwb_write_issue(), it
  97  * will be protected by the zilog's "zl_issuer_lock". Basically, prior
  98  * to it being issued, it will only be accessed by the thread that's
  99  * holding the "zl_issuer_lock". After the lwb is issued, the zilog's
 100  * "zl_lock" is used to protect the lwb against concurrent access.
 101  */
 102 typedef struct lwb {
 103         zilog_t         *lwb_zilog;     /* back pointer to log struct */
 104         blkptr_t        lwb_blk;        /* on disk address of this log blk */
 105         boolean_t       lwb_slim;       /* log block has slim format */
 106         boolean_t       lwb_slog;       /* lwb_blk is on SLOG device */
 107         int             lwb_error;      /* log block allocation error */
 108         int             lwb_nmax;       /* max bytes in the buffer */
 109         int             lwb_nused;      /* # used bytes in buffer */
 110         int             lwb_nfilled;    /* # filled bytes in buffer */
 111         int             lwb_sz;         /* size of block and buffer */
 112         lwb_state_t     lwb_state;      /* the state of this lwb */
 113         char            *lwb_buf;       /* log write buffer */
 114         zio_t           *lwb_child_zio; /* parent zio for children */
 115         zio_t           *lwb_write_zio; /* zio for the lwb buffer */
 116         zio_t           *lwb_root_zio;  /* root zio for lwb write and flushes */
 117         hrtime_t        lwb_issued_timestamp; /* when was the lwb issued? */
 118         uint64_t        lwb_issued_txg; /* the txg when the write is issued */
 119         uint64_t        lwb_alloc_txg;  /* the txg when lwb_blk is allocated */
 120         uint64_t        lwb_max_txg;    /* highest txg in this lwb */
 121         list_node_t     lwb_node;       /* zilog->zl_lwb_list linkage */
 122         list_node_t     lwb_issue_node; /* linkage of lwbs ready for issue */
 123         list_t          lwb_itxs;       /* list of itx's */
 124         list_t          lwb_waiters;    /* list of zil_commit_waiter's */
 125         avl_tree_t      lwb_vdev_tree;  /* vdevs to flush after lwb write */
 126         kmutex_t        lwb_vdev_lock;  /* protects lwb_vdev_tree */
 127 } lwb_t;
 128
 129 /*
 130  * ZIL commit waiter.
 131  *
 132  * This structure is allocated each time zil_commit() is called, and is
 133  * used by zil_commit() to communicate with other parts of the ZIL, such
 134  * that zil_commit() can know when it safe for it return. For more
 135  * details, see the comment above zil_commit().
 136  *
 137  * The "zcw_lock" field is used to protect the commit waiter against
 138  * concurrent access. This lock is often acquired while already holding
 139  * the zilog's "zl_issuer_lock" or "zl_lock"; see the functions
 140  * zil_process_commit_list() and zil_lwb_flush_vdevs_done() as examples
 141  * of this. Thus, one must be careful not to acquire the
 142  * "zl_issuer_lock" or "zl_lock" when already holding the "zcw_lock";
 143  * e.g. see the zil_commit_waiter_timeout() function.
 144  */
 145 typedef struct zil_commit_waiter {
 146         kcondvar_t      zcw_cv;         /* signalled when "done" */
 147         kmutex_t        zcw_lock;       /* protects fields of this struct */
 148         list_node_t     zcw_node;       /* linkage in lwb_t:lwb_waiter list */
 149         lwb_t           *zcw_lwb;       /* back pointer to lwb when linked */
 150         boolean_t       zcw_done;       /* B_TRUE when "done", else B_FALSE */
 151         int             zcw_zio_error;  /* contains the zio io_error value */
 152 } zil_commit_waiter_t;
 153
 154 /*
 155  * Intent log transaction lists
 156  */
 157 typedef struct itxs {
 158         list_t          i_sync_list;    /* list of synchronous itxs */
 159         avl_tree_t      i_async_tree;   /* tree of foids for async itxs */
 160 } itxs_t;
 161
 162 typedef struct itxg {
 163         kmutex_t        itxg_lock;      /* lock for this structure */
 164         uint64_t        itxg_txg;       /* txg for this chain */
 165         itxs_t          *itxg_itxs;     /* sync and async itxs */
 166 } itxg_t;
 167
 168 /* for async nodes we build up an AVL tree of lists of async itxs per file */
 169 typedef struct itx_async_node {
 170         uint64_t        ia_foid;        /* file object id */
 171         list_t          ia_list;        /* list of async itxs for this foid */
 172         avl_node_t      ia_node;        /* AVL tree linkage */
 173 } itx_async_node_t;
 174
 175 /*
 176  * Vdev flushing: during a zil_commit(), we build up an AVL tree of the vdevs
 177  * we've touched so we know which ones need a write cache flush at the end.
