Line data Source code
1 : #include "fd_replay_tile.h"
2 : #include "fd_sched.h"
3 : #include "fd_execrp.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/shred/fd_fec_set.h"
18 : #include "../../disco/pack/fd_pack.h"
19 : #include "../../discof/fd_accdb_topo.h"
20 : #include "../../discof/reasm/fd_reasm.h"
21 : #include "../../disco/keyguard/fd_keyload.h"
22 : #include "../../disco/keyguard/fd_keyswitch.h"
23 : #include "../../disco/genesis/fd_genesis_cluster.h"
24 : #include "../../discof/genesis/genesis_hash.h"
25 : #include "../../util/pod/fd_pod.h"
26 : #include "../../flamenco/accdb/fd_accdb_admin_v1.h"
27 : #include "../../flamenco/accdb/fd_accdb_admin_v2.h"
28 : #include "../../flamenco/accdb/fd_accdb_impl_v1.h"
29 : #include "../../flamenco/accdb/fd_accdb_sync.h"
30 : #include "../../flamenco/accdb/fd_vinyl_req_pool.h"
31 : #include "../../flamenco/rewards/fd_rewards.h"
32 : #include "../../flamenco/leaders/fd_multi_epoch_leaders.h"
33 : #include "../../flamenco/progcache/fd_progcache_admin.h"
34 : #include "../../disco/topo/fd_wksp_mon.h"
35 : #include "../../disco/metrics/fd_metrics.h"
36 :
37 : #include "../../flamenco/fd_flamenco_base.h"
38 : #include "../../flamenco/runtime/fd_runtime.h"
39 : #include "../../flamenco/runtime/fd_runtime_stack.h"
40 : #include "../../flamenco/runtime/fd_genesis_parse.h"
41 : #include "../../flamenco/runtime/sysvar/fd_sysvar_epoch_schedule.h"
42 : #include "../../flamenco/runtime/program/fd_precompiles.h"
43 : #include "../../flamenco/runtime/program/vote/fd_vote_state_versioned.h"
44 : #include "../../flamenco/runtime/program/vote/fd_vote_codec.h"
45 : #include "../../flamenco/runtime/tests/fd_dump_pb.h"
46 :
47 : #include <stdio.h>
48 :
49 : /* Replay concepts:
50 :
51 : - Blocks are aggregations of entries aka. microblocks which are
52 : groupings of txns and are constructed by the block producer (see
53 : fd_pack).
54 :
55 : - Entries are grouped into entry batches by the block producer (see
56 : fd_pack / fd_shredder).
57 :
58 : - Entry batches are divided into chunks known as shreds by the block
59 : producer (see fd_shredder).
60 :
61 : - Shreds are grouped into forward-error-correction sets (FEC sets) by
62 : the block producer (see fd_shredder).
63 :
64 : - Shreds are transmitted to the rest of the cluster via the Turbine
65 : protocol (see fd_shredder / fd_shred).
66 :
67 : - Once enough shreds within a FEC set are received to recover the
68 : entirety of the shred data encoded by that FEC set, the receiver
69 : can "complete" the FEC set (see fd_fec_resolver).
70 :
71 : - If shreds in the FEC set are missing such that it can't complete,
72 : the receiver can use the Repair protocol to request missing shreds
73 : in FEC set (see fd_repair).
74 :
75 : - The current Repair protocol does not support requesting coding
76 : shreds. As a result, some FEC sets might be actually complete
77 : (contain all data shreds). Repair currently hacks around this by
78 : forcing completion but the long-term solution is to add support for
79 : fec_repairing coding shreds via Repair.
80 :
81 : - FEC sets are delivered in partial-order to the Replay tile by the
82 : Repair tile. Currently Replay only supports replaying entry batches
83 : so FEC sets need to reassembled into an entry batch before they can
84 : be replayed. The new Dispatcher will change this by taking a FEC
85 : set as input instead. */
86 :
87 0 : #define IN_KIND_SNAP ( 0)
88 0 : #define IN_KIND_GENESIS ( 1)
89 0 : #define IN_KIND_IPECHO ( 2)
90 0 : #define IN_KIND_TOWER ( 3)
91 0 : #define IN_KIND_RESOLV ( 4)
92 0 : #define IN_KIND_POH ( 5)
93 0 : #define IN_KIND_EXECRP ( 6)
94 0 : #define IN_KIND_REPAIR ( 7)
95 0 : #define IN_KIND_TXSEND ( 8)
96 0 : #define IN_KIND_RPC ( 9)
97 0 : #define IN_KIND_GOSSIP_OUT (10)
98 :
99 : #define DEBUG_LOGGING 0
100 :
101 : /* The first bank that the replay tile produces either for genesis
102 : or the snapshot boot will always be at bank index 0. */
103 0 : #define FD_REPLAY_BOOT_BANK_IDX (0UL)
104 :
105 : struct fd_replay_in_link {
106 : fd_wksp_t * mem;
107 : ulong chunk0;
108 : ulong wmark;
109 : ulong mtu;
110 : };
111 :
112 : typedef struct fd_replay_in_link fd_replay_in_link_t;
113 :
114 : struct fd_replay_out_link {
115 : ulong idx;
116 : fd_wksp_t * mem;
117 : ulong chunk0;
118 : ulong wmark;
119 : ulong chunk;
120 : };
121 :
122 : typedef struct fd_replay_out_link fd_replay_out_link_t;
123 :
124 : /* fd_block_id_map is a simple map of block-ids to bank indices. The
125 : map sits on top of an array of fd_block_id_ele_t. This serves as a
126 : translation layer between block ids to bank indices. The data
127 : array is indexed by bank index and the latest observed merkle root
128 : for the bank index is stored in the array. Once the block id has
129 : been observed, the entry is keyed by the latest merkle root (aka the
130 : block id). */
131 :
132 : struct fd_block_id_ele {
133 : fd_hash_t latest_mr;
134 : uint latest_fec_idx;
135 : int block_id_seen;
136 : ulong slot;
137 : ulong next_;
138 : };
139 : typedef struct fd_block_id_ele fd_block_id_ele_t;
140 :
141 : #define MAP_NAME fd_block_id_map
142 : #define MAP_ELE_T fd_block_id_ele_t
143 : #define MAP_KEY_T fd_hash_t
144 0 : #define MAP_KEY latest_mr
145 0 : #define MAP_NEXT next_
146 0 : #define MAP_KEY_EQ(k0,k1) (!memcmp((k0),(k1), sizeof(fd_hash_t)))
147 0 : #define MAP_KEY_HASH(key,seed) (fd_hash((seed),(key),sizeof(fd_hash_t)))
148 : #include "../../util/tmpl/fd_map_chain.c"
149 :
150 : static inline ulong
151 0 : fd_block_id_ele_get_idx( fd_block_id_ele_t * ele_arr, fd_block_id_ele_t * ele ) {
152 0 : return (ulong)(ele - ele_arr);
153 0 : }
154 :
155 : struct fd_replay_tile {
156 : fd_wksp_t * wksp;
157 :
158 : uint rng_seed;
159 : fd_rng_t rng[ 1 ];
160 :
161 : fd_accdb_admin_t accdb_admin[1];
162 : fd_accdb_user_t accdb[1];
163 : fd_progcache_join_t progcache[1];
164 : fd_wksp_mon_t progcache_wksp_mon[1];
165 : fd_wksp_mon_t accdb_cache_wksp_mon[1];
166 :
167 : fd_txncache_t * txncache;
168 : fd_store_t * store;
169 : fd_banks_t * banks;
170 : ulong frontier_indices[ FD_BANKS_MAX_BANKS ];
171 : ulong frontier_cnt;
172 :
173 : /* This flag is 1 If we have seen a vote signature that our node has
174 : sent out get rooted at least one time. The value is 0 otherwise.
175 : We can't become leader and pack blocks until this flag has been
176 : set. This parallels the Agave 'has_new_vote_been_rooted'. */
177 : int identity_vote_rooted;
178 : int wait_for_vote_to_start_leader;
179 :
180 : /* wfs_enabled is 1 if the validator is booted in
181 : wait_for_supermajority mode. In this mode replay (and, by extension,
182 : downstream consumers) is not allowed to make progress until 80% of
183 : the cluster has published their ContactInfo in Gossip with a
184 : shred version matching expected_shred_version. When this happens,
185 : wfs_complete will be set to 1. */
186 : int wfs_enabled;
187 : int wfs_complete;
188 :
189 : fd_hash_t expected_bank_hash;
190 :
191 : ulong reasm_seed;
192 : fd_reasm_t * reasm;
193 : fd_reasm_fec_t * reasm_evicted; /* evicted FEC by reasm_insert must be stored in returnable_frag, and then drained in after_credit */
194 :
195 : fd_sched_t * sched;
196 :
197 : ulong vote_tracker_seed;
198 : fd_vote_tracker_t * vote_tracker;
199 :
200 : int has_genesis_hash;
201 : char genesis_path[ PATH_MAX ];
202 : fd_hash_t genesis_hash[1];
203 : fd_genesis_t genesis[1];
204 : ulong cluster_type;
205 :
206 : int has_genesis_timestamp;
207 : ulong genesis_timestamp;
208 : int has_expected_genesis_timestamp;
209 : ulong expected_genesis_timestamp;
210 :
211 : #define FD_REPLAY_HARD_FORKS_MAX (64UL)
212 : ulong hard_forks_cnt;
213 : ulong hard_forks[ FD_REPLAY_HARD_FORKS_MAX ];
214 : ulong hard_forks_cnts[ FD_REPLAY_HARD_FORKS_MAX ];
215 :
216 : ushort expected_shred_version;
217 : ushort ipecho_shred_version;
218 :
219 : /* A note on publishing ...
220 :
221 : The watermarks are used to publish our fork-aware structures. For
222 : example, store, banks, and txncache need to be published to release
223 : resources occupied by rooted or dead blocks. In general,
224 : publishing has the effect of pruning forks in those structures,
225 : indicating that it is ok to release the memory being occupied by
226 : the blocks on said forks. Tower is responsible for informing us of
227 : the latest block on the consensus rooted fork. As soon as we can,
228 : we should move the published root as close as possible to the
229 : latest consensus root, publishing/pruning everything on the fork
230 : tree along the way. That is, all the blocks that directly descend
231 : from the current published root (inclusive) to the new published
232 : root (exclusive) on the rooted fork, as well as all the minority
233 : forks that branch from said blocks.
234 :
235 : Ideally, we'd move the published root to the consensus root
236 : immediately upon receiving a new consensus root. However, that's
237 : not always safe to do. One thing we need to be careful about is
238 : making sure that there are no more users/consumers of
239 : soon-to-be-pruned blocks, lest a use-after-free occurs. This can
240 : be done by using a reference counter for each block. Any
241 : concurrent activity, such as transaction execution in the exec
242 : tiles, should retain a refcnt on the block for as
243 : long as it needs access to the shared fork-aware structures related
244 : to that block. Eventually, refcnt on a given block will drop down
245 : to 0 as the block either finishes replaying or gets marked as dead,
246 : and any other tile that has retained a refcnt on the block releases
247 : it. At that point, it becomes a candidate for pruning. The key to
248 : safe publishing then becomes figuring out how far we could advance
249 : the published root, such that every minority fork branching off of
250 : blocks in between the current published root (inclusive) and the
251 : new published root (exclusive) is safe to be pruned. This is a
252 : straightforward tree traversal, where if a block B on the rooted
253 : fork has refcnt 0, and all minority forks branching off of B also
254 : have refcnt 0, then B is safe to be pruned. We advance the
255 : published root to the farthest consecutively prunable block on the
256 : rooted fork. Note that reasm presents the replay tile with a clean
257 : view of the world where every block is chained off of a parent
258 : block. So there are no orpahned/dangling tree nodes to worry
259 : about. The world is a nice single tree as far as replay is
260 : concerned.
261 :
262 : In the following fork tree, every node is a block and the number in
263 : parentheses is the refcnt on the block. The chain marked with
264 : double slashes is the rooted fork. Suppose the published root is
265 : at block P, and consensus root is at block T. We can't publish
266 : past block P because Q has refcnt 1.
267 :
268 :
269 : P(0)
270 : / \\
271 : Q(1) A(0)
272 : / || \
273 : X(0) B(0) C(0)
274 : / || \
275 : Y(0) M(0) R(0)
276 : / || / \
277 : D(2) T(0) J(0) L(0)
278 : ||
279 : ..
280 : ..
281 : ..
282 : ||
283 : blocks we might be actively replaying
284 :
285 :
286 : When refcnt on Q drops to 0, we would be able to advance the
287 : published root to block M, because blocks P, A, and B, as well as
288 : all subtrees branching off of them, have refcnt 0, and therefore
289 : can be pruned. Block M itself cannot be pruned yet because its
290 : child block D has refcnt 2. After publishing/pruning, the fork
291 : tree would be:
292 :
293 :
294 : M(0)
295 : / ||
296 : D(2) T(0)
297 : ||
298 : ..
299 : ..
300 : ..
301 : ||
302 : blocks we might be actively replaying
303 :
304 :
305 : As a result, the shared fork-aware structures can free resources
306 : for blocks P, A, B, and all subtrees branching off of them.
307 :
308 : For the reference counting part, the replay tile is the sole entity
309 : that can update the refcnt. This ensures that all refcnt increment
310 : and decrement attempts are serialized at the replay tile, and that
311 : there are no racy resurrection of a soon-to-be-pruned block. If a
312 : refcnt increment request arrives after a block has been pruned,
313 : replay simply rejects the request.
314 :
315 : A note on the implementation of the above ...
316 :
317 : Upon receiving a new consensus root, we descend down the rooted
318 : fork from the current published root to the new consensus root. On
319 : each node/block of the rooted fork, we do a summation of the refcnt
320 : on the block and all the minority fork blocks branching from the
321 : block. If the summation is 0, the block is safe for pruning. We
322 : advance the published root to the far end of the consecutive run of
323 : 0 refcnt sums originating from the current published root. On our
324 : descent down the minority forks, we also mark any block that hasn't
325 : finished replaying as dead, so we don't waste time executing them.
326 : No more transactions shall be dispatched for execution from dead
327 : blocks.
328 :
329 : Blocks start out with a refcnt of 0. Other tiles may send a
330 : request to the replay tile for a reference on a block. The
331 : transaction dispatcher is another source of refcnt updates. On
332 : every dispatch of a transaction for block B, we increment the
333 : refcnt for B. And on every transaction finalization, we decrement
334 : the refcnt for B. This means that whenever the refcnt on a block
335 : is 0, there is no more reference on that block from the execution
336 : pipeline. While it might be tempting to simply increment the
337 : refcnt once when we start replaying a block, and decrement the
338 : refcnt once when we finish a block, this more fine-grained refcnt
339 : update strategy allows for aborting and potentially immediate
340 : pruning of blocks under interleaved block replay. Upon receiving a
341 : new consensus root, we can simply look at the refcnt on minority
342 : fork blocks, and a refcnt of 0 would imply that the block is safe
343 : for pruning, even if we haven't finished replaying it. Without the
344 : fine-grained refcnt, we would need to first stop dispatching from
345 : the aborted block, and then wait for a full drain of the execution
346 : pipeline to know for sure that there are no more in-flight
347 : transactions executing on the aborted block. Note that this will
348 : allow the refcnt on any block to transiently drop down to 0. We
349 : will not mistakenly prune an actively replaying block, aka a leaf
350 : node, that is chaining off of the rooted fork, because the
351 : consensus root is always an ancestor of the actively replaying tip.
352 : */
353 : fd_hash_t consensus_root; /* The most recent block to have reached max lockout in the tower. */
354 : ulong consensus_root_slot; /* slot number of the above. */
355 : ulong consensus_root_bank_idx; /* bank index of the above. */
356 : ulong published_root_slot; /* slot number of the published root. */
357 : ulong published_root_bank_idx; /* bank index of the published root. */
358 :
359 : /* Randomly generated block id for the initial genesis/snapshot slot.
360 : To be replaced with block id in the snapshot manifest when SIMD-333
361 : is activated. */
362 :
363 : fd_hash_t initial_block_id;
364 :
365 : /* We need to maintain a tile-local mapping of block-ids to bank index
366 : and vice versa. This translation layer is needed for conversion
367 : since tower operates on block-ids and downstream consumers of FEC
368 : sets operate on bank indices. This mapping must happen both ways:
369 : 1. tower sends us block ids and we must map them to bank indices.
370 : 2. when a block is completed, we must map the bank index to a block
371 : id to send a slot complete message to tower. */
372 : ulong block_id_len;
373 : fd_block_id_ele_t * block_id_arr;
374 : ulong block_id_map_seed;
375 : fd_block_id_map_t * block_id_map;
376 :
377 : /* Capture-related configs */
378 : fd_capture_ctx_t * capture_ctx;
379 : FILE * capture_file;
380 : fd_capture_link_buf_t cap_repl_out[1];
381 :
382 : /* Protobuf dumping context for debugging runtime execution and
383 : collecting seed corpora. */
384 : fd_dump_proto_ctx_t * dump_proto_ctx;
385 :
386 : /* Whether the runtime has been booted either from snapshot loading
387 : or from genesis. */
388 : int is_booted;
389 :
390 : /* Buffer to store vote towers that need to be published to the Tower
391 : tile. */
392 :
393 : fd_multi_epoch_leaders_t * mleaders;
394 :
395 : int larger_max_cost_per_block;
396 :
397 : /* When we transition to becoming leader, we can only unbecome the
398 : leader if we have received a block id from the FEC reassembler, and
399 : a message from PoH that the leader slot has ended. After both of
400 : these conditions are met, then we are free to unbecome the leader.
401 : */
402 : uint is_leader : 1;
403 : uint supports_leader : 1;
404 : int recv_poh;
405 : ulong next_leader_slot;
406 : long next_leader_tickcount;
407 : ulong highwater_leader_slot;
408 : ulong reset_slot;
409 : fd_bank_t * reset_bank;
410 : fd_hash_t reset_block_id;
411 : long reset_timestamp_nanos;
412 : double slot_duration_nanos;
413 : double slot_duration_ticks;
414 : fd_bank_t * leader_bank;
415 :
416 : fd_pubkey_t identity_pubkey[1];
417 : ulong identity_idx;
418 :
419 : fd_keyswitch_t * keyswitch;
420 : int halt_leader;
421 :
422 : ulong resolv_tile_cnt;
423 :
424 : int in_kind[ 128 ];
425 : fd_replay_in_link_t in[ 128 ];
426 :
427 : fd_replay_out_link_t exec_out[ 1 ];
428 :
429 : fd_replay_out_link_t replay_out[1];
430 :
431 : fd_replay_out_link_t epoch_out[1];
432 :
433 : /* The rpc tile needs to occasionally own a reference to a live bank.
