Line data Source code
1 : #include "fd_replay_tile.h"
2 : #include "fd_sched.h"
3 : #include "fd_exec.h"
4 : #include "fd_vote_tracker.h"
5 : #include "generated/fd_replay_tile_seccomp.h"
6 :
7 : #include "../genesis/fd_genesi_tile.h"
8 : #include "../poh/fd_poh.h"
9 : #include "../poh/fd_poh_tile.h"
10 : #include "../tower/fd_tower_tile.h"
11 : #include "../resolv/fd_resolv_tile.h"
12 : #include "../restore/utils/fd_ssload.h"
13 :
14 : #include "../../disco/tiles.h"
15 : #include "../../disco/fd_txn_m.h"
16 : #include "../../disco/store/fd_store.h"
17 : #include "../../disco/pack/fd_pack.h"
18 : #include "../../discof/reasm/fd_reasm.h"
19 : #include "../../disco/keyguard/fd_keyload.h"
20 : #include "../../disco/genesis/fd_genesis_cluster.h"
21 : #include "../../util/pod/fd_pod.h"
22 : #include "../../flamenco/accdb/fd_accdb_admin.h"
23 : #include "../../flamenco/accdb/fd_accdb_impl_v1.h"
24 : #include "../../flamenco/rewards/fd_rewards.h"
25 : #include "../../flamenco/leaders/fd_multi_epoch_leaders.h"
26 : #include "../../flamenco/progcache/fd_progcache_admin.h"
27 : #include "../../disco/metrics/fd_metrics.h"
28 :
29 : #include "../../flamenco/runtime/fd_runtime.h"
30 : #include "../../flamenco/runtime/fd_runtime_stack.h"
31 : #include "../../flamenco/fd_flamenco_base.h"
32 : #include "../../flamenco/runtime/sysvar/fd_sysvar_epoch_schedule.h"
33 :
34 : #include "../../flamenco/runtime/tests/fd_dump_pb.h"
35 :
36 : #include <errno.h>
37 : #include <stdio.h>
38 :
39 : /* Replay concepts:
40 :
41 : - Blocks are aggregations of entries aka. microblocks which are
42 : groupings of txns and are constructed by the block producer (see
43 : fd_pack).
44 :
45 : - Entries are grouped into entry batches by the block producer (see
46 : fd_pack / fd_shredder).
47 :
48 : - Entry batches are divided into chunks known as shreds by the block
49 : producer (see fd_shredder).
50 :
51 : - Shreds are grouped into forward-error-correction sets (FEC sets) by
52 : the block producer (see fd_shredder).
53 :
54 : - Shreds are transmitted to the rest of the cluster via the Turbine
55 : protocol (see fd_shredder / fd_shred).
56 :
57 : - Once enough shreds within a FEC set are received to recover the
58 : entirety of the shred data encoded by that FEC set, the receiver
59 : can "complete" the FEC set (see fd_fec_resolver).
60 :
61 : - If shreds in the FEC set are missing such that it can't complete,
62 : the receiver can use the Repair protocol to request missing shreds
63 : in FEC set (see fd_repair).
64 :
65 : - The current Repair protocol does not support requesting coding
66 : shreds. As a result, some FEC sets might be actually complete
67 : (contain all data shreds). Repair currently hacks around this by
68 : forcing completion but the long-term solution is to add support for
69 : fec_repairing coding shreds via Repair.
70 :
71 : - FEC sets are delivered in partial-order to the Replay tile by the
72 : Repair tile. Currently Replay only supports replaying entry batches
73 : so FEC sets need to reassembled into an entry batch before they can
74 : be replayed. The new Dispatcher will change this by taking a FEC
75 : set as input instead. */
76 :
77 0 : #define IN_KIND_SNAP ( 0)
78 0 : #define IN_KIND_GENESIS ( 1)
79 0 : #define IN_KIND_IPECHO ( 2)
80 0 : #define IN_KIND_TOWER ( 3)
81 0 : #define IN_KIND_RESOLV ( 4)
82 0 : #define IN_KIND_POH ( 5)
83 0 : #define IN_KIND_EXEC ( 6)
84 0 : #define IN_KIND_SHRED ( 7)
85 0 : #define IN_KIND_VTXN ( 8)
86 0 : #define IN_KIND_GUI ( 9)
87 0 : #define IN_KIND_RPC (10)
88 :
89 : #define DEBUG_LOGGING 0
90 :
91 : /* The first bank that that the replay tile produces either for genesis
92 : or the snapshot boot will always be at bank index 0. */
93 0 : #define FD_REPLAY_BOOT_BANK_IDX (0UL)
94 :
95 : struct fd_replay_in_link {
96 : fd_wksp_t * mem;
97 : ulong chunk0;
98 : ulong wmark;
99 : ulong mtu;
100 : };
101 :
102 : typedef struct fd_replay_in_link fd_replay_in_link_t;
103 :
104 : struct fd_replay_out_link {
105 : ulong idx;
106 : fd_wksp_t * mem;
107 : ulong chunk0;
108 : ulong wmark;
109 : ulong chunk;
110 : };
111 :
112 : typedef struct fd_replay_out_link fd_replay_out_link_t;
113 :
114 : /* fd_block_id_map is a simple map of block-ids to bank indices. The
115 : map sits on top of an array of fd_block_id_ele_t. This serves as a
116 : translation layer between block ids to bank indices. */
117 :
118 : struct fd_block_id_ele {
119 : fd_hash_t block_id;
120 : ulong slot; /* = FD_SLOT_NULL if not initialized */
121 : ulong next_;
122 : };
123 : typedef struct fd_block_id_ele fd_block_id_ele_t;
124 :
125 : #define MAP_NAME fd_block_id_map
126 : #define MAP_ELE_T fd_block_id_ele_t
127 : #define MAP_KEY_T fd_hash_t
128 0 : #define MAP_KEY block_id
129 0 : #define MAP_NEXT next_
130 0 : #define MAP_KEY_EQ(k0,k1) (!memcmp((k0),(k1), sizeof(fd_hash_t)))
131 0 : #define MAP_KEY_HASH(key,seed) (fd_hash((seed),(key),sizeof(fd_hash_t)))
132 : #include "../../util/tmpl/fd_map_chain.c"
133 :
134 : static inline ulong
135 0 : fd_block_id_ele_get_idx( fd_block_id_ele_t * ele_arr, fd_block_id_ele_t * ele ) {
136 0 : return (ulong)(ele - ele_arr);
137 0 : }
138 :
139 : struct fd_replay_tile {
140 : fd_wksp_t * wksp;
141 :
142 : fd_accdb_admin_t accdb_admin[1];
143 : fd_accdb_user_t accdb[1];
144 : fd_progcache_admin_t progcache_admin[1];
145 :
146 : fd_txncache_t * txncache;
147 : fd_store_t * store;
148 : fd_banks_t * banks;
149 :
150 : /* This flag is 1 If we have seen a vote signature that our node has
151 : sent out get rooted at least one time. The value is 0 otherwise.
152 : We can't become leader and pack blocks until this flag has been
153 : set. This parallels the Agave 'has_new_vote_been_rooted'.
154 :
155 : TODO: Add a flag to the toml to make this optional. */
156 : int has_identity_vote_rooted;
157 :
158 : fd_reasm_t * reasm;
159 :
160 : /* Replay state machine. */
161 : fd_sched_t * sched;
162 : ulong exec_cnt;
163 : fd_replay_out_link_t exec_out[ 1 ]; /* Sending work down to exec tiles */
164 :
165 : fd_vote_tracker_t * vote_tracker;
166 :
167 : int has_genesis_hash;
168 : char genesis_path[ PATH_MAX ];
169 : uchar genesis_hash[ 32UL ];
170 : ulong cluster_type;
171 :
172 : #define FD_REPLAY_HARD_FORKS_MAX (64UL)
173 : ulong hard_forks_cnt;
174 : ulong hard_forks[ FD_REPLAY_HARD_FORKS_MAX ];
175 :
176 : ushort expected_shred_version;
177 : ushort ipecho_shred_version;
178 :
179 : /* A note on publishing ...
180 :
181 : The watermarks are used to publish our fork-aware structures. For
182 : example, store, banks, and txncache need to be published to release
183 : resources occupied by rooted or dead blocks. In general,
184 : publishing has the effect of pruning forks in those structures,
185 : indicating that it is ok to release the memory being occupied by
186 : the blocks on said forks. Tower is responsible for informing us of
187 : the latest block on the consensus rooted fork. As soon as we can,
188 : we should move the published root as close as possible to the
189 : latest consensus root, publishing/pruning everything on the fork
190 : tree along the way. That is, all the blocks that directly descend
191 : from the current published root (inclusive) to the new published
192 : root (exclusive) on the rooted fork, as well as all the minority
193 : forks that branch from said blocks.
194 :
195 : Ideally, we'd move the published root to the consensus root
196 : immediately upon receiving a new consensus root. However, that's
197 : not always safe to do. One thing we need to be careful about is
198 : making sure that there are no more users/consumers of
199 : soon-to-be-pruned blocks, lest a use-after-free occurs. This can
200 : be done by using a reference counter for each block. Any
201 : concurrent activity, such as transaction execution in the exec
202 : tiles, should retain a refcnt on the block for as
203 : long as it needs access to the shared fork-aware structures related
204 : to that block. Eventually, refcnt on a given block will drop down
205 : to 0 as the block either finishes replaying or gets marked as dead,
206 : and any other tile that has retained a refcnt on the block releases
207 : it. At that point, it becomes a candidate for pruning. The key to
208 : safe publishing then becomes figuring out how far we could advance
209 : the published root, such that every minority fork branching off of
210 : blocks in between the current published root (inclusive) and the
211 : new published root (exclusive) is safe to be pruned. This is a
212 : straightforward tree traversal, where if a block B on the rooted
213 : fork has refcnt 0, and all minority forks branching off of B also
214 : have refcnt 0, then B is safe to be pruned. We advance the
215 : published root to the farthest consecutively prunable block on the
216 : rooted fork. Note that reasm presents the replay tile with a clean
217 : view of the world where every block is chained off of a parent
218 : block. So there are no orpahned/dangling tree nodes to worry
219 : about. The world is a nice single tree as far as replay is
220 : concerned.
221 :
222 : In the following fork tree, every node is a block and the number in
223 : parentheses is the refcnt on the block. The chain marked with
224 : double slashes is the rooted fork. Suppose the published root is
225 : at block P, and consensus root is at block T. We can't publish
226 : past block P because Q has refcnt 1.
227 :
228 :
229 : P(0)
230 : / \\
231 : Q(1) A(0)
232 : / || \
233 : X(0) B(0) C(0)
234 : / || \
235 : Y(0) M(0) R(0)
236 : / || / \
237 : D(2) T(0) J(0) L(0)
238 : ||
239 : ..
240 : ..
241 : ..
242 : ||
243 : blocks we might be actively replaying
244 :
245 :
246 : When refcnt on Q drops to 0, we would be able to advance the
247 : published root to block M, because blocks P, A, and B, as well as
248 : all subtrees branching off of them, have refcnt 0, and therefore
249 : can be pruned. Block M itself cannot be pruned yet because its
250 : child block D has refcnt 2. After publishing/pruning, the fork
251 : tree would be:
252 :
253 :
254 : M(0)
255 : / ||
256 : D(2) T(0)
257 : ||
258 : ..
259 : ..
260 : ..
261 : ||
262 : blocks we might be actively replaying
263 :
264 :
265 : As a result, the shared fork-aware structures can free resources
266 : for blocks P, A, B, and all subtrees branching off of them.
267 :
268 : For the reference counting part, the replay tile is the sole entity
269 : that can update the refcnt. This ensures that all refcnt increment
270 : and decrement attempts are serialized at the replay tile, and that
271 : there are no racy resurrection of a soon-to-be-pruned block. If a
272 : refcnt increment request arrives after a block has been pruned,
273 : replay simply rejects the request.
274 :
275 : A note on the implementation of the above ...
276 :
277 : Upon receiving a new consensus root, we descend down the rooted
278 : fork from the current published root to the new consensus root. On
279 : each node/block of the rooted fork, we do a summation of the refcnt
280 : on the block and all the minority fork blocks branching from the
281 : block. If the summation is 0, the block is safe for pruning. We
282 : advance the published root to the far end of the consecutive run of
283 : 0 refcnt sums originating from the current published root. On our
284 : descent down the minority forks, we also mark any block that hasn't
285 : finished replaying as dead, so we don't waste time executing them.
286 : No more transactions shall be dispatched for execution from dead
287 : blocks.
288 :
289 : Blocks start out with a refcnt of 0. Other tiles may send a
290 : request to the replay tile for a reference on a block. The
291 : transaction dispatcher is another source of refcnt updates. On
292 : every dispatch of a transaction for block B, we increment the
293 : refcnt for B. And on every transaction finalization, we decrement
294 : the refcnt for B. This means that whenever the refcnt on a block
295 : is 0, there is no more reference on that block from the execution
296 : pipeline. While it might be tempting to simply increment the
297 : refcnt once when we start replaying a block, and decrement the
298 : refcnt once when we finish a block, this more fine-grained refcnt
299 : update strategy allows for aborting and potentially immediate
300 : pruning of blocks under interleaved block replay. Upon receiving a
301 : new consensus root, we can simply look at the refcnt on minority
302 : fork blocks, and a refcnt of 0 would imply that the block is safe
303 : for pruning, even if we haven't finished replaying it. Without the
304 : fine-grained refcnt, we would need to first stop dispatching from
305 : the aborted block, and then wait for a full drain of the execution
306 : pipeline to know for sure that there are no more in-flight
307 : transactions executing on the aborted block. Note that this will
308 : allow the refcnt on any block to transiently drop down to 0. We
309 : will not mistakenly prune an actively replaying block, aka a leaf
310 : node, that is chaining off of the rooted fork, because the
311 : consensus root is always an ancestor of the actively replaying tip.
312 : */
313 : fd_hash_t consensus_root; /* The most recent block to have reached max lockout in the tower. */
314 : ulong consensus_root_slot; /* slot number of the above. */
315 : ulong consensus_root_bank_idx; /* bank index of the above. */
316 : ulong published_root_slot; /* slot number of the published root. */
317 : ulong published_root_bank_idx; /* bank index of the published root. */
318 :
319 : /* We need to maintain a tile-local mapping of block-ids to bank index
320 : and vice versa. This translation layer is needed for conversion
321 : since tower operates on block-ids and downstream consumers of FEC
322 : sets operate on bank indices. This mapping must happen both ways:
323 : 1. tower sends us block ids and we must map them to bank indices.
324 : 2. when a block is completed, we must map the bank index to a block
325 : id to send a slot complete message to tower. */
326 : ulong block_id_len;
327 : fd_block_id_ele_t * block_id_arr;
328 : fd_block_id_map_t * block_id_map;
329 :
330 : /* Capture-related configs */
331 : fd_capture_ctx_t * capture_ctx;
332 : FILE * capture_file;
333 :
334 : /* Whether the runtime has been booted either from snapshot loading
335 : or from genesis. */
336 : int is_booted;
337 :
338 : /* Buffer to store vote towers that need to be published to the Tower
339 : tile. */
340 :
341 : fd_multi_epoch_leaders_t * mleaders;
342 :
343 : int larger_max_cost_per_block;
344 :
345 : fd_pubkey_t identity_pubkey[1]; /* TODO: Keyswitch */
346 :
347 : /* When we transition to becoming leader, we can only unbecome the
348 : leader if we have received a block id from the FEC reassembler, and
349 : a message from PoH that the leader slot has ended. After both of
350 : these conditions are met, then we are free to unbecome the leader.