 178  */
 179 typedef struct zil_vdev_node {
 180         uint64_t        zv_vdev;        /* vdev to be flushed */
 181         avl_node_t      zv_node;        /* AVL tree linkage */
 182 } zil_vdev_node_t;
 183
 184 #define ZIL_BURSTS 8
 185
 186 /*
 187  * Stable storage intent log management structure.  One per dataset.
 188  */
 189 struct zilog {
 190         kmutex_t        zl_lock;        /* protects most zilog_t fields */
 191         struct dsl_pool *zl_dmu_pool;   /* DSL pool */
 192         spa_t           *zl_spa;        /* handle for read/write log */
 193         const zil_header_t *zl_header;  /* log header buffer */
 194         objset_t        *zl_os;         /* object set we're logging */
 195         zil_get_data_t  *zl_get_data;   /* callback to get object content */
 196         lwb_t           *zl_last_lwb_opened; /* most recent lwb opened */
 197         hrtime_t        zl_last_lwb_latency; /* zio latency of last lwb done */
 198         uint64_t        zl_lr_seq;      /* on-disk log record sequence number */
 199         uint64_t        zl_commit_lr_seq; /* last committed on-disk lr seq */
 200         uint64_t        zl_destroy_txg; /* txg of last zil_destroy() */
 201         uint64_t        zl_replayed_seq[TXG_SIZE]; /* last replayed rec seq */
 202         uint64_t        zl_replaying_seq; /* current replay seq number */
 203         uint32_t        zl_suspend;     /* log suspend count */
 204         kcondvar_t      zl_cv_suspend;  /* log suspend completion */
 205         uint8_t         zl_suspending;  /* log is currently suspending */
 206         uint8_t         zl_keep_first;  /* keep first log block in destroy */
 207         uint8_t         zl_replay;      /* replaying records while set */
 208         uint8_t         zl_stop_sync;   /* for debugging */
 209         kmutex_t        zl_issuer_lock; /* single writer, per ZIL, at a time */
 210         uint8_t         zl_logbias;     /* latency or throughput */
 211         uint8_t         zl_sync;        /* synchronous or asynchronous */
 212         int             zl_parse_error; /* last zil_parse() error */
 213         uint64_t        zl_parse_blk_seq; /* highest blk seq on last parse */
 214         uint64_t        zl_parse_lr_seq; /* highest lr seq on last parse */
 215         uint64_t        zl_parse_blk_count; /* number of blocks parsed */
 216         uint64_t        zl_parse_lr_count; /* number of log records parsed */
 217         itxg_t          zl_itxg[TXG_SIZE]; /* intent log txg chains */
 218         list_t          zl_itx_commit_list; /* itx list to be committed */
 219         uint64_t        zl_cur_size;    /* current burst full size */
 220         uint64_t        zl_cur_left;    /* current burst remaining size */
 221         uint64_t        zl_cur_max;     /* biggest record in current burst */
 222         list_t          zl_lwb_list;    /* in-flight log write list */
 223         avl_tree_t      zl_bp_tree;     /* track bps during log parse */
 224         clock_t         zl_replay_time; /* lbolt of when replay started */
 225         uint64_t        zl_replay_blks; /* number of log blocks replayed */
 226         zil_header_t    zl_old_header;  /* debugging aid */
 227         uint_t          zl_parallel;    /* workload is multi-threaded */
 228         uint_t          zl_prev_rotor;  /* rotor for zl_prev[] */
 229         uint_t          zl_prev_opt[ZIL_BURSTS]; /* optimal block size */
 230         uint_t          zl_prev_min[ZIL_BURSTS]; /* minimal first block size */
 231         txg_node_t      zl_dirty_link;  /* protected by dp_dirty_zilogs list */
 232         uint64_t        zl_dirty_max_txg; /* highest txg used to dirty zilog */
 233
 234         kmutex_t        zl_lwb_io_lock; /* protect following members */
 235         uint64_t        zl_lwb_inflight[TXG_SIZE]; /* io issued, but not done */
 236         kcondvar_t      zl_lwb_io_cv;   /* signal when the flush is done */
 237         uint64_t        zl_lwb_max_issued_txg; /* max txg when lwb io issued */
 238
 239         /*
 240          * Max block size for this ZIL.  Note that this can not be changed
 241          * while the ZIL is in use because consumers (ZPL/zvol) need to take
 242          * this into account when deciding between WR_COPIED and WR_NEED_COPY
 243          * (see zil_max_copied_data()).
 244          */
 245         uint64_t        zl_max_block_size;
 246
 247         /* Pointer for per dataset zil sums */
 248         zil_sums_t *zl_sums;
 249 };
 250
 251 typedef struct zil_bp_node {
 252         dva_t           zn_dva;
 253         avl_node_t      zn_node;
 254 } zil_bp_node_t;
 255
 256 #ifdef  __cplusplus
 257 }
 258 #endif
 259
 260 #endif  /* _SYS_ZIL_IMPL_H */