434 : Replay needs to know if the rpc as a consumer is enabled so it can
435 : increment the bank's refcnt before publishing bank_idx. */
436 : int rpc_enabled;
437 :
438 : # if FD_HAS_FLATCC
439 : /* For dumping blocks to protobuf. For backtest only. */
440 : fd_block_dump_ctx_t * block_dump_ctx;
441 : # endif
442 :
443 : /* We need a few pieces of information to compute the right addresses
444 : for bundle crank information that we need to send to pack. */
445 : struct {
446 : int enabled;
447 : fd_pubkey_t vote_account;
448 : fd_bundle_crank_gen_t gen[1];
449 : } bundle;
450 :
451 : struct {
452 : ulong store_query_acquire;
453 : ulong store_query_release;
454 : fd_histf_t store_query_wait[1];
455 : fd_histf_t store_query_work[1];
456 : ulong store_query_cnt;
457 : ulong store_query_missing_cnt;
458 : ulong store_query_mr;
459 : ulong store_query_missing_mr;
460 :
461 : ulong slots_total;
462 : ulong transactions_total;
463 :
464 : ulong reasm_latest_slot;
465 : ulong reasm_latest_fec_idx;
466 :
467 : ulong sched_full;
468 : ulong reasm_empty;
469 : ulong leader_bid_wait;
470 : ulong banks_full;
471 : ulong storage_root_behind;
472 :
473 : fd_histf_t root_slot_dur[1];
474 : fd_histf_t root_account_dur[1];
475 : } metrics;
476 :
477 : uchar __attribute__((aligned(FD_MULTI_EPOCH_LEADERS_ALIGN))) mleaders_mem[ FD_MULTI_EPOCH_LEADERS_FOOTPRINT ];
478 :
479 : ulong runtime_stack_seed;
480 : fd_runtime_stack_t * runtime_stack;
481 : };
482 :
483 : typedef struct fd_replay_tile fd_replay_tile_t;
484 :
485 : FD_FN_CONST static inline ulong
486 0 : scratch_align( void ) {
487 0 : return 128UL;
488 0 : }
489 : FD_FN_PURE static inline ulong
490 0 : scratch_footprint( fd_topo_tile_t const * tile ) {
491 0 : ulong chain_cnt = fd_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
492 :
493 0 : ulong l = FD_LAYOUT_INIT;
494 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
495 0 : l = FD_LAYOUT_APPEND( l, fd_runtime_stack_align(), fd_runtime_stack_footprint( FD_RUNTIME_MAX_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_STAKE_ACCOUNTS ) );
496 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_block_id_ele_t), sizeof(fd_block_id_ele_t) * tile->replay.max_live_slots );
497 0 : l = FD_LAYOUT_APPEND( l, fd_block_id_map_align(), fd_block_id_map_footprint( chain_cnt ) );
498 0 : l = FD_LAYOUT_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
499 0 : l = FD_LAYOUT_APPEND( l, fd_reasm_align(), fd_reasm_footprint( tile->replay.fec_max ) );
500 0 : l = FD_LAYOUT_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.sched_depth, tile->replay.max_live_slots ) );
501 0 : l = FD_LAYOUT_APPEND( l, fd_vinyl_req_pool_align(), fd_vinyl_req_pool_footprint( 1UL, 1UL ) );
502 0 : l = FD_LAYOUT_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
503 0 : l = FD_LAYOUT_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
504 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_dump_proto_ctx_t), sizeof(fd_dump_proto_ctx_t) );
505 :
506 0 : # if FD_HAS_FLATCC
507 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
508 0 : l = FD_LAYOUT_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
509 0 : }
510 0 : # endif
511 :
512 0 : l = FD_LAYOUT_FINI( l, scratch_align() );
513 :
514 0 : return l;
515 0 : }
516 :
517 : static inline void
518 0 : metrics_write( fd_replay_tile_t * ctx ) {
519 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_ACQUIRE, ctx->metrics.store_query_acquire );
520 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_RELEASE, ctx->metrics.store_query_release );
521 0 : FD_MHIST_COPY( REPLAY, STORE_QUERY_WAIT, ctx->metrics.store_query_wait );
522 0 : FD_MHIST_COPY( REPLAY, STORE_QUERY_WORK, ctx->metrics.store_query_work );
523 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_CNT, ctx->metrics.store_query_cnt );
524 0 : FD_MCNT_SET ( REPLAY, STORE_QUERY_MISSING_CNT, ctx->metrics.store_query_missing_cnt );
525 0 : FD_MGAUGE_SET( REPLAY, STORE_QUERY_MR, ctx->metrics.store_query_mr );
526 0 : FD_MGAUGE_SET( REPLAY, STORE_QUERY_MISSING_MR, ctx->metrics.store_query_missing_mr );
527 :
528 0 : FD_MGAUGE_SET( REPLAY, ROOT_SLOT, ctx->consensus_root_slot==ULONG_MAX ? 0UL : ctx->consensus_root_slot );
529 0 : ulong leader_slot = ctx->leader_bank ? ctx->leader_bank->f.slot : 0UL;
530 :
531 0 : if( FD_LIKELY( ctx->leader_bank ) ) {
532 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, leader_slot );
533 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, leader_slot );
534 0 : } else {
535 0 : FD_MGAUGE_SET( REPLAY, NEXT_LEADER_SLOT, ctx->next_leader_slot==ULONG_MAX ? 0UL : ctx->next_leader_slot );
536 0 : FD_MGAUGE_SET( REPLAY, LEADER_SLOT, 0UL );
537 0 : }
538 0 : FD_MGAUGE_SET( REPLAY, RESET_SLOT, ctx->reset_slot==ULONG_MAX ? 0UL : ctx->reset_slot );
539 :
540 0 : FD_MGAUGE_SET( REPLAY, LIVE_BANKS, fd_banks_pool_used_cnt( ctx->banks ) );
541 :
542 0 : ulong reasm_free = fd_reasm_free( ctx->reasm );
543 0 : FD_MGAUGE_SET( REPLAY, REASM_FREE, reasm_free );
544 :
545 0 : FD_MCNT_SET( REPLAY, SLOTS_TOTAL, ctx->metrics.slots_total );
546 0 : FD_MCNT_SET( REPLAY, TRANSACTIONS_TOTAL, ctx->metrics.transactions_total );
547 :
548 0 : FD_MGAUGE_SET( REPLAY, REASM_LATEST_SLOT, ctx->metrics.reasm_latest_slot );
549 0 : FD_MGAUGE_SET( REPLAY, REASM_LATEST_FEC_IDX, ctx->metrics.reasm_latest_fec_idx );
550 :
551 0 : fd_sched_metrics_write( ctx->sched );
552 :
553 0 : FD_MCNT_SET( REPLAY, SCHED_FULL, ctx->metrics.sched_full );
554 0 : FD_MCNT_SET( REPLAY, REASM_EMPTY, ctx->metrics.reasm_empty );
555 0 : FD_MCNT_SET( REPLAY, LEADER_BID_WAIT, ctx->metrics.leader_bid_wait );
556 0 : FD_MCNT_SET( REPLAY, BANKS_FULL, ctx->metrics.banks_full );
557 0 : FD_MCNT_SET( REPLAY, STORAGE_ROOT_BEHIND, ctx->metrics.storage_root_behind );
558 :
559 0 : FD_MCNT_SET( REPLAY, ACCDB_CREATED, ctx->accdb->base.created_cnt );
560 0 : FD_MCNT_SET( REPLAY, ACCDB_REVERTED, ctx->accdb_admin->base.revert_cnt );
561 0 : FD_MCNT_SET( REPLAY, ACCDB_ROOTED, ctx->accdb_admin->base.root_cnt );
562 0 : FD_MCNT_SET( REPLAY, ACCDB_ROOTED_BYTES, ctx->accdb_admin->base.root_tot_sz );
563 0 : FD_MCNT_SET( REPLAY, ACCDB_GC_ROOT, ctx->accdb_admin->base.gc_root_cnt );
564 0 : FD_MCNT_SET( REPLAY, ACCDB_RECLAIMED, ctx->accdb_admin->base.reclaim_cnt );
565 0 : FD_MHIST_COPY( REPLAY, ROOT_SLOT_DURATION_SECONDS, ctx->metrics.root_slot_dur );
566 0 : FD_MHIST_COPY( REPLAY, ROOT_ACCOUNT_DURATION_SECONDS, ctx->metrics.root_account_dur );
567 0 : FD_MCNT_SET( REPLAY, ROOT_ELAPSED_SECONDS_DB, (ulong)ctx->accdb_admin->base.dt_vinyl );
568 0 : FD_MCNT_SET( REPLAY, ROOT_ELAPSED_SECONDS_COPY, (ulong)ctx->accdb_admin->base.dt_copy );
569 0 : FD_MCNT_SET( REPLAY, ROOT_ELAPSED_SECONDS_GC, (ulong)ctx->accdb_admin->base.dt_gc );
570 :
571 0 : fd_progcache_admin_metrics_t const * pcm = &fd_progcache_admin_metrics_g;
572 0 : FD_MCNT_SET( REPLAY, PROGCACHE_ROOTED, pcm->root_cnt );
573 0 : FD_MCNT_SET( REPLAY, PROGCACHE_GC_ROOT, pcm->gc_root_cnt );
574 :
575 0 : fd_wksp_mon_t * wm = fd_wksp_mon_tick( ctx->progcache_wksp_mon, fd_tickcount() );
576 0 : FD_MGAUGE_SET( REPLAY, PROGCACHE_FREE_PARTS, wm->free_cnt );
577 0 : FD_MGAUGE_SET( REPLAY, PROGCACHE_FREE_BYTES, wm->free_sz );
578 0 : FD_MGAUGE_SET( REPLAY, PROGCACHE_SIZE_BYTES, wm->wksp->data_max );
579 0 : FD_MGAUGE_SET( REPLAY, PROGCACHE_FREE_PART_MAX_BYTES, wm->free_max_sz );
580 0 : FD_MGAUGE_SET( REPLAY, PROGCACHE_USED_PART_MEDIAN_BYTES, wm->part_median_sz );
581 0 : FD_MGAUGE_SET( REPLAY, PROGCACHE_USED_PART_MEAN_BYTES, wm->part_mean_sz );
582 :
583 0 : fd_wksp_mon_t * am = fd_wksp_mon_tick( ctx->accdb_cache_wksp_mon, fd_tickcount() );
584 0 : FD_MGAUGE_SET( REPLAY, ACCDB_CACHE_FREE_PARTS, am->free_cnt );
585 0 : FD_MGAUGE_SET( REPLAY, ACCDB_CACHE_FREE_BYTES, am->free_sz );
586 0 : FD_MGAUGE_SET( REPLAY, ACCDB_CACHE_SIZE_BYTES, am->wksp->data_max );
587 0 : FD_MGAUGE_SET( REPLAY, ACCDB_CACHE_FREE_PART_MAX_BYTES, am->free_max_sz );
588 0 : FD_MGAUGE_SET( REPLAY, ACCDB_CACHE_USED_PART_MEDIAN_BYTES, am->part_median_sz );
589 0 : FD_MGAUGE_SET( REPLAY, ACCDB_CACHE_USED_PART_MEAN_BYTES, am->part_mean_sz );
590 0 : }
591 :
592 : static void
593 : publish_epoch_info( fd_replay_tile_t * ctx,
594 : fd_stem_context_t * stem,
595 : fd_bank_t * bank,
596 0 : int current_epoch ) {
597 0 : fd_epoch_schedule_t const * schedule = &bank->f.epoch_schedule;
598 0 : ulong epoch = fd_slot_to_epoch( schedule, bank->f.slot, NULL ) + fd_ulong_if( current_epoch, 1UL, 0UL );
599 :
600 0 : fd_features_t const * features = &bank->f.features;
601 :
602 0 : fd_runtime_stack_t * runtime_stack = ctx->runtime_stack;
603 :
604 0 : fd_epoch_info_msg_t * epoch_info_msg = fd_chunk_to_laddr( ctx->epoch_out->mem, ctx->epoch_out->chunk );
605 :
606 0 : epoch_info_msg->staked_vote_cnt = current_epoch ? runtime_stack->epoch_weights.next_stake_weights_cnt : runtime_stack->epoch_weights.stake_weights_cnt;
607 0 : epoch_info_msg->staked_id_cnt = current_epoch ? runtime_stack->epoch_weights.next_id_weights_cnt : runtime_stack->epoch_weights.id_weights_cnt;
608 0 : epoch_info_msg->epoch_schedule = *schedule;
609 0 : epoch_info_msg->features = *features;
610 0 : epoch_info_msg->epoch = epoch;
611 0 : epoch_info_msg->start_slot = fd_epoch_slot0( schedule, epoch );
612 0 : epoch_info_msg->slot_cnt = fd_epoch_slot_cnt( schedule, epoch );
613 0 : epoch_info_msg->excluded_id_stake = current_epoch ? runtime_stack->epoch_weights.next_id_weights_excluded : runtime_stack->epoch_weights.id_weights_excluded;
614 :
615 0 : fd_vote_stake_weight_t * stake_weights = fd_type_pun( epoch_info_msg + 1 );
616 0 : fd_vote_stake_weight_t * src_stake_weights = current_epoch ? runtime_stack->epoch_weights.next_stake_weights : runtime_stack->epoch_weights.stake_weights;
617 0 : memcpy( stake_weights, src_stake_weights, epoch_info_msg->staked_vote_cnt * sizeof(fd_vote_stake_weight_t) );
618 :
619 0 : fd_stake_weight_t * id_weights = fd_epoch_info_msg_id_weights( epoch_info_msg );
620 0 : fd_stake_weight_t * src_id_weights = current_epoch ? runtime_stack->epoch_weights.next_id_weights : runtime_stack->epoch_weights.id_weights;
621 0 : fd_memcpy( id_weights, src_id_weights, epoch_info_msg->staked_id_cnt * sizeof(fd_stake_weight_t) );
622 :
623 0 : ulong epoch_info_sz = fd_epoch_info_msg_sz( epoch_info_msg->staked_vote_cnt , epoch_info_msg->staked_id_cnt );
624 :
625 0 : ulong epoch_info_sig = 4UL;
626 0 : fd_stem_publish( stem, ctx->epoch_out->idx, epoch_info_sig, ctx->epoch_out->chunk, epoch_info_sz, 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
627 0 : ctx->epoch_out->chunk = fd_dcache_compact_next( ctx->epoch_out->chunk, epoch_info_sz, ctx->epoch_out->chunk0, ctx->epoch_out->wmark );
628 :
629 0 : fd_multi_epoch_leaders_epoch_msg_init( ctx->mleaders, epoch_info_msg );
630 0 : fd_multi_epoch_leaders_epoch_msg_fini( ctx->mleaders );
631 0 : }
632 :
633 : /**********************************************************************/
634 : /* Transaction execution state machine helpers */
635 : /**********************************************************************/
636 :
637 : static void
638 : replay_block_start( fd_replay_tile_t * ctx,
639 : fd_stem_context_t * stem,
640 : ulong bank_idx,
641 : ulong parent_bank_idx,
642 0 : ulong slot ) {
643 0 : long before = fd_log_wallclock();
644 :
645 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
646 0 : FD_CRIT( bank, "invariant violation: bank is NULL" );
647 0 : FD_CRIT( bank->state==FD_BANK_STATE_INIT, "invariant violation: bank is not in correct state" );
648 :
649 0 : bank->preparation_begin_nanos = before;
650 :
651 0 : fd_bank_t * parent_bank = fd_banks_bank_query( ctx->banks, parent_bank_idx );
652 0 : FD_CRIT( parent_bank, "invariant violation: parent bank is NULL" );
653 0 : FD_CRIT( parent_bank->state==FD_BANK_STATE_FROZEN, "invariant violation: parent bank is not in correct state" );
654 :
655 0 : ulong parent_slot = parent_bank->f.slot;
656 :
657 : /* Clone the bank from the parent. We must special case the first
658 : slot that is executed as the snapshot does not provide a parent
659 : block id. */
660 :
661 0 : bank = fd_banks_clone_from_parent( ctx->banks, bank_idx );
662 0 : if( FD_UNLIKELY( !bank ) ) {
663 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", bank_idx ));
664 0 : }
665 0 : bank->f.slot = slot;
666 0 : bank->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, parent_bank->txncache_fork_id );
667 :
668 : /* Create a new funk txn for the block. */
669 :
670 0 : fd_funk_txn_xid_t xid = { .ul = { slot, bank_idx } };
671 0 : fd_funk_txn_xid_t parent_xid = { .ul = { parent_slot, parent_bank_idx } };
672 0 : fd_accdb_attach_child ( ctx->accdb_admin, &parent_xid, &xid );
673 0 : fd_progcache_attach_child( ctx->progcache, &parent_xid, &xid );
674 :
675 : /* Update required runtime state and handle potential boundary. */
676 :
677 0 : int is_epoch_boundary = 0;
678 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
679 0 : if( FD_UNLIKELY( is_epoch_boundary ) ) publish_epoch_info( ctx, stem, bank, 0 );
680 :
681 0 : ulong max_tick_height;
682 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS!=fd_runtime_compute_max_tick_height( parent_bank->f.ticks_per_slot, slot, &max_tick_height ) ) ) {
683 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, parent_bank->f.ticks_per_slot ));
684 0 : }
685 0 : bank->f.max_tick_height = max_tick_height;
686 0 : fd_sched_set_poh_params( ctx->sched, bank->idx, bank->f.tick_height, bank->f.max_tick_height, bank->f.hashes_per_tick, &parent_bank->f.poh );
687 :
688 0 : FD_LOG_DEBUG(( "replay_block_start: bank_idx=%lu slot=%lu parent_bank_idx=%lu", bank_idx, slot, parent_bank_idx ));
689 0 : }
690 :
691 : static void
692 0 : cost_tracker_snap( fd_bank_t * bank, fd_replay_slot_completed_t * slot_info ) {
693 0 : if( FD_LIKELY( bank->cost_tracker_pool_idx!=ULONG_MAX ) ) {
694 0 : fd_cost_tracker_t const * cost_tracker = fd_bank_cost_tracker_query( bank );
695 0 : if( FD_UNLIKELY( cost_tracker->block_cost_limit==0UL ) ) {
696 0 : memset( &slot_info->cost_tracker, -1 /* ULONG_MAX */, sizeof(slot_info->cost_tracker) );
697 0 : } else {
698 0 : slot_info->cost_tracker.block_cost = cost_tracker->block_cost;
699 0 : slot_info->cost_tracker.vote_cost = cost_tracker->vote_cost;
700 0 : slot_info->cost_tracker.allocated_accounts_data_size = cost_tracker->allocated_accounts_data_size;
701 0 : slot_info->cost_tracker.block_cost_limit = cost_tracker->block_cost_limit;
702 0 : slot_info->cost_tracker.vote_cost_limit = cost_tracker->vote_cost_limit;
703 0 : slot_info->cost_tracker.account_cost_limit = cost_tracker->account_cost_limit;
704 0 : }
705 0 : } else {
706 0 : memset( &slot_info->cost_tracker, -1 /* ULONG_MAX */, sizeof(slot_info->cost_tracker) );
707 0 : }
708 0 : }
709 :
710 : static ulong
711 0 : get_identity_balance( fd_replay_tile_t * ctx, fd_funk_txn_xid_t xid ) {
712 0 : ulong identity_balance = ULONG_MAX;
713 0 : fd_accdb_ro_t identity_acc[1];
714 0 : if( FD_LIKELY( fd_accdb_open_ro( ctx->accdb, identity_acc, &xid, ctx->identity_pubkey ) ) ) {
715 0 : identity_balance = identity_acc->meta->lamports;
716 0 : fd_accdb_close_ro( ctx->accdb, identity_acc );
717 0 : }
718 0 : return identity_balance;
719 0 : }
720 :
721 : static void
722 : publish_slot_completed( fd_replay_tile_t * ctx,
723 : fd_stem_context_t * stem,
724 : fd_bank_t * bank,
725 : int is_initial,
726 0 : int is_leader ) {
727 :
728 0 : ulong slot = bank->f.slot;
729 :
730 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank->idx ];
731 :
732 : /* HACKY: hacky way of checking if we should send a null parent block
733 : id */
734 0 : fd_hash_t parent_block_id = {0};
735 0 : if( FD_UNLIKELY( !is_initial ) ) {
736 0 : parent_block_id = ctx->block_id_arr[ bank->parent_idx ].latest_mr;
737 0 : }
738 :
739 0 : fd_hash_t const * bank_hash = &bank->f.bank_hash;
740 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( &bank->f.block_hash_queue );
741 0 : FD_TEST( block_hash );
742 :
743 0 : if( FD_LIKELY( !is_initial ) ) fd_txncache_finalize_fork( ctx->txncache, bank->txncache_fork_id, 0UL, block_hash->uc );
744 :
745 0 : fd_epoch_schedule_t const * epoch_schedule = &bank->f.epoch_schedule;
746 0 : ulong slot_idx;
747 0 : ulong epoch = fd_slot_to_epoch( epoch_schedule, slot, &slot_idx );
748 :
749 0 : ctx->metrics.slots_total++;
750 0 : ctx->metrics.transactions_total = bank->f.txn_count;
751 :
752 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
753 0 : slot_info->slot = slot;