351 : */
352 : int is_leader;
353 : int recv_poh;
354 : int recv_block_id;
355 : ulong next_leader_slot;
356 : long next_leader_tickcount;
357 : ulong highwater_leader_slot;
358 : ulong reset_slot;
359 : fd_bank_t * reset_bank;
360 : fd_hash_t reset_block_id;
361 : long reset_timestamp_nanos;
362 : double slot_duration_nanos;
363 : double slot_duration_ticks;
364 : fd_bank_t * leader_bank; /* ==NULL if not currently the leader */
365 :
366 : ulong resolv_tile_cnt;
367 :
368 : int in_kind[ 64 ];
369 : fd_replay_in_link_t in[ 64 ];
370 :
371 : fd_replay_out_link_t replay_out[1];
372 :
373 : fd_replay_out_link_t stake_out[1];
374 :
375 : /* The gui tile needs to reliably own a reference to the most recent
376 : completed active bank. Replay needs to know if the gui as a
377 : consumer is enabled so it can increment the bank's refcnt before
378 : publishing the bank_idx to the gui. */
379 : int gui_enabled;
380 : int rpc_enabled;
381 :
382 : # if FD_HAS_FLATCC
383 : /* For dumping blocks to protobuf. For backtest only. */
384 : fd_block_dump_ctx_t * block_dump_ctx;
385 : # endif
386 :
387 : /* We need a few pieces of information to compute the right addresses
388 : for bundle crank information that we need to send to pack. */
389 : struct {
390 : int enabled;
391 : fd_pubkey_t vote_account;
392 : fd_bundle_crank_gen_t gen[1];
393 : } bundle;
394 :
395 : struct {
396 : fd_histf_t store_read_wait[ 1 ];
397 : fd_histf_t store_read_work[ 1 ];
398 : fd_histf_t store_publish_wait[ 1 ];
399 : fd_histf_t store_publish_work[ 1 ];
400 : fd_histf_t store_link_wait[ 1 ];
401 : fd_histf_t store_link_work[ 1 ];
402 :
403 : ulong slots_total;
404 : ulong transactions_total;
405 : } metrics;
406 :
407 : uchar __attribute__((aligned(FD_MULTI_EPOCH_LEADERS_ALIGN))) mleaders_mem[ FD_MULTI_EPOCH_LEADERS_FOOTPRINT ];
408 :
409 : fd_runtime_stack_t runtime_stack;
410 : };
411 :
412 : typedef struct fd_replay_tile fd_replay_tile_t;
413 :
414 : FD_FN_CONST static inline ulong
415 0 : scratch_align( void ) {
416 0 : return 128UL;
417 0 : }
418 : FD_FN_PURE static inline ulong
419 0 : scratch_footprint( fd_topo_tile_t const * tile ) {
420 0 : ulong chain_cnt = fd_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
421 :
422 0 : ulong l = FD_LAYOUT_INIT;
423 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
424 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_block_id_ele_t), sizeof(fd_block_id_ele_t) * tile->replay.max_live_slots );
425 0 : l = FD_LAYOUT_APPEND( l, fd_block_id_map_align(), fd_block_id_map_footprint( chain_cnt ) );
426 0 : l = FD_LAYOUT_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
427 0 : l = FD_LAYOUT_APPEND( l, fd_reasm_align(), fd_reasm_footprint( 1 << 20 ) );
428 0 : l = FD_LAYOUT_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.max_live_slots ) );
429 0 : l = FD_LAYOUT_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
430 0 : l = FD_LAYOUT_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
431 :
432 0 : # if FD_HAS_FLATCC
433 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
434 0 : l = FD_LAYOUT_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
435 0 : }
436 0 : # endif
437 :
438 0 : l = FD_LAYOUT_FINI( l, scratch_align() );
439 :
440 0 : return l;
441 0 : }
442 :
443 : static inline void
444 0 : metrics_write( fd_replay_tile_t * ctx ) {
445 0 : FD_MHIST_COPY( REPLAY, STORE_LINK_WAIT, ctx->metrics.store_link_wait );
446 0 : FD_MHIST_COPY( REPLAY, STORE_LINK_WORK, ctx->metrics.store_link_work );
447 0 : FD_MHIST_COPY( REPLAY, STORE_READ_WAIT, ctx->metrics.store_read_wait );
448 0 : FD_MHIST_COPY( REPLAY, STORE_READ_WORK, ctx->metrics.store_read_work );
449 0 : FD_MHIST_COPY( REPLAY, STORE_PUBLISH_WAIT, ctx->metrics.store_publish_wait );
450 0 : FD_MHIST_COPY( REPLAY, STORE_PUBLISH_WORK, ctx->metrics.store_publish_work );
451 :
452 0 : FD_MGAUGE_SET( REPLAY, ROOT_SLOT, ctx->consensus_root_slot==ULONG_MAX ? 0UL : ctx->consensus_root_slot );
453 0 : ulong leader_slot = ctx->leader_bank ? fd_bank_slot_get( ctx->leader_bank ) : 0UL;
454 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, leader_slot );
455 :
456 0 : if( FD_LIKELY( ctx->leader_bank ) ) {
457 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, leader_slot );
458 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, leader_slot );
459 0 : } else {
460 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, ctx->next_leader_slot==ULONG_MAX ? 0UL : ctx->next_leader_slot );
461 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, 0UL );
462 0 : }
463 0 : FD_MGAUGE_SET( REPLAY, RESET_SLOT, ctx->reset_slot==ULONG_MAX ? 0UL : ctx->reset_slot );
464 :
465 0 : fd_bank_t * bank_pool = fd_banks_get_bank_pool( ctx->banks );
466 0 : ulong live_banks = fd_banks_pool_max( bank_pool ) - fd_banks_pool_free( bank_pool );
467 0 : FD_MGAUGE_SET( REPLAY, LIVE_BANKS, live_banks );
468 :
469 0 : FD_MCNT_SET( REPLAY, SLOTS_TOTAL, ctx->metrics.slots_total );
470 0 : FD_MCNT_SET( REPLAY, TRANSACTIONS_TOTAL, ctx->metrics.transactions_total );
471 :
472 0 : FD_MCNT_SET( REPLAY, PROGCACHE_ROOTED, ctx->progcache_admin->metrics.root_cnt );
473 0 : FD_MCNT_SET( REPLAY, PROGCACHE_GC_ROOT, ctx->progcache_admin->metrics.gc_root_cnt );
474 0 : }
475 :
476 : static inline ulong
477 : generate_stake_weight_msg( ulong epoch,
478 : fd_epoch_schedule_t const * epoch_schedule,
479 : fd_vote_states_t const * epoch_stakes,
480 0 : ulong * stake_weight_msg_out ) {
481 0 : fd_stake_weight_msg_t * stake_weight_msg = (fd_stake_weight_msg_t *)fd_type_pun( stake_weight_msg_out );
482 0 : fd_vote_stake_weight_t * stake_weights = stake_weight_msg->weights;
483 :
484 0 : stake_weight_msg->epoch = epoch;
485 0 : stake_weight_msg->start_slot = fd_epoch_slot0( epoch_schedule, epoch );
486 0 : stake_weight_msg->slot_cnt = epoch_schedule->slots_per_epoch;
487 0 : stake_weight_msg->excluded_stake = 0UL;
488 0 : stake_weight_msg->vote_keyed_lsched = 1UL;
489 :
490 : /* FIXME: SIMD-0180 - hack to (de)activate in testnet vs mainnet.
491 : This code can be removed once the feature is active. */
492 0 : if( (1==epoch_schedule->warmup && epoch<FD_SIMD0180_ACTIVE_EPOCH_TESTNET) ||
493 0 : (0==epoch_schedule->warmup && epoch<FD_SIMD0180_ACTIVE_EPOCH_MAINNET) ) {
494 0 : stake_weight_msg->vote_keyed_lsched = 0UL;
495 0 : }
496 :
497 : /* epoch_stakes from manifest are already filtered (stake>0), but not sorted */
498 0 : fd_vote_states_iter_t iter_[1];
499 0 : ulong idx = 0UL;
500 0 : for( fd_vote_states_iter_t * iter = fd_vote_states_iter_init( iter_, epoch_stakes ); !fd_vote_states_iter_done( iter ); fd_vote_states_iter_next( iter ) ) {
501 0 : fd_vote_state_ele_t * vote_state = fd_vote_states_iter_ele( iter );
502 0 : if( FD_UNLIKELY( !vote_state->stake ) ) continue;
503 :
504 0 : stake_weights[ idx ].stake = vote_state->stake;
505 0 : memcpy( stake_weights[ idx ].id_key.uc, &vote_state->node_account, sizeof(fd_pubkey_t) );
506 0 : memcpy( stake_weights[ idx ].vote_key.uc, &vote_state->vote_account, sizeof(fd_pubkey_t) );
507 0 : idx++;
508 0 : }
509 0 : stake_weight_msg->staked_cnt = idx;
510 0 : sort_vote_weights_by_stake_vote_inplace( stake_weights, idx );
511 :
512 0 : return fd_stake_weight_msg_sz( idx );
513 0 : }
514 :
515 : static void
516 : publish_stake_weights( fd_replay_tile_t * ctx,
517 : fd_stem_context_t * stem,
518 : fd_bank_t * bank,
519 0 : int current_epoch ) {
520 0 : fd_epoch_schedule_t const * schedule = fd_bank_epoch_schedule_query( bank );
521 0 : ulong epoch = fd_slot_to_epoch( schedule, fd_bank_slot_get( bank ), NULL );
522 :
523 0 : fd_vote_states_t const * vote_states_prev;
524 0 : if( FD_LIKELY( current_epoch ) ) vote_states_prev = fd_bank_vote_states_prev_locking_query( bank );
525 0 : else vote_states_prev = fd_bank_vote_states_prev_prev_locking_query( bank );
526 :
527 0 : ulong * stake_weights_msg = fd_chunk_to_laddr( ctx->stake_out->mem, ctx->stake_out->chunk );
528 0 : ulong stake_weights_sz = generate_stake_weight_msg( epoch+fd_ulong_if( current_epoch, 1UL, 0UL), schedule, vote_states_prev, stake_weights_msg );
529 0 : ulong stake_weights_sig = 4UL;
530 0 : fd_stem_publish( stem, ctx->stake_out->idx, stake_weights_sig, ctx->stake_out->chunk, stake_weights_sz, 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
531 0 : ctx->stake_out->chunk = fd_dcache_compact_next( ctx->stake_out->chunk, stake_weights_sz, ctx->stake_out->chunk0, ctx->stake_out->wmark );
532 :
533 0 : if( FD_LIKELY( current_epoch ) ) fd_bank_vote_states_prev_end_locking_query( bank );
534 0 : else fd_bank_vote_states_prev_prev_end_locking_query( bank );
535 :
536 0 : fd_multi_epoch_leaders_stake_msg_init( ctx->mleaders, fd_type_pun_const( stake_weights_msg ) );
537 0 : fd_multi_epoch_leaders_stake_msg_fini( ctx->mleaders );
538 0 : }
539 :
540 : /**********************************************************************/
541 : /* Transaction execution state machine helpers */
542 : /**********************************************************************/
543 :
544 : static fd_bank_t *
545 : replay_block_start( fd_replay_tile_t * ctx,
546 : fd_stem_context_t * stem,
547 : ulong bank_idx,
548 : ulong parent_bank_idx,
549 0 : ulong slot ) {
550 0 : long before = fd_log_wallclock();
551 :
552 : /* Switch to a new block that we don't have a bank for. */
553 :
554 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
555 0 : if( FD_UNLIKELY( !bank ) ) {
556 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", bank_idx ));
557 0 : }
558 0 : if( FD_UNLIKELY( bank->flags!=FD_BANK_FLAGS_INIT ) ) {
559 0 : FD_LOG_CRIT(( "invariant violation: bank is not in correct state for bank index %lu", bank_idx ));
560 0 : }
561 :
562 0 : bank->preparation_begin_nanos = before;
563 :
564 0 : fd_bank_t * parent_bank = fd_banks_bank_query( ctx->banks, parent_bank_idx );
565 0 : if( FD_UNLIKELY( !parent_bank ) ) {
566 0 : FD_LOG_CRIT(( "invariant violation: parent bank is NULL for bank index %lu", parent_bank_idx ));
567 0 : }
568 0 : if( FD_UNLIKELY( !(parent_bank->flags&FD_BANK_FLAGS_FROZEN) ) ) {
569 0 : FD_LOG_CRIT(( "invariant violation: parent bank is not frozen for bank index %lu", parent_bank_idx ));
570 0 : }
571 0 : ulong parent_slot = fd_bank_slot_get( parent_bank );
572 :
573 : /* Clone the bank from the parent. We must special case the first
574 : slot that is executed as the snapshot does not provide a parent
575 : block id. */
576 :
577 0 : bank = fd_banks_clone_from_parent( ctx->banks, bank_idx, parent_bank_idx );
578 0 : if( FD_UNLIKELY( !bank ) ) {
579 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", bank_idx ));
580 0 : }
581 0 : fd_bank_slot_set( bank, slot );
582 0 : fd_bank_parent_slot_set( bank, parent_slot );
583 0 : bank->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, parent_bank->txncache_fork_id );
584 :
585 : /* Create a new funk txn for the block. */
586 :
587 0 : fd_funk_txn_xid_t xid = { .ul = { slot, bank_idx } };
588 0 : fd_funk_txn_xid_t parent_xid = { .ul = { parent_slot, parent_bank_idx } };
589 0 : fd_accdb_attach_child( ctx->accdb_admin, &parent_xid, &xid );
590 0 : fd_progcache_txn_attach_child( ctx->progcache_admin, &parent_xid, &xid );
591 :
592 : /* Update any required runtime state and handle any potential epoch
593 : boundary change. */
594 :
595 0 : if( ctx->capture_ctx ) {
596 0 : fd_solcap_writer_set_slot( ctx->capture_ctx->capture, slot );
597 0 : }
598 :
599 0 : fd_bank_shred_cnt_set( bank, 0UL );
600 0 : fd_bank_execution_fees_set( bank, 0UL );
601 0 : fd_bank_priority_fees_set( bank, 0UL );
602 0 : fd_bank_tips_set( bank, 0UL );
603 :
604 0 : fd_bank_has_identity_vote_set( bank, 0 );
605 :
606 : /* Set the tick height. */
607 0 : fd_bank_tick_height_set( bank, fd_bank_max_tick_height_get( bank ) );
608 :
609 : /* Update block height. */
610 0 : fd_bank_block_height_set( bank, fd_bank_block_height_get( bank ) + 1UL );
611 :
612 0 : ulong * max_tick_height = fd_bank_max_tick_height_modify( bank );
613 0 : ulong ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
614 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS != fd_runtime_compute_max_tick_height( ticks_per_slot, slot, max_tick_height ) ) ) {
615 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, ticks_per_slot ));
616 0 : }
617 :
618 0 : int is_epoch_boundary = 0;
619 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, &ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
620 0 : if( FD_UNLIKELY( is_epoch_boundary ) ) publish_stake_weights( ctx, stem, bank, 1 );
621 :
622 0 : return bank;
623 0 : }
624 :
625 : static void
626 0 : cost_tracker_snap( fd_bank_t * bank, fd_replay_slot_completed_t * slot_info ) {
627 0 : if( bank->cost_tracker_pool_idx!=fd_bank_cost_tracker_pool_idx_null( fd_bank_get_cost_tracker_pool( bank ) ) ) {
628 0 : fd_cost_tracker_t const * cost_tracker = fd_bank_cost_tracker_locking_query( bank );
629 0 : slot_info->cost_tracker.block_cost = cost_tracker->block_cost;
630 0 : slot_info->cost_tracker.vote_cost = cost_tracker->vote_cost;
631 0 : slot_info->cost_tracker.allocated_accounts_data_size = cost_tracker->allocated_accounts_data_size;
632 0 : slot_info->cost_tracker.block_cost_limit = cost_tracker->block_cost_limit;
633 0 : slot_info->cost_tracker.vote_cost_limit = cost_tracker->vote_cost_limit;
634 0 : slot_info->cost_tracker.account_cost_limit = cost_tracker->account_cost_limit;
635 0 : fd_bank_cost_tracker_end_locking_query( bank );
636 0 : } else {
637 0 : memset( &slot_info->cost_tracker, 0, sizeof(slot_info->cost_tracker) );
638 0 : }
639 0 : }
640 :
641 : static ulong
642 0 : get_identity_balance( fd_replay_tile_t * ctx, fd_funk_txn_xid_t xid ) {
643 0 : ulong identity_balance = ULONG_MAX;
644 0 : fd_txn_account_t identity_acc[1];
645 0 : fd_funk_t * funk = fd_accdb_user_v1_funk( ctx->accdb );
646 0 : int err = fd_txn_account_init_from_funk_readonly( identity_acc,
647 0 : ctx->identity_pubkey,
648 0 : funk,
649 0 : &xid );
650 0 : if( FD_LIKELY( !err && identity_acc->meta ) ) identity_balance = identity_acc->meta->lamports;
651 :
652 0 : return identity_balance;
653 0 : }
654 :
655 : static void
656 : publish_slot_completed( fd_replay_tile_t * ctx,
657 : fd_stem_context_t * stem,
658 : fd_bank_t * bank,
659 : int is_initial,
660 0 : int is_leader ) {
661 :
662 0 : ulong slot = fd_bank_slot_get( bank );
663 :
664 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank->idx ];
665 :
666 : /* HACKY: hacky way of checking if we should send a null parent block
667 : id */
668 0 : fd_hash_t parent_block_id = {0};
669 0 : if( FD_UNLIKELY( !is_initial ) ) {
670 0 : parent_block_id = ctx->block_id_arr[ bank->parent_idx ].block_id;
671 0 : }
672 :
673 0 : fd_hash_t const * bank_hash = fd_bank_bank_hash_query( bank );
674 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( fd_bank_block_hash_queue_query( bank ) );
675 0 : FD_TEST( bank_hash );
676 0 : FD_TEST( block_hash );
677 :
678 0 : if( FD_LIKELY( !is_initial ) ) fd_txncache_finalize_fork( ctx->txncache, bank->txncache_fork_id, 0UL, block_hash->uc );
679 :
680 0 : fd_epoch_schedule_t const * epoch_schedule = fd_bank_epoch_schedule_query( bank );
681 0 : ulong slot_idx;
682 0 : ulong epoch = fd_slot_to_epoch( epoch_schedule, slot, &slot_idx );
683 :
684 0 : ctx->metrics.slots_total++;
685 0 : ctx->metrics.transactions_total = fd_bank_txn_count_get( bank );
686 :
687 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
688 0 : slot_info->slot = slot;
689 0 : slot_info->root_slot = ctx->consensus_root_slot;
690 0 : slot_info->storage_slot = ctx->published_root_slot;
691 0 : slot_info->epoch = epoch;
692 0 : slot_info->slot_in_epoch = slot_idx;
693 0 : slot_info->block_height = fd_bank_block_height_get( bank );
694 0 : slot_info->parent_slot = fd_bank_parent_slot_get( bank );
695 0 : slot_info->block_id = block_id_ele->block_id;
696 0 : slot_info->parent_block_id = parent_block_id;
697 0 : slot_info->bank_hash = *bank_hash;
698 0 : slot_info->block_hash = *block_hash;
699 0 : slot_info->transaction_count = fd_bank_txn_count_get( bank );
700 :
701 0 : fd_inflation_t inflation = fd_bank_inflation_get( bank );
702 0 : slot_info->inflation.foundation = inflation.foundation;
703 0 : slot_info->inflation.foundation_term = inflation.foundation_term;
704 0 : slot_info->inflation.terminal = inflation.terminal;
705 0 : slot_info->inflation.initial = inflation.initial;
706 0 : slot_info->inflation.taper = inflation.taper;
707 :
708 0 : fd_rent_t rent = fd_bank_rent_get( bank );
709 0 : slot_info->rent.burn_percent = rent.burn_percent;
710 0 : slot_info->rent.lamports_per_uint8_year = rent.lamports_per_uint8_year;
711 0 : slot_info->rent.exemption_threshold = rent.exemption_threshold;
712 :
713 0 : slot_info->first_fec_set_received_nanos = bank->first_fec_set_received_nanos;
714 0 : slot_info->preparation_begin_nanos = bank->preparation_begin_nanos;
715 0 : slot_info->first_transaction_scheduled_nanos = bank->first_transaction_scheduled_nanos;
716 0 : slot_info->last_transaction_finished_nanos = bank->last_transaction_finished_nanos;
717 0 : slot_info->completion_time_nanos = fd_log_wallclock();
718 :
719 : /* refcnt should be incremented by 1 for each consumer that uses
720 : `bank_idx`. Each consumer should decrement the bank's refcnt once
721 : they are done usin the bank. */
722 0 : bank->refcnt++; /* tower_tile */
723 0 : if( FD_LIKELY( ctx->gui_enabled ) ) bank->refcnt++; /* gui tile */
724 0 : slot_info->bank_idx = bank->idx;
725 :
726 0 : slot_info->parent_bank_idx = ULONG_MAX;
727 0 : fd_bank_t * parent_bank = fd_banks_get_parent( ctx->banks, bank );
728 0 : if( FD_LIKELY( parent_bank && ctx->gui_enabled ) ) {
729 0 : parent_bank->refcnt++;
730 0 : slot_info->parent_bank_idx = parent_bank->idx;
731 0 : }
732 :
733 0 : slot_info->is_leader = is_leader;
734 :
735 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_SLOT_COMPLETED, ctx->replay_out->chunk, sizeof(fd_replay_slot_completed_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