754 0 : slot_info->root_slot = ctx->consensus_root_slot;
755 0 : slot_info->storage_slot = ctx->published_root_slot;
756 0 : slot_info->epoch = epoch;
757 0 : slot_info->slot_in_epoch = slot_idx;
758 0 : slot_info->slots_per_epoch = fd_epoch_slot_cnt( epoch_schedule, epoch );
759 0 : slot_info->block_height = bank->f.block_height;
760 0 : slot_info->parent_slot = bank->f.parent_slot;
761 0 : slot_info->block_id = block_id_ele->latest_mr;
762 0 : slot_info->parent_block_id = parent_block_id;
763 0 : slot_info->bank_hash = *bank_hash;
764 0 : slot_info->block_hash = *block_hash;
765 0 : slot_info->transaction_count = bank->f.txn_count;
766 :
767 0 : fd_inflation_t inflation = bank->f.inflation;
768 0 : slot_info->inflation.foundation = inflation.foundation;
769 0 : slot_info->inflation.foundation_term = inflation.foundation_term;
770 0 : slot_info->inflation.terminal = inflation.terminal;
771 0 : slot_info->inflation.initial = inflation.initial;
772 0 : slot_info->inflation.taper = inflation.taper;
773 :
774 0 : fd_rent_t rent = bank->f.rent;
775 0 : slot_info->rent.burn_percent = rent.burn_percent;
776 0 : slot_info->rent.lamports_per_uint8_year = rent.lamports_per_uint8_year;
777 0 : slot_info->rent.exemption_threshold = rent.exemption_threshold;
778 :
779 0 : slot_info->first_fec_set_received_nanos = bank->first_fec_set_received_nanos;
780 0 : slot_info->preparation_begin_nanos = bank->preparation_begin_nanos;
781 0 : slot_info->first_transaction_scheduled_nanos = bank->first_transaction_scheduled_nanos;
782 0 : slot_info->last_transaction_finished_nanos = bank->last_transaction_finished_nanos;
783 0 : slot_info->completion_time_nanos = fd_log_wallclock();
784 0 : if( !slot_info->first_transaction_scheduled_nanos ) { /* edge case: empty slot */
785 0 : slot_info->first_transaction_scheduled_nanos = slot_info->last_transaction_finished_nanos;
786 0 : }
787 :
788 : /* refcnt should be incremented by 1 for each consumer that uses
789 : `bank_idx`. Each consumer should decrement the bank's refcnt once
790 : they are done using the bank. */
791 0 : bank->refcnt++; /* tower_tile */
792 0 : if( FD_LIKELY( ctx->rpc_enabled ) ) bank->refcnt++; /* rpc tile */
793 0 : slot_info->bank_idx = bank->idx;
794 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for tower, rpc", bank->idx, slot, bank->refcnt ));
795 :
796 0 : fd_bank_t * parent_bank = fd_banks_get_parent( ctx->banks, bank );
797 0 : if( FD_LIKELY( parent_bank ) ) {
798 0 : ulong total_txn_cnt = bank->f.txn_count;
799 0 : ulong nonvote_txn_cnt = bank->f.nonvote_txn_count;
800 0 : ulong failed_txn_cnt = bank->f.failed_txn_count;
801 0 : ulong nonvote_failed_txn_cnt = bank->f.nonvote_failed_txn_count;
802 :
803 0 : slot_info->nonvote_success = nonvote_txn_cnt - nonvote_failed_txn_cnt;
804 0 : slot_info->nonvote_failed = nonvote_failed_txn_cnt;
805 0 : slot_info->vote_failed = failed_txn_cnt - nonvote_failed_txn_cnt;
806 0 : slot_info->vote_success = total_txn_cnt - nonvote_txn_cnt - slot_info->vote_failed;
807 0 : } else {
808 0 : slot_info->vote_failed = ULONG_MAX;
809 0 : slot_info->vote_success = ULONG_MAX;
810 0 : slot_info->nonvote_success = ULONG_MAX;
811 0 : slot_info->nonvote_failed = ULONG_MAX;
812 0 : }
813 :
814 0 : slot_info->is_leader = is_leader;
815 0 : slot_info->transaction_fee = bank->f.execution_fees;
816 0 : slot_info->transaction_fee -= (slot_info->transaction_fee>>1); /* burn */
817 0 : slot_info->priority_fee = bank->f.priority_fees;
818 0 : slot_info->tips = bank->f.tips;
819 0 : slot_info->shred_cnt = bank->f.shred_cnt;
820 :
821 0 : FD_BASE58_ENCODE_32_BYTES( ctx->block_id_arr[ bank->idx ].latest_mr.uc, block_id_b58 );
822 0 : FD_BASE58_ENCODE_32_BYTES( bank->f.bank_hash.uc, bank_hash_b58 );
823 0 : FD_LOG_DEBUG(( "finished replaying slot %lu with (block id %s, bank hash %s, transactions %lu, votes %lu, shreds %lu, CUs used %lu, fees %lu)"
824 0 : "and timings [since parent fini %ld ns, started prepare %ld ns, started dispatching transactions %ld ns, finished executing transactions %ld ns, finished block %ld ns]",
825 0 : bank->f.slot, block_id_b58,
826 0 : bank_hash_b58,
827 0 : bank->f.transaction_count,
828 0 : bank->f.transaction_count - bank->f.nonvote_txn_count,
829 0 : bank->f.shred_cnt,
830 0 : bank->f.total_compute_units_used,
831 0 : bank->f.execution_fees + bank->f.priority_fees,
832 0 : !!parent_bank ? parent_bank->block_completed_nanos - bank->first_fec_set_received_nanos : LONG_MAX,
833 0 : bank->preparation_begin_nanos - bank->first_fec_set_received_nanos,
834 0 : bank->first_transaction_scheduled_nanos - bank->preparation_begin_nanos,
835 0 : bank->last_transaction_finished_nanos - bank->first_transaction_scheduled_nanos,
836 0 : bank->block_completed_nanos - bank->last_transaction_finished_nanos ));
837 :
838 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() ) );
839 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 );
840 0 : }
841 :
842 : static void
843 : publish_slot_dead( fd_replay_tile_t * ctx,
844 : fd_stem_context_t * stem,
845 : ulong slot,
846 0 : fd_hash_t const * block_id ) {
847 0 : fd_replay_slot_dead_t * slot_dead = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
848 0 : slot_dead->slot = slot;
849 0 : slot_dead->block_id = *block_id;
850 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() ) );
851 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 );
852 0 : }
853 :
854 : static void
855 : publish_txn_executed( fd_replay_tile_t * ctx,
856 : fd_stem_context_t * stem,
857 0 : ulong txn_idx ) {
858 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, txn_idx );
859 0 : fd_replay_txn_executed_t * txn_executed = fd_type_pun( fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk ) );
860 0 : *txn_executed->txn = *fd_sched_get_txn( ctx->sched, txn_idx );
861 0 : txn_executed->txn_err = txn_info->txn_err;
862 0 : txn_executed->is_committable = !!(txn_info->flags&FD_SCHED_TXN_IS_COMMITTABLE);
863 0 : txn_executed->is_fees_only = !!(txn_info->flags&FD_SCHED_TXN_IS_FEES_ONLY);
864 0 : txn_executed->tick_parsed = txn_info->tick_parsed;
865 0 : txn_executed->tick_sigverify_disp = txn_info->tick_sigverify_disp;
866 0 : txn_executed->tick_sigverify_done = txn_info->tick_sigverify_done;
867 0 : txn_executed->tick_exec_disp = txn_info->tick_exec_disp;
868 0 : txn_executed->tick_exec_done = txn_info->tick_exec_done;
869 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_TXN_EXECUTED, ctx->replay_out->chunk, sizeof(*txn_executed), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
870 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(*txn_executed), ctx->replay_out->chunk0, ctx->replay_out->wmark );
871 0 : }
872 :
873 : static void
874 : replay_block_finalize( fd_replay_tile_t * ctx,
875 : fd_stem_context_t * stem,
876 0 : fd_bank_t * bank ) {
877 0 : bank->last_transaction_finished_nanos = fd_log_wallclock();
878 :
879 : /* Set poh hash in bank. */
880 0 : fd_hash_t * poh = fd_sched_get_poh( ctx->sched, bank->idx );
881 0 : bank->f.poh = *poh;
882 :
883 : /* Set shred count in bank. */
884 0 : bank->f.shred_cnt = fd_sched_get_shred_cnt( ctx->sched, bank->idx );
885 :
886 : /* Do hashing and other end-of-block processing. */
887 0 : fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx );
888 :
889 : /* Copy out cost tracker fields before freezing */
890 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
891 0 : cost_tracker_snap( bank, slot_info );
892 :
893 : /* fetch identity / vote balance updates infrequently */
894 0 : ulong slot = bank->f.slot;
895 0 : fd_funk_txn_xid_t xid = { .ul = { slot, bank->idx } };
896 0 : slot_info->identity_balance = FD_UNLIKELY( slot%4096==0UL ) ? get_identity_balance( ctx, xid ) : ULONG_MAX;
897 :
898 : /* Mark the bank as frozen. */
899 0 : fd_banks_mark_bank_frozen( bank );
900 0 : bank->block_completed_nanos = fd_log_wallclock();
901 :
902 : /**********************************************************************/
903 : /* Bank hash comparison, and halt if there's a mismatch after replay */
904 : /**********************************************************************/
905 :
906 : /* Must be last so we can measure completion time correctly, even
907 : though we could technically do this before the hash cmp and vote
908 : tower stuff. */
909 0 : publish_slot_completed( ctx, stem, bank, 0, 0 /* is_leader */ );
910 :
911 0 : # if FD_HAS_FLATCC
912 : /* If enabled, dump the block to a file and reset the dumping
913 : context state */
914 0 : if( FD_UNLIKELY( ctx->dump_proto_ctx && ctx->dump_proto_ctx->dump_block_to_pb ) ) {
915 0 : fd_dump_block_to_protobuf( ctx->block_dump_ctx, ctx->banks, bank, ctx->accdb, ctx->dump_proto_ctx, ctx->runtime_stack );
916 0 : fd_block_dump_context_reset( ctx->block_dump_ctx );
917 0 : }
918 0 : # endif
919 0 : }
920 :
921 : /**********************************************************************/
922 : /* Leader bank management */
923 : /**********************************************************************/
924 :
925 : static fd_bank_t *
926 : prepare_leader_bank( fd_replay_tile_t * ctx,
927 : fd_stem_context_t * stem,
928 : ulong slot,
929 : long now,
930 0 : fd_hash_t const * parent_block_id ) {
931 0 : long before = fd_log_wallclock();
932 :
933 : /* Make sure that we are not already leader. */
934 0 : FD_TEST( ctx->leader_bank==NULL );
935 :
936 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 );
937 0 : if( FD_UNLIKELY( !parent_ele ) ) {
938 0 : FD_BASE58_ENCODE_32_BYTES( parent_block_id->key, parent_block_id_b58 );
939 0 : FD_LOG_CRIT(( "invariant violation: parent bank index not found for merkle root %s", parent_block_id_b58 ));
940 0 : }
941 0 : ulong parent_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, parent_ele );
942 :
943 0 : fd_bank_t * parent_bank = fd_banks_bank_query( ctx->banks, parent_bank_idx );
944 0 : if( FD_UNLIKELY( !parent_bank ) ) {
945 0 : FD_LOG_CRIT(( "invariant violation: parent bank not found for bank index %lu", parent_bank_idx ));
946 0 : }
947 0 : ulong parent_slot = parent_bank->f.slot;
948 :
949 0 : ctx->leader_bank = fd_banks_new_bank( ctx->banks, parent_bank_idx, now );
950 0 : if( FD_UNLIKELY( !ctx->leader_bank ) ) {
951 0 : FD_LOG_CRIT(( "invariant violation: leader bank is NULL for slot %lu", slot ));
952 0 : }
953 :
954 0 : ctx->leader_bank = fd_banks_clone_from_parent( ctx->banks, ctx->leader_bank->idx );
955 0 : if( FD_UNLIKELY( !ctx->leader_bank ) ) {
956 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for slot %lu", slot ));
957 0 : }
958 :
959 0 : ctx->leader_bank->preparation_begin_nanos = before;
960 :
961 0 : ctx->leader_bank->f.slot = slot;
962 0 : ctx->leader_bank->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, parent_bank->txncache_fork_id );
963 : /* prepare the funk transaction for the leader bank */
964 0 : fd_funk_txn_xid_t xid = { .ul = { slot, ctx->leader_bank->idx } };
965 0 : fd_funk_txn_xid_t parent_xid = { .ul = { parent_slot, parent_bank_idx } };
966 0 : fd_accdb_attach_child ( ctx->accdb_admin, &parent_xid, &xid );
967 0 : fd_progcache_attach_child( ctx->progcache, &parent_xid, &xid );
968 :
969 0 : int is_epoch_boundary = 0;
970 0 : fd_runtime_block_execute_prepare( ctx->banks, ctx->leader_bank, ctx->accdb, ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
971 0 : if( FD_UNLIKELY( is_epoch_boundary ) ) publish_epoch_info( ctx, stem, ctx->leader_bank, 0 );
972 :
973 0 : ulong max_tick_height;
974 0 : if( FD_UNLIKELY( FD_RUNTIME_EXECUTE_SUCCESS!=fd_runtime_compute_max_tick_height( parent_bank->f.ticks_per_slot, slot, &max_tick_height ) ) ) {
975 0 : FD_LOG_CRIT(( "couldn't compute tick height/max tick height slot %lu ticks_per_slot %lu", slot, parent_bank->f.ticks_per_slot ));
976 0 : }
977 0 : ctx->leader_bank->f.max_tick_height = max_tick_height;
978 :
979 : /* Now that a bank has been created for the leader slot, increment the
980 : reference count until we are done with the leader slot. */
981 0 : ctx->leader_bank->refcnt++;
982 :
983 0 : return ctx->leader_bank;
984 0 : }
985 :
986 : static inline void
987 0 : maybe_switch_identity( fd_replay_tile_t * ctx ) {
988 :
989 0 : if( FD_LIKELY( fd_keyswitch_state_query( ctx->keyswitch )!=FD_KEYSWITCH_STATE_SWITCH_PENDING ) ) return;
990 :
991 : /* Switch identity */
992 :
993 0 : FD_LOG_DEBUG(( "keyswitch: switching identity" ));
994 :
995 0 : memcpy( ctx->identity_pubkey, ctx->keyswitch->bytes, 32UL );
996 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_COMPLETED );
997 :
998 : /* The next leader slot will be incorrect now that the identity has
999 : switched. The next leader slot normally gets updated based on the
1000 : reset slot returned by tower. */
1001 0 : ulong min_leader_slot = fd_ulong_max( ctx->reset_slot+1UL, fd_ulong_if( ctx->highwater_leader_slot==ULONG_MAX, 0UL, ctx->highwater_leader_slot+1UL ) );
1002 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, min_leader_slot, ctx->identity_pubkey );
1003 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1004 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1005 0 : } else {
1006 0 : ctx->next_leader_tickcount = LONG_MAX;
1007 0 : }
1008 :
1009 0 : ctx->identity_vote_rooted = 0;
1010 0 : ctx->identity_idx++;
1011 0 : fd_vote_tracker_reset( ctx->vote_tracker );
1012 0 : }
1013 :
1014 : static void
1015 : fini_leader_bank( fd_replay_tile_t * ctx,
1016 0 : fd_stem_context_t * stem ) {
1017 :
1018 0 : FD_TEST( ctx->leader_bank!=NULL );
1019 0 : FD_TEST( ctx->is_leader );
1020 0 : FD_TEST( ctx->block_id_arr[ ctx->leader_bank->idx ].block_id_seen );
1021 0 : FD_TEST( ctx->recv_poh );
1022 :
1023 0 : ctx->leader_bank->last_transaction_finished_nanos = fd_log_wallclock();
1024 :
1025 0 : ulong curr_slot = ctx->leader_bank->f.slot;
1026 :
1027 0 : fd_sched_block_add_done( ctx->sched, ctx->leader_bank->idx, ctx->leader_bank->parent_idx, curr_slot );
1028 :
1029 0 : fd_runtime_block_execute_finalize( ctx->leader_bank, ctx->accdb, ctx->capture_ctx );
1030 :
1031 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1032 0 : cost_tracker_snap( ctx->leader_bank, slot_info );
1033 0 : fd_funk_txn_xid_t xid = { .ul = { curr_slot, ctx->leader_bank->idx } };
1034 0 : slot_info->identity_balance = FD_UNLIKELY( curr_slot%4096==0UL ) ? get_identity_balance( ctx, xid ) : ULONG_MAX;
1035 :
1036 0 : fd_banks_mark_bank_frozen( ctx->leader_bank );
1037 0 : ctx->leader_bank->block_completed_nanos = fd_log_wallclock();
1038 :
1039 0 : publish_slot_completed( ctx, stem, ctx->leader_bank, 0, 1 /* is_leader */ );
1040 :
1041 : /* The reference on the bank is finally no longer needed. */
1042 0 : ctx->leader_bank->refcnt--;
1043 :
1044 : /* We are no longer leader so we can clear the bank index we use for
1045 : being the leader. */
1046 0 : ctx->leader_bank = NULL;
1047 0 : ctx->recv_poh = 0;
1048 0 : ctx->is_leader = 0;
1049 :
1050 0 : maybe_switch_identity( ctx );
1051 :
1052 0 : }
1053 :
1054 : static void
1055 : publish_root_advanced( fd_replay_tile_t * ctx,
1056 0 : fd_stem_context_t * stem ) {
1057 :
1058 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, ctx->consensus_root_bank_idx );
1059 0 : if( FD_UNLIKELY( !bank ) ) {
1060 0 : FD_LOG_CRIT(( "invariant violation: consensus root bank is NULL at bank index %lu", ctx->consensus_root_bank_idx ));
1061 0 : }
1062 :
1063 0 : if( FD_UNLIKELY( bank->f.epoch>fd_slot_to_epoch( &bank->f.epoch_schedule, bank->f.parent_slot, NULL ) )) {
1064 0 : publish_epoch_info( ctx, stem, bank, 1 );
1065 0 : }
1066 :
1067 0 : if( ctx->rpc_enabled ) {
1068 0 : bank->refcnt++;
1069 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for rpc", bank->idx, bank->f.slot, bank->refcnt ));
1070 0 : }
1071 :
1072 : /* Increment the reference count on the consensus root bank to account
1073 : for the number of resolv tiles that are waiting on it. */
1074 0 : bank->refcnt += ctx->resolv_tile_cnt;
1075 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for resolv", bank->idx, bank->f.slot, bank->refcnt ));
1076 :
1077 0 : fd_replay_root_advanced_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1078 0 : msg->bank_idx = bank->idx;
1079 :
1080 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() ) );
1081 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 );
1082 0 : }
1083 :
1084 : /* init_funk performs pre-flight checks for the account database and
1085 : program cache. Ensures that the account database was set up
1086 : correctly by bootstrap components (e.g. genesis or snapshot loader).
1087 : Mirrors the account database's fork tree down to the program cache. */
1088 :
1089 : static void
1090 : init_funk( fd_replay_tile_t * ctx,
1091 0 : ulong bank_slot ) {
1092 : /* Ensure that the loaded bank root corresponds to the account
1093 : database's root. */
1094 0 : fd_funk_t * funk = fd_accdb_user_v1_funk( ctx->accdb );
1095 0 : if( FD_UNLIKELY( !funk->shmem ) ) {
1096 0 : FD_LOG_CRIT(( "failed to initialize account database: replay tile is not joined to database shared memory objects" ));
1097 0 : }
1098 0 : fd_funk_txn_xid_t const * accdb_pub = fd_funk_last_publish( funk );
1099 0 : if( FD_UNLIKELY( accdb_pub->ul[0]!=bank_slot ) ) {
1100 0 : FD_LOG_CRIT(( "failed to initialize account database: accdb is at slot %lu, but chain state is at slot %lu\n"
1101 0 : "This is a bug in startup components.",
1102 0 : accdb_pub->ul[0], bank_slot ));
1103 0 : }
1104 0 : if( FD_UNLIKELY( fd_funk_last_publish_is_frozen( funk ) ) ) {
1105 0 : FD_LOG_CRIT(( "failed to initialize account database: accdb fork graph is not clean.\n"
1106 0 : "The account database should only contain state for the root slot at this point,\n"
1107 0 : "but there are incomplete database transactions leftover.\n"
1108 0 : "This is a bug in startup components." ));
1109 0 : }
1110 :
1111 : /* The program cache tracks the account database's fork graph at all
1112 : times. Perform initial synchronization: pivot from funk 'root' (a
1113 : sentinel XID) to 'last publish' (the bootstrap root slot). */
1114 0 : if( FD_UNLIKELY( !ctx->progcache->shmem ) ) {
1115 0 : FD_LOG_CRIT(( "failed to initialize account database: replay tile is not joined to program cache" ));
1116 0 : }
1117 0 : fd_progcache_clear( ctx->progcache );
1118 :
1119 0 : fd_funk_txn_xid_t last_publish = fd_accdb_root_get( ctx->accdb_admin );
1120 0 : fd_funk_txn_xid_t root = { .ul = { ULONG_MAX, ULONG_MAX } };
1121 0 : fd_progcache_attach_child( ctx->progcache, &root, &last_publish );
1122 0 : fd_progcache_advance_root( ctx->progcache, &last_publish );
1123 0 : }
1124 :
1125 : static void
1126 0 : init_after_snapshot( fd_replay_tile_t * ctx ) {
1127 : /* Now that the snapshot has been loaded in, we have to refresh the
1128 : stake delegations since the manifest does not contain the full set
1129 : of data required for the stake delegations. See
1130 : fd_stake_delegations.h for why this is required. */
1131 :
1132 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
1133 0 : if( FD_UNLIKELY( !bank ) ) {
1134 0 : FD_LOG_CRIT(( "invariant violation: replay bank is NULL at bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
1135 0 : }
1136 :
1137 0 : fd_funk_txn_xid_t xid = { .ul = { bank->f.slot, bank->idx } };
1138 0 : init_funk( ctx, bank->f.slot );
1139 :
1140 0 : bank->f.warmup_cooldown_rate_epoch = fd_slot_to_epoch( &bank->f.epoch_schedule, bank->f.features.reduce_stake_warmup_cooldown, NULL );
1141 0 : fd_stake_delegations_t * root_delegations = fd_banks_stake_delegations_root_query( ctx->banks );
1142 0 : fd_stake_delegations_refresh(
1143 0 : root_delegations,
1144 0 : bank->f.epoch,
1145 0 : fd_sysvar_cache_stake_history_join_const( &bank->f.sysvar_cache ),
1146 0 : &bank->f.warmup_cooldown_rate_epoch,
1147 0 : ctx->accdb,
1148 0 : &xid );
1149 :
1150 0 : fd_top_votes_t * top_votes_t_2 = fd_bank_top_votes_t_2_modify( bank );
1151 0 : fd_top_votes_refresh( top_votes_t_2, ctx->accdb, &xid );
1152 :
1153 : /* After both snapshots have been loaded in, we can determine if we should
1154 : start distributing rewards. */
1155 :
1156 0 : fd_rewards_recalculate_partitioned_rewards( ctx->banks, bank, ctx->accdb, &xid, ctx->runtime_stack, ctx->capture_ctx );
1157 :
1158 0 : ulong snapshot_slot = bank->f.slot;
1159 0 : if( FD_UNLIKELY( !snapshot_slot ) ) {
1160 : /* Genesis-specific setup. */
1161 : /* FIXME: This branch does not set up a new block exec ctx
1162 : properly. Needs to do whatever prepare_new_block_execution
1163 : does, but just hacking that in breaks stuff. */
1164 0 : fd_runtime_update_leaders( bank, ctx->runtime_stack );
1165 :
1166 0 : ulong hashcnt_per_slot = bank->f.hashes_per_tick * bank->f.ticks_per_slot;
1167 0 : fd_hash_t * poh = &bank->f.poh;
1168 0 : while( hashcnt_per_slot-- ) {
1169 0 : fd_sha256_hash( poh->hash, 32UL, poh->hash );
1170 0 : }
1171 :
1172 0 : int is_epoch_boundary = 0;
1173 0 : fd_runtime_block_execute_prepare( ctx->banks, bank, ctx->accdb, ctx->runtime_stack, ctx->capture_ctx, &is_epoch_boundary );
1174 0 : FD_TEST( !is_epoch_boundary );
1175 0 : fd_runtime_block_execute_finalize( bank, ctx->accdb, ctx->capture_ctx );
1176 :
1177 0 : snapshot_slot = 0UL;
1178 0 : }
1179 0 : }
1180 :
1181 : static inline int
1182 : maybe_become_leader( fd_replay_tile_t * ctx,
1183 0 : fd_stem_context_t * stem ) {
1184 0 : FD_TEST( ctx->is_booted );
1185 0 : if( FD_LIKELY( ctx->next_leader_slot==ULONG_MAX || ctx->is_leader || (!ctx->identity_vote_rooted && ctx->wait_for_vote_to_start_leader) || ctx->replay_out->idx==ULONG_MAX || !ctx->wfs_complete ) ) return 0;
1186 0 : if( FD_UNLIKELY( ctx->halt_leader ) ) return 0;
1187 0 : if( !ctx->supports_leader ) return 0;
1188 :
1189 0 : FD_TEST( ctx->next_leader_slot>ctx->reset_slot );
1190 0 : long now = fd_tickcount();
1191 0 : if( FD_LIKELY( now<ctx->next_leader_tickcount ) ) return 0;
1192 :
1193 : /* If a prior leader is still in the process of publishing their slot,
1194 : delay ours to let them finish ... unless they are so delayed that
1195 : we risk getting skipped by the leader following us. 1.2 seconds
1196 : is a reasonable default here, although any value between 0 and 1.6
1197 : seconds could be considered reasonable. This is arbitrary and
1198 : chosen due to intuition. */
1199 0 : if( FD_UNLIKELY( now<ctx->next_leader_tickcount+(long)(3.0*ctx->slot_duration_ticks) ) ) {
1200 0 : FD_TEST( ctx->reset_bank );
1201 :
1202 : /* TODO: Make the max_active_descendant calculation more efficient
1203 : by caching it in the bank structure and updating it as banks are
1204 : created and completed. */
1205 0 : ulong max_active_descendant = 0UL;
1206 0 : ulong child_idx = ctx->reset_bank->child_idx;
1207 0 : while( child_idx!=ULONG_MAX ) {
1208 0 : fd_bank_t * child_bank = fd_banks_bank_query( ctx->banks, child_idx );
1209 0 : max_active_descendant = fd_ulong_max( max_active_descendant, child_bank->f.slot );
1210 0 : child_idx = child_bank->sibling_idx;
1211 0 : }
1212 :
1213 : /* If the max_active_descendant is >= next_leader_slot, we waited
1214 : too long and a leader after us started publishing to try and skip
1215 : us. Just start our leader slot immediately, we might win ... */
1216 0 : if( FD_LIKELY( max_active_descendant>=ctx->reset_slot && max_active_descendant<ctx->next_leader_slot ) ) {
1217 : /* If one of the leaders between the reset slot and our leader
1218 : slot is in the process of publishing (they have a descendant
1219 : bank that is in progress of being replayed), then keep waiting.
1220 : We probably wouldn't get a leader slot out before they
1221 : finished.
1222 :
1223 : Unless... we are past the deadline to start our slot by more
1224 : than 1.2 seconds, in which case we should probably start it to
1225 : avoid getting skipped by the leader behind us. */
1226 0 : return 0;
1227 0 : }
1228 0 : }
1229 :
1230 : /* If we haven't started replaying the prior block, but we have
1231 : finished replaying the second to last slot of the prior
1232 : leader (and that leader is not us), we should give the prior leader
1233 : a little more time. */
1234 0 : if( FD_UNLIKELY( ctx->next_leader_slot==ctx->reset_slot+2UL && now<ctx->next_leader_tickcount+(long)(1.0*ctx->slot_duration_ticks) ) ) {
1235 :
1236 0 : fd_pubkey_t const * reset_leader = fd_multi_epoch_leaders_get_leader_for_slot( ctx->mleaders, ctx->reset_slot );
1237 0 : if( FD_UNLIKELY( reset_leader && !fd_memeq( reset_leader, ctx->identity_pubkey, 32UL ) ) ) return 0;
1238 0 : }
1239 :
1240 :
1241 0 : long now_nanos = fd_log_wallclock();
1242 :
1243 0 : ctx->is_leader = 1;
1244 0 : ctx->recv_poh = 0;
1245 :
1246 0 : FD_TEST( ctx->highwater_leader_slot==ULONG_MAX || ctx->highwater_leader_slot<ctx->next_leader_slot );
1247 0 : ctx->highwater_leader_slot = ctx->next_leader_slot;
1248 :
1249 0 : FD_LOG_INFO(( "becoming leader for slot %lu, parent slot is %lu", ctx->next_leader_slot, ctx->reset_slot ));
1250 :
1251 : /* Acquires bank, sets up initial state, and refcnts it. */
1252 0 : fd_bank_t * bank = prepare_leader_bank( ctx, stem, ctx->next_leader_slot, now_nanos, &ctx->reset_block_id );
1253 0 : fd_funk_txn_xid_t xid = { .ul = { ctx->next_leader_slot, ctx->leader_bank->idx } };
1254 :
1255 0 : fd_bundle_crank_tip_payment_config_t config[1] = { 0 };
1256 0 : fd_pubkey_t tip_receiver_owner = {0};
1257 :
1258 0 : if( FD_UNLIKELY( ctx->bundle.enabled ) ) {
1259 0 : fd_acct_addr_t tip_payment_config[1];
1260 0 : fd_acct_addr_t tip_receiver[1];
1261 0 : fd_bundle_crank_get_addresses( ctx->bundle.gen, bank->f.epoch, tip_payment_config, tip_receiver );
1262 :
1263 0 : fd_accdb_ro_t tip_config_acc[1];
1264 0 : if( FD_UNLIKELY( !fd_accdb_open_ro( ctx->accdb, tip_config_acc, &xid, tip_payment_config ) ) ) {
1265 0 : FD_BASE58_ENCODE_32_BYTES( tip_payment_config->b, tip_config_acc_b58 );
1266 0 : FD_LOG_WARNING(( "tip payment config account %s does not exist", tip_config_acc_b58 ));
1267 0 : } else if( FD_UNLIKELY( fd_accdb_ref_data_sz( tip_config_acc )<sizeof(fd_bundle_crank_tip_payment_config_t) ) ) {
1268 0 : FD_LOG_HEXDUMP_WARNING(( "invalid tip payment config account data", fd_accdb_ref_data_const( tip_config_acc ), fd_accdb_ref_data_sz( tip_config_acc ) ));
1269 0 : fd_accdb_close_ro( ctx->accdb, tip_config_acc );
1270 0 : } else {
1271 0 : memcpy( config, fd_accdb_ref_data_const( tip_config_acc ), sizeof(fd_bundle_crank_tip_payment_config_t) );
1272 0 : fd_accdb_close_ro( ctx->accdb, tip_config_acc );
1273 0 : }
1274 :
1275 : /* It is possible that the tip receiver account does not exist yet
1276 : if it is the first time in an epoch. */
1277 0 : fd_accdb_ro_t tip_receiver_acc[1];
1278 0 : if( FD_LIKELY( fd_accdb_open_ro( ctx->accdb, tip_receiver_acc, &xid, tip_receiver ) ) ) {
1279 0 : tip_receiver_owner = *fd_accdb_ref_owner( tip_receiver_acc );
1280 0 : fd_accdb_close_ro( ctx->accdb, tip_receiver_acc );
1281 0 : }
1282 0 : }
1283 :
1284 :
1285 0 : fd_became_leader_t * msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1286 0 : msg->slot = ctx->next_leader_slot;
1287 0 : msg->slot_start_ns = now_nanos;
1288 0 : msg->slot_end_ns = now_nanos+(long)ctx->slot_duration_nanos;
1289 0 : msg->bank = NULL;
1290 0 : msg->bank_idx = bank->idx;
1291 0 : msg->ticks_per_slot = bank->f.ticks_per_slot;
1292 0 : msg->hashcnt_per_tick = bank->f.hashes_per_tick;
1293 0 : msg->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)msg->ticks_per_slot);
1294 0 : msg->bundle->config[0] = config[0];
1295 0 : memcpy( msg->bundle->last_blockhash, bank->f.poh.hash, sizeof(fd_hash_t) );
1296 0 : memcpy( msg->bundle->tip_receiver_owner, tip_receiver_owner.uc, sizeof(fd_pubkey_t) );
1297 :
1298 0 : if( FD_UNLIKELY( msg->hashcnt_per_tick==1UL ) ) {
1299 : /* Low power producer, maximum of one microblock per tick in the slot */
1300 0 : msg->max_microblocks_in_slot = msg->ticks_per_slot;
1301 0 : } else {
1302 : /* See the long comment in after_credit for this limit */
1303 0 : msg->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, msg->ticks_per_slot*(msg->hashcnt_per_tick-1UL) );
1304 0 : }
1305 :
1306 0 : msg->total_skipped_ticks = msg->ticks_per_slot*(ctx->next_leader_slot-ctx->reset_slot);
1307 0 : msg->epoch = fd_slot_to_epoch( &bank->f.epoch_schedule, ctx->next_leader_slot, NULL );
1308 :
1309 0 : fd_cost_tracker_t const * cost_tracker = fd_bank_cost_tracker_query( bank );
1310 :
1311 0 : msg->limits.slot_max_cost = ctx->larger_max_cost_per_block ? LARGER_MAX_COST_PER_BLOCK : cost_tracker->block_cost_limit;
1312 0 : msg->limits.slot_max_vote_cost = cost_tracker->vote_cost_limit;
1313 0 : msg->limits.slot_max_write_cost_per_acct = cost_tracker->account_cost_limit;
1314 :
1315 0 : if( FD_UNLIKELY( msg->ticks_per_slot+msg->total_skipped_ticks>USHORT_MAX ) ) {
1316 : /* There can be at most USHORT_MAX skipped ticks, because the
1317 : parent_offset field in the shred data is only 2 bytes wide. */
1318 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 ));
1319 0 : }
1320 :
1321 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() ) );
1322 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 );
1323 :
1324 0 : ctx->next_leader_slot = ULONG_MAX;
1325 0 : ctx->next_leader_tickcount = LONG_MAX;
1326 :
1327 0 : return 1;
1328 0 : }
1329 :
1330 : static void
1331 : process_poh_message( fd_replay_tile_t * ctx,
1332 0 : fd_poh_leader_slot_ended_t const * slot_ended ) {
1333 :
1334 0 : FD_TEST( ctx->is_booted );
1335 0 : FD_TEST( ctx->is_leader );
1336 0 : FD_TEST( ctx->leader_bank!=NULL );
1337 :
1338 0 : FD_TEST( ctx->highwater_leader_slot>=slot_ended->slot );
1339 0 : FD_TEST( ctx->next_leader_slot>ctx->highwater_leader_slot );
1340 :
1341 : /* Update the poh hash in the bank. We will want to maintain a refcnt
1342 : on the bank until we have recieved the block id for the block after
1343 : it has been shredded. */
1344 :
1345 0 : memcpy( &ctx->leader_bank->f.poh, slot_ended->blockhash, sizeof(fd_hash_t) );
1346 :
1347 0 : ctx->recv_poh = 1;
1348 0 : }
1349 :
1350 : static void
1351 : publish_reset( fd_replay_tile_t * ctx,
1352 : fd_stem_context_t * stem,
1353 0 : fd_bank_t * bank ) {
1354 0 : if( FD_UNLIKELY( ctx->replay_out->idx==ULONG_MAX ) ) return;
1355 :
1356 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( &bank->f.block_hash_queue );
1357 0 : FD_TEST( block_hash );
1358 :
1359 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1360 :
1361 0 : reset->bank_idx = bank->idx;
1362 0 : reset->timestamp = fd_log_wallclock();
1363 0 : reset->completed_slot = bank->f.slot;
1364 0 : reset->hashcnt_per_tick = bank->f.hashes_per_tick;
1365 0 : reset->ticks_per_slot = bank->f.ticks_per_slot;
1366 0 : reset->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)reset->ticks_per_slot);
1367 0 : fd_memcpy( reset->completed_block_id, ctx->reset_block_id.uc, sizeof(fd_hash_t) );
1368 0 : fd_memcpy( reset->completed_blockhash, block_hash->uc, sizeof(fd_hash_t) );
1369 :
1370 0 : ulong ticks_per_slot = bank->f.ticks_per_slot;
1371 0 : if( FD_UNLIKELY( reset->hashcnt_per_tick==1UL ) ) {
1372 : /* Low power producer, maximum of one microblock per tick in the slot */
1373 0 : reset->max_microblocks_in_slot = ticks_per_slot;
1374 0 : } else {
1375 : /* See the long comment in after_credit for this limit */
1376 0 : reset->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ticks_per_slot*(reset->hashcnt_per_tick-1UL) );
1377 0 : }
1378 0 : reset->next_leader_slot = ctx->next_leader_slot;
1379 :
1380 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() ) );
1381 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 );
1382 0 : }
1383 :
1384 : static void
1385 : store_xinsert( fd_store_t * store,
1386 0 : fd_hash_t const * merkle_root ) {
1387 0 : fd_store_pool_t pool = {
1388 0 : .pool = fd_wksp_laddr_fast( fd_store_wksp( store ), store->pool_mem_gaddr ),
1389 0 : .ele = fd_wksp_laddr_fast( fd_store_wksp( store ), store->pool_ele_gaddr ),
1390 0 : .ele_max = store->fec_max
1391 0 : };
1392 0 : int err; fd_store_fec_t * fec = fd_store_pool_acquire( &pool, NULL, 1 /* blocking */, &err );
1393 0 : if( FD_UNLIKELY( err!=FD_POOL_SUCCESS ) ) FD_LOG_CRIT(( "store pool: %s", fd_store_pool_strerror( err ) ));
1394 0 : fec->key.merkle_root = *merkle_root;
1395 0 : fec->key.part_idx = 0;
1396 0 : fec->cmr = (fd_hash_t){ 0 };
1397 0 : fec->next = fd_store_pool_idx_null();
1398 0 : fec->data_sz = 0UL;
1399 :
1400 0 : FD_STORE_XLOCK_BEGIN( store ) {
1401 0 : fd_store_map_ele_insert( fd_wksp_laddr_fast( fd_store_wksp( store ), store->map_gaddr ), fec, pool.ele );
1402 0 : } FD_STORE_XLOCK_END;
1403 0 : }
1404 :
1405 : static void
1406 : boot_genesis( fd_replay_tile_t * ctx,
1407 : fd_stem_context_t * stem,
1408 0 : fd_genesis_meta_t const * meta ) {
1409 : /* If we are bootstrapping, we can't wait to wait for our identity
1410 : vote to be rooted as this creates a circular dependency. */
1411 0 : ctx->identity_vote_rooted = 1;
1412 :
1413 0 : uchar const * genesis_blob = (uchar const *)( meta+1 );
1414 0 : FD_TEST( meta->bootstrap && meta->has_lthash );
1415 0 : FD_TEST( fd_genesis_parse( ctx->genesis, genesis_blob, meta->blob_sz ) );
1416 :
1417 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
1418 0 : FD_TEST( bank );
1419 0 : fd_funk_txn_xid_t xid = { .ul = { 0UL, FD_REPLAY_BOOT_BANK_IDX } };
1420 :
1421 : /* Do genesis-related processing in a non-rooted transaction */
1422 0 : fd_funk_txn_xid_t root_xid = { .ul = { LONG_MAX, LONG_MAX } };
1423 0 : fd_funk_txn_xid_t target_xid = { .ul = { 0UL, 0UL } };
1424 0 : fd_accdb_attach_child( ctx->accdb_admin, &root_xid, &target_xid );
1425 0 : fd_runtime_read_genesis( ctx->banks, bank, ctx->accdb, &xid, NULL, &meta->genesis_hash, &meta->lthash, ctx->genesis, genesis_blob, ctx->runtime_stack );
1426 0 : fd_accdb_advance_root( ctx->accdb_admin, &target_xid );
1427 :
1428 0 : static const fd_txncache_fork_id_t txncache_root = { .val = USHORT_MAX };
1429 0 : bank->txncache_fork_id = fd_txncache_attach_child( ctx->txncache, txncache_root );
1430 :
1431 0 : fd_hash_t const * block_hash = fd_blockhashes_peek_last_hash( &bank->f.block_hash_queue );
1432 0 : fd_txncache_finalize_fork( ctx->txncache, bank->txncache_fork_id, 0UL, block_hash->uc );
1433 :
1434 0 : publish_epoch_info( ctx, stem, bank, 0 );
1435 0 : publish_epoch_info( ctx, stem, bank, 1 );
1436 :
1437 : /* We call this after fd_runtime_read_genesis, which sets up the
1438 : slot_bank needed in blockstore_init. */
1439 0 : init_after_snapshot( ctx );
1440 :
1441 0 : ctx->published_root_slot = 0UL;
1442 0 : fd_sched_block_add_done( ctx->sched, bank->idx, ULONG_MAX, 0UL );
1443 :
1444 0 : bank->f.block_height = 1UL;
1445 :
1446 0 : ctx->consensus_root = ctx->initial_block_id;
1447 0 : ctx->consensus_root_slot = 0UL;
1448 0 : ctx->consensus_root_bank_idx = 0UL;
1449 0 : ctx->published_root_slot = 0UL;
1450 0 : ctx->published_root_bank_idx = 0UL;
1451 :
1452 0 : ctx->reset_slot = 0UL;
1453 0 : ctx->reset_bank = bank;
1454 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1455 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, 1UL, ctx->identity_pubkey );
1456 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1457 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1458 0 : } else {
1459 0 : ctx->next_leader_tickcount = LONG_MAX;
1460 0 : }
1461 :
1462 0 : ctx->is_booted = 1;
1463 0 : maybe_become_leader( ctx, stem );
1464 :
1465 0 : fd_hash_t initial_block_id = ctx->initial_block_id;
1466 0 : fd_reasm_fec_t * fec = fd_reasm_insert( ctx->reasm, &initial_block_id, NULL, 0 /* genesis slot */, 0, 0, 0, 0, 1, 0, ctx->store, &ctx->reasm_evicted ); /* FIXME manifest block_id */
1467 0 : fec->bank_idx = bank->idx;
1468 0 : fec->bank_seq = bank->bank_seq;
1469 0 : store_xinsert( ctx->store, &initial_block_id );
1470 :
1471 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
1472 0 : block_id_ele->latest_mr = initial_block_id;
1473 0 : block_id_ele->slot = 0UL;
1474 :
1475 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1476 :
1477 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1478 0 : cost_tracker_snap( bank, slot_info );
1479 0 : slot_info->identity_balance = get_identity_balance( ctx, xid );
1480 :
1481 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */ );
1482 0 : publish_root_advanced( ctx, stem );
1483 0 : publish_reset( ctx, stem, bank );
1484 0 : }
1485 :
1486 : static inline void
1487 0 : maybe_verify_cluster_type( fd_replay_tile_t * ctx ) {
1488 0 : if( FD_UNLIKELY( !ctx->is_booted || !ctx->has_genesis_hash ) ) {
1489 0 : return;
1490 0 : }
1491 :
1492 0 : FD_BASE58_ENCODE_32_BYTES( ctx->genesis_hash->uc, hash_cstr );
1493 0 : ulong cluster = fd_genesis_cluster_identify( hash_cstr );
1494 : /* Map pyth-related clusters to unkwown. */
1495 0 : switch( cluster ) {
1496 0 : case FD_CLUSTER_PYTHNET:
1497 0 : case FD_CLUSTER_PYTHTEST:
1498 0 : cluster = FD_CLUSTER_UNKNOWN;
1499 0 : }
1500 :
1501 0 : if( FD_UNLIKELY( cluster!=ctx->cluster_type ) ) {
1502 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` has a genesis hash of `%s` which means the cluster is %s "
1503 0 : "but the snapshot you loaded is for a different cluster %s. If you are trying to join the "
1504 0 : "%s cluster, you can delete the genesis.bin file and restart the node to download the correct "
1505 0 : "genesis file automatically.",
1506 0 : ctx->genesis_path,
1507 0 : hash_cstr,
1508 0 : fd_genesis_cluster_name( cluster ),
1509 0 : fd_genesis_cluster_name( ctx->cluster_type ),
1510 0 : fd_genesis_cluster_name( cluster ) ));
1511 0 : }
1512 0 : }
1513 :
1514 : static void
1515 : on_snapshot_message( fd_replay_tile_t * ctx,
1516 : fd_stem_context_t * stem,
1517 : ulong in_idx,
1518 : ulong chunk,
1519 0 : ulong sig ) {
1520 0 : ulong msg = fd_ssmsg_sig_message( sig );
1521 0 : if( FD_LIKELY( msg==FD_SSMSG_DONE ) ) {
1522 : /* An end of message notification indicates the snapshot is loaded.