736 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_replay_slot_completed_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
737 0 : }
738 :
739 : static void
740 : publish_slot_dead( fd_replay_tile_t * ctx,
741 : fd_stem_context_t * stem,
742 0 : fd_bank_t * bank ) {
743 0 : fd_replay_slot_dead_t * slot_dead = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
744 0 : slot_dead->slot = fd_bank_slot_get( bank );
745 0 : slot_dead->block_id = ctx->block_id_arr[ bank->idx ].block_id;
746 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_SLOT_DEAD, ctx->replay_out->chunk, sizeof(fd_replay_slot_dead_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
747 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_replay_slot_dead_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
748 0 : }
749 :
750 : static void
751 : replay_block_finalize( fd_replay_tile_t * ctx,
752 : fd_stem_context_t * stem,
753 0 : fd_bank_t * bank ) {
754 :
755 0 : bank->last_transaction_finished_nanos = fd_log_wallclock();
756 :
757 0 : if( FD_UNLIKELY( ctx->capture_ctx ) ) fd_solcap_writer_flush( ctx->capture_ctx->capture );
758 :
759 0 : FD_TEST( !(bank->flags&FD_BANK_FLAGS_FROZEN) );
760 :
761 : /* Set poh hash in bank. */
762 0 : fd_hash_t * poh = fd_sched_get_poh( ctx->sched, bank->idx );
763 0 : fd_bank_poh_set( bank, *poh );
764 :
765 : /* Set shred count in bank. */
766 0 : fd_bank_shred_cnt_set( bank, fd_sched_get_shred_cnt( ctx->sched, bank->idx ) );
767 :
768 : /* Do hashing and other end-of-block processing. */
769 0 : fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx );
770 :
771 : /* Copy out cost tracker fields before freezing */
772 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
773 0 : cost_tracker_snap( bank, slot_info );
774 :
775 : /* fetch identity / vote balance updates infrequently */
776 0 : ulong slot = fd_bank_slot_get( bank );
777 0 : fd_funk_txn_xid_t xid = { .ul = { slot, bank->idx } };
778 0 : slot_info->identity_balance = FD_UNLIKELY( slot%4096==0UL ) ? get_identity_balance( ctx, xid ) : ULONG_MAX;
779 :
780 : /* Mark the bank as frozen. */
781 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
782 :
783 : /**********************************************************************/
784 : /* Bank hash comparison, and halt if there's a mismatch after replay */
785 : /**********************************************************************/
786 :
787 0 : fd_hash_t const * bank_hash = fd_bank_bank_hash_query( bank );
788 0 : FD_TEST( bank_hash );
789 :
790 : /* Must be last so we can measure completion time correctly, even
791 : though we could technically do this before the hash cmp and vote
792 : tower stuff. */
793 0 : publish_slot_completed( ctx, stem, bank, 0, 0 /* is_leader */ );
794 :
795 0 : # if FD_HAS_FLATCC
796 : /* If enabled, dump the block to a file and reset the dumping
797 : context state */
798 0 : if( FD_UNLIKELY( ctx->capture_ctx && ctx->capture_ctx->dump_block_to_pb ) ) {
799 0 : fd_funk_t * funk = fd_accdb_user_v1_funk( ctx->accdb );
800 0 : fd_dump_block_to_protobuf( ctx->block_dump_ctx, ctx->banks, bank, funk, ctx->capture_ctx );
801 0 : fd_block_dump_context_reset( ctx->block_dump_ctx );
802 0 : }
803 0 : # endif
804 0 : }
805 :
806 : /**********************************************************************/
807 : /* Leader bank management */
808 : /**********************************************************************/
809 :
810 : static fd_bank_t *
811 : prepare_leader_bank( fd_replay_tile_t * ctx,
812 : ulong slot,
813 : long now,
814 : fd_hash_t const * parent_block_id,
815 0 : fd_stem_context_t * stem ) {
816 0 : long before = fd_log_wallclock();
817 :
818 : /* Make sure that we are not already leader. */
819 0 : FD_TEST( ctx->leader_bank==NULL );
820 :
821 0 : fd_block_id_ele_t * parent_ele = fd_block_id_map_ele_query( ctx->block_id_map, parent_block_id, NULL, ctx->block_id_arr );
822 0 : if( FD_UNLIKELY( !parent_ele ) ) {
823 0 : FD_LOG_CRIT(( "invariant violation: parent bank index not found for merkle root %s", FD_BASE58_ENC_32_ALLOCA( parent_block_id->uc ) ));
824 0 : }
825 0 : ulong parent_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, parent_ele );
826 :
827 0 : fd_bank_t * parent_bank = fd_banks_bank_query( ctx->banks, parent_bank_idx );
828 0 : if( FD_UNLIKELY( !parent_bank ) ) {
829 0 : FD_LOG_CRIT(( "invariant violation: parent bank not found for bank index %lu", parent_bank_idx ));
830 0 : }
831 0 : ulong parent_slot = fd_bank_slot_get( parent_bank );
832 :
833 0 : ctx->leader_bank = fd_banks_new_bank( ctx->banks, parent_bank_idx, now );
834 0 : if( FD_UNLIKELY( !ctx->leader_bank ) ) {
835 0 : FD_LOG_CRIT(( "invariant violation: leader bank is NULL for slot %lu", slot ));
836 0 : }
837 :
838 0 : if( FD_UNLIKELY( !fd_banks_clone_from_parent( ctx->banks, ctx->leader_bank->idx, parent_bank_idx ) ) ) {
839 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for slot %lu", slot ));
840 0 : }
841 :
842 0 : ctx->leader_bank->preparation_begin_nanos = before;
843 :
844 0 : fd_bank_slot_set( ctx->leader_bank, slot );
845 0 : fd_bank_parent_slot_set( ctx->leader_bank, parent_slot );
846 0 : ctx->leader_bank->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, parent_bank->txncache_fork_id );
847 : /* prepare the funk transaction for the leader bank */
848 0 : fd_funk_txn_xid_t xid = { .ul = { slot, ctx->leader_bank->idx } };
849 0 : fd_funk_txn_xid_t parent_xid = { .ul = { parent_slot, parent_bank_idx } };
850 0 : fd_accdb_attach_child( ctx->accdb_admin, &parent_xid, &xid );
851 0 : fd_progcache_txn_attach_child( ctx->progcache_admin, &parent_xid, &xid );
852 :
853 0 : fd_bank_execution_fees_set( ctx->leader_bank, 0UL );
854 0 : fd_bank_priority_fees_set( ctx->leader_bank, 0UL );
855 0 : fd_bank_shred_cnt_set( ctx->leader_bank, 0UL );
856 0 : fd_bank_tips_set( ctx->leader_bank, 0UL );
857 :
858 : /* Set the tick height. */
859 0 : fd_bank_tick_height_set( ctx->leader_bank, fd_bank_max_tick_height_get( ctx->leader_bank ) );
860 :
861 : /* Update block height. */
862 0 : fd_bank_block_height_set( ctx->leader_bank, fd_bank_block_height_get( ctx->leader_bank ) + 1UL );
863 :
864 0 : ulong * max_tick_height = fd_bank_max_tick_height_modify( ctx->leader_bank );
865 0 : ulong ticks_per_slot = fd_bank_ticks_per_slot_get( ctx->leader_bank );
866 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS != fd_runtime_compute_max_tick_height( ticks_per_slot, slot, max_tick_height ) ) ) {
867 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, ticks_per_slot ));
868 0 : }
869 :
870 0 : int is_epoch_boundary = 0;
871 0 : fd_runtime_block_execute_prepare( ctx->banks, ctx->leader_bank, ctx->accdb, &ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
872 0 : if( FD_UNLIKELY( is_epoch_boundary ) ) publish_stake_weights( ctx, stem, ctx->leader_bank, 1 );
873 :
874 : /* Now that a bank has been created for the leader slot, increment the
875 : reference count until we are done with the leader slot. */
876 0 : ctx->leader_bank->refcnt++;
877 :
878 0 : return ctx->leader_bank;
879 0 : }
880 :
881 : static void
882 : fini_leader_bank( fd_replay_tile_t * ctx,
883 0 : fd_stem_context_t * stem ) {
884 :
885 0 : FD_TEST( ctx->leader_bank!=NULL );
886 0 : FD_TEST( ctx->is_leader );
887 0 : FD_TEST( ctx->recv_block_id );
888 0 : FD_TEST( ctx->recv_poh );
889 :
890 0 : ctx->leader_bank->last_transaction_finished_nanos = fd_log_wallclock();
891 :
892 0 : ulong curr_slot = fd_bank_slot_get( ctx->leader_bank );
893 :
894 0 : fd_sched_block_add_done( ctx->sched, ctx->leader_bank->idx, ctx->leader_bank->parent_idx, curr_slot );
895 :
896 : /* Do hashing and other end-of-block processing */
897 0 : fd_funk_t * funk = fd_accdb_user_v1_funk( ctx->accdb );
898 0 : fd_funk_txn_map_t * txn_map = fd_funk_txn_map( funk );
899 0 : if( FD_UNLIKELY( !txn_map->map ) ) {
900 0 : FD_LOG_ERR(( "Could not find valid funk transaction map" ));
901 0 : }
902 :
903 0 : fd_runtime_block_execute_finalize( ctx->leader_bank, ctx->accdb, ctx->capture_ctx );
904 :
905 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
906 0 : cost_tracker_snap( ctx->leader_bank, slot_info );
907 0 : fd_funk_txn_xid_t xid = { .ul = { curr_slot, ctx->leader_bank->idx } };
908 0 : slot_info->identity_balance = FD_UNLIKELY( curr_slot%4096==0UL ) ? get_identity_balance( ctx, xid ) : ULONG_MAX;
909 :
910 0 : fd_banks_mark_bank_frozen( ctx->banks, ctx->leader_bank );
911 :
912 0 : fd_hash_t const * bank_hash = fd_bank_bank_hash_query( ctx->leader_bank );
913 0 : FD_TEST( bank_hash );
914 :
915 0 : publish_slot_completed( ctx, stem, ctx->leader_bank, 0, 1 /* is_leader */ );
916 :
917 : /* The reference on the bank is finally no longer needed. */
918 0 : ctx->leader_bank->refcnt--;
919 :
920 : /* We are no longer leader so we can clear the bank index we use for
921 : being the leader. */
922 0 : ctx->leader_bank = NULL;
923 0 : ctx->recv_block_id = 0;
924 0 : ctx->recv_poh = 0;
925 0 : ctx->is_leader = 0;
926 0 : }
927 :
928 : static void
929 : publish_root_advanced( fd_replay_tile_t * ctx,
930 0 : fd_stem_context_t * stem ) {
931 :
932 : /* FIXME: for now we want to send the child of the consensus root to
933 : avoid data races with funk root advancing. This is a temporary
934 : hack because currently it is not safe to query against the xid for
935 : the root that is being advanced in funk. This doesn't eliminate
936 : the data race that exists in funk, but reduces how often it occurs.
937 :
938 : Case that causes a data race:
939 : replay: we are advancing the root from slot A->B
940 : resolv: we are resolving ALUTs against slot B */
941 :
942 0 : fd_bank_t * consensus_root_bank = fd_banks_bank_query( ctx->banks, ctx->consensus_root_bank_idx );
943 0 : if( FD_UNLIKELY( !consensus_root_bank ) ) {
944 0 : FD_LOG_CRIT(( "invariant violation: consensus root bank is NULL at bank index %lu", ctx->consensus_root_bank_idx ));
945 0 : }
946 :
947 0 : if( FD_UNLIKELY( consensus_root_bank->child_idx==ULONG_MAX ) ) {
948 0 : return;
949 0 : }
950 :
951 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, consensus_root_bank->child_idx );
952 0 : if( FD_UNLIKELY( !bank ) ) {
953 0 : FD_LOG_CRIT(( "invariant violation: consensus root bank child is NULL at bank index %lu", consensus_root_bank->child_idx ));
954 0 : }
955 :
956 : /* Increment the reference count on the consensus root bank to account
957 : for the number of exec tiles that are waiting on it. */
958 0 : bank->refcnt += ctx->resolv_tile_cnt;
959 :
960 0 : fd_replay_root_advanced_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
961 0 : msg->bank_idx = bank->idx;
962 :
963 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_ROOT_ADVANCED, ctx->replay_out->chunk, sizeof(fd_replay_root_advanced_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
964 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_replay_root_advanced_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
965 0 : }
966 :
967 : /* init_funk performs pre-flight checks for the account database and
968 : program cache. Ensures that the account database was set up
969 : correctly by bootstrap components (e.g. genesis or snapshot loader).
970 : Mirrors the account database's fork tree down to the program cache. */
971 :
972 : static void
973 : init_funk( fd_replay_tile_t * ctx,
974 0 : ulong bank_slot ) {
975 : /* Ensure that the loaded bank root corresponds to the account
976 : database's root. */
977 0 : fd_funk_t * funk = ctx->accdb_admin->funk;
978 0 : if( FD_UNLIKELY( !funk->shmem ) ) {
979 0 : FD_LOG_CRIT(( "failed to initialize account database: replay tile is not joined to database shared memory objects" ));
980 0 : }
981 0 : fd_funk_txn_xid_t const * accdb_pub = fd_funk_last_publish( funk );
982 0 : if( FD_UNLIKELY( accdb_pub->ul[0]!=bank_slot ) ) {
983 0 : FD_LOG_CRIT(( "failed to initialize account database: accdb is at slot %lu, but chain state is at slot %lu\n"
984 0 : "This is a bug in startup components.",
985 0 : accdb_pub->ul[0], bank_slot ));
986 0 : }
987 0 : if( FD_UNLIKELY( fd_funk_last_publish_is_frozen( funk ) ) ) {
988 0 : FD_LOG_CRIT(( "failed to initialize account database: accdb fork graph is not clean.\n"
989 0 : "The account database should only contain state for the root slot at this point,\n"
990 0 : "but there are incomplete database transactions leftover.\n"
991 0 : "This is a bug in startup components." ));
992 0 : }
993 :
994 : /* The program cache tracks the account database's fork graph at all
995 : times. Perform initial synchronization: pivot from funk 'root' (a
996 : sentinel XID) to 'last publish' (the bootstrap root slot). */
997 0 : if( FD_UNLIKELY( !ctx->progcache_admin->funk->shmem ) ) {
998 0 : FD_LOG_CRIT(( "failed to initialize account database: replay tile is not joined to program cache" ));
999 0 : }
1000 0 : fd_progcache_clear( ctx->progcache_admin );
1001 0 : fd_progcache_txn_attach_child( ctx->progcache_admin, fd_funk_root( ctx->progcache_admin->funk ), fd_funk_last_publish( ctx->accdb_admin->funk ) );
1002 0 : fd_progcache_txn_advance_root( ctx->progcache_admin, fd_funk_last_publish( ctx->accdb_admin->funk ) );
1003 0 : }
1004 :
1005 : static void
1006 0 : init_after_snapshot( fd_replay_tile_t * ctx ) {
1007 : /* Now that the snapshot has been loaded in, we have to refresh the
1008 : stake delegations since the manifest does not contain the full set
1009 : of data required for the stake delegations. See
1010 : fd_stake_delegations.h for why this is required. */
1011 :
1012 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
1013 0 : if( FD_UNLIKELY( !bank ) ) {
1014 0 : FD_LOG_CRIT(( "invariant violation: replay bank is NULL at bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
1015 0 : }
1016 :
1017 0 : fd_funk_t * funk = fd_accdb_user_v1_funk( ctx->accdb );
1018 0 : fd_funk_txn_xid_t xid = { .ul = { fd_bank_slot_get( bank ), bank->idx } };
1019 0 : init_funk( ctx, fd_bank_slot_get( bank ) );
1020 :
1021 0 : fd_stake_delegations_t * root_delegations = fd_banks_stake_delegations_root_query( ctx->banks );
1022 :
1023 0 : fd_stake_delegations_refresh( root_delegations, funk, &xid );
1024 :
1025 : /* After both snapshots have been loaded in, we can determine if we should
1026 : start distributing rewards. */
1027 :
1028 0 : fd_rewards_recalculate_partitioned_rewards( ctx->banks, bank, funk, &xid, &ctx->runtime_stack, ctx->capture_ctx );
1029 :
1030 0 : ulong snapshot_slot = fd_bank_slot_get( bank );
1031 0 : if( FD_UNLIKELY( !snapshot_slot ) ) {
1032 : /* Genesis-specific setup. */
1033 : /* FIXME: This branch does not set up a new block exec ctx
1034 : properly. Needs to do whatever prepare_new_block_execution
1035 : does, but just hacking that in breaks stuff. */
1036 0 : fd_runtime_update_leaders( bank, &ctx->runtime_stack );
1037 :
1038 0 : ulong hashcnt_per_slot = fd_bank_hashes_per_tick_get( bank ) * fd_bank_ticks_per_slot_get( bank );
1039 0 : fd_hash_t * poh = fd_bank_poh_modify( bank );
1040 0 : while( hashcnt_per_slot-- ) {
1041 0 : fd_sha256_hash( poh->hash, 32UL, poh->hash );
1042 0 : }
1043 :
1044 0 : int is_epoch_boundary = 0;
1045 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, &ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
1046 0 : FD_TEST( !is_epoch_boundary );
1047 0 : fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx );
1048 :
1049 0 : snapshot_slot = 0UL;
1050 0 : }
1051 :
1052 0 : if( FD_UNLIKELY( ctx->capture_ctx ) ) fd_solcap_writer_flush( ctx->capture_ctx->capture );
1053 0 : }
1054 :
1055 : static inline int
1056 : maybe_become_leader( fd_replay_tile_t * ctx,
1057 0 : fd_stem_context_t * stem ) {
1058 0 : FD_TEST( ctx->is_booted );
1059 0 : if( FD_LIKELY( ctx->next_leader_slot==ULONG_MAX || ctx->is_leader || !ctx->has_identity_vote_rooted || ctx->replay_out->idx==ULONG_MAX ) ) return 0;
1060 :
1061 0 : FD_TEST( ctx->next_leader_slot>ctx->reset_slot );
1062 0 : long now = fd_tickcount();
1063 0 : if( FD_LIKELY( now<ctx->next_leader_tickcount ) ) return 0;
1064 :
1065 : /* TODO:
1066 : if( FD_UNLIKELY( ctx->halted_switching_key ) ) return 0; */
1067 :
1068 : /* If a prior leader is still in the process of publishing their slot,
1069 : delay ours to let them finish ... unless they are so delayed that
1070 : we risk getting skipped by the leader following us. 1.2 seconds
1071 : is a reasonable default here, although any value between 0 and 1.6
1072 : seconds could be considered reasonable. This is arbitrary and
1073 : chosen due to intuition. */
1074 0 : if( FD_UNLIKELY( now<ctx->next_leader_tickcount+(long)(3.0*ctx->slot_duration_ticks) ) ) {
1075 0 : FD_TEST( ctx->reset_bank );
1076 :
1077 : /* TODO: Make the max_active_descendant calculation more efficient
1078 : by caching it in the bank structure and updating it as banks are
1079 : created and completed. */
1080 0 : ulong max_active_descendant = 0UL;
1081 0 : ulong child_idx = ctx->reset_bank->child_idx;
1082 0 : while( child_idx!=ULONG_MAX ) {
1083 0 : fd_bank_t const * child_bank = fd_banks_bank_query( ctx->banks, child_idx );
1084 0 : max_active_descendant = fd_ulong_max( max_active_descendant, fd_bank_slot_get( child_bank ) );
1085 0 : child_idx = child_bank->sibling_idx;
1086 0 : }
1087 :
1088 : /* If the max_active_descendant is >= next_leader_slot, we waited
1089 : too long and a leader after us started publishing to try and skip
1090 : us. Just start our leader slot immediately, we might win ... */
1091 0 : if( FD_LIKELY( max_active_descendant>=ctx->reset_slot && max_active_descendant<ctx->next_leader_slot ) ) {
1092 : /* If one of the leaders between the reset slot and our leader
1093 : slot is in the process of publishing (they have a descendant
1094 : bank that is in progress of being replayed), then keep waiting.