1523 : Replay is able to start executing from this point onwards. */
1524 : /* TODO: replay should finish booting. Could make replay a
1525 : state machine and set the state here accordingly. */
1526 0 : ctx->is_booted = 1;
1527 :
1528 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX );
1529 0 : if( FD_UNLIKELY( !bank ) ) {
1530 0 : FD_LOG_CRIT(( "invariant violation: bank is NULL for bank index %lu", FD_REPLAY_BOOT_BANK_IDX ));
1531 0 : }
1532 :
1533 0 : ulong snapshot_slot = bank->f.slot;
1534 :
1535 0 : fd_hash_t bank_hash = bank->f.bank_hash;
1536 0 : if( FD_UNLIKELY( ctx->wfs_enabled && memcmp( ctx->expected_bank_hash.uc, bank_hash.uc, sizeof(fd_hash_t) ) ) ) {
1537 0 : FD_BASE58_ENCODE_32_BYTES( ctx->expected_bank_hash.uc, expected_bank_hash_cstr );
1538 0 : FD_BASE58_ENCODE_32_BYTES( bank_hash.uc, actual_bank_hash_cstr );
1539 0 : FD_LOG_ERR(( "[consensus.wait_for_supermajority_with_bank_hash] expected_bank_hash=%s does not match snapshot slot"
1540 0 : "=%lu bank_hash=%s. If you are loading a snapshot from the network, check that the slot matches the "
1541 0 : "cluster restart slot. ", expected_bank_hash_cstr, snapshot_slot, actual_bank_hash_cstr ));
1542 0 : }
1543 0 : if( FD_UNLIKELY( ctx->wfs_enabled ) ) {
1544 0 : FD_LOG_NOTICE(( "waiting for supermajority at snapshot slot %lu", snapshot_slot ));
1545 0 : }
1546 :
1547 : /* FIXME: This is a hack because the block id of the snapshot slot
1548 : is not provided in the snapshot. A possible solution is to get
1549 : the block id of the snapshot slot from repair. */
1550 0 : fd_hash_t manifest_block_id = ctx->initial_block_id;
1551 :
1552 0 : fd_funk_txn_xid_t xid = { .ul = { snapshot_slot, FD_REPLAY_BOOT_BANK_IDX } };
1553 0 : fd_features_restore( bank, ctx->accdb, &xid );
1554 :
1555 0 : FD_TEST( fd_sysvar_cache_restore( bank, ctx->accdb, &xid ) );
1556 :
1557 0 : ctx->consensus_root = manifest_block_id;
1558 0 : ctx->consensus_root_slot = snapshot_slot;
1559 0 : ctx->consensus_root_bank_idx = 0UL;
1560 0 : ctx->published_root_slot = ctx->consensus_root_slot;
1561 0 : ctx->published_root_bank_idx = 0UL;
1562 :
1563 0 : ctx->reset_slot = snapshot_slot;
1564 0 : ctx->reset_bank = bank;
1565 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
1566 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, 1UL, ctx->identity_pubkey );
1567 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
1568 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
1569 0 : } else {
1570 0 : ctx->next_leader_tickcount = LONG_MAX;
1571 0 : }
1572 :
1573 0 : fd_sched_block_add_done( ctx->sched, bank->idx, ULONG_MAX, snapshot_slot );
1574 0 : FD_TEST( bank->idx==0UL );
1575 :
1576 0 : fd_runtime_update_leaders( bank, ctx->runtime_stack );
1577 :
1578 : /* Typically, when we cross an epoch boundary during normal
1579 : operation, we publish the stake weights for the new epoch. But
1580 : since we are starting from a snapshot, we need to publish two
1581 : epochs worth of stake weights: the previous epoch (which is
1582 : needed for voting on the current epoch), and the current epoch
1583 : (which is needed for voting on the next epoch). */
1584 0 : publish_epoch_info( ctx, stem, bank, 0 );
1585 0 : publish_epoch_info( ctx, stem, bank, 1 );
1586 :
1587 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ 0 ];
1588 0 : block_id_ele->latest_mr = manifest_block_id;
1589 0 : block_id_ele->slot = snapshot_slot;
1590 0 : block_id_ele->block_id_seen = 1;
1591 0 : block_id_ele->latest_fec_idx = 0U;
1592 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1593 :
1594 : /* We call this after fd_runtime_read_genesis, which sets up the
1595 : slot_bank needed in blockstore_init. */
1596 0 : init_after_snapshot( ctx );
1597 :
1598 0 : fd_replay_slot_completed_t * slot_info = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
1599 0 : cost_tracker_snap( bank, slot_info );
1600 0 : slot_info->identity_balance = get_identity_balance( ctx, xid );
1601 :
1602 0 : publish_slot_completed( ctx, stem, bank, 1, 0 /* is_leader */ );
1603 0 : publish_root_advanced( ctx, stem );
1604 :
1605 0 : fd_reasm_fec_t * fec = fd_reasm_insert( ctx->reasm, &manifest_block_id, NULL, snapshot_slot, 0, 0, 0, 0, 1, 0, ctx->store, &ctx->reasm_evicted ); /* FIXME manifest block_id */
1606 0 : fec->bank_idx = bank->idx;
1607 0 : fec->bank_seq = bank->bank_seq;
1608 0 : store_xinsert( ctx->store, &manifest_block_id );
1609 :
1610 0 : ctx->cluster_type = bank->f.cluster_type;
1611 :
1612 0 : maybe_verify_cluster_type( ctx );
1613 :
1614 0 : return;
1615 0 : }
1616 :
1617 0 : switch( msg ) {
1618 0 : case FD_SSMSG_MANIFEST_FULL:
1619 0 : case FD_SSMSG_MANIFEST_INCREMENTAL: {
1620 : /* We may either receive a full snapshot manifest or an
1621 : incremental snapshot manifest. Note that this external message
1622 : id is only used temporarily because replay cannot yet receive
1623 : the firedancer-internal snapshot manifest message. */
1624 0 : if( FD_UNLIKELY( chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark ) )
1625 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 ));
1626 :
1627 0 : fd_ssload_recover( fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ),
1628 0 : ctx->banks,
1629 0 : fd_banks_bank_query( ctx->banks, FD_REPLAY_BOOT_BANK_IDX ),
1630 0 : ctx->runtime_stack,
1631 0 : msg==FD_SSMSG_MANIFEST_INCREMENTAL );
1632 :
1633 0 : fd_snapshot_manifest_t const * manifest = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
1634 0 : ctx->hard_forks_cnt = manifest->hard_forks_len;
1635 0 : for( ulong i=0UL; i<manifest->hard_forks_len; i++ ) {
1636 0 : ctx->hard_forks[ i ] = manifest->hard_forks[ i ];
1637 0 : ctx->hard_forks_cnts[ i ] = manifest->hard_forks_cnts[ i ];
1638 0 : }
1639 0 : ctx->has_expected_genesis_timestamp = 1;
1640 0 : ctx->expected_genesis_timestamp = manifest->creation_time_seconds;
1641 0 : break;
1642 0 : }
1643 0 : default: {
1644 0 : FD_LOG_ERR(( "Received unknown snapshot message with msg %lu", msg ));
1645 0 : return;
1646 0 : }
1647 0 : }
1648 :
1649 0 : return;
1650 0 : }
1651 :
1652 : static void
1653 : dispatch_task( fd_replay_tile_t * ctx,
1654 : fd_stem_context_t * stem,
1655 0 : fd_sched_task_t * task ) {
1656 :
1657 0 : switch( task->task_type ) {
1658 0 : case FD_SCHED_TT_TXN_EXEC: {
1659 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_exec->txn_idx );
1660 :
1661 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->txn_exec->bank_idx );
1662 0 : FD_TEST( bank );
1663 :
1664 0 : # if FD_HAS_FLATCC
1665 : /* Add the transaction to the block dumper if necessary. This
1666 : logic doesn't need to be fork-aware since it's only meant to
1667 : be used in backtest. */
1668 0 : if( FD_UNLIKELY( ctx->dump_proto_ctx && ctx->dump_proto_ctx->dump_block_to_pb ) ) {
1669 0 : fd_dump_block_to_protobuf_collect_tx( ctx->block_dump_ctx, txn_p );
1670 0 : }
1671 0 : # endif
1672 :
1673 0 : bank->refcnt++;
1674 :
1675 0 : if( FD_UNLIKELY( !bank->first_transaction_scheduled_nanos ) ) bank->first_transaction_scheduled_nanos = fd_log_wallclock();
1676 :
1677 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1678 0 : fd_execrp_txn_exec_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1679 0 : memcpy( exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
1680 0 : exec_msg->bank_idx = task->txn_exec->bank_idx;
1681 0 : exec_msg->txn_idx = task->txn_exec->txn_idx;
1682 0 : if( FD_UNLIKELY( ctx->capture_ctx ) ) {
1683 0 : exec_msg->capture_txn_idx = ctx->capture_ctx->current_txn_idx++;
1684 0 : }
1685 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXECRP_TT_TXN_EXEC<<32) | task->txn_exec->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
1686 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1687 0 : break;
1688 0 : }
1689 0 : case FD_SCHED_TT_TXN_SIGVERIFY: {
1690 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, task->txn_sigverify->txn_idx );
1691 :
1692 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->txn_sigverify->bank_idx );
1693 0 : FD_TEST( bank );
1694 0 : bank->refcnt++;
1695 :
1696 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1697 0 : fd_execrp_txn_sigverify_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1698 0 : memcpy( exec_msg->txn, txn_p, sizeof(fd_txn_p_t) );
1699 0 : exec_msg->bank_idx = task->txn_sigverify->bank_idx;
1700 0 : exec_msg->txn_idx = task->txn_sigverify->txn_idx;
1701 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXECRP_TT_TXN_SIGVERIFY<<32) | task->txn_sigverify->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, 0UL );
1702 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1703 0 : break;
1704 0 : };
1705 0 : case FD_SCHED_TT_POH_HASH: {
1706 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->poh_hash->bank_idx );
1707 0 : FD_TEST( bank );
1708 0 : bank->refcnt++;
1709 :
1710 0 : fd_replay_out_link_t * exec_out = ctx->exec_out;
1711 0 : fd_execrp_poh_hash_msg_t * exec_msg = fd_chunk_to_laddr( exec_out->mem, exec_out->chunk );
1712 0 : exec_msg->bank_idx = task->poh_hash->bank_idx;
1713 0 : exec_msg->mblk_idx = task->poh_hash->mblk_idx;
1714 0 : exec_msg->hashcnt = task->poh_hash->hashcnt;
1715 0 : memcpy( exec_msg->hash, task->poh_hash->hash, sizeof(fd_hash_t) );
1716 0 : fd_stem_publish( stem, exec_out->idx, (FD_EXECRP_TT_POH_HASH<<32) | task->poh_hash->exec_idx, exec_out->chunk, sizeof(*exec_msg), 0UL, 0UL, 0UL );
1717 0 : exec_out->chunk = fd_dcache_compact_next( exec_out->chunk, sizeof(*exec_msg), exec_out->chunk0, exec_out->wmark );
1718 0 : break;
1719 0 : };
1720 0 : default: {
1721 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
1722 0 : }
1723 0 : }
1724 0 : }
1725 :
1726 : static void
1727 : mark_bank_dead( fd_replay_tile_t * ctx,
1728 : fd_stem_context_t * stem,
1729 0 : ulong bank_idx ) {
1730 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
1731 0 : FD_TEST( bank );
1732 0 : fd_banks_mark_bank_dead( ctx->banks, bank_idx );
1733 :
1734 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank_idx ];
1735 0 : if( block_id_ele->block_id_seen ) publish_slot_dead( ctx, stem, block_id_ele->slot, &block_id_ele->latest_mr );
1736 :
1737 0 : fd_reasm_fec_t * fec = fd_reasm_query( ctx->reasm, &block_id_ele->latest_mr );
1738 0 : if( FD_UNLIKELY( !fec ) ) return;
1739 0 : fec->bank_dead = 1;
1740 :
1741 0 : }
1742 :
1743 : /* Returns 1 if charge_busy. */
1744 : static int
1745 : replay( fd_replay_tile_t * ctx,
1746 0 : fd_stem_context_t * stem ) {
1747 :
1748 0 : if( FD_UNLIKELY( !ctx->is_booted ) ) return 0;
1749 :
1750 0 : int charge_busy = 0;
1751 0 : fd_sched_task_t task[ 1 ];
1752 0 : if( FD_UNLIKELY( !fd_sched_task_next_ready( ctx->sched, task ) ) ) {
1753 0 : return charge_busy; /* Nothing to execute or do. */
1754 0 : }
1755 :
1756 0 : charge_busy = 1;
1757 :
1758 0 : switch( task->task_type ) {
1759 0 : case FD_SCHED_TT_BLOCK_START: {
1760 0 : replay_block_start( ctx, stem, task->block_start->bank_idx, task->block_start->parent_bank_idx, task->block_start->slot );
1761 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_START, ULONG_MAX, ULONG_MAX, NULL );
1762 0 : break;
1763 0 : }
1764 0 : case FD_SCHED_TT_BLOCK_END: {
1765 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->block_end->bank_idx );
1766 0 : if( FD_LIKELY( bank->state==FD_BANK_STATE_REPLAYABLE ) ) replay_block_finalize( ctx, stem, bank );
1767 0 : fd_sched_task_done( ctx->sched, FD_SCHED_TT_BLOCK_END, ULONG_MAX, ULONG_MAX, NULL );
1768 0 : break;
1769 0 : }
1770 0 : case FD_SCHED_TT_TXN_EXEC:
1771 0 : case FD_SCHED_TT_TXN_SIGVERIFY:
1772 0 : case FD_SCHED_TT_POH_HASH: {
1773 : /* Common case: we have a transaction we need to execute. */
1774 0 : dispatch_task( ctx, stem, task );
1775 0 : break;
1776 0 : }
1777 0 : case FD_SCHED_TT_MARK_DEAD: {
1778 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, task->mark_dead->bank_idx );
1779 0 : FD_TEST( bank );
1780 0 : mark_bank_dead( ctx, stem, task->mark_dead->bank_idx );
1781 0 : break;
1782 0 : }
1783 0 : default: {
1784 0 : FD_LOG_CRIT(( "unexpected task type %lu", task->task_type ));
1785 0 : }
1786 0 : }
1787 :
1788 0 : return charge_busy;
1789 0 : }
1790 :
1791 : static int
1792 : can_process_fec( fd_replay_tile_t * ctx,
1793 0 : int * evict_banks_out ) {
1794 0 : fd_reasm_fec_t * fec;
1795 0 : if( FD_UNLIKELY( fd_sched_can_ingest_cnt( ctx->sched )==0UL ) ) {
1796 0 : ctx->metrics.sched_full++;
1797 0 : return 0;
1798 0 : }
1799 :
1800 0 : if( FD_UNLIKELY( (fec = fd_reasm_peek( ctx->reasm ))==NULL ) ) {
1801 0 : ctx->metrics.reasm_empty++;
1802 0 : return 0;
1803 0 : }
1804 :
1805 0 : ctx->metrics.reasm_latest_slot = fec->slot;
1806 0 : ctx->metrics.reasm_latest_fec_idx = fec->fec_set_idx;
1807 :
1808 0 : if( FD_UNLIKELY( ctx->is_leader && fec->fec_set_idx==0U && fd_reasm_parent( ctx->reasm, fec )->bank_idx==ctx->leader_bank->idx ) ) {
1809 : /* This guards against a rare race where we receive the FEC set for
1810 : the slot right after our leader rotation before we freeze the
1811 : bank for the last slot in our leader rotation. Leader slot
1812 : freezing happens only after if we've received the final PoH hash
1813 : from the poh tile as well as the final FEC set for the leader
1814 : slot. So the race happens when FEC sets are delivered and
1815 : processed sooner than the PoH hash, aka when the
1816 : poh=>shred=>replay path for the block id beats the poh=>replay
1817 : path for the poh hash. To mitigate this race, we must block on
1818 : ingesting the FEC set for the ensuing slot before the leader
1819 : bank freezes, because that would violate ordering invariants in
1820 : banks and sched. */
1821 0 : FD_TEST( ctx->block_id_arr[ ctx->leader_bank->idx ].block_id_seen );
1822 0 : FD_TEST( !ctx->recv_poh );
1823 0 : ctx->metrics.leader_bid_wait++;
1824 0 : return 0;
1825 0 : }
1826 :
1827 : /* If fec_set_idx is 0, we need a new bank for a new slot. Banks must
1828 : not be full in this case. */
1829 0 : if( FD_UNLIKELY( fd_banks_is_full( ctx->banks ) && fec->fec_set_idx==0 ) ) {
1830 0 : ctx->metrics.banks_full++;
1831 : /* We only want to evict banks if sched is drained and banks is no
1832 : longer making progress. Otherwise, sched might not release
1833 : refcnts on the frontier/leaf banks immediately, and the eviction
1834 : will have to wait for sched to drain anyways. */
1835 0 : if( FD_UNLIKELY( fd_sched_is_drained( ctx->sched ) ) ) *evict_banks_out = 1;
1836 0 : return 0;
1837 0 : }
1838 :
1839 : /* Otherwise, banks may not be full, so we can always create a new
1840 : bank if needed. Or, if banks are full, the current fec set's
1841 : ancestor (idx 0) already created a bank for this slot.*/
1842 0 : return 1;
1843 0 : }
1844 :
1845 : /* Returns 0 on successful FEC ingestion, 1 if the block got marked
1846 : dead. */
1847 : static int
1848 : insert_fec_set( fd_replay_tile_t * ctx,
1849 : fd_stem_context_t * stem,
1850 0 : fd_reasm_fec_t * reasm_fec ) {
1851 :
1852 0 : long now = fd_log_wallclock();
1853 :
1854 0 : reasm_fec->parent_bank_idx = fd_reasm_parent( ctx->reasm, reasm_fec )->bank_idx;
1855 :
1856 0 : fd_bank_t * parent_bank = fd_banks_bank_query( ctx->banks, reasm_fec->parent_bank_idx );
1857 0 : FD_TEST( parent_bank );
1858 0 : reasm_fec->parent_bank_seq = parent_bank->bank_seq;
1859 :
1860 0 : if( FD_UNLIKELY( reasm_fec->fec_set_idx==0U ) ) {
1861 : /* If the first FEC set for a slot is observed, provision a new bank
1862 : if you are not the leader. Remove any stale block id map entry
1863 : and update the block id entry. */
1864 0 : fd_bank_t * bank = NULL;
1865 0 : if( FD_UNLIKELY( reasm_fec->is_leader ) ) {
1866 0 : bank = ctx->leader_bank;
1867 0 : } else {
1868 0 : bank = fd_banks_new_bank( ctx->banks, reasm_fec->parent_bank_idx, now );
1869 0 : }
1870 :
1871 0 : reasm_fec->bank_idx = bank->idx;
1872 0 : reasm_fec->bank_seq = bank->bank_seq;
1873 :
1874 : /* At this point remove any stale entry in the block id map if it
1875 : exists and set the block id as not having been seen yet. This is
1876 : safe because we know that the old entry for this bank index has
1877 : already been pruned away. */
1878 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1879 0 : if( FD_LIKELY( fd_block_id_map_ele_query( ctx->block_id_map, &block_id_ele->latest_mr, NULL, ctx->block_id_arr )==block_id_ele ) ) {
1880 0 : FD_TEST( fd_block_id_map_ele_remove( ctx->block_id_map, &block_id_ele->latest_mr, NULL, ctx->block_id_arr ) );
1881 0 : }
1882 0 : block_id_ele->block_id_seen = 0;
1883 0 : block_id_ele->slot = reasm_fec->slot;
1884 0 : block_id_ele->latest_fec_idx = 0U;
1885 0 : block_id_ele->latest_mr = reasm_fec->key;
1886 0 : } else {
1887 : /* We are continuing to execute through a slot that we already have
1888 : a bank index for. */
1889 0 : reasm_fec->bank_idx = reasm_fec->parent_bank_idx;
1890 0 : reasm_fec->bank_seq = reasm_fec->parent_bank_seq;
1891 :
1892 0 : FD_TEST( reasm_fec->bank_idx!=ULONG_MAX );
1893 :
1894 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1895 0 : if( FD_UNLIKELY( block_id_ele->latest_fec_idx>=reasm_fec->fec_set_idx ) ) {
1896 0 : FD_LOG_WARNING(( "dropping FEC set (slot=%lu, fec_set_idx=%u) because it is at least as old as the latest FEC set (slot=%lu, fec_set_idx=%u)", reasm_fec->slot, reasm_fec->fec_set_idx, block_id_ele->slot, block_id_ele->latest_fec_idx ));
1897 0 : return 0;
1898 0 : }
1899 0 : block_id_ele->latest_fec_idx = reasm_fec->fec_set_idx;
1900 0 : block_id_ele->latest_mr = reasm_fec->key;
1901 0 : }
1902 :
1903 0 : if( FD_UNLIKELY( reasm_fec->slot_complete ) ) {
1904 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ reasm_fec->bank_idx ];
1905 0 : block_id_ele->block_id_seen = 1;
1906 0 : block_id_ele->latest_mr = reasm_fec->key;
1907 0 : block_id_ele->latest_fec_idx = reasm_fec->fec_set_idx;
1908 0 : FD_TEST( fd_block_id_map_ele_insert( ctx->block_id_map, block_id_ele, ctx->block_id_arr ) );
1909 0 : }
1910 :
1911 : /* If we are the leader, we don't need to process the FEC set. */
1912 0 : if( FD_UNLIKELY( reasm_fec->is_leader ) ) return 0;
1913 :
1914 : /* Forks form a partial ordering over FEC sets. The Repair tile
1915 : delivers FEC sets in-order per fork, but FEC set ordering across
1916 : forks is arbitrary */
1917 0 : fd_sched_fec_t sched_fec[ 1 ];
1918 :
1919 : # if DEBUG_LOGGING
1920 : FD_BASE58_ENCODE_32_BYTES( reasm_fec->key.key, key_b58 );
1921 : FD_BASE58_ENCODE_32_BYTES( reasm_fec->cmr.key, cmr_b58 );
1922 : 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, key_b58, cmr_b58 ));
1923 : # endif
1924 :
1925 0 : sched_fec->shred_cnt = reasm_fec->data_cnt;
1926 0 : sched_fec->is_last_in_batch = !!reasm_fec->data_complete;
1927 0 : sched_fec->is_last_in_block = !!reasm_fec->slot_complete;
1928 0 : sched_fec->bank_idx = reasm_fec->bank_idx;
1929 0 : sched_fec->parent_bank_idx = reasm_fec->parent_bank_idx;
1930 0 : sched_fec->slot = reasm_fec->slot;
1931 0 : sched_fec->parent_slot = reasm_fec->slot - reasm_fec->parent_off;
1932 0 : sched_fec->is_first_in_block = reasm_fec->fec_set_idx==0U;
1933 0 : fd_funk_txn_xid_t const root = fd_accdb_root_get( ctx->accdb_admin );
1934 0 : fd_funk_txn_xid_copy( sched_fec->alut_ctx->xid, &root );
1935 0 : sched_fec->alut_ctx->accdb[0] = ctx->accdb[0];
1936 0 : sched_fec->alut_ctx->els = ctx->published_root_slot;
1937 :
1938 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, sched_fec->bank_idx );
1939 0 : FD_TEST( bank );
1940 0 : if( sched_fec->is_first_in_block ) {
1941 0 : bank->refcnt++;
1942 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for sched", bank->idx, sched_fec->slot, bank->refcnt ));
1943 0 : }
1944 :
1945 : /* Read FEC set from the store. This should happen before we try to
1946 : ingest the FEC set. This allows us to filter out frags that were
1947 : in-flight when we published away minority forks that the frags land
1948 : on. These frags would have no bank to execute against, because
1949 : their corresponding banks, or parent banks, have also been pruned
1950 : during publishing. A query against store will rightfully tell us
1951 : that the underlying data is not found, implying that this is for a
1952 : minority fork that we can safeljy ignore. */
1953 :
1954 0 : ulong wait = (ulong)fd_log_wallclock();
1955 0 : ulong work = wait;
1956 0 : FD_STORE_SLOCK_BEGIN( ctx->store ) {
1957 0 : ctx->metrics.store_query_acquire++;
1958 0 : work = (ulong)fd_log_wallclock();
1959 0 : fd_histf_sample( ctx->metrics.store_query_wait, work - wait );
1960 :
1961 0 : fd_store_fec_t * store_fec = fd_store_query( ctx->store, &reasm_fec->key );
1962 0 : ctx->metrics.store_query_cnt++;
1963 0 : if( FD_UNLIKELY( !store_fec ) ) {
1964 :
1965 : /* The only case in which a FEC is not found in the store after
1966 : repair has notified is if the FEC was on a minority fork that
1967 : has already been published away. In this case we abandon the
1968 : entire slice because it is no longer relevant. */
1969 :
1970 0 : ctx->metrics.store_query_missing_cnt++;
1971 0 : ctx->metrics.store_query_missing_mr = reasm_fec->key.ul[0];
1972 0 : FD_BASE58_ENCODE_32_BYTES( reasm_fec->key.key, key_b58 );
1973 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, key_b58 ));
1974 0 : return 0;
1975 0 : }
1976 0 : sched_fec->fec = store_fec;
1977 0 : if( FD_UNLIKELY( !fd_sched_fec_ingest( ctx->sched, sched_fec ) ) ) { /* FIXME this critical section is unnecessarily complex. should refactor to just be held for the memcpy and block_offs. */
1978 0 : mark_bank_dead( ctx, stem, sched_fec->bank_idx );
1979 0 : return 1;
1980 0 : }
1981 0 : } FD_STORE_SLOCK_END;
1982 :
1983 0 : ctx->metrics.store_query_release++;
1984 0 : fd_histf_sample( ctx->metrics.store_query_work, (ulong)fd_log_wallclock() - work );
1985 0 : return 0;
1986 0 : }
1987 :
1988 : static void
1989 : process_fec_set( fd_replay_tile_t * ctx,
1990 : fd_stem_context_t * stem,
1991 0 : fd_reasm_fec_t * reasm_fec ) {
1992 :
1993 0 : fd_reasm_fec_t * parent = fd_reasm_parent( ctx->reasm, reasm_fec );
1994 0 : if( FD_UNLIKELY( !parent ) ) {
1995 0 : FD_LOG_WARNING(( "dropping FEC set (slot=%lu, fec_set_idx=%u) because it is unconnected in reasm", reasm_fec->slot, reasm_fec->fec_set_idx ));
1996 0 : return;
1997 0 : }
1998 :
1999 0 : if( FD_UNLIKELY( parent->bank_dead ) ) {
2000 : /* Inherit the dead flag from the parent. If a dead slot is
2001 : completed, we publish the slot as dead. Don't insert FECs for
2002 : dead slots. */
2003 0 : reasm_fec->bank_dead = 1;
2004 0 : if( FD_UNLIKELY( reasm_fec->slot_complete ) ) publish_slot_dead( ctx, stem, reasm_fec->slot, &reasm_fec->key );
2005 0 : FD_LOG_DEBUG(( "dropping FEC set (slot=%lu, fec_set_idx=%u) because parent bank is marked dead", reasm_fec->slot, reasm_fec->fec_set_idx ));
2006 0 : return;
2007 0 : }
2008 :
2009 : /* Standard case, the parent FEC has a valid corresponding bank. */
2010 0 : fd_bank_t * parent_fec_bank = fd_banks_bank_query( ctx->banks, parent->bank_idx );
2011 0 : if( FD_LIKELY( parent_fec_bank &&
2012 0 : parent_fec_bank->bank_seq==parent->bank_seq ) ) {
2013 0 : insert_fec_set( ctx, stem, reasm_fec );
2014 0 : return;
2015 0 : }
2016 :
2017 : /* In the case the FEC doesn't directly connect, iterate up the reasm
2018 : tree to find the closest valid slot complete that corresponds to a
2019 : valid bank. */
2020 :
2021 : /* First keep track of all of the slot completes up to and including
2022 : the fec we want to insert off of. */
2023 0 : fd_reasm_fec_t * path[ FD_BANKS_MAX_BANKS ];
2024 0 : ulong path_cnt = 0UL;
2025 0 : path[ path_cnt++ ] = reasm_fec;
2026 :
2027 0 : for( fd_reasm_fec_t * curr = reasm_fec;; ) {
2028 0 : curr = fd_reasm_parent( ctx->reasm, curr );
2029 0 : FD_TEST( curr );
2030 0 : if( FD_LIKELY( !curr->slot_complete ) ) continue;
2031 :
2032 0 : fd_bank_t * curr_bank = fd_banks_bank_query( ctx->banks, curr->bank_idx );
2033 0 : if( FD_LIKELY( curr_bank && curr_bank->bank_seq==curr->bank_seq ) ) break;
2034 :
2035 0 : FD_TEST( path_cnt<FD_BANKS_MAX_BANKS );
2036 0 : path[ path_cnt++ ] = curr;
2037 0 : }
2038 :
2039 0 : for( ulong i=path_cnt; i>0UL; i-- ) {
2040 0 : fd_reasm_fec_t * leaf = path[ i-1 ];
2041 :
2042 : /* If there's not capacity in the sched or banks, return early and
2043 : drop the FEC. We have inserted as much as we can for now. */
2044 0 : if( FD_UNLIKELY( fd_sched_can_ingest_cnt( ctx->sched ) < (leaf->fec_set_idx/FD_FEC_SHRED_CNT + 1) ) ) return;
2045 0 : if( FD_UNLIKELY( fd_banks_is_full( ctx->banks ) ) ) return;
2046 :
2047 : /* Gather all FECs for this slot; */
2048 0 : fd_reasm_fec_t * slot_fecs[ FD_FEC_BLK_MAX ];
2049 0 : fd_reasm_fec_t * curr = leaf;
2050 0 : for(;;) {
2051 0 : slot_fecs[ curr->fec_set_idx/FD_FEC_SHRED_CNT ] = curr;
2052 0 : if( curr->fec_set_idx==0U ) break;
2053 0 : curr = fd_reasm_parent( ctx->reasm, curr );
2054 0 : FD_TEST( curr );
2055 0 : }
2056 0 : FD_LOG_NOTICE(( "backfilling FEC sets for slot %lu from fec_set_idx %u to fec_set_idx %u", leaf->slot, leaf->fec_set_idx, curr->fec_set_idx ));
2057 :
2058 0 : for( ulong j=0UL; j<=leaf->fec_set_idx/FD_FEC_SHRED_CNT; j++ ) {
2059 0 : if( FD_UNLIKELY( insert_fec_set( ctx, stem, slot_fecs[ j ] ) ) ) return;
2060 0 : }
2061 0 : }
2062 0 : }
2063 :
2064 : /* accdb_advance_root moves account records from the unrooted to the
2065 : rooted database. */
2066 :
2067 : static inline ulong
2068 0 : accdb_root_op_total( fd_replay_tile_t const * ctx ) {
2069 0 : return ctx->accdb_admin->base.root_cnt +
2070 0 : ctx->accdb_admin->base.reclaim_cnt;
2071 0 : }
2072 :
2073 : static void
2074 : accdb_advance_root( fd_replay_tile_t * ctx,
2075 : ulong slot,
2076 0 : ulong bank_idx ) {
2077 0 : fd_funk_txn_xid_t xid = { .ul[0] = slot, .ul[1] = bank_idx };
2078 0 : FD_LOG_DEBUG(( "advancing root to slot=%lu", slot ));
2079 :
2080 0 : long rooted_accounts = -(long)accdb_root_op_total( ctx );
2081 0 : long t0 = fd_tickcount();
2082 0 : fd_accdb_advance_root( ctx->accdb_admin, &xid );
2083 0 : rooted_accounts += (long)accdb_root_op_total( ctx );
2084 0 : long t1 = fd_tickcount();
2085 0 : long root_accounts_dt = t1 - t0;
2086 0 : fd_histf_sample( ctx->metrics.root_slot_dur, (ulong)root_accounts_dt );
2087 0 : fd_histf_sample( ctx->metrics.root_account_dur, (ulong)root_accounts_dt / (ulong)fd_long_max( rooted_accounts, 1L ) );
2088 :
2089 0 : fd_progcache_advance_root( ctx->progcache, &xid );
2090 0 : long t2 = fd_tickcount();
2091 0 : FD_MCNT_INC( REPLAY, PROGCACHE_TIME_SECONDS, (ulong)( t2-t1 ) );
2092 0 : }
2093 :
2094 : static int
2095 0 : advance_published_root( fd_replay_tile_t * ctx ) {
2096 :
2097 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 );
2098 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2099 0 : FD_BASE58_ENCODE_32_BYTES( ctx->consensus_root.key, consensus_root_b58 );
2100 0 : FD_LOG_CRIT(( "invariant violation: block id ele not found for consensus root %s", consensus_root_b58 ));
2101 0 : }
2102 0 : ulong target_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2103 :
2104 : /* If the identity vote has been seen on a bank that should be rooted,
2105 : then we are now ready to produce blocks. */
2106 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
2107 0 : fd_bank_t * root_bank = fd_banks_bank_query( ctx->banks, target_bank_idx );
2108 0 : if( FD_UNLIKELY( !root_bank ) ) FD_LOG_CRIT(( "invariant violation: root bank not found for bank index %lu", target_bank_idx ));
2109 0 : if( root_bank->f.identity_vote_idx==ctx->identity_idx ) ctx->identity_vote_rooted = 1;
2110 0 : }
2111 :
2112 0 : ulong advanceable_root_idx = ULONG_MAX;
2113 0 : if( FD_UNLIKELY( !fd_banks_advance_root_prepare( ctx->banks, target_bank_idx, &advanceable_root_idx ) ) ) {
2114 0 : ctx->metrics.storage_root_behind++;
2115 0 : return 0;
2116 0 : }
2117 :
2118 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, advanceable_root_idx );
2119 0 : FD_TEST( bank );
2120 :
2121 0 : if( FD_UNLIKELY( advanceable_root_idx >= ctx->block_id_len ) ) {
2122 0 : FD_LOG_CRIT(( "invariant violation: advanceable root ele out of bounds [0, %lu) index %lu", ctx->block_id_len, advanceable_root_idx ));
2123 0 : }
2124 0 : fd_block_id_ele_t * advanceable_root_ele = &ctx->block_id_arr[ advanceable_root_idx ];
2125 :
2126 0 : ulong advanceable_root_slot = bank->f.slot;
2127 0 : accdb_advance_root( ctx, advanceable_root_slot, bank->idx );
2128 :
2129 0 : fd_txncache_advance_root( ctx->txncache, bank->txncache_fork_id );
2130 0 : fd_sched_advance_root( ctx->sched, advanceable_root_idx );
2131 0 : fd_banks_advance_root( ctx->banks, advanceable_root_idx );
2132 :
2133 : /* Reasm also prunes from the store during its publish. */
2134 :
2135 0 : fd_reasm_publish( ctx->reasm, &advanceable_root_ele->latest_mr, ctx->store );
2136 :
2137 0 : ctx->published_root_slot = advanceable_root_slot;
2138 0 : ctx->published_root_bank_idx = advanceable_root_idx;
2139 :
2140 0 : return 1;
2141 0 : }
2142 :
2143 : static void
2144 : after_credit( fd_replay_tile_t * ctx,
2145 : fd_stem_context_t * stem,
2146 : int * opt_poll_in,
2147 0 : int * charge_busy ) {
2148 0 : if( FD_UNLIKELY( !ctx->is_booted || !ctx->wfs_complete ) ) return;
2149 :
2150 0 : if( FD_UNLIKELY( maybe_become_leader( ctx, stem ) ) ) {
2151 0 : *charge_busy = 1;
2152 0 : *opt_poll_in = 0;
2153 0 : return;
2154 0 : }
2155 :
2156 : /* If we are leader, we can only unbecome the leader iff we have
2157 : received the poh hash from the poh tile and block id from reasm.