1095 : We probably wouldn't get a leader slot out before they
1096 : finished.
1097 :
1098 : Unless... we are past the deadline to start our slot by more
1099 : than 1.2 seconds, in which case we should probably start it to
1100 : avoid getting skipped by the leader behind us. */
1101 0 : return 0;
1102 0 : }
1103 0 : }
1104 :
1105 0 : long now_nanos = fd_log_wallclock();
1106 :
1107 0 : ctx->is_leader = 1;
1108 0 : ctx->recv_poh = 0;
1109 0 : ctx->recv_block_id = 0;
1110 :
1111 0 : FD_TEST( ctx->highwater_leader_slot==ULONG_MAX || ctx->highwater_leader_slot<ctx->next_leader_slot );
1112 0 : ctx->highwater_leader_slot = ctx->next_leader_slot;
1113 :
1114 0 : FD_LOG_INFO(( "becoming leader for slot %lu, parent slot is %lu", ctx->next_leader_slot, ctx->reset_slot ));
1115 :
1116 : /* Acquires bank, sets up initial state, and refcnts it. */
1117 0 : fd_bank_t * bank = prepare_leader_bank( ctx, ctx->next_leader_slot, now_nanos, &ctx->reset_block_id, stem );
1118 0 : fd_funk_txn_xid_t xid = { .ul = { ctx->next_leader_slot, ctx->leader_bank->idx } };
1119 :
1120 0 : fd_bundle_crank_tip_payment_config_t config[1] = { 0 };
1121 0 : fd_acct_addr_t tip_receiver_owner[1] = { 0 };
1122 :
1123 0 : if( FD_UNLIKELY( ctx->bundle.enabled ) ) {
1124 0 : fd_acct_addr_t tip_payment_config[1];
1125 0 : fd_acct_addr_t tip_receiver[1];
1126 0 : fd_bundle_crank_get_addresses( ctx->bundle.gen, fd_bank_epoch_get( bank ), tip_payment_config, tip_receiver );
1127 :
1128 0 : fd_funk_t * funk = fd_accdb_user_v1_funk( ctx->accdb );
1129 0 : fd_txn_account_t tip_config_acc[1];
1130 0 : int err = fd_txn_account_init_from_funk_readonly( tip_config_acc,
1131 0 : (fd_hash_t *)tip_payment_config->b,
1132 0 : funk,
1133 0 : &xid );
1134 0 : if( FD_UNLIKELY( err ) ) {
1135 0 : FD_LOG_CRIT(( "failed to initialize tip payment config account: err=%d", err ));
1136 0 : }
1137 0 : memcpy( config, fd_txn_account_get_data( tip_config_acc ), sizeof(fd_bundle_crank_tip_payment_config_t) );
1138 :
1139 : /* It is possible that the tip receiver account does not exist yet
1140 : if it is the first time in an epoch. */
1141 0 : fd_txn_account_t tip_receiver_acc[1];
1142 0 : err = fd_txn_account_init_from_funk_readonly( tip_receiver_acc,
1143 0 : (fd_hash_t *)tip_receiver->b,
1144 0 : funk,
1145 0 : &xid );
1146 0 : if( FD_LIKELY( !err ) ) {
1147 0 : memcpy( tip_receiver_owner, tip_receiver_acc->meta->owner, sizeof(fd_acct_addr_t) );
1148 0 : }
1149 0 : }
1150 :
1151 :
1152 0 : fd_became_leader_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1153 0 : msg->slot = ctx->next_leader_slot;
1154 0 : msg->slot_start_ns = now_nanos;
1155 0 : msg->slot_end_ns = now_nanos+(long)ctx->slot_duration_nanos;
1156 0 : msg->bank = NULL;
1157 0 : msg->bank_idx = bank->idx;
1158 0 : msg->ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
1159 0 : msg->hashcnt_per_tick = fd_bank_hashes_per_tick_get( bank );
1160 0 : msg->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)msg->ticks_per_slot);
1161 0 : msg->bundle->config[0] = config[0];
1162 0 : memcpy( msg->bundle->last_blockhash, (fd_hash_t *)fd_bank_poh_query( bank )->hash, 32UL );
1163 0 : memcpy( msg->bundle->tip_receiver_owner, tip_receiver_owner, 32UL );
1164 :
1165 :
1166 0 : if( FD_UNLIKELY( msg->hashcnt_per_tick==1UL ) ) {
1167 : /* Low power producer, maximum of one microblock per tick in the slot */
1168 0 : msg->max_microblocks_in_slot = msg->ticks_per_slot;
1169 0 : } else {
1170 : /* See the long comment in after_credit for this limit */
1171 0 : msg->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, msg->ticks_per_slot*(msg->hashcnt_per_tick-1UL) );
1172 0 : }
1173 :
1174 0 : msg->total_skipped_ticks = msg->ticks_per_slot*(ctx->next_leader_slot-ctx->reset_slot);
1175 0 : msg->epoch = fd_slot_to_epoch( fd_bank_epoch_schedule_query( bank ), ctx->next_leader_slot, NULL );
1176 :
1177 0 : fd_cost_tracker_t const * cost_tracker = fd_bank_cost_tracker_locking_query( bank );
1178 :
1179 0 : msg->limits.slot_max_cost = ctx->larger_max_cost_per_block ? LARGER_MAX_COST_PER_BLOCK : cost_tracker->block_cost_limit;
1180 0 : msg->limits.slot_max_vote_cost = cost_tracker->vote_cost_limit;
1181 0 : msg->limits.slot_max_write_cost_per_acct = cost_tracker->account_cost_limit;
1182 :
1183 0 : fd_bank_cost_tracker_end_locking_query( bank );
1184 :
1185 0 : if( FD_UNLIKELY( msg->ticks_per_slot+msg->total_skipped_ticks>USHORT_MAX ) ) {
1186 : /* There can be at most USHORT_MAX skipped ticks, because the
1187 : parent_offset field in the shred data is only 2 bytes wide. */
1188 0 : FD_LOG_ERR(( "too many skipped ticks %lu for slot %lu, chain must halt", msg->ticks_per_slot+msg->total_skipped_ticks, ctx->next_leader_slot ));
1189 0 : }
1190 :
1191 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_BECAME_LEADER, ctx->replay_out->chunk, sizeof(fd_became_leader_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
1192 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_became_leader_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
1193 :
1194 0 : ctx->next_leader_slot = ULONG_MAX;
1195 0 : ctx->next_leader_tickcount = LONG_MAX;
1196 :
1197 0 : return 1;
1198 0 : }
1199 :
1200 : static void
1201 : process_poh_message( fd_replay_tile_t * ctx,
1202 0 : fd_poh_leader_slot_ended_t const * slot_ended ) {
1203 :
1204 0 : FD_TEST( ctx->is_booted );
1205 0 : FD_TEST( ctx->is_leader );
1206 0 : FD_TEST( ctx->leader_bank!=NULL );
1207 :
1208 0 : FD_TEST( ctx->highwater_leader_slot>=slot_ended->slot );
1209 0 : FD_TEST( ctx->next_leader_slot>ctx->highwater_leader_slot );
1210 :
1211 : /* Update the poh hash in the bank. We will want to maintain a refcnt
1212 : on the bank until we have recieved the block id for the block after
1213 : it has been shredded. */
1214 :
1215 0 : memcpy( fd_bank_poh_modify( ctx->leader_bank ), slot_ended->blockhash, sizeof(fd_hash_t) );
1216 :
1217 0 : ctx->recv_poh = 1;
1218 0 : }
1219 :
1220 : static void
1221 : publish_reset( fd_replay_tile_t * ctx,
1222 : fd_stem_context_t * stem,
1223 0 : fd_bank_t * bank ) {
1224 0 : if( FD_UNLIKELY( ctx->replay_out->idx==ULONG_MAX ) ) return;
1225 :
1226 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( fd_bank_block_hash_queue_query( bank ) );
1227 0 : FD_TEST( block_hash );
1228 :
1229 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1230 :
1231 0 : reset->bank_idx = bank->idx;
1232 0 : reset->timestamp = fd_log_wallclock();
1233 0 : reset->completed_slot = fd_bank_slot_get( bank );
1234 0 : reset->hashcnt_per_tick = fd_bank_hashes_per_tick_get( bank );
1235 0 : reset->ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
1236 0 : reset->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)reset->ticks_per_slot);
1237 0 : fd_memcpy( reset->completed_blockhash, block_hash->uc, sizeof(fd_hash_t) );
1238 :
1239 0 : ulong ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
1240 0 : if( FD_UNLIKELY( reset->hashcnt_per_tick==1UL ) ) {
1241 : /* Low power producer, maximum of one microblock per tick in the slot */
1242 0 : reset->max_microblocks_in_slot = ticks_per_slot;
1243 0 : } else {
1244 : /* See the long comment in after_credit for this limit */
1245 0 : reset->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ticks_per_slot*(reset->hashcnt_per_tick-1UL) );
1246 0 : }
1247 0 : reset->next_leader_slot = ctx->next_leader_slot;
1248 :
1249 0 : if( FD_LIKELY( ctx->rpc_enabled ) ) bank->refcnt++;
1250 :
1251 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_RESET, ctx->replay_out->chunk, sizeof(fd_poh_reset_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
1252 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_poh_reset_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
1253 0 : }
1254 :
1255 : static void
1256 : boot_genesis( fd_replay_tile_t * ctx,
1257 : fd_stem_context_t * stem,
1258 : ulong in_idx,
1259 0 : ulong chunk ) {
1260 : /* If we are bootstrapping, we can't wait to wait for our identity
1261 : vote to be rooted as this creates a circular dependency. */
1262 0 : ctx->has_identity_vote_rooted = 1;
1263 :
1264 0 : uchar const * lthash = (uchar*)fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
1265 0 : uchar const * genesis_hash = (uchar*)fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk )+sizeof(fd_lthash_value_t);
1266 :
1267 : // TODO: Do not pass the fd_types type between tiles, it have offsets
1268 : // that are unsafe and can't be validated as being in-bounds. Need to
1269 : // pass an actual owned genesis type.
1270 0 : fd_genesis_solana_global_t const * genesis = fd_type_pun( (uchar*)fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk )+sizeof(fd_hash_t)+sizeof(fd_lthash_value_t) );
1271 :
1272 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
1273 0 : FD_TEST( bank );
1274 0 : fd_funk_txn_xid_t xid = { .ul = { 0UL, FD_REPLAY_BOOT_BANK_IDX } };
1275 :
1276 : /* Do genesis-related processing in a non-rooted transaction */
1277 0 : fd_funk_txn_xid_t root_xid; fd_funk_txn_xid_set_root( &root_xid );
1278 0 : fd_funk_txn_xid_t target_xid = { .ul = { 0UL, 0UL } };
1279 0 : fd_accdb_attach_child( ctx->accdb_admin, &root_xid, &target_xid );
1280 0 : fd_runtime_read_genesis( ctx->banks, bank, ctx->accdb, &xid, NULL, fd_type_pun_const( genesis_hash ), fd_type_pun_const( lthash ), genesis, &ctx->runtime_stack );
1281 0 : fd_accdb_advance_root( ctx->accdb_admin, &target_xid );
1282 :
1283 0 : static const fd_txncache_fork_id_t txncache_root = { .val = USHORT_MAX };
1284 0 : bank->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, txncache_root );
1285 :
1286 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( fd_bank_block_hash_queue_query( bank ) );
1287 0 : fd_txncache_finalize_fork( ctx->txncache, bank->txncache_fork_id, 0UL, block_hash->uc );
1288 :
1289 0 : publish_stake_weights( ctx, stem, bank, 0 );
1290 0 : publish_stake_weights( ctx, stem, bank, 1 );
1291 :
1292 : /* We call this after fd_runtime_read_genesis, which sets up the
1293 : slot_bank needed in blockstore_init. */
1294 0 : init_after_snapshot( ctx );
1295 :
1296 : /* Initialize store for genesis case, similar to snapshot case */
1297 0 : fd_hash_t genesis_block_id = { .ul[0] = FD_RUNTIME_INITIAL_BLOCK_ID };
1298 0 : fd_store_exacq( ctx->store );
1299 0 : if( FD_UNLIKELY( fd_store_root( ctx->store ) ) ) {
1300 0 : FD_LOG_CRIT(( "invariant violation: store root is not 0 for genesis" ));
1301 0 : }
1302 0 : fd_store_insert( ctx->store, 0, &genesis_block_id );
1303 0 : ctx->store->slot0 = 0UL; /* Genesis slot */
1304 0 : fd_store_exrel( ctx->store );
1305 :
1306 0 : ctx->published_root_slot = 0UL;
1307 0 : fd_sched_block_add_done( ctx->sched, bank->idx, ULONG_MAX, 0UL );
1308 :
1309 0 : fd_bank_block_height_set( bank, 1UL );
1310 :
1311 0 : ctx->consensus_root = (fd_hash_t){ .ul[0] = FD_RUNTIME_INITIAL_BLOCK_ID };
1312 0 : ctx->consensus_root_slot = 0UL;
1313 0 : ctx->consensus_root_bank_idx = 0UL;
1314 0 : ctx->published_root_slot = 0UL;
1315 0 : ctx->published_root_bank_idx = 0UL;
1316 :
1317 0 : ctx->reset_slot = 0UL;
1318 0 : ctx->reset_bank = bank;
1319 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1320 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, 1UL, ctx->identity_pubkey );
1321 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1322 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1323 0 : } else {
1324 0 : ctx->next_leader_tickcount = LONG_MAX;
1325 0 : }
1326 :
1327 0 : ctx->is_booted = 1;
1328 0 : maybe_become_leader( ctx, stem );
1329 :
1330 0 : fd_hash_t initial_block_id = { .ul = { FD_RUNTIME_INITIAL_BLOCK_ID } };
1331 0 : fd_reasm_fec_t * fec = fd_reasm_insert( ctx->reasm, &initial_block_id, NULL, 0 /* genesis slot */, 0, 0, 0, 0, 1, 0 ); /* FIXME manifest block_id */
1332 0 : fec->bank_idx = 0UL;
1333 :
1334 :
1335 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
1336 0 : FD_TEST( block_id_ele );
1337 0 : block_id_ele->block_id = initial_block_id;
1338 0 : block_id_ele->slot = 0UL;
1339 :
1340 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1341 :
1342 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1343 0 : slot_info->identity_balance = get_identity_balance( ctx, xid );
1344 :
1345 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */ );
1346 0 : publish_root_advanced( ctx, stem );
1347 0 : publish_reset( ctx, stem, bank );
1348 0 : }
1349 :
1350 : static inline void
1351 0 : maybe_verify_cluster_type( fd_replay_tile_t * ctx ) {
1352 0 : if( FD_UNLIKELY( !ctx->is_booted || !ctx->has_genesis_hash ) ) {
1353 0 : return;
1354 0 : }
1355 :
1356 0 : FD_BASE58_ENCODE_32_BYTES( ctx->genesis_hash, hash_cstr );
1357 0 : ulong cluster = fd_genesis_cluster_identify( hash_cstr );
1358 : /* Map pyth-related clusters to unkwown. */
1359 0 : switch( cluster ) {
1360 0 : case FD_CLUSTER_PYTHNET:
1361 0 : case FD_CLUSTER_PYTHTEST:
1362 0 : cluster = FD_CLUSTER_UNKNOWN;
1363 0 : }
1364 :
1365 0 : if( FD_UNLIKELY( cluster!=ctx->cluster_type ) ) {
1366 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` has a genesis hash of `%s` which means the cluster is %s "
1367 0 : "but the snapshot you loaded is for a different cluster %s. If you are trying to join the "
1368 0 : "%s cluster, you can delete the genesis.bin file and restart the node to download the correct "
1369 0 : "genesis file automatically.",
1370 0 : ctx->genesis_path,
1371 0 : hash_cstr,
1372 0 : fd_genesis_cluster_name( cluster ),
1373 0 : fd_genesis_cluster_name( ctx->cluster_type ),
1374 0 : fd_genesis_cluster_name( cluster ) ));
1375 0 : }
1376 0 : }
1377 :
1378 : static void
1379 : on_snapshot_message( fd_replay_tile_t * ctx,
1380 : fd_stem_context_t * stem,
1381 : ulong in_idx,
1382 : ulong chunk,
1383 0 : ulong sig ) {
1384 0 : ulong msg = fd_ssmsg_sig_message( sig );
1385 0 : if( FD_LIKELY( msg==FD_SSMSG_DONE ) ) {
1386 : /* An end of message notification indicates the snapshot is loaded.