2158 : We have to do an additional check against the slot of the leader
2159 : bank because we lazily remove entries from the block id arr. */
2160 0 : if( FD_UNLIKELY( ctx->is_leader &&
2161 0 : ctx->recv_poh &&
2162 0 : ctx->block_id_arr[ ctx->leader_bank->idx ].block_id_seen &&
2163 0 : ctx->block_id_arr[ ctx->leader_bank->idx ].slot==ctx->leader_bank->f.slot ) ) {
2164 :
2165 0 : fini_leader_bank( ctx, stem );
2166 0 : *charge_busy = 1;
2167 0 : *opt_poll_in = 0;
2168 0 : return;
2169 0 : }
2170 :
2171 0 : ulong bank_idx;
2172 0 : while( (bank_idx=fd_sched_pruned_block_next( ctx->sched ))!=ULONG_MAX ) {
2173 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
2174 0 : FD_TEST( bank );
2175 0 : bank->refcnt--;
2176 0 : FD_LOG_DEBUG(( "bank (idx=%lu) refcnt decremented to %lu for sched", bank->idx, bank->refcnt ));
2177 0 : }
2178 :
2179 : /* If the published_root is not caught up to the consensus root, then
2180 : we should try to advance the published root. */
2181 0 : if( FD_UNLIKELY( ctx->consensus_root_bank_idx!=ctx->published_root_bank_idx && advance_published_root( ctx ) ) ) {
2182 0 : *charge_busy = 1;
2183 0 : *opt_poll_in = 0;
2184 0 : return;
2185 0 : }
2186 :
2187 0 : fd_banks_prune_cancel_info_t cancel_info[ 1 ];
2188 0 : int pruned = fd_banks_prune_one_dead_bank( ctx->banks, cancel_info );
2189 0 : if( FD_UNLIKELY( pruned==2 ) ) {
2190 0 : fd_txncache_cancel_fork( ctx->txncache, cancel_info->txncache_fork_id );
2191 0 : fd_funk_txn_xid_t xid = { .ul = { cancel_info->slot, cancel_info->bank_idx } };
2192 0 : fd_accdb_cancel ( ctx->accdb_admin, &xid );
2193 0 : fd_progcache_cancel( ctx->progcache, &xid );
2194 0 : *charge_busy = 1;
2195 0 : *opt_poll_in = 0;
2196 0 : return;
2197 0 : }
2198 :
2199 : /* if reasm evicted is set, publish starting from reasm_evicted down
2200 : to the leaf node to repair so repair can re-request for it */
2201 :
2202 0 : if( FD_UNLIKELY( ctx->reasm_evicted ) ) {
2203 0 : fd_replay_fec_evicted_t evicted = (fd_replay_fec_evicted_t){ .mr = ctx->reasm_evicted->key, .slot = ctx->reasm_evicted->slot, .fec_set_idx = ctx->reasm_evicted->fec_set_idx, .bank_idx = ctx->reasm_evicted->bank_idx };
2204 0 : fd_memcpy( fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk ), &evicted, sizeof(fd_replay_fec_evicted_t) );
2205 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_REASM_EVICTED, ctx->replay_out->chunk, sizeof(fd_replay_fec_evicted_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
2206 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_replay_fec_evicted_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
2207 :
2208 : /* eviction policy only evicts chains of nodes until there is a
2209 : fork, so guaranteed that the evict path is always the left-child */
2210 0 : fd_reasm_pool_release( ctx->reasm, ctx->reasm_evicted );
2211 0 : ctx->reasm_evicted = fd_reasm_child( ctx->reasm, ctx->reasm_evicted ); /* indexes into pool, safe to use */
2212 :
2213 0 : *charge_busy = 1;
2214 0 : *opt_poll_in = 0;
2215 0 : return;
2216 0 : }
2217 :
2218 : /* Mark a frontier eviction victim bank as dead. As refcnts on said
2219 : banks are drained, they will be pruned away. */
2220 0 : if( FD_UNLIKELY( ctx->frontier_cnt ) ) {
2221 0 : *charge_busy = 1;
2222 0 : *opt_poll_in = 0;
2223 0 : bank_idx = ctx->frontier_indices[ --ctx->frontier_cnt ];
2224 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
2225 0 : FD_TEST( bank );
2226 0 : if( FD_UNLIKELY( ctx->is_leader && bank_idx==ctx->leader_bank->idx ) ) return;
2227 0 : mark_bank_dead( ctx, stem, bank->idx );
2228 0 : fd_sched_block_abandon( ctx->sched, bank->idx );
2229 :
2230 : /* evict it from reasm */
2231 :
2232 0 : fd_block_id_ele_t * block_id_ele = &ctx->block_id_arr[ bank->idx ];
2233 0 : fd_reasm_fec_t * fec = fd_reasm_query( ctx->reasm, &block_id_ele->latest_mr );
2234 0 : FD_TEST( fec );
2235 0 : fd_reasm_fec_t * evicted_head = fd_reasm_remove( ctx->reasm, fec, ctx->store );
2236 0 : if( FD_UNLIKELY( ctx->reasm_evicted ) ) {
2237 : /* already have a chain we are evicting. Prepend the new chain to the existing chain */
2238 0 : fec->child = fd_reasm_pool_idx( ctx->reasm, ctx->reasm_evicted );
2239 0 : }
2240 0 : ctx->reasm_evicted = evicted_head;
2241 0 : return;
2242 0 : }
2243 :
2244 : /* If the reassembler has a fec that is ready, we should process it
2245 : and pass it to the scheduler. */
2246 0 : int evict_banks = 0;
2247 0 : if( FD_LIKELY( can_process_fec( ctx, &evict_banks ) ) ) {
2248 0 : fd_reasm_fec_t * fec = fd_reasm_pop( ctx->reasm );
2249 0 : process_fec_set( ctx, stem, fec );
2250 0 : *charge_busy = 1;
2251 0 : *opt_poll_in = 0;
2252 0 : return;
2253 0 : }
2254 :
2255 0 : if( FD_UNLIKELY( evict_banks ) ) {
2256 0 : FD_LOG_WARNING(( "banks are full and partially executed frontier banks are being evicted" ));
2257 0 : fd_banks_get_frontier( ctx->banks, ctx->frontier_indices, &ctx->frontier_cnt );
2258 0 : *charge_busy = 1;
2259 0 : *opt_poll_in = 0;
2260 0 : return;
2261 0 : }
2262 :
2263 0 : *charge_busy = replay( ctx, stem );
2264 0 : *opt_poll_in = !*charge_busy;
2265 0 : }
2266 :
2267 : static int
2268 : before_frag( fd_replay_tile_t * ctx,
2269 : ulong in_idx,
2270 : ulong seq FD_PARAM_UNUSED,
2271 0 : ulong sig ) {
2272 :
2273 0 : if( FD_UNLIKELY( ctx->in_kind[ in_idx ]==IN_KIND_GOSSIP_OUT && sig!=FD_GOSSIP_UPDATE_TAG_WFS_DONE ) ) return 1;
2274 0 : return 0;
2275 0 : }
2276 :
2277 : static void
2278 : process_exec_task_done( fd_replay_tile_t * ctx,
2279 : fd_stem_context_t * stem,
2280 : fd_execrp_task_done_msg_t * msg,
2281 0 : ulong sig ) {
2282 :
2283 0 : ulong exec_tile_idx = sig&0xFFFFFFFFUL;
2284 :
2285 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, msg->bank_idx );
2286 0 : FD_TEST( bank );
2287 0 : bank->refcnt--;
2288 :
2289 0 : switch( sig>>32 ) {
2290 0 : case FD_EXECRP_TT_TXN_EXEC: {
2291 0 : ulong txn_idx = msg->txn_exec->txn_idx;
2292 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
2293 : /* Query the txn signature against our recently generated vote
2294 : txn signatures. If the query is successful, then we have
2295 : seen our own vote transaction land and this should be marked
2296 : in the bank. We go through this exercise until we've seen
2297 : our vote rooted. */
2298 0 : fd_txn_p_t * txn_p = fd_sched_get_txn( ctx->sched, txn_idx );
2299 :
2300 0 : fd_pubkey_t * identity_pubkey_out = NULL;
2301 0 : if( fd_vote_tracker_query_sig( ctx->vote_tracker, fd_type_pun_const( txn_p->payload+TXN( txn_p )->signature_off ), &identity_pubkey_out ) && fd_pubkey_eq( identity_pubkey_out, ctx->identity_pubkey ) ) {
2302 0 : bank->f.identity_vote_idx = ctx->identity_idx;
2303 0 : }
2304 0 : }
2305 0 : if( FD_UNLIKELY( !msg->txn_exec->is_committable && bank->state!=FD_BANK_STATE_DEAD) ) {
2306 : /* Every transaction in a valid block has to execute.
2307 : Otherwise, we should mark the block as dead. */
2308 0 : mark_bank_dead( ctx, stem, bank->idx );
2309 0 : fd_sched_block_abandon( ctx->sched, bank->idx );
2310 0 : }
2311 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_EXEC, txn_idx, exec_tile_idx, NULL );
2312 0 : FD_TEST( res==0 );
2313 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, txn_idx );
2314 0 : txn_info->flags |= FD_SCHED_TXN_EXEC_DONE;
2315 0 : if( FD_LIKELY( !(txn_info->flags&FD_SCHED_TXN_SIGVERIFY_DONE)||!txn_info->txn_err ) ) { /* Set execution status if sigverify hasn't happened yet or if sigverify was a success. */
2316 0 : txn_info->txn_err = msg->txn_exec->txn_err;
2317 0 : txn_info->flags |= fd_ulong_if( msg->txn_exec->is_committable, FD_SCHED_TXN_IS_COMMITTABLE, 0UL );
2318 0 : txn_info->flags |= fd_ulong_if( msg->txn_exec->is_fees_only, FD_SCHED_TXN_IS_FEES_ONLY, 0UL );
2319 0 : }
2320 0 : if( FD_UNLIKELY( (txn_info->flags&FD_SCHED_TXN_REPLAY_DONE)==FD_SCHED_TXN_REPLAY_DONE ) ) { /* UNLIKELY because generally exec happens before sigverify. */
2321 0 : publish_txn_executed( ctx, stem, txn_idx );
2322 0 : }
2323 0 : break;
2324 0 : }
2325 0 : case FD_EXECRP_TT_TXN_SIGVERIFY: {
2326 0 : ulong txn_idx = msg->txn_sigverify->txn_idx;
2327 0 : fd_sched_txn_info_t * txn_info = fd_sched_get_txn_info( ctx->sched, txn_idx );
2328 0 : txn_info->flags |= FD_SCHED_TXN_SIGVERIFY_DONE;
2329 0 : if( FD_UNLIKELY( msg->txn_sigverify->err ) ) {
2330 0 : txn_info->txn_err = FD_RUNTIME_TXN_ERR_SIGNATURE_FAILURE;
2331 0 : txn_info->flags &= ~FD_SCHED_TXN_IS_COMMITTABLE;
2332 0 : txn_info->flags &= ~FD_SCHED_TXN_IS_FEES_ONLY;
2333 0 : }
2334 0 : if( FD_UNLIKELY( msg->txn_sigverify->err && bank->state!=FD_BANK_STATE_DEAD ) ) {
2335 : /* Every transaction in a valid block has to sigverify.
2336 : Otherwise, we should mark the block as dead. Also freeze the
2337 : bank if possible. */
2338 0 : mark_bank_dead( ctx, stem, bank->idx );
2339 0 : fd_sched_block_abandon( ctx->sched, bank->idx );
2340 0 : }
2341 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_TXN_SIGVERIFY, txn_idx, exec_tile_idx, NULL );
2342 0 : FD_TEST( res==0 );
2343 0 : if( FD_LIKELY( (txn_info->flags&FD_SCHED_TXN_REPLAY_DONE)==FD_SCHED_TXN_REPLAY_DONE ) ) {
2344 0 : publish_txn_executed( ctx, stem, txn_idx );
2345 0 : }
2346 0 : break;
2347 0 : }
2348 0 : case FD_EXECRP_TT_POH_HASH: {
2349 0 : int res = fd_sched_task_done( ctx->sched, FD_SCHED_TT_POH_HASH, ULONG_MAX, exec_tile_idx, msg->poh_hash );
2350 0 : if( FD_UNLIKELY( res<0 && bank->state!=FD_BANK_STATE_DEAD ) ) {
2351 0 : mark_bank_dead( ctx, stem, bank->idx );
2352 0 : }
2353 0 : break;
2354 0 : }
2355 0 : default: FD_LOG_CRIT(( "unexpected sig 0x%lx", sig ));
2356 0 : }
2357 :
2358 : /* Reference counter just decreased, and an exec tile just got freed
2359 : up. If there's a need to be more aggressively pruning, we could
2360 : check here if more slots just became publishable and publish. Not
2361 : publishing here shouldn't bloat the fork tree too much though. We
2362 : mark minority forks dead as soon as we can, and execution dispatch
2363 : stops on dead blocks. So shortly afterwards, dead blocks should be
2364 : eligible for pruning as in-flight transactions retire from the
2365 : execution pipeline. */
2366 :
2367 0 : }
2368 :
2369 : static void
2370 : process_tower_slot_done( fd_replay_tile_t * ctx,
2371 : fd_stem_context_t * stem,
2372 : fd_tower_slot_done_t const * msg,
2373 0 : ulong seq ) {
2374 0 : fd_bank_t * replay_bank = fd_banks_bank_query( ctx->banks, msg->replay_bank_idx );
2375 0 : if( FD_UNLIKELY( !replay_bank ) ) FD_LOG_CRIT(( "invariant violation: bank not found for bank index %lu", msg->replay_bank_idx ));
2376 0 : replay_bank->refcnt--;
2377 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt decremented to %lu for tower", replay_bank->idx, msg->replay_slot, replay_bank->refcnt ));
2378 :
2379 0 : ctx->reset_block_id = msg->reset_block_id;
2380 0 : ctx->reset_slot = msg->reset_slot;
2381 0 : ctx->reset_timestamp_nanos = fd_log_wallclock();
2382 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 ) );
2383 0 : ctx->next_leader_slot = fd_multi_epoch_leaders_get_next_slot( ctx->mleaders, min_leader_slot, ctx->identity_pubkey );
2384 0 : if( FD_LIKELY( ctx->next_leader_slot != ULONG_MAX ) ) {
2385 0 : ctx->next_leader_tickcount = (long)((double)(ctx->next_leader_slot-ctx->reset_slot-1UL)*ctx->slot_duration_ticks) + fd_tickcount();
2386 0 : } else {
2387 0 : ctx->next_leader_tickcount = LONG_MAX;
2388 0 : }
2389 :
2390 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 );
2391 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2392 0 : FD_BASE58_ENCODE_32_BYTES( msg->reset_block_id.key, reset_block_id_b58 );
2393 0 : FD_LOG_CRIT(( "invariant violation: block id ele doesn't exist for reset block id: %s, slot: %lu", reset_block_id_b58, msg->reset_slot ));
2394 0 : }
2395 0 : ulong reset_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2396 :
2397 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, reset_bank_idx );
2398 0 : if( FD_UNLIKELY( !bank ) ) {
2399 0 : FD_LOG_CRIT(( "invariant violation: bank not found for bank index %lu", reset_bank_idx ));
2400 0 : }
2401 :
2402 0 : if( FD_LIKELY( msg->root_slot!=ULONG_MAX ) ) FD_TEST( msg->root_slot<=msg->reset_slot );
2403 0 : ctx->reset_bank = bank;
2404 :
2405 0 : if( FD_LIKELY( ctx->replay_out->idx!=ULONG_MAX ) ) {
2406 0 : fd_poh_reset_t * reset = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2407 :
2408 0 : reset->bank_idx = bank->idx;
2409 0 : reset->timestamp = ctx->reset_timestamp_nanos;
2410 0 : reset->completed_slot = ctx->reset_slot;
2411 0 : reset->hashcnt_per_tick = bank->f.hashes_per_tick;
2412 0 : reset->ticks_per_slot = bank->f.ticks_per_slot;
2413 0 : reset->tick_duration_ns = (ulong)(ctx->slot_duration_nanos/(double)reset->ticks_per_slot);
2414 :
2415 0 : fd_memcpy( reset->completed_block_id, &block_id_ele->latest_mr, sizeof(fd_hash_t) );
2416 :
2417 0 : fd_blockhashes_t const * block_hash_queue = &bank->f.block_hash_queue;
2418 0 : fd_hash_t const * last_hash = fd_blockhashes_peek_last_hash( block_hash_queue );
2419 0 : FD_TEST( last_hash );
2420 0 : fd_memcpy( reset->completed_blockhash, last_hash->uc, sizeof(fd_hash_t) );
2421 :
2422 0 : ulong ticks_per_slot = bank->f.ticks_per_slot;
2423 0 : if( FD_UNLIKELY( reset->hashcnt_per_tick==1UL ) ) {
2424 : /* Low power producer, maximum of one microblock per tick in the slot */
2425 0 : reset->max_microblocks_in_slot = ticks_per_slot;
2426 0 : } else {
2427 : /* See the long comment in after_credit for this limit */
2428 0 : reset->max_microblocks_in_slot = fd_ulong_min( MAX_MICROBLOCKS_PER_SLOT, ticks_per_slot*(reset->hashcnt_per_tick-1UL) );
2429 0 : }
2430 0 : reset->next_leader_slot = ctx->next_leader_slot;
2431 :
2432 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() ) );
2433 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 );
2434 0 : }
2435 :
2436 0 : FD_LOG_INFO(( "tower_slot_done(reset_slot=%lu, next_leader_slot=%lu, vote_slot=%lu, replay_slot=%lu, root_slot=%lu, seqno=%lu)", msg->reset_slot, ctx->next_leader_slot, msg->vote_slot, msg->replay_slot, msg->root_slot, seq ));
2437 0 : maybe_become_leader( ctx, stem );
2438 :
2439 0 : if( FD_LIKELY( msg->root_slot!=ULONG_MAX ) ) {
2440 :
2441 0 : FD_TEST( msg->root_slot>=ctx->consensus_root_slot );
2442 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 );
2443 0 : FD_TEST( block_id_ele );
2444 0 : ctx->consensus_root_slot = msg->root_slot;
2445 0 : ctx->consensus_root = msg->root_block_id;
2446 0 : ctx->consensus_root_bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2447 :
2448 0 : publish_root_advanced( ctx, stem );
2449 :
2450 0 : fd_sched_root_notify( ctx->sched, ctx->consensus_root_bank_idx );
2451 0 : }
2452 :
2453 0 : ulong distance = 0UL;
2454 0 : fd_bank_t * parent = bank;
2455 0 : while( parent ) {
2456 0 : if( FD_UNLIKELY( parent->idx==ctx->consensus_root_bank_idx ) ) break;
2457 0 : parent = fd_banks_get_parent( ctx->banks, parent );
2458 0 : distance++;
2459 0 : }
2460 :
2461 0 : FD_MGAUGE_SET( REPLAY, ROOT_DISTANCE, distance );
2462 0 : }
2463 :
2464 : static void
2465 : process_fec_complete( fd_replay_tile_t * ctx,
2466 : fd_stem_context_t * stem,
2467 0 : uchar const * shred_buf ) {
2468 0 : fd_shred_t const * shred = (fd_shred_t const *)fd_type_pun_const( shred_buf );
2469 :
2470 0 : fd_hash_t const * merkle_root = (fd_hash_t const *)fd_type_pun_const( shred_buf + FD_SHRED_DATA_HEADER_SZ );
2471 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) );
2472 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) );
2473 :
2474 0 : int data_complete = !!( shred->data.flags & FD_SHRED_DATA_FLAG_DATA_COMPLETE );
2475 0 : int slot_complete = !!( shred->data.flags & FD_SHRED_DATA_FLAG_SLOT_COMPLETE );
2476 :
2477 0 : if( FD_UNLIKELY( shred->slot - shred->data.parent_off == fd_reasm_slot0( ctx->reasm ) && shred->fec_set_idx == 0) ) {
2478 0 : chained_merkle_root = &fd_reasm_root( ctx->reasm )->key;
2479 0 : }
2480 :
2481 0 : if( FD_UNLIKELY( fd_reasm_query( ctx->reasm, merkle_root ) ) ) return;
2482 0 : fd_reasm_fec_t * fec = 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, ctx->store, &ctx->reasm_evicted );
2483 :
2484 0 : if( FD_UNLIKELY( !fec ) ) {
2485 : /* reasm failed to insert. We don't want to just put this back on
2486 : the returnable_frag queue because it's unclear whether this FEC
2487 : is truly something we want to process. Therefore our best option
2488 : is to punt it and "go around." reasm_insert populates it's last
2489 : pool element with the data of the failed insert, so we make sure
2490 : to publish the failed insert data to repair in after_credit. */
2491 0 : return;
2492 0 : }
2493 :
2494 0 : if( FD_UNLIKELY( ctx->reasm_evicted && ctx->reasm_evicted->bank_idx != ULONG_MAX ) ) {
2495 0 : mark_bank_dead( ctx, stem, ctx->reasm_evicted->bank_idx );
2496 0 : fd_sched_block_abandon( ctx->sched, ctx->reasm_evicted->bank_idx );
2497 0 : }
2498 0 : }
2499 :
2500 : static void
2501 0 : process_resolv_slot_completed( fd_replay_tile_t * ctx, ulong bank_idx ) {
2502 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
2503 0 : FD_TEST( bank );
2504 0 : bank->refcnt--;
2505 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt decremented to %lu for resolv", bank->idx, bank->f.slot, bank->refcnt ));