1387 : Replay is able to start executing from this point onwards. */
1388 : /* TODO: replay should finish booting. Could make replay a
1389 : state machine and set the state here accordingly. */
1390 0 : ctx->is_booted = 1;
1391 :
1392 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
1393 0 : if( FD_UNLIKELY( !bank ) ) {
1394 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
1395 0 : }
1396 :
1397 0 : ulong snapshot_slot = fd_bank_slot_get( bank );
1398 : /* FIXME: This is a hack because the block id of the snapshot slot
1399 : is not provided in the snapshot. A possible solution is to get
1400 : the block id of the snapshot slot from repair. */
1401 0 : fd_hash_t manifest_block_id = { .ul = { FD_RUNTIME_INITIAL_BLOCK_ID } };
1402 :
1403 0 : fd_store_exacq( ctx->store );
1404 0 : FD_TEST( !fd_store_root( ctx->store ) );
1405 0 : fd_store_insert( ctx->store, 0, &manifest_block_id );
1406 0 : ctx->store->slot0 = snapshot_slot; /* FIXME manifest_block_id */
1407 0 : fd_store_exrel( ctx->store );
1408 :
1409 : /* Typically, when we cross an epoch boundary during normal
1410 : operation, we publish the stake weights for the new epoch. But
1411 : since we are starting from a snapshot, we need to publish two
1412 : epochs worth of stake weights: the previous epoch (which is
1413 : needed for voting on the current epoch), and the current epoch
1414 : (which is needed for voting on the next epoch). */
1415 0 : publish_stake_weights( ctx, stem, bank, 0 );
1416 0 : publish_stake_weights( ctx, stem, bank, 1 );
1417 :
1418 0 : ctx->consensus_root = manifest_block_id;
1419 0 : ctx->consensus_root_slot = snapshot_slot;
1420 0 : ctx->consensus_root_bank_idx = 0UL;
1421 0 : ctx->published_root_slot = ctx->consensus_root_slot;
1422 0 : ctx->published_root_bank_idx = 0UL;
1423 :
1424 0 : ctx->reset_slot = snapshot_slot;
1425 0 : ctx->reset_bank = bank;
1426 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1427 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, 1UL, ctx->identity_pubkey );
1428 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1429 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1430 0 : } else {
1431 0 : ctx->next_leader_tickcount = LONG_MAX;
1432 0 : }
1433 :
1434 0 : fd_sched_block_add_done( ctx->sched, bank->idx, ULONG_MAX, snapshot_slot );
1435 0 : FD_TEST( bank->idx==0UL );
1436 :
1437 0 : fd_funk_txn_xid_t xid = { .ul = { snapshot_slot, FD_REPLAY_BOOT_BANK_IDX } };
1438 :
1439 0 : fd_funk_t * funk = fd_accdb_user_v1_funk( ctx->accdb );
1440 0 : fd_features_restore( bank, funk, &xid );
1441 :
1442 0 : fd_runtime_update_leaders( bank, &ctx->runtime_stack );
1443 :
1444 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
1445 0 : FD_TEST( block_id_ele );
1446 0 : block_id_ele->block_id = manifest_block_id;
1447 0 : block_id_ele->slot = snapshot_slot;
1448 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1449 :
1450 : /* We call this after fd_runtime_read_genesis, which sets up the
1451 : slot_bank needed in blockstore_init. */
1452 0 : init_after_snapshot( ctx );
1453 :
1454 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1455 0 : slot_info->identity_balance = get_identity_balance( ctx, xid );
1456 :
1457 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */ );
1458 0 : publish_root_advanced( ctx, stem );
1459 :
1460 0 : fd_reasm_fec_t * fec = fd_reasm_insert( ctx->reasm, &manifest_block_id, NULL, snapshot_slot, 0, 0, 0, 0, 1, 0 ); /* FIXME manifest block_id */
1461 0 : fec->bank_idx = 0UL;
1462 :
1463 0 : ctx->cluster_type = fd_bank_cluster_type_get( bank );
1464 :
1465 0 : maybe_verify_cluster_type( ctx );
1466 :
1467 0 : return;
1468 0 : }
1469 :
1470 0 : switch( msg ) {
1471 0 : case FD_SSMSG_MANIFEST_FULL:
1472 0 : case FD_SSMSG_MANIFEST_INCREMENTAL: {
1473 : /* We may either receive a full snapshot manifest or an
1474 : incremental snapshot manifest. Note that this external message
1475 : id is only used temporarily because replay cannot yet receive
1476 : the firedancer-internal snapshot manifest message. */
1477 0 : if( FD_UNLIKELY( chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark ) )
1478 0 : FD_LOG_ERR(( "chunk %lu from in %d corrupt, not in range [%lu,%lu]", chunk, ctx->in_kind[ in_idx ], ctx->in[ in_idx ].chunk0, ctx->in[ in_idx ].wmark ));
1479 :
1480 0 : fd_ssload_recover( fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ),
1481 0 : ctx->banks,
1482 0 : fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX ),
1483 0 : ctx->runtime_stack.stakes.vote_credits );
1484 :
1485 0 : fd_snapshot_manifest_t const * manifest = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
1486 0 : ctx->hard_forks_cnt = manifest->hard_forks_len;
1487 0 : for( ulong i=0UL; i<manifest->hard_forks_len; i++ ) ctx->hard_forks[ i ] = manifest->hard_forks[ i ];
1488 0 : break;
1489 0 : }
1490 0 : default: {
1491 0 : FD_LOG_ERR(( "Received unknown snapshot message with msg %lu", msg ));
1492 0 : return;
1493 0 : }
1494 0 : }
1495 :
1496 0 : return;
1497 0 : }
1498 :
1499 : static void
1500 : dispatch_task( fd_replay_tile_t * ctx,
1501 : fd_stem_context_t * stem,
1502 0 : fd_sched_task_t * task ) {
1503 :
1504 0 : switch( task->task_type ) {
1505 0 : case FD_SCHED_TT_TXN_EXEC: {
1506 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_exec->txn_idx );
1507 :
1508 : /* FIXME: this should be done during txn parsing so that we don't
1509 : have to loop over all accounts a second time. */
1510 : /* Insert or reverify invoked programs for this epoch, if needed. */
1511 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->txn_exec->bank_idx );
1512 :
1513 0 : # if FD_HAS_FLATCC
1514 : /* Add the transaction to the block dumper if necessary. This
1515 : logic doesn't need to be fork-aware since it's only meant to
1516 : be used in backtest. */
1517 0 : if( FD_UNLIKELY( ctx->capture_ctx && ctx->capture_ctx->dump_block_to_pb ) ) {
1518 0 : fd_dump_block_to_protobuf_collect_tx( ctx->block_dump_ctx, txn_p );
1519 0 : }
1520 0 : # endif
1521 :
1522 0 : bank->refcnt++;
1523 :
1524 0 : if( FD_UNLIKELY( !bank->first_transaction_scheduled_nanos ) ) bank->first_transaction_scheduled_nanos = fd_log_wallclock();
1525 :
1526 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1527 0 : fd_exec_txn_exec_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1528 0 : memcpy( &exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
1529 0 : exec_msg->bank_idx = task->txn_exec->bank_idx;
1530 0 : exec_msg->txn_idx = task->txn_exec->txn_idx;
1531 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXEC_TT_TXN_EXEC<<32) | task->txn_exec->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
1532 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1533 0 : break;
1534 0 : }
1535 0 : case FD_SCHED_TT_TXN_SIGVERIFY: {
1536 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_sigverify->txn_idx );
1537 :
1538 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->txn_sigverify->bank_idx );
1539 0 : bank->refcnt++;
1540 :
1541 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1542 0 : fd_exec_txn_sigverify_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1543 0 : memcpy( &exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
1544 0 : exec_msg->bank_idx = task->txn_sigverify->bank_idx;
1545 0 : exec_msg->txn_idx = task->txn_sigverify->txn_idx;
1546 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXEC_TT_TXN_SIGVERIFY<<32) | task->txn_sigverify->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, 0UL );
1547 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1548 0 : break;
1549 0 : };
1550 0 : default: {
1551 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
1552 0 : }
1553 0 : }
1554 0 : }
1555 :
1556 : /* Returns 1 if charge_busy. */
1557 : static int
1558 : replay( fd_replay_tile_t * ctx,
1559 0 : fd_stem_context_t * stem ) {
1560 :
1561 0 : if( FD_UNLIKELY( !ctx->is_booted ) ) return 0;
1562 :
1563 0 : int charge_busy = 0;
1564 0 : fd_sched_task_t task[ 1 ];
1565 0 : if( FD_UNLIKELY( !fd_sched_task_next_ready( ctx->sched, task ) ) ) {
1566 0 : return charge_busy; /* Nothing to execute or do. */
1567 0 : }
1568 :
1569 0 : charge_busy = 1;
1570 :
1571 0 : switch( task->task_type ) {
1572 0 : case FD_SCHED_TT_BLOCK_START: {
1573 0 : replay_block_start( ctx, stem, task->block_start->bank_idx, task->block_start->parent_bank_idx, task->block_start->slot );
1574 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_START, ULONG_MAX, ULONG_MAX );
1575 0 : break;
1576 0 : }
1577 0 : case FD_SCHED_TT_BLOCK_END: {
1578 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->block_end->bank_idx );
1579 0 : if( FD_LIKELY( !(bank->flags&FD_BANK_FLAGS_DEAD) ) ) replay_block_finalize( ctx, stem, bank );
1580 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_END, ULONG_MAX, ULONG_MAX );
1581 0 : break;
1582 0 : }
1583 0 : case FD_SCHED_TT_TXN_EXEC:
1584 0 : case FD_SCHED_TT_TXN_SIGVERIFY: {
1585 : /* Likely/common case: we have a transaction we actually need to
1586 : execute. */
1587 0 : dispatch_task( ctx, stem, task );
1588 0 : break;
1589 0 : }
1590 0 : default: {
1591 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
1592 0 : }
1593 0 : }
1594 :
1595 0 : return charge_busy;
1596 0 : }
1597 :
1598 : static int
1599 0 : can_process_fec( fd_replay_tile_t * ctx ) {
1600 0 : fd_reasm_fec_t * fec;
1601 0 : if( FD_UNLIKELY( !fd_sched_can_ingest( ctx->sched, 1UL ) ) ) return 0;
1602 0 : if( FD_UNLIKELY( (fec = fd_reasm_peek( ctx->reasm ))==NULL ) ) return 0;
1603 :
1604 0 : if( FD_UNLIKELY( ctx->is_leader && fec->fec_set_idx==0U && fd_reasm_parent( ctx->reasm, fec )->bank_idx==ctx->leader_bank->idx ) ) {
1605 : /* There's a race that's exceedingly rare, where we receive the
1606 : FEC set for the slot right after our leader rotation before we
1607 : freeze the bank for the last slot in our leader rotation.
1608 : Leader slot freezing happens only after if we've received the
1609 : final PoH hash from the poh tile as well as the final FEC set
1610 : for the leader slot. So the race happens when FEC sets are
1611 : delivered and processed sooner than the PoH hash, aka when the
1612 : poh=>shred=>replay path for the block id somehow beats the
1613 : poh=>replay path for the poh hash. To mitigate this race,
1614 : we must block on ingesting the FEC set for the ensuing slot
1615 : before the leader bank freezes, because that would violate
1616 : ordering invariants in banks and sched. */
1617 0 : FD_TEST( ctx->recv_block_id );
1618 0 : FD_TEST( !ctx->recv_poh );
1619 0 : return 0;
1620 0 : }
1621 :
1622 : /* If fec_set_idx is 0, we need a new bank for a new slot. Banks must
1623 : not be full in this case. */
1624 0 : if( FD_UNLIKELY( fd_banks_is_full( ctx->banks ) && fec->fec_set_idx==0 ) ) return 0;
1625 :
1626 : /* Otherwise, banks may not be full, so we can always create a new
1627 : bank if needed. Or, if banks are full, the current fec set's
1628 : ancestor (idx 0) already created a bank for this slot.*/
1629 0 : return 1;
1630 0 : }
1631 :
1632 : static void
1633 : process_fec_set( fd_replay_tile_t * ctx,
1634 : fd_stem_context_t * stem,
1635 0 : fd_reasm_fec_t * reasm_fec ) {
1636 0 : long now = fd_log_wallclock();
1637 :
1638 : /* Linking only requires a shared lock because the fields that are
1639 : modified are only read on publish which uses exclusive lock. */
1640 :
1641 0 : long shacq_start, shacq_end, shrel_end;
1642 :
1643 0 : FD_STORE_SHARED_LOCK( ctx->store, shacq_start, shacq_end, shrel_end ) {
1644 0 : if( FD_UNLIKELY( !fd_store_link( ctx->store, &reasm_fec->key, &reasm_fec->cmr ) ) ) FD_LOG_WARNING(( "failed to link %s %s. slot %lu fec_set_idx %u", FD_BASE58_ENC_32_ALLOCA( &reasm_fec->key ), FD_BASE58_ENC_32_ALLOCA( &reasm_fec->cmr ), reasm_fec->slot, reasm_fec->fec_set_idx ));
1645 0 : } FD_STORE_SHARED_LOCK_END;
1646 0 : fd_histf_sample( ctx->metrics.store_link_wait, (ulong)fd_long_max( shacq_end - shacq_start, 0L ) );
1647 0 : fd_histf_sample( ctx->metrics.store_link_work, (ulong)fd_long_max( shrel_end - shacq_end, 0L ) );
1648 :
1649 : /* Update the reasm_fec with the correct bank index and parent bank
1650 : index. If the FEC belongs to a leader, we have already allocated
1651 : a bank index for the FEC and it just needs to be propagated to the
1652 : reasm_fec. */
1653 :
1654 0 : reasm_fec->parent_bank_idx = fd_reasm_parent( ctx->reasm, reasm_fec )->bank_idx;
1655 :
1656 0 : if( FD_UNLIKELY( reasm_fec->leader ) ) {
1657 : /* If we are the leader we just need to copy in the bank index that
1658 : the leader slot is using. */
1659 0 : FD_TEST( ctx->leader_bank!=NULL );
1660 0 : reasm_fec->bank_idx = ctx->leader_bank->idx;
1661 0 : } else if( FD_UNLIKELY( reasm_fec->fec_set_idx==0U ) ) {
1662 : /* If we are seeing a FEC with fec set idx 0, this means that we are
1663 : starting a new slot, and we need a new bank index. */
1664 0 : reasm_fec->bank_idx = fd_banks_new_bank( ctx->banks, reasm_fec->parent_bank_idx, now )->idx;
1665 0 : } else {
1666 : /* We are continuing to execute through a slot that we already have
1667 : a bank index for. */
1668 0 : reasm_fec->bank_idx = reasm_fec->parent_bank_idx;
1669 0 : }
1670 :
1671 0 : if( FD_UNLIKELY( reasm_fec->slot_complete ) ) {
1672 : /* Once the block id for a block is known it must be added to the
1673 : leader block mapping. */
1674 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1675 0 : FD_TEST( block_id_ele );
1676 :
1677 : /* If an entry already exists for this bank index in the block id
1678 : map, we can safely remove it and replace it with the new entry.
1679 : This is safe because we know that the old entry for this fork
1680 : index has already been pruned away. */
1681 0 : if( FD_LIKELY( block_id_ele->slot!=FD_SLOT_NULL && fd_block_id_map_ele_query( ctx->block_id_map, &block_id_ele->block_id, NULL, ctx->block_id_arr ) ) ) {
1682 0 : FD_TEST( fd_block_id_map_ele_remove( ctx->block_id_map, &block_id_ele->block_id, NULL, ctx->block_id_arr ) );
1683 0 : }
1684 :
1685 0 : block_id_ele->block_id = reasm_fec->key;
1686 0 : block_id_ele->slot = reasm_fec->slot;
1687 :
1688 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1689 :
1690 0 : if( FD_UNLIKELY( reasm_fec->leader ) ) {
1691 0 : ctx->recv_block_id = 1;
1692 0 : }
1693 0 : }
1694 :
1695 0 : if( FD_UNLIKELY( reasm_fec->leader ) ) {
1696 0 : return;
1697 0 : }
1698 :
1699 : /* Forks form a partial ordering over FEC sets. The Repair tile
1700 : delivers FEC sets in-order per fork, but FEC set ordering across
1701 : forks is arbitrary */
1702 0 : fd_sched_fec_t sched_fec[ 1 ];
1703 :
1704 : # if DEBUG_LOGGING
1705 : FD_LOG_INFO(( "replay processing FEC set for slot %lu fec_set_idx %u, mr %s cmr %s", reasm_fec->slot, reasm_fec->fec_set_idx, FD_BASE58_ENC_32_ALLOCA( &reasm_fec->key ), FD_BASE58_ENC_32_ALLOCA( &reasm_fec->cmr ) ));
1706 : # endif
1707 :
1708 : /* Read FEC set from the store. This should happen before we try to
1709 : ingest the FEC set. This allows us to filter out frags that were
1710 : in-flight when we published away minority forks that the frags land
1711 : on. These frags would have no bank to execute against, because
1712 : their corresponding banks, or parent banks, have also been pruned
1713 : during publishing. A query against store will rightfully tell us
1714 : that the underlying data is not found, implying that this is for a
1715 : minority fork that we can safely ignore. */
1716 0 : FD_STORE_SHARED_LOCK( ctx->store, shacq_start, shacq_end, shrel_end ) {
1717 0 : fd_store_fec_t * store_fec = fd_store_query( ctx->store, &reasm_fec->key );
1718 0 : if( FD_UNLIKELY( !store_fec ) ) {
1719 : /* The only case in which a FEC is not found in the store after
1720 : repair has notified is if the FEC was on a minority fork that
1721 : has already been published away. In this case we abandon the
1722 : entire slice because it is no longer relevant. */
1723 0 : FD_LOG_WARNING(( "store fec for slot: %lu is on minority fork already pruned by publish. abandoning slice. root: %lu. pruned merkle: %s", reasm_fec->slot, ctx->consensus_root_slot, FD_BASE58_ENC_32_ALLOCA( &reasm_fec->key ) ));
1724 0 : return;
1725 0 : }
1726 0 : FD_TEST( store_fec );
1727 0 : sched_fec->fec = store_fec;
1728 0 : sched_fec->shred_cnt = reasm_fec->data_cnt;
1729 0 : } FD_STORE_SHARED_LOCK_END;
1730 :
1731 0 : fd_histf_sample( ctx->metrics.store_read_wait, (ulong)fd_long_max( shacq_end - shacq_start, 0UL ) );
1732 0 : fd_histf_sample( ctx->metrics.store_read_work, (ulong)fd_long_max( shrel_end - shacq_end, 0UL ) );
1733 :
1734 0 : sched_fec->is_last_in_batch = !!reasm_fec->data_complete;
1735 0 : sched_fec->is_last_in_block = !!reasm_fec->slot_complete;
1736 0 : sched_fec->bank_idx = reasm_fec->bank_idx;
1737 0 : sched_fec->parent_bank_idx = reasm_fec->parent_bank_idx;
1738 0 : sched_fec->slot = reasm_fec->slot;
1739 0 : sched_fec->parent_slot = reasm_fec->slot - reasm_fec->parent_off;
1740 0 : sched_fec->is_first_in_block = reasm_fec->fec_set_idx==0U;
1741 0 : fd_funk_txn_xid_copy( sched_fec->alut_ctx->xid, fd_funk_last_publish( ctx->accdb_admin->funk ) );
1742 0 : sched_fec->alut_ctx->accdb[0] = ctx->accdb[0];
1743 0 : sched_fec->alut_ctx->els = ctx->published_root_slot;
1744 :
1745 0 : if( FD_UNLIKELY( !fd_sched_fec_ingest( ctx->sched, sched_fec ) ) ) {
1746 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, sched_fec->bank_idx );
1747 0 : publish_slot_dead( ctx, stem, bank );
1748 0 : fd_banks_mark_bank_dead( ctx->banks, bank );
1749 0 : }
1750 0 : }
1751 :
1752 : static void
1753 : funk_publish( fd_replay_tile_t * ctx,
1754 : ulong slot,
1755 0 : ulong bank_idx ) {
1756 0 : fd_funk_txn_xid_t xid = { .ul[0] = slot, .ul[1] = bank_idx };
1757 0 : FD_LOG_DEBUG(( "publishing slot=%lu", slot ));
1758 :
1759 : /* This is the standard case. Publish all transactions up to and
1760 : including the watermark. This will publish any in-prep ancestors
1761 : of root_txn as well. */
1762 0 : fd_accdb_advance_root( ctx->accdb_admin, &xid );
1763 0 : fd_progcache_txn_advance_root( ctx->progcache_admin, &xid );
1764 0 : }
1765 :
1766 : static int
1767 0 : advance_published_root( fd_replay_tile_t * ctx ) {
1768 :
1769 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_map_ele_query( ctx->block_id_map, &ctx->consensus_root, NULL, ctx->block_id_arr );
1770 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
1771 0 : FD_LOG_CRIT(( "invariant violation: block id ele not found for consensus root %s", FD_BASE58_ENC_32_ALLOCA( &ctx->consensus_root ) ));
1772 0 : }
1773 0 : ulong target_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
1774 :
1775 0 : fd_sched_root_notify( ctx->sched, target_bank_idx );
1776 :
1777 : /* If the identity vote has been seen on a bank that should be rooted,
1778 : then we are now ready to produce blocks. */
1779 0 : if( FD_UNLIKELY( !ctx->has_identity_vote_rooted ) ) {
1780 0 : fd_bank_t * root_bank = fd_banks_bank_query( ctx->banks, target_bank_idx );
1781 0 : if( FD_UNLIKELY( !root_bank ) ) FD_LOG_CRIT(( "invariant violation: root bank not found for bank index %lu", target_bank_idx ));
1782 0 : if( FD_LIKELY( fd_bank_has_identity_vote_get( root_bank ) ) ) ctx->has_identity_vote_rooted = 1;
1783 0 : }
1784 :
1785 0 : ulong advanceable_root_idx = ULONG_MAX;
1786 0 : if( FD_UNLIKELY( !fd_banks_advance_root_prepare( ctx->banks, target_bank_idx, &advanceable_root_idx ) ) ) return 0;
1787 :
1788 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, advanceable_root_idx );
1789 0 : FD_TEST( bank );
1790 :
1791 0 : fd_block_id_ele_t * advanceable_root_ele = &ctx->block_id_arr[ advanceable_root_idx ];
1792 0 : if( FD_UNLIKELY( !advanceable_root_ele ) ) {
1793 0 : FD_LOG_CRIT(( "invariant violation: advanceable root ele not found for bank index %lu", advanceable_root_idx ));
1794 0 : }
1795 :
1796 0 : long exacq_start, exacq_end, exrel_end;
1797 0 : FD_STORE_EXCLUSIVE_LOCK( ctx->store, exacq_start, exacq_end, exrel_end ) {
1798 0 : fd_store_publish( ctx->store, &advanceable_root_ele->block_id );
1799 0 : } FD_STORE_EXCLUSIVE_LOCK_END;
1800 :
1801 0 : fd_histf_sample( ctx->metrics.store_publish_wait, (ulong)fd_long_max( exacq_end-exacq_start, 0UL ) );
1802 0 : fd_histf_sample( ctx->metrics.store_publish_work, (ulong)fd_long_max( exrel_end-exacq_end, 0UL ) );
1803 :
1804 0 : ulong advanceable_root_slot = fd_bank_slot_get( bank );
1805 0 : funk_publish( ctx, advanceable_root_slot, bank->idx );
1806 :
1807 0 : fd_txncache_advance_root( ctx->txncache, bank->txncache_fork_id );
1808 0 : fd_sched_advance_root( ctx->sched, advanceable_root_idx );
1809 0 : fd_banks_advance_root( ctx->banks, advanceable_root_idx );
1810 0 : fd_reasm_publish( ctx->reasm, &advanceable_root_ele->block_id );
1811 :
1812 0 : ctx->published_root_slot = advanceable_root_slot;
1813 0 : ctx->published_root_bank_idx = advanceable_root_idx;
1814 :
1815 0 : return 1;
1816 0 : }
1817 :
1818 : static void
1819 : after_credit( fd_replay_tile_t * ctx,
1820 : fd_stem_context_t * stem,
1821 : int * opt_poll_in,
1822 0 : int * charge_busy ) {
1823 0 : if( FD_UNLIKELY( !ctx->is_booted ) ) return;
1824 :
1825 0 : if( FD_UNLIKELY( maybe_become_leader( ctx, stem ) ) ) {
1826 0 : *charge_busy = 1;
1827 0 : *opt_poll_in = 0;
1828 0 : return;
1829 0 : }
1830 :
1831 : /* If we are leader, we can only unbecome the leader iff we have
1832 : received the poh hash from the poh tile and block id from reasm. */
1833 0 : if( FD_UNLIKELY( ctx->is_leader && ctx->recv_block_id && ctx->recv_poh ) ) {
1834 0 : fini_leader_bank( ctx, stem );
1835 0 : *charge_busy = 1;
1836 0 : *opt_poll_in = 0;
1837 0 : return;
1838 0 : }
1839 :
1840 : /* If the published_root is not caught up to the consensus root, then
1841 : we should try to advance the published root. */
1842 0 : if( FD_UNLIKELY( ctx->consensus_root_bank_idx!=ctx->published_root_bank_idx && advance_published_root( ctx ) ) ) {
1843 0 : *charge_busy = 1;
1844 0 : *opt_poll_in = 0;
1845 0 : return;
1846 0 : }
1847 :
1848 : /* If the reassembler has a fec that is ready, we should process it
1849 : and pass it to the scheduler. */
1850 :
1851 : /* FIXME: The reasm logic needs to get reworked to support
1852 : equivocation more robustly. */
1853 0 : if( FD_LIKELY( can_process_fec( ctx ) ) ) {
1854 0 : fd_reasm_fec_t * fec = fd_reasm_peek( ctx->reasm );
1855 :
1856 : /* If fec->eqvoc is set that means that equivocation mid-block was
1857 : detected in fd_reasm_t. We need to replay up to and including
1858 : the equivocating FEC on a new bank. */
1859 :
1860 0 : if( FD_UNLIKELY( fec->eqvoc ) ) {
1861 0 : FD_LOG_WARNING(( "Block equivocation detected at slot %lu", fec->slot ));
1862 :
1863 : /* We need to figure out which and how many FECs we need to
1864 : (re)insert into the scheduler. We work backwards from the
1865 : equivocating FEC, querying for chained merkle roots until we
1866 : reach the first FEC in the slot.
1867 : TODO: replace the magic number with a constant for the max
1868 : number of fecs possible in a slot with fix-32. */
1869 0 : fd_reasm_fec_t * fecs[ 1024 ] = { [0] = fec };
1870 0 : ulong fec_cnt = 1UL;
1871 0 : while( fecs[ fec_cnt-1UL ]->fec_set_idx!=0UL ) {
1872 0 : fec = fd_reasm_query( ctx->reasm, &fecs[ fec_cnt-1UL ]->cmr );
1873 0 : fecs[ fec_cnt++ ] = fec;
1874 0 : }
1875 :
1876 : /* If we don't have enough space in the scheduler to ingest all of
1877 : FECs, we can't proceed yet. */
1878 0 : if( FD_UNLIKELY( !fd_sched_can_ingest( ctx->sched, fec_cnt ) ) ) return;
1879 :
1880 : /* Now that we have validated that sched can ingest all of the
1881 : required FECs, it is finally safe to remove the equivocating
1882 : fec from the reasm deque. */
1883 0 : fd_reasm_out( ctx->reasm );
1884 :
1885 : /* Now we can process all of the FECs. */
1886 0 : for( ulong i=fec_cnt; i>0UL; i-- ) {
1887 0 : process_fec_set( ctx, stem, fecs[i-1UL] );
1888 0 : }
1889 0 : } else {
1890 : /* Standard case. */
1891 0 : fec = fd_reasm_out( ctx->reasm );
1892 0 : process_fec_set( ctx, stem, fec );
1893 0 : }
1894 :
1895 0 : *charge_busy = 1;
1896 0 : *opt_poll_in = 0;
1897 0 : return;
1898 0 : }
1899 :
1900 0 : *charge_busy = replay( ctx, stem );
1901 0 : *opt_poll_in = !*charge_busy;
1902 0 : }
1903 :
1904 : static int
1905 : before_frag( fd_replay_tile_t * ctx,
1906 : ulong in_idx,
1907 : ulong seq FD_PARAM_UNUSED,
1908 0 : ulong sig FD_PARAM_UNUSED ) {
1909 :
1910 0 : if( FD_UNLIKELY( ctx->in_kind[ in_idx ]==IN_KIND_SHRED ) ) {
1911 : /* If reasm is full, we can not insert any more FEC sets. We must
1912 : not consume any frags from shred_out until reasm can process more
1913 : FEC sets. */
1914 :
1915 0 : if( FD_UNLIKELY( !fd_reasm_free( ctx->reasm ) ) ) {
1916 0 : return -1;
1917 0 : }
1918 0 : }
1919 :
1920 0 : return 0;
1921 0 : }
1922 :
1923 : static void
1924 : process_solcap_account_update( fd_replay_tile_t * ctx,
1925 0 : fd_capture_ctx_account_update_msg_t const * msg ) {
1926 :
1927 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, msg->bank_idx );
1928 0 : if( FD_UNLIKELY( !bank ) ) {
1929 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", msg->bank_idx ));
1930 0 : }
1931 :
1932 0 : if( FD_UNLIKELY( !ctx->capture_ctx || !ctx->capture_ctx->capture ) ) return;
1933 0 : if( FD_UNLIKELY( fd_bank_slot_get( bank )<ctx->capture_ctx->solcap_start_slot ) ) return;
1934 :
1935 0 : uchar const * account_data = (uchar const *)fd_type_pun_const( msg )+sizeof(fd_capture_ctx_account_update_msg_t);
1936 0 : fd_solcap_write_account( ctx->capture_ctx->capture, &msg->pubkey, &msg->info, account_data, msg->data_sz );
1937 0 : }
1938 :
1939 : static void
1940 : process_exec_task_done( fd_replay_tile_t * ctx,
1941 : fd_stem_context_t * stem,
1942 : fd_exec_task_done_msg_t * msg,
1943 0 : ulong sig ) {
1944 0 : if( FD_UNLIKELY( sig==0UL ) ) {
1945 : // FIXME remove this branch with new solcap
1946 0 : process_solcap_account_update( ctx, fd_type_pun( msg ) );
1947 0 : return;
1948 0 : }
1949 :
1950 0 : ulong exec_tile_idx = sig&0xFFFFFFFFUL;
1951 :
1952 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, msg->bank_idx );
1953 0 : bank->refcnt--;
1954 :
1955 0 : switch( sig>>32 ) {
1956 0 : case FD_EXEC_TT_TXN_EXEC: {
1957 0 : if( FD_UNLIKELY( !ctx->has_identity_vote_rooted ) ) {
1958 : /* Query the txn signature against our recently generated vote
1959 : txn signatures. If the query is successful, then we have
1960 : seen our own vote transaction land and this should be marked
1961 : in the bank. We go through this exercise until we've seen
1962 : our vote rooted. */
1963 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, msg->txn_exec->txn_idx );
1964 0 : if( fd_vote_tracker_query_sig( ctx->vote_tracker, fd_type_pun_const( txn_p->payload+TXN( txn_p )->signature_off ) ) ) {
1965 0 : *fd_bank_has_identity_vote_modify( bank ) += 1;
1966 0 : }
1967 0 : }
1968 0 : if( FD_UNLIKELY( msg->txn_exec->err && !(bank->flags&FD_BANK_FLAGS_DEAD) ) ) {
1969 : /* Every transaction in a valid block has to execute.
1970 : Otherwise, we should mark the block as dead. Also freeze the
1971 : bank if possible. */
1972 0 : publish_slot_dead( ctx, stem, bank );
1973 0 : fd_banks_mark_bank_dead( ctx->banks, bank );
1974 0 : fd_sched_block_abandon( ctx->sched, bank->idx );
1975 0 : }
1976 0 : if( FD_UNLIKELY( (bank->flags&FD_BANK_FLAGS_DEAD) && bank->refcnt==0UL ) ) {
1977 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
1978 0 : }
1979 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_EXEC, msg->txn_exec->txn_idx, exec_tile_idx );
1980 0 : break;
1981 0 : }
1982 0 : case FD_EXEC_TT_TXN_SIGVERIFY: {
1983 0 : if( FD_UNLIKELY( msg->txn_sigverify->err && !(bank->flags&FD_BANK_FLAGS_DEAD) ) ) {
1984 : /* Every transaction in a valid block has to sigverify.
1985 : Otherwise, we should mark the block as dead. Also freeze the
1986 : bank if possible. */
1987 0 : publish_slot_dead( ctx, stem, bank );
1988 0 : fd_banks_mark_bank_dead( ctx->banks, bank );
1989 0 : fd_sched_block_abandon( ctx->sched, bank->idx );
1990 0 : }
1991 0 : if( FD_UNLIKELY( (bank->flags&FD_BANK_FLAGS_DEAD) && bank->refcnt==0UL ) ) {
1992 0 : fd_banks_mark_bank_frozen( ctx->banks, bank );
1993 0 : }
1994 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_SIGVERIFY, msg->txn_sigverify->txn_idx, exec_tile_idx );
1995 0 : break;
1996 0 : }
1997 0 : default: FD_LOG_CRIT(( "unexpected sig 0x%lx", sig ));
1998 0 : }
1999 :
2000 : /* Reference counter just decreased, and an exec tile just got freed
2001 : up. If there's a need to be more aggressively pruning, we could
2002 : check here if more slots just became publishable and publish. Not
2003 : publishing here shouldn't bloat the fork tree too much though. We
2004 : mark minority forks dead as soon as we can, and execution dispatch
2005 : stops on dead blocks. So shortly afterwards, dead blocks should be
2006 : eligible for pruning as in-flight transactions retire from the
2007 : execution pipeline. */
2008 :
2009 0 : }
2010 :
2011 : static void
2012 : process_tower_slot_done( fd_replay_tile_t * ctx,
2013 : fd_stem_context_t * stem,
2014 0 : fd_tower_slot_done_t const * msg ) {
2015 0 : fd_bank_t * replay_bank = fd_banks_bank_query( ctx->banks, msg->replay_bank_idx );
2016 0 : if( FD_UNLIKELY( !replay_bank ) ) FD_LOG_CRIT(( "invariant violation: bank not found for bank index %lu", msg->replay_bank_idx ));
2017 0 : replay_bank->refcnt--;
2018 :
2019 0 : ctx->reset_block_id = msg->reset_block_id;
2020 0 : ctx->reset_slot = msg->reset_slot;
2021 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
2022 0 : ulong min_leader_slot = fd_ulong_max( msg->reset_slot+1UL, fd_ulong_if( ctx->highwater_leader_slot==ULONG_MAX, 0UL, ctx->highwater_leader_slot+1UL ) );
2023 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, min_leader_slot, ctx->identity_pubkey );
2024 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
2025 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
2026 0 : } else {
2027 0 : ctx->next_leader_tickcount = LONG_MAX;
2028 0 : }
2029 :
2030 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_map_ele_query( ctx->block_id_map, &msg->reset_block_id, NULL, ctx->block_id_arr );
2031 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2032 0 : FD_LOG_CRIT(( "invariant violation: block id ele doesn't exist for reset block id: %s, slot: %lu", FD_BASE58_ENC_32_ALLOCA( &msg->reset_block_id ), msg->reset_slot ));
2033 0 : }
2034 0 : ulong reset_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2035 :
2036 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, reset_bank_idx );
2037 0 : if( FD_UNLIKELY( !bank ) ) {
2038 0 : FD_LOG_CRIT(( "invariant violation: bank not found for bank index %lu", reset_bank_idx ));
2039 0 : }
2040 :
2041 0 : if( FD_LIKELY( msg->root_slot!=ULONG_MAX ) ) FD_TEST( msg->root_slot<=msg->reset_slot );
2042 0 : ctx->reset_bank = bank;
2043 :
2044 0 : if( FD_LIKELY( ctx->replay_out->idx!=ULONG_MAX ) ) {
2045 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2046 :
2047 0 : reset->bank_idx = bank->idx;
2048 0 : reset->timestamp = ctx->reset_timestamp_nanos;
2049 0 : reset->completed_slot = ctx->reset_slot;
2050 0 : reset->hashcnt_per_tick = fd_bank_hashes_per_tick_get( bank );
2051 0 : reset->ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
2052 0 : reset->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)reset->ticks_per_slot);
2053 :
2054 0 : fd_memcpy( reset->completed_block_id, &block_id_ele->block_id, sizeof(fd_hash_t) );
2055 :
2056 0 : fd_blockhashes_t const * block_hash_queue = fd_bank_block_hash_queue_query( bank );
2057 0 : fd_hash_t const * last_hash = fd_blockhashes_peek_last_hash( block_hash_queue );
2058 0 : FD_TEST( last_hash );
2059 0 : fd_memcpy( reset->completed_blockhash, last_hash->uc, sizeof(fd_hash_t) );
2060 :
2061 0 : ulong ticks_per_slot = fd_bank_ticks_per_slot_get( bank );
2062 0 : if( FD_UNLIKELY( reset->hashcnt_per_tick==1UL ) ) {
2063 : /* Low power producer, maximum of one microblock per tick in the slot */
2064 0 : reset->max_microblocks_in_slot = ticks_per_slot;
2065 0 : } else {
2066 : /* See the long comment in after_credit for this limit */
2067 0 : reset->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ticks_per_slot*(reset->hashcnt_per_tick-1UL) );
2068 0 : }
2069 0 : reset->next_leader_slot = ctx->next_leader_slot;
2070 :
2071 0 : if( FD_LIKELY( ctx->rpc_enabled ) ) bank->refcnt++;
2072 :
2073 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_RESET, ctx->replay_out->chunk, sizeof(fd_poh_reset_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
2074 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_poh_reset_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
2075 0 : }
2076 :
2077 0 : FD_LOG_INFO(( "tower_slot_done(reset_slot=%lu, next_leader_slot=%lu, vote_slot=%lu)", msg->reset_slot, ctx->next_leader_slot, msg->vote_slot ));
2078 0 : maybe_become_leader( ctx, stem );
2079 :
2080 0 : if( FD_LIKELY( msg->root_slot!=ULONG_MAX ) ) {
2081 :
2082 0 : FD_TEST( msg->root_slot>=ctx->consensus_root_slot );
2083 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_map_ele_query( ctx->block_id_map, &msg->root_block_id, NULL, ctx->block_id_arr );
2084 0 : FD_TEST( block_id_ele );
2085 :
2086 0 : ctx->consensus_root_slot = msg->root_slot;
2087 0 : ctx->consensus_root = msg->root_block_id;
2088 0 : ctx->consensus_root_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2089 :
2090 0 : publish_root_advanced( ctx, stem );
2091 0 : }
2092 :
2093 0 : ulong distance = 0UL;
2094 0 : fd_bank_t * parent = bank;
2095 0 : while( parent ) {
2096 0 : if( FD_UNLIKELY( parent->idx==ctx->consensus_root_bank_idx ) ) break;
2097 0 : parent = fd_banks_get_parent( ctx->banks, parent );
2098 0 : distance++;
2099 0 : }
2100 :
2101 0 : FD_MGAUGE_SET( REPLAY, ROOT_DISTANCE, distance );
2102 0 : }
2103 :
2104 : static void
2105 : process_fec_complete( fd_replay_tile_t * ctx,
2106 0 : uchar const * shred_buf ) {
2107 0 : fd_shred_t const * shred = (fd_shred_t const *)fd_type_pun_const( shred_buf );
2108 :
2109 0 : fd_hash_t const * merkle_root = (fd_hash_t const *)fd_type_pun_const( shred_buf + FD_SHRED_DATA_HEADER_SZ );
2110 0 : fd_hash_t const * chained_merkle_root = (fd_hash_t const *)fd_type_pun_const( shred_buf + FD_SHRED_DATA_HEADER_SZ + sizeof(fd_hash_t) );
2111 0 : int is_leader_fec = *(int const *) fd_type_pun_const( shred_buf + FD_SHRED_DATA_HEADER_SZ + sizeof(fd_hash_t) + sizeof(fd_hash_t) );
2112 :
2113 0 : int data_complete = !!( shred->data.flags & FD_SHRED_DATA_FLAG_DATA_COMPLETE );
2114 0 : int slot_complete = !!( shred->data.flags & FD_SHRED_DATA_FLAG_SLOT_COMPLETE );
2115 :
2116 0 : FD_TEST( !fd_reasm_query( ctx->reasm, merkle_root ) );
2117 0 : if( FD_UNLIKELY( shred->slot - shred->data.parent_off == fd_reasm_slot0( ctx->reasm ) && shred->fec_set_idx == 0) ) {
2118 0 : chained_merkle_root = &fd_reasm_root( ctx->reasm )->key;
2119 0 : }
2120 :
2121 0 : FD_TEST( fd_reasm_free( ctx->reasm ) );
2122 :
2123 0 : FD_TEST( fd_reasm_insert( ctx->reasm, merkle_root, chained_merkle_root, shred->slot, shred->fec_set_idx, shred->data.parent_off, (ushort)(shred->idx - shred->fec_set_idx + 1), data_complete, slot_complete, is_leader_fec ) );
2124 0 : }
2125 :
2126 : static void
2127 0 : process_resolv_slot_completed( fd_replay_tile_t * ctx, ulong bank_idx ) {
2128 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
2129 0 : FD_TEST( bank );
2130 :
2131 0 : bank->refcnt--;
2132 0 : }
2133 :
2134 : static void
2135 : process_vote_txn_sent( fd_replay_tile_t * ctx,
2136 0 : fd_txn_m_t * txnm ) {
2137 : /* The send tile has signed and sent a vote. Add this vote to the
2138 : vote tracker. We go through this exercise until we've seen our
2139 : vote rooted. */
2140 0 : if( FD_UNLIKELY( !ctx->has_identity_vote_rooted ) ) {
2141 0 : uchar * payload = ((uchar *)txnm) + sizeof(fd_txn_m_t);
2142 0 : uchar txn_mem[ FD_TXN_MAX_SZ ] __attribute__((aligned(alignof(fd_txn_t))));
2143 0 : fd_txn_t * txn = (fd_txn_t *)txn_mem;
2144 0 : if( FD_UNLIKELY( !fd_txn_parse( payload, txnm->payload_sz, txn_mem, NULL ) ) ) {
2145 0 : FD_LOG_CRIT(( "Could not parse txn from send tile" ));
2146 0 : }
2147 0 : fd_vote_tracker_insert( ctx->vote_tracker, fd_type_pun_const( payload+txn->signature_off ) );
2148 0 : }
2149 0 : }
2150 :
2151 : static inline void
2152 0 : maybe_verify_shred_version( fd_replay_tile_t * ctx ) {
2153 0 : if( FD_LIKELY( ctx->expected_shred_version && ctx->ipecho_shred_version ) ) {
2154 0 : if( FD_UNLIKELY( ctx->expected_shred_version!=ctx->ipecho_shred_version ) ) {
2155 0 : FD_LOG_ERR(( "shred version mismatch: expected %u but got %u from ipecho", ctx->expected_shred_version, ctx->ipecho_shred_version ) );
2156 0 : }
2157 0 : }
2158 :
2159 0 : if( FD_LIKELY( ctx->has_genesis_hash && ctx->hard_forks_cnt!=ULONG_MAX && (ctx->expected_shred_version || ctx->ipecho_shred_version) ) ) {
2160 0 : ushort expected_shred_version = ctx->expected_shred_version ? ctx->expected_shred_version : ctx->ipecho_shred_version;
2161 :
2162 0 : union {
2163 0 : uchar c[ 32 ];
2164 0 : ushort s[ 16 ];
2165 0 : } running_hash;
2166 0 : fd_memcpy( running_hash.c, ctx->genesis_hash, sizeof(fd_hash_t) );
2167 :
2168 0 : ulong processed = 0UL;
2169 0 : ulong min_value = 0UL;
2170 0 : while( processed<ctx->hard_forks_cnt ) {
2171 0 : ulong min_index = ULONG_MAX;
2172 0 : for( ulong i=0UL; i<ctx->hard_forks_cnt; i++ ) {
2173 0 : if( ctx->hard_forks[ i ]>=min_value && (min_index==ULONG_MAX || ctx->hard_forks[ i ]<ctx->hard_forks[ min_index ] ) ) {
2174 0 : min_index = i;
2175 0 : }
2176 0 : }
2177 :
2178 0 : FD_TEST( min_index!=ULONG_MAX );
2179 0 : min_value = ctx->hard_forks[ min_index ];
2180 0 : ulong min_count = 0UL;
2181 0 : for( ulong i=0UL; i<ctx->hard_forks_cnt; i++ ) {
2182 0 : if( ctx->hard_forks[ i ]==min_value ) min_count++;
2183 0 : }
2184 :
2185 0 : uchar data[ 48UL ];
2186 0 : fd_memcpy( data, running_hash.c, sizeof(fd_hash_t) );
2187 0 : fd_memcpy( data+32UL, &min_value, sizeof(ulong) );
2188 0 : fd_memcpy( data+40UL, &min_count, sizeof(ulong) );
2189 :
2190 0 : FD_TEST( fd_sha256_hash( data, 48UL, running_hash.c ) );
2191 0 : processed += min_count;
2192 0 : min_value += 1UL;
2193 0 : }
2194 :
2195 0 : ushort xor = 0;
2196 0 : for( ulong i=0UL; i<16UL; i++ ) xor ^= running_hash.s[ i ];
2197 :
2198 0 : xor = fd_ushort_bswap( xor );
2199 0 : xor = fd_ushort_if( xor<USHORT_MAX, (ushort)(xor + 1), USHORT_MAX );
2200 :
2201 0 : if( FD_UNLIKELY( expected_shred_version!=xor ) ) {
2202 0 : FD_LOG_ERR(( "shred version mismatch: expected %u but got %u from genesis hash %s and hard forks", expected_shred_version, xor, FD_BASE58_ENC_32_ALLOCA( &ctx->genesis_hash ) ));
2203 0 : }
2204 0 : }
2205 0 : }
2206 :
2207 : static inline int
2208 : returnable_frag( fd_replay_tile_t * ctx,
2209 : ulong in_idx,
2210 : ulong seq,
2211 : ulong sig,
2212 : ulong chunk,
2213 : ulong sz,
2214 : ulong ctl,
2215 : ulong tsorig,
2216 : ulong tspub,
2217 0 : fd_stem_context_t * stem ) {
2218 0 : (void)seq;
2219 0 : (void)ctl;
2220 0 : (void)tsorig;
2221 0 : (void)tspub;
2222 :
2223 0 : if( FD_UNLIKELY( sz!=0UL && (chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark || sz>ctx->in[ in_idx ].mtu ) ) )
2224 0 : FD_LOG_ERR(( "chunk %lu %lu from in %d corrupt, not in range [%lu,%lu]", chunk, sz, ctx->in_kind[ in_idx ], ctx->in[ in_idx ].chunk0, ctx->in[ in_idx ].wmark ));
2225 :
2226 0 : switch( ctx->in_kind[in_idx] ) {
2227 0 : case IN_KIND_GENESIS: {
2228 0 : uchar const * src = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2229 0 : ctx->has_genesis_hash = 1;
2230 0 : if( FD_LIKELY( sig==GENESI_SIG_BOOTSTRAP_COMPLETED ) ) {
2231 0 : boot_genesis( ctx, stem, in_idx, chunk );
2232 0 : fd_memcpy( ctx->genesis_hash, src+sizeof(fd_lthash_value_t), sizeof(fd_hash_t) );
2233 0 : } else {
2234 0 : fd_memcpy( ctx->genesis_hash, src, sizeof(fd_hash_t) );
2235 0 : }
2236 :
2237 0 : maybe_verify_cluster_type( ctx );
2238 0 : maybe_verify_shred_version( ctx );
2239 0 : break;
2240 0 : }
2241 0 : case IN_KIND_IPECHO: {
2242 0 : FD_TEST( sig && sig<=USHORT_MAX );
2243 0 : ctx->ipecho_shred_version = (ushort)sig;
2244 0 : maybe_verify_shred_version( ctx );
2245 0 : break;
2246 0 : }
2247 0 : case IN_KIND_SNAP:
2248 0 : on_snapshot_message( ctx, stem, in_idx, chunk, sig );
2249 0 : maybe_verify_shred_version( ctx );
2250 0 : break;
2251 0 : case IN_KIND_EXEC: {
2252 0 : process_exec_task_done( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ), sig );
2253 0 : break;
2254 0 : }
2255 0 : case IN_KIND_POH: {
2256 0 : process_poh_message( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2257 0 : break;
2258 0 : }
2259 0 : case IN_KIND_RESOLV: {
2260 0 : fd_resolv_slot_exchanged_t * exchanged_slot = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2261 0 : process_resolv_slot_completed( ctx, exchanged_slot->bank_idx );
2262 0 : break;
2263 0 : }
2264 0 : case IN_KIND_TOWER: {
2265 0 : if ( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_DONE ) ) process_tower_slot_done( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2266 0 : else if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_CONFIRMED ) ) {
2267 0 : fd_tower_slot_confirmed_t const * msg = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2268 :
2269 : /* Implement replay plugin API here */
2270 :
2271 0 : switch( msg->kind ) {
2272 0 : case FD_TOWER_SLOT_CONFIRMED_OPTIMISTIC: break;
2273 0 : case FD_TOWER_SLOT_CONFIRMED_ROOTED: break;
2274 0 : }
2275 0 : };
2276 0 : break;
2277 0 : }
2278 0 : case IN_KIND_SHRED: {
2279 : /* TODO: This message/sz should be defined. */
2280 0 : if( sz==FD_SHRED_DATA_HEADER_SZ + sizeof(fd_hash_t) + sizeof(fd_hash_t) + sizeof(int) ) {
2281 : /* If receive a FEC complete message. */
2282 0 : process_fec_complete( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2283 0 : }
2284 0 : break;
2285 0 : }
2286 0 : case IN_KIND_VTXN: {
2287 0 : process_vote_txn_sent( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2288 0 : break;
2289 0 : }
2290 0 : case IN_KIND_RPC:
2291 0 : case IN_KIND_GUI: {
2292 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, sig );
2293 0 : FD_TEST( bank );
2294 0 : bank->refcnt--;
2295 0 : break;
2296 0 : }
2297 0 : default:
2298 0 : FD_LOG_ERR(( "unhandled kind %d", ctx->in_kind[ in_idx ] ));
2299 0 : }
2300 :
2301 0 : return 0;
2302 0 : }
2303 :
2304 : static inline fd_replay_out_link_t
2305 : out1( fd_topo_t const * topo,
2306 : fd_topo_tile_t const * tile,
2307 0 : char const * name ) {
2308 0 : ulong idx = ULONG_MAX;
2309 :
2310 0 : for( ulong i=0UL; i<tile->out_cnt; i++ ) {
2311 0 : fd_topo_link_t const * link = &topo->links[ tile->out_link_id[ i ] ];
2312 0 : if( !strcmp( link->name, name ) ) {
2313 0 : if( FD_UNLIKELY( idx!=ULONG_MAX ) ) FD_LOG_ERR(( "tile %s:%lu had multiple output links named %s but expected one", tile->name, tile->kind_id, name ));
2314 0 : idx = i;
2315 0 : }
2316 0 : }
2317 :
2318 0 : if( FD_UNLIKELY( idx==ULONG_MAX ) ) return (fd_replay_out_link_t){ .idx = ULONG_MAX, .mem = NULL, .chunk0 = 0, .wmark = 0, .chunk = 0 };
2319 :
2320 0 : void * mem = topo->workspaces[ topo->objs[ topo->links[ tile->out_link_id[ idx ] ].dcache_obj_id ].wksp_id ].wksp;
2321 0 : ulong chunk0 = fd_dcache_compact_chunk0( mem, topo->links[ tile->out_link_id[ idx ] ].dcache );
2322 0 : ulong wmark = fd_dcache_compact_wmark ( mem, topo->links[ tile->out_link_id[ idx ] ].dcache, topo->links[ tile->out_link_id[ idx ] ].mtu );
2323 :
2324 0 : return (fd_replay_out_link_t){ .idx = idx, .mem = mem, .chunk0 = chunk0, .wmark = wmark, .chunk = chunk0 };
2325 0 : }
2326 :
2327 : static void
2328 : privileged_init( fd_topo_t * topo,
2329 0 : fd_topo_tile_t * tile ) {
2330 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2331 :
2332 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2333 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
2334 :
2335 0 : if( FD_UNLIKELY( !strcmp( tile->replay.identity_key_path, "" ) ) ) FD_LOG_ERR(( "identity_key_path not set" ));
2336 :
2337 0 : ctx->identity_pubkey[ 0 ] = *(fd_pubkey_t const *)fd_type_pun_const( fd_keyload_load( tile->replay.identity_key_path, /* pubkey only: */ 1 ) );
2338 :
2339 0 : if( FD_UNLIKELY( !tile->replay.bundle.vote_account_path[0] ) ) {
2340 0 : tile->replay.bundle.enabled = 0;
2341 0 : }
2342 0 : if( FD_UNLIKELY( tile->replay.bundle.enabled ) ) {
2343 0 : if( FD_UNLIKELY( !fd_base58_decode_32( tile->replay.bundle.vote_account_path, ctx->bundle.vote_account.uc ) ) ) {
2344 0 : const uchar * vote_key = fd_keyload_load( tile->replay.bundle.vote_account_path, /* pubkey only: */ 1 );
2345 0 : fd_memcpy( ctx->bundle.vote_account.uc, vote_key, 32UL );
2346 0 : }
2347 0 : }
2348 0 : }
2349 :
2350 : static void
2351 : unprivileged_init( fd_topo_t * topo,
2352 0 : fd_topo_tile_t * tile ) {
2353 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2354 :
2355 0 : ulong chain_cnt = fd_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
2356 :
2357 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2358 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
2359 0 : void * block_id_arr_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_block_id_ele_t), sizeof(fd_block_id_ele_t) * tile->replay.max_live_slots );
2360 0 : void * block_id_map_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_block_id_map_align(), fd_block_id_map_footprint( chain_cnt ) );
2361 0 : void * _txncache = FD_SCRATCH_ALLOC_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
2362 0 : void * reasm_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_reasm_align(), fd_reasm_footprint( 1 << 20 ) );
2363 0 : void * sched_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.max_live_slots ) );
2364 0 : void * vote_tracker_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
2365 0 : void * _capture_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
2366 0 : # if FD_HAS_FLATCC
2367 0 : void * block_dump_ctx = NULL;
2368 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
2369 0 : block_dump_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
2370 0 : }
2371 0 : # endif
2372 :
2373 0 : ulong store_obj_id = fd_pod_query_ulong( topo->props, "store", ULONG_MAX );
2374 0 : FD_TEST( store_obj_id!=ULONG_MAX );
2375 0 : ctx->store = fd_store_join( fd_topo_obj_laddr( topo, store_obj_id ) );
2376 0 : FD_TEST( ctx->store );
2377 :
2378 0 : ulong banks_obj_id = fd_pod_query_ulong( topo->props, "banks", ULONG_MAX );
2379 0 : FD_TEST( banks_obj_id!=ULONG_MAX );
2380 0 : ctx->banks = fd_banks_join( fd_topo_obj_laddr( topo, banks_obj_id ) );
2381 0 : FD_TEST( ctx->banks );
2382 :
2383 0 : fd_bank_t * bank_pool = fd_banks_get_bank_pool( ctx->banks );
2384 0 : FD_MGAUGE_SET( REPLAY, MAX_LIVE_BANKS, fd_banks_pool_max( bank_pool ) );
2385 :
2386 0 : fd_bank_t * bank = fd_banks_init_bank( ctx->banks );
2387 0 : fd_bank_slot_set( bank, 0UL );
2388 0 : FD_TEST( bank );
2389 0 : FD_TEST( bank->idx==FD_REPLAY_BOOT_BANK_IDX );
2390 :
2391 0 : ctx->consensus_root_slot = ULONG_MAX;
2392 0 : ctx->consensus_root = (fd_hash_t){ .ul[0] = FD_RUNTIME_INITIAL_BLOCK_ID };
2393 0 : ctx->published_root_slot = ULONG_MAX;
2394 :
2395 0 : ctx->expected_shred_version = tile->replay.expected_shred_version;
2396 0 : ctx->ipecho_shred_version = 0;
2397 0 : fd_memcpy( ctx->genesis_path, tile->replay.genesis_path, sizeof(ctx->genesis_path) );
2398 0 : ctx->has_genesis_hash = 0;
2399 0 : ctx->cluster_type = FD_CLUSTER_UNKNOWN;
2400 0 : ctx->hard_forks_cnt = ULONG_MAX;
2401 :
2402 0 : if( FD_UNLIKELY( tile->replay.bundle.enabled ) ) {
2403 0 : ctx->bundle.enabled = 1;
2404 0 : if( FD_UNLIKELY( !fd_bundle_crank_gen_init( ctx->bundle.gen,
2405 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_distribution_program_addr,
2406 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_payment_program_addr,
2407 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc,
2408 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc, "NAN", 0UL ) ) ) {
2409 0 : FD_LOG_ERR(( "failed to initialize bundle crank gen" ));
2410 0 : }
2411 0 : } else {
2412 0 : ctx->bundle.enabled = 0;
2413 0 : }
2414 :
2415 0 : fd_features_t * features = fd_bank_features_modify( bank );
2416 0 : fd_features_enable_cleaned_up( features, &FD_RUNTIME_CLUSTER_VERSION );
2417 :
2418 0 : char const * one_off_features[ 16UL ];
2419 0 : FD_TEST( tile->replay.enable_features_cnt<=sizeof(one_off_features)/sizeof(one_off_features[0]) );
2420 0 : for( ulong i=0UL; i<tile->replay.enable_features_cnt; i++ ) one_off_features[ i ] = tile->replay.enable_features[i];
2421 0 : fd_features_enable_one_offs( features, one_off_features, (uint)tile->replay.enable_features_cnt, 0UL );
2422 :
2423 0 : FD_TEST( fd_accdb_admin_join ( ctx->accdb_admin, fd_topo_obj_laddr( topo, tile->replay.funk_obj_id ) ) );
2424 0 : FD_TEST( fd_accdb_user_v1_init ( ctx->accdb, fd_topo_obj_laddr( topo, tile->replay.funk_obj_id ) ) );
2425 0 : FD_TEST( fd_progcache_admin_join( ctx->progcache_admin, fd_topo_obj_laddr( topo, tile->replay.progcache_obj_id ) ) );
2426 :
2427 0 : void * _txncache_shmem = fd_topo_obj_laddr( topo, tile->replay.txncache_obj_id );
2428 0 : fd_txncache_shmem_t * txncache_shmem = fd_txncache_shmem_join( _txncache_shmem );
2429 0 : FD_TEST( txncache_shmem );
2430 0 : ctx->txncache = fd_txncache_join( fd_txncache_new( _txncache, txncache_shmem ) );
2431 0 : FD_TEST( ctx->txncache );
2432 :
2433 0 : ctx->capture_ctx = NULL;
2434 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) || strcmp( "", tile->replay.dump_proto_dir ) ) ) {
2435 0 : ctx->capture_ctx = fd_capture_ctx_join( fd_capture_ctx_new( _capture_ctx ) );
2436 0 : }
2437 :
2438 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) ) ) {
2439 0 : ctx->capture_ctx->checkpt_freq = ULONG_MAX;
2440 0 : ctx->capture_file = fopen( tile->replay.solcap_capture, "w+" );
2441 0 : if( FD_UNLIKELY( !ctx->capture_file ) ) FD_LOG_ERR(( "fopen(%s) failed (%d-%s)", tile->replay.solcap_capture, errno, fd_io_strerror( errno ) ));
2442 :
2443 0 : ctx->capture_ctx->capture_txns = 0;
2444 0 : ctx->capture_ctx->solcap_start_slot = tile->replay.capture_start_slot;
2445 0 : fd_solcap_writer_init( ctx->capture_ctx->capture, ctx->capture_file );
2446 0 : }
2447 :
2448 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.dump_proto_dir ) ) ) {
2449 0 : ctx->capture_ctx->dump_proto_output_dir = tile->replay.dump_proto_dir;
2450 0 : if( FD_LIKELY( tile->replay.dump_block_to_pb ) ) ctx->capture_ctx->dump_block_to_pb = tile->replay.dump_block_to_pb;
2451 0 : }
2452 :
2453 0 : # if FD_HAS_FLATCC
2454 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
2455 0 : ctx->block_dump_ctx = fd_block_dump_context_join( fd_block_dump_context_new( block_dump_ctx ) );
2456 0 : } else {
2457 0 : ctx->block_dump_ctx = NULL;
2458 0 : }
2459 0 : # endif
2460 :
2461 0 : ctx->exec_cnt = fd_topo_tile_name_cnt( topo, "exec" );
2462 :
2463 0 : ctx->is_booted = 0;
2464 :
2465 0 : ctx->larger_max_cost_per_block = tile->replay.larger_max_cost_per_block;
2466 :
2467 0 : ctx->reasm = fd_reasm_join( fd_reasm_new( reasm_mem, 1 << 20, 0 ) );
2468 0 : FD_TEST( ctx->reasm );
2469 :
2470 0 : ctx->sched = fd_sched_join( fd_sched_new( sched_mem, tile->replay.max_live_slots, ctx->exec_cnt ), tile->replay.max_live_slots );
2471 0 : FD_TEST( ctx->sched );
2472 :
2473 0 : ctx->vote_tracker = fd_vote_tracker_join( fd_vote_tracker_new( vote_tracker_mem, 0UL ) );
2474 0 : FD_TEST( ctx->vote_tracker );
2475 :
2476 0 : ctx->has_identity_vote_rooted = 0;
2477 :
2478 0 : ctx->mleaders = fd_multi_epoch_leaders_join( fd_multi_epoch_leaders_new( ctx->mleaders_mem ) );
2479 0 : FD_TEST( ctx->mleaders );
2480 :
2481 0 : ctx->is_leader = 0;
2482 0 : ctx->reset_slot = 0UL;
2483 0 : ctx->reset_bank = NULL;
2484 0 : ctx->reset_block_id = (fd_hash_t){ .ul[0] = FD_RUNTIME_INITIAL_BLOCK_ID };
2485 0 : ctx->reset_timestamp_nanos = 0UL;
2486 0 : ctx->next_leader_slot = ULONG_MAX;
2487 0 : ctx->next_leader_tickcount = LONG_MAX;
2488 0 : ctx->highwater_leader_slot = ULONG_MAX;
2489 0 : ctx->slot_duration_nanos = 350L*1000L*1000L; /* TODO: Not fixed ... not always 400ms ... */
2490 0 : ctx->slot_duration_ticks = (double)ctx->slot_duration_nanos*fd_tempo_tick_per_ns( NULL );
2491 0 : ctx->leader_bank = NULL;
2492 :
2493 : /* TODO: We need a real seed here. */
2494 0 : ctx->block_id_len = tile->replay.max_live_slots;
2495 0 : ctx->block_id_arr = (fd_block_id_ele_t *)block_id_arr_mem;
2496 0 : ctx->block_id_map = fd_block_id_map_join( fd_block_id_map_new( block_id_map_mem, chain_cnt, 999UL ) );
2497 0 : FD_TEST( ctx->block_id_map );
2498 :
2499 0 : for( ulong i=0UL; i<tile->replay.max_live_slots; i++ ) {
2500 0 : ctx->block_id_arr[ i ].slot = FD_SLOT_NULL;
2501 0 : }
2502 :
2503 0 : ctx->resolv_tile_cnt = fd_topo_tile_name_cnt( topo, "resolv" );
2504 :
2505 0 : FD_TEST( tile->in_cnt<=sizeof(ctx->in)/sizeof(ctx->in[0]) );
2506 0 : for( ulong i=0UL; i<tile->in_cnt; i++ ) {
2507 0 : fd_topo_link_t * link = &topo->links[ tile->in_link_id[ i ] ];
2508 0 : fd_topo_wksp_t * link_wksp = &topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ];
2509 :
2510 0 : if( FD_LIKELY( link->dcache ) ) {
2511 0 : ctx->in[ i ].mem = link_wksp->wksp;
2512 0 : ctx->in[ i ].chunk0 = fd_dcache_compact_chunk0( ctx->in[ i ].mem, link->dcache );
2513 0 : ctx->in[ i ].wmark = fd_dcache_compact_wmark ( ctx->in[ i ].mem, link->dcache, link->mtu );
2514 0 : ctx->in[ i ].mtu = link->mtu;
2515 0 : }
2516 :
2517 0 : if( !strcmp( link->name, "genesi_out" ) ) ctx->in_kind[ i ] = IN_KIND_GENESIS;
2518 0 : else if( !strcmp( link->name, "ipecho_out" ) ) ctx->in_kind[ i ] = IN_KIND_IPECHO;
2519 0 : else if( !strcmp( link->name, "snapin_manif" ) ) ctx->in_kind[ i ] = IN_KIND_SNAP;
2520 0 : else if( !strcmp( link->name, "exec_replay" ) ) ctx->in_kind[ i ] = IN_KIND_EXEC;
2521 0 : else if( !strcmp( link->name, "tower_out" ) ) ctx->in_kind[ i ] = IN_KIND_TOWER;
2522 0 : else if( !strcmp( link->name, "poh_replay" ) ) ctx->in_kind[ i ] = IN_KIND_POH;
2523 0 : else if( !strcmp( link->name, "resolv_repla" ) ) ctx->in_kind[ i ] = IN_KIND_RESOLV;
2524 0 : else if( !strcmp( link->name, "shred_out" ) ) ctx->in_kind[ i ] = IN_KIND_SHRED;
2525 0 : else if( !strcmp( link->name, "send_out" ) ) ctx->in_kind[ i ] = IN_KIND_VTXN;
2526 0 : else if( !strcmp( link->name, "gui_replay" ) ) ctx->in_kind[ i ] = IN_KIND_GUI;
2527 0 : else if( !strcmp( link->name, "rpc_replay" ) ) ctx->in_kind[ i ] = IN_KIND_RPC;
2528 0 : else FD_LOG_ERR(( "unexpected input link name %s", link->name ));
2529 0 : }
2530 :
2531 0 : *ctx->stake_out = out1( topo, tile, "replay_stake" ); FD_TEST( ctx->stake_out->idx!=ULONG_MAX );
2532 0 : *ctx->replay_out = out1( topo, tile, "replay_out" ); FD_TEST( ctx->replay_out->idx!=ULONG_MAX );
2533 :
2534 0 : ulong idx = fd_topo_find_tile_out_link( topo, tile, "replay_exec", 0UL );
2535 0 : FD_TEST( idx!=ULONG_MAX );
2536 0 : fd_topo_link_t * link = &topo->links[ tile->out_link_id[ idx ] ];
2537 :
2538 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
2539 0 : exec_out->idx = idx;
2540 0 : exec_out->mem = topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ].wksp;
2541 0 : exec_out->chunk0 = fd_dcache_compact_chunk0( exec_out->mem, link->dcache );
2542 0 : exec_out->wmark = fd_dcache_compact_wmark( exec_out->mem, link->dcache, link->mtu );
2543 0 : exec_out->chunk = exec_out->chunk0;
2544 :
2545 0 : ctx->gui_enabled = fd_topo_find_tile( topo, "gui", 0UL )!=ULONG_MAX;
2546 0 : ctx->rpc_enabled = fd_topo_find_tile( topo, "rpc", 0UL )!=ULONG_MAX;
2547 :
2548 0 : fd_memset( &ctx->metrics, 0, sizeof(ctx->metrics) );
2549 :
2550 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_link_wait, FD_MHIST_SECONDS_MIN( REPLAY, STORE_LINK_WAIT ),
2551 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_LINK_WAIT ) ) );
2552 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_link_work, FD_MHIST_SECONDS_MIN( REPLAY, STORE_LINK_WORK ),
2553 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_LINK_WORK ) ) );
2554 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_read_wait, FD_MHIST_SECONDS_MIN( REPLAY, STORE_READ_WAIT ),
2555 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_READ_WAIT ) ) );
2556 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_read_work, FD_MHIST_SECONDS_MIN( REPLAY, STORE_READ_WORK ),
2557 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_READ_WORK ) ) );
2558 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_publish_wait, FD_MHIST_SECONDS_MIN( REPLAY, STORE_PUBLISH_WAIT ),
2559 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_PUBLISH_WAIT ) ) );
2560 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_publish_work, FD_MHIST_SECONDS_MIN( REPLAY, STORE_PUBLISH_WORK ),
2561 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_PUBLISH_WORK ) ) );
2562 :
2563 0 : ulong scratch_top = FD_SCRATCH_ALLOC_FINI( l, 1UL );
2564 0 : if( FD_UNLIKELY( scratch_top > (ulong)scratch + scratch_footprint( tile ) ) )
2565 0 : FD_LOG_ERR(( "scratch overflow %lu %lu %lu", scratch_top - (ulong)scratch - scratch_footprint( tile ), scratch_top, (ulong)scratch + scratch_footprint( tile ) ));
2566 0 : }
2567 :
2568 : static ulong
2569 : populate_allowed_seccomp( fd_topo_t const * topo FD_FN_UNUSED,
2570 : fd_topo_tile_t const * tile FD_FN_UNUSED,
2571 : ulong out_cnt,
2572 0 : struct sock_filter * out ) {
2573 :
2574 0 : populate_sock_filter_policy_fd_replay_tile( out_cnt, out, (uint)fd_log_private_logfile_fd() );
2575 0 : return sock_filter_policy_fd_replay_tile_instr_cnt;
2576 0 : }
2577 :
2578 : static ulong
2579 : populate_allowed_fds( fd_topo_t const * topo FD_FN_UNUSED,
2580 : fd_topo_tile_t const * tile FD_FN_UNUSED,
2581 : ulong out_fds_cnt,
2582 0 : int * out_fds ) {
2583 :
2584 0 : if( FD_UNLIKELY( out_fds_cnt<2UL ) ) FD_LOG_ERR(( "out_fds_cnt %lu", out_fds_cnt ));
2585 :
2586 0 : ulong out_cnt = 0UL;
2587 0 : out_fds[ out_cnt++ ] = 2; /* stderr */
2588 0 : if( FD_LIKELY( -1!=fd_log_private_logfile_fd() ) )
2589 0 : out_fds[ out_cnt++ ] = fd_log_private_logfile_fd(); /* logfile */
2590 0 : return out_cnt;
2591 0 : }
2592 :
2593 : #undef DEBUG_LOGGING
2594 :
2595 : /* counting carefully, after_credit can generate at most 7 frags and
2596 : returnable_frag boot_genesis can also generate at most 7 frags, so 14
2597 : is a conservative bound. */
2598 0 : #define STEM_BURST (14UL)
2599 :
2600 : /* TODO: calculate this properly/fix stem to work with larger numbers of links */
2601 : /* 1000 chosen empirically as anything larger slowed down replay times. Need to calculate
2602 : this properly. */
2603 0 : #define STEM_LAZY ((long)10e3)
2604 :
2605 0 : #define STEM_CALLBACK_CONTEXT_TYPE fd_replay_tile_t
2606 0 : #define STEM_CALLBACK_CONTEXT_ALIGN alignof(fd_replay_tile_t)
2607 :
2608 0 : #define STEM_CALLBACK_METRICS_WRITE metrics_write
2609 0 : #define STEM_CALLBACK_AFTER_CREDIT after_credit
2610 0 : #define STEM_CALLBACK_BEFORE_FRAG before_frag
2611 0 : #define STEM_CALLBACK_RETURNABLE_FRAG returnable_frag
2612 :
2613 : #include "../../disco/stem/fd_stem.c"
2614 :
2615 : fd_topo_run_tile_t fd_tile_replay = {
2616 : .name = "replay",
2617 : .populate_allowed_seccomp = populate_allowed_seccomp,
2618 : .populate_allowed_fds = populate_allowed_fds,
2619 : .scratch_align = scratch_align,
2620 : .scratch_footprint = scratch_footprint,
2621 : .privileged_init = privileged_init,
2622 : .unprivileged_init = unprivileged_init,
2623 : .run = stem_run,
2624 : };
|