2506 0 : }
2507 :
2508 : static void
2509 : process_vote_txn_sent( fd_replay_tile_t * ctx,
2510 0 : fd_txn_m_t * txnm ) {
2511 : /* The send tile has signed and sent a vote. Add this vote to the
2512 : vote tracker. We go through this exercise until the client has
2513 : seen a vote corresponding to the current identity rooted. */
2514 0 : if( FD_UNLIKELY( !ctx->identity_vote_rooted ) ) {
2515 0 : uchar * payload = (uchar *)txnm + sizeof(fd_txn_m_t);
2516 0 : uchar txn_mem[ FD_TXN_MAX_SZ ] __attribute__((aligned(alignof(fd_txn_t))));
2517 0 : fd_txn_t * txn = (fd_txn_t *)txn_mem;
2518 0 : if( FD_UNLIKELY( !fd_txn_parse( payload, txnm->payload_sz, txn_mem, NULL ) ) ) {
2519 0 : FD_LOG_CRIT(( "Could not parse txn from send tile" ));
2520 0 : }
2521 : /* The identity of the validator that the signed the vote will
2522 : always be the first signer in the vote transaction. */
2523 0 : fd_pubkey_t * vote_identity = fd_type_pun( payload+txn->acct_addr_off );
2524 0 : fd_vote_tracker_insert( ctx->vote_tracker, vote_identity, fd_type_pun_const( payload+txn->signature_off ) );
2525 0 : }
2526 0 : }
2527 :
2528 : static inline void
2529 0 : maybe_verify_shred_version( fd_replay_tile_t * ctx ) {
2530 0 : if( FD_LIKELY( ctx->expected_shred_version && ctx->ipecho_shred_version ) ) {
2531 0 : if( FD_UNLIKELY( ctx->expected_shred_version!=ctx->ipecho_shred_version ) ) {
2532 0 : FD_LOG_ERR(( "shred version mismatch: expected %u but got %u from ipecho", ctx->expected_shred_version, ctx->ipecho_shred_version ) );
2533 0 : }
2534 0 : }
2535 :
2536 0 : if( FD_LIKELY( ctx->has_genesis_hash && ctx->hard_forks_cnt!=ULONG_MAX && (ctx->expected_shred_version || ctx->ipecho_shred_version) ) ) {
2537 0 : ushort expected_shred_version = ctx->expected_shred_version ? ctx->expected_shred_version : ctx->ipecho_shred_version;
2538 :
2539 0 : ushort actual_shred_version = compute_shred_version( ctx->genesis_hash->uc, ctx->hard_forks, ctx->hard_forks_cnts, ctx->hard_forks_cnt );
2540 :
2541 0 : if( FD_UNLIKELY( expected_shred_version!=actual_shred_version ) ) {
2542 0 : FD_BASE58_ENCODE_32_BYTES( ctx->genesis_hash->uc, genesis_hash_b58 );
2543 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` combined with the hard_forks from the loaded snapshot have produced "
2544 0 : "a shred version of %hu but the entrypoint you connected to on boot reported a shred version of %hu. "
2545 0 : "This likely means that the genesis.bin file you have is for a different cluster than the one you "
2546 0 : "are trying to connect to, you can delete it and restart the node to download the correct genesis "
2547 0 : "file automatically.", ctx->genesis_path, actual_shred_version, expected_shred_version ));
2548 0 : }
2549 0 : }
2550 0 : }
2551 :
2552 : static inline void
2553 0 : maybe_verify_genesis_timestamp( fd_replay_tile_t * ctx ) {
2554 0 : if( FD_LIKELY( !ctx->has_expected_genesis_timestamp || !ctx->has_genesis_timestamp ) ) return;
2555 0 : if( FD_LIKELY( ctx->genesis_timestamp==ctx->expected_genesis_timestamp ) ) return;
2556 :
2557 0 : FD_LOG_ERR(( "Your genesis.bin file at `%s` has a genesis timestamp of %lu but the snapshot you loaded has a genesis "
2558 0 : "timestamp of %lu. This either means that the genesis.bin file you have is for a different cluster than "
2559 0 : "the one you are trying to connect to, or you have loaded a snapshot for the wrong cluster. In either "
2560 0 : "case, you can delete the problematic file and restart the node to download the correct one automatically.",
2561 0 : ctx->genesis_path, ctx->genesis_timestamp, ctx->expected_genesis_timestamp ));
2562 0 : }
2563 :
2564 : static void
2565 : process_tower_optimistic_confirmed( fd_replay_tile_t * ctx,
2566 : fd_stem_context_t * stem,
2567 0 : fd_tower_slot_confirmed_t const * msg ) {
2568 :
2569 0 : fd_block_id_ele_t * block_id_ele = fd_block_id_map_ele_query( ctx->block_id_map, &msg->block_id, NULL, ctx->block_id_arr );
2570 0 : if( FD_UNLIKELY( !block_id_ele ) ) {
2571 0 : FD_BASE58_ENCODE_32_BYTES( msg->block_id.key, block_id_b58 );
2572 0 : FD_LOG_WARNING(( "missing bank for confirmed block_id: %s level %d", block_id_b58, msg->level ));
2573 0 : return;
2574 0 : }
2575 :
2576 0 : ulong bank_idx = fd_block_id_ele_get_idx( ctx->block_id_arr, block_id_ele );
2577 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, bank_idx );
2578 :
2579 :
2580 0 : if( FD_UNLIKELY( !bank ) ) {
2581 0 : FD_BASE58_ENCODE_32_BYTES( msg->block_id.key, block_id_cstr );
2582 0 : FD_LOG_WARNING(( "failed to query optimistically confirmed bank for block id %s", block_id_cstr ));
2583 0 : return;
2584 0 : }
2585 :
2586 0 : if( ctx->rpc_enabled ) {
2587 0 : bank->refcnt++;
2588 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt incremented to %lu for rpc", bank->idx, bank->f.slot, bank->refcnt ));
2589 0 : }
2590 :
2591 0 : fd_replay_oc_advanced_t * replay_msg = fd_chunk_to_laddr( ctx->replay_out->mem, ctx->replay_out->chunk );
2592 0 : replay_msg->bank_idx = bank_idx;
2593 0 : replay_msg->slot = msg->slot;
2594 :
2595 0 : fd_stem_publish( stem, ctx->replay_out->idx, REPLAY_SIG_OC_ADVANCED, ctx->replay_out->chunk, sizeof(fd_replay_oc_advanced_t), 0UL, 0UL, fd_frag_meta_ts_comp( fd_tickcount() ) );
2596 0 : ctx->replay_out->chunk = fd_dcache_compact_next( ctx->replay_out->chunk, sizeof(fd_replay_oc_advanced_t), ctx->replay_out->chunk0, ctx->replay_out->wmark );
2597 0 : }
2598 :
2599 : static inline int
2600 : returnable_frag( fd_replay_tile_t * ctx,
2601 : ulong in_idx,
2602 : ulong seq,
2603 : ulong sig,
2604 : ulong chunk,
2605 : ulong sz,
2606 : ulong ctl,
2607 : ulong tsorig,
2608 : ulong tspub,
2609 0 : fd_stem_context_t * stem ) {
2610 0 : (void)seq;
2611 0 : (void)ctl;
2612 0 : (void)tsorig;
2613 0 : (void)tspub;
2614 :
2615 0 : if( FD_UNLIKELY( sz!=0UL && (chunk<ctx->in[ in_idx ].chunk0 || chunk>ctx->in[ in_idx ].wmark || sz>ctx->in[ in_idx ].mtu ) ) )
2616 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 ));
2617 :
2618 0 : switch( ctx->in_kind[in_idx] ) {
2619 0 : case IN_KIND_GENESIS: {
2620 0 : fd_genesis_meta_t const * meta = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2621 0 : ctx->has_genesis_hash = 1;
2622 0 : ctx->has_genesis_timestamp = 1;
2623 0 : ctx->genesis_timestamp = meta->creation_time_seconds;
2624 0 : *ctx->genesis_hash = meta->genesis_hash;
2625 0 : if( FD_LIKELY( meta->bootstrap ) ) {
2626 0 : boot_genesis( ctx, stem, meta );
2627 0 : } else {
2628 0 : uchar const * genesis_blob = (uchar const *)( meta+1 );
2629 0 : FD_TEST( fd_genesis_parse( ctx->genesis, genesis_blob, meta->blob_sz ) );
2630 0 : }
2631 0 : ctx->has_genesis_timestamp = 1;
2632 0 : ctx->genesis_timestamp = ctx->genesis->creation_time;
2633 :
2634 0 : maybe_verify_cluster_type( ctx );
2635 0 : maybe_verify_shred_version( ctx );
2636 0 : maybe_verify_genesis_timestamp( ctx );
2637 0 : break;
2638 0 : }
2639 0 : case IN_KIND_IPECHO: {
2640 0 : FD_TEST( sig && sig<=USHORT_MAX );
2641 0 : ctx->ipecho_shred_version = (ushort)sig;
2642 0 : maybe_verify_shred_version( ctx );
2643 0 : break;
2644 0 : }
2645 0 : case IN_KIND_SNAP: {
2646 0 : on_snapshot_message( ctx, stem, in_idx, chunk, sig );
2647 0 : maybe_verify_shred_version( ctx );
2648 0 : maybe_verify_genesis_timestamp( ctx );
2649 0 : break;
2650 0 : }
2651 0 : case IN_KIND_EXECRP: {
2652 0 : process_exec_task_done( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ), sig );
2653 0 : break;
2654 0 : }
2655 0 : case IN_KIND_POH: {
2656 0 : process_poh_message( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2657 0 : break;
2658 0 : }
2659 0 : case IN_KIND_RESOLV: {
2660 0 : fd_resolv_slot_exchanged_t * exchanged_slot = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2661 0 : process_resolv_slot_completed( ctx, exchanged_slot->bank_idx );
2662 0 : break;
2663 0 : }
2664 0 : case IN_KIND_TOWER: {
2665 0 : if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_DONE ) ) {
2666 0 : process_tower_slot_done( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ), seq );
2667 0 : } else if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_CONFIRMED ) ) {
2668 0 : fd_tower_slot_confirmed_t const * msg = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2669 0 : if( msg->level==FD_TOWER_SLOT_CONFIRMED_OPTIMISTIC && !msg->fwd ) process_tower_optimistic_confirmed( ctx, stem, msg );
2670 0 : if( msg->level==FD_TOWER_SLOT_CONFIRMED_DUPLICATE ) fd_reasm_confirm( ctx->reasm, &msg->block_id );
2671 0 : } else if( FD_LIKELY( sig==FD_TOWER_SIG_SLOT_IGNORED ) ) {
2672 0 : fd_tower_slot_ignored_t const * msg = fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk );
2673 0 : fd_tower_slot_done_t ignored = {
2674 0 : .replay_slot = msg->slot,
2675 0 : .replay_bank_idx = msg->bank_idx,
2676 0 : .vote_slot = ULONG_MAX,
2677 0 : .reset_slot = ctx->reset_slot, /* Use most recent reset slot */
2678 0 : .reset_block_id = ctx->reset_block_id,
2679 0 : .root_slot = ULONG_MAX
2680 0 : };
2681 0 : process_tower_slot_done( ctx, stem, &ignored, seq );
2682 0 : }
2683 0 : break;
2684 0 : }
2685 0 : case IN_KIND_REPAIR: {
2686 : /* TODO: This message/sz should be defined. */
2687 0 : if( sz!=0 && fd_disco_shred_out_msg_type( sig )==FD_SHRED_OUT_MSG_TYPE_FEC ) {
2688 : /* If receive a FEC complete message. */
2689 0 : process_fec_complete( ctx, stem, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2690 0 : }
2691 0 : break;
2692 0 : }
2693 0 : case IN_KIND_TXSEND: {
2694 0 : process_vote_txn_sent( ctx, fd_chunk_to_laddr( ctx->in[ in_idx ].mem, chunk ) );
2695 0 : break;
2696 0 : }
2697 0 : case IN_KIND_GOSSIP_OUT: {
2698 0 : FD_TEST( sig==FD_GOSSIP_UPDATE_TAG_WFS_DONE );
2699 0 : ctx->wfs_complete = 1;
2700 0 : FD_LOG_NOTICE(( "Done waiting for supermajority. More than 80 percent of cluster stake has joined." ));
2701 0 : break;
2702 0 : }
2703 0 : case IN_KIND_RPC: {
2704 0 : fd_bank_t * bank = fd_banks_bank_query( ctx->banks, sig );
2705 0 : FD_TEST( bank );
2706 0 : bank->refcnt--;
2707 0 : FD_LOG_DEBUG(( "bank (idx=%lu, slot=%lu) refcnt decremented to %lu for %s", bank->idx, bank->f.slot, bank->refcnt, ctx->in_kind[ in_idx ]==IN_KIND_RPC ? "rpc" : "gui" ));
2708 0 : break;
2709 0 : }
2710 0 : default:
2711 0 : FD_LOG_ERR(( "unhandled kind %d", ctx->in_kind[ in_idx ] ));
2712 0 : }
2713 :
2714 0 : return 0;
2715 0 : }
2716 :
2717 : static inline fd_replay_out_link_t
2718 : out1( fd_topo_t const * topo,
2719 : fd_topo_tile_t const * tile,
2720 0 : char const * name ) {
2721 0 : ulong idx = ULONG_MAX;
2722 :
2723 0 : for( ulong i=0UL; i<tile->out_cnt; i++ ) {
2724 0 : fd_topo_link_t const * link = &topo->links[ tile->out_link_id[ i ] ];
2725 0 : if( !strcmp( link->name, name ) ) {
2726 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 ));
2727 0 : idx = i;
2728 0 : }
2729 0 : }
2730 :
2731 0 : if( FD_UNLIKELY( idx==ULONG_MAX ) ) return (fd_replay_out_link_t){ .idx = ULONG_MAX, .mem = NULL, .chunk0 = 0, .wmark = 0, .chunk = 0 };
2732 :
2733 0 : void * mem = topo->workspaces[ topo->objs[ topo->links[ tile->out_link_id[ idx ] ].dcache_obj_id ].wksp_id ].wksp;
2734 0 : ulong chunk0 = fd_dcache_compact_chunk0( mem, topo->links[ tile->out_link_id[ idx ] ].dcache );
2735 0 : ulong wmark = fd_dcache_compact_wmark ( mem, topo->links[ tile->out_link_id[ idx ] ].dcache, topo->links[ tile->out_link_id[ idx ] ].mtu );
2736 :
2737 0 : return (fd_replay_out_link_t){ .idx = idx, .mem = mem, .chunk0 = chunk0, .wmark = wmark, .chunk = chunk0 };
2738 0 : }
2739 :
2740 : static void
2741 : privileged_init( fd_topo_t * topo,
2742 0 : fd_topo_tile_t * tile ) {
2743 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2744 :
2745 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2746 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
2747 :
2748 0 : if( FD_UNLIKELY( !strcmp( tile->replay.identity_key_path, "" ) ) ) FD_LOG_ERR(( "identity_key_path not set" ));
2749 :
2750 0 : ctx->identity_pubkey[ 0 ] = *(fd_pubkey_t const *)fd_type_pun_const( fd_keyload_load( tile->replay.identity_key_path, /* pubkey only: */ 1 ) );
2751 0 : ctx->identity_idx = 0UL;
2752 :
2753 0 : if( FD_UNLIKELY( !tile->replay.bundle.vote_account_path[0] ) ) {
2754 0 : tile->replay.bundle.enabled = 0;
2755 0 : }
2756 :
2757 0 : if( FD_UNLIKELY( tile->replay.bundle.enabled ) ) {
2758 0 : if( FD_UNLIKELY( !fd_base58_decode_32( tile->replay.bundle.vote_account_path, ctx->bundle.vote_account.uc ) ) ) {
2759 0 : const uchar * vote_key = fd_keyload_load( tile->replay.bundle.vote_account_path, /* pubkey only: */ 1 );
2760 0 : fd_memcpy( ctx->bundle.vote_account.uc, vote_key, 32UL );
2761 0 : }
2762 0 : }
2763 :
2764 0 : FD_TEST( fd_rng_secure( &ctx->rng_seed, sizeof(ctx->rng_seed) ) );
2765 :
2766 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->reasm_seed, sizeof(ulong) ) ) ) {
2767 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2768 0 : }
2769 :
2770 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->vote_tracker_seed, sizeof(ulong) ) ) ) {
2771 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2772 0 : }
2773 :
2774 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->block_id_map_seed, sizeof(ulong) ) ) ) {
2775 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2776 0 : }
2777 :
2778 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->initial_block_id, sizeof(fd_hash_t) ) ) ) {
2779 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2780 0 : }
2781 :
2782 0 : if( FD_UNLIKELY( !fd_rng_secure( &ctx->runtime_stack_seed, sizeof(ulong) ) ) ) {
2783 0 : FD_LOG_CRIT(( "fd_rng_secure failed" ));
2784 0 : }
2785 0 : }
2786 :
2787 : static void
2788 : unprivileged_init( fd_topo_t * topo,
2789 0 : fd_topo_tile_t * tile ) {
2790 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
2791 :
2792 0 : ulong chain_cnt = fd_block_id_map_chain_cnt_est( tile->replay.max_live_slots );
2793 :
2794 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
2795 0 : fd_replay_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_replay_tile_t), sizeof(fd_replay_tile_t) );
2796 0 : void * runtime_stack_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_runtime_stack_align(), fd_runtime_stack_footprint( FD_RUNTIME_MAX_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_STAKE_ACCOUNTS ) );
2797 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 );
2798 0 : void * block_id_map_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_block_id_map_align(), fd_block_id_map_footprint( chain_cnt ) );
2799 0 : void * _txncache = FD_SCRATCH_ALLOC_APPEND( l, fd_txncache_align(), fd_txncache_footprint( tile->replay.max_live_slots ) );
2800 0 : void * reasm_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_reasm_align(), fd_reasm_footprint( tile->replay.fec_max ) );
2801 0 : void * sched_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_sched_align(), fd_sched_footprint( tile->replay.sched_depth, tile->replay.max_live_slots ) );
2802 0 : void * vinyl_req_pool_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_vinyl_req_pool_align(), fd_vinyl_req_pool_footprint( 1UL, 1UL ) );
2803 0 : void * vote_tracker_mem = FD_SCRATCH_ALLOC_APPEND( l, fd_vote_tracker_align(), fd_vote_tracker_footprint() );
2804 0 : void * _capture_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_capture_ctx_align(), fd_capture_ctx_footprint() );
2805 0 : void * dump_proto_ctx_mem = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_dump_proto_ctx_t), sizeof(fd_dump_proto_ctx_t) );
2806 0 : # if FD_HAS_FLATCC
2807 0 : void * block_dump_ctx = NULL;
2808 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
2809 0 : block_dump_ctx = FD_SCRATCH_ALLOC_APPEND( l, fd_block_dump_context_align(), fd_block_dump_context_footprint() );
2810 0 : }
2811 0 : # endif
2812 :
2813 0 : ctx->runtime_stack = fd_runtime_stack_join( fd_runtime_stack_new( runtime_stack_mem, FD_RUNTIME_MAX_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_VOTE_ACCOUNTS, FD_RUNTIME_EXPECTED_STAKE_ACCOUNTS, ctx->runtime_stack_seed ) );
2814 0 : FD_TEST( ctx->runtime_stack );
2815 :
2816 0 : ctx->wksp = topo->workspaces[ topo->objs[ tile->tile_obj_id ].wksp_id ].wksp;
2817 :
2818 0 : ulong store_obj_id = fd_pod_query_ulong( topo->props, "store", ULONG_MAX );
2819 0 : FD_TEST( store_obj_id!=ULONG_MAX );
2820 0 : ctx->store = fd_store_join( fd_topo_obj_laddr( topo, store_obj_id ) );
2821 0 : FD_TEST( ctx->store );
2822 :
2823 0 : ulong banks_obj_id = fd_pod_query_ulong( topo->props, "banks", ULONG_MAX );
2824 0 : FD_TEST( banks_obj_id!=ULONG_MAX );
2825 :
2826 0 : ctx->banks = fd_banks_join( fd_topo_obj_laddr( topo, banks_obj_id ) );
2827 0 : FD_TEST( ctx->banks );
2828 :
2829 0 : FD_MGAUGE_SET( REPLAY, MAX_LIVE_BANKS, fd_banks_pool_max_cnt( ctx->banks ) );
2830 :
2831 0 : ctx->frontier_cnt = 0UL;
2832 :
2833 0 : fd_bank_t * bank = fd_banks_init_bank( ctx->banks );
2834 0 : FD_TEST( bank );
2835 0 : bank->f.slot = 0UL;
2836 0 : FD_TEST( bank->idx==FD_REPLAY_BOOT_BANK_IDX );
2837 :
2838 0 : ctx->consensus_root_slot = ULONG_MAX;
2839 0 : ctx->consensus_root = ctx->initial_block_id;
2840 0 : ctx->published_root_slot = ULONG_MAX;
2841 :
2842 0 : ctx->expected_shred_version = tile->replay.expected_shred_version;
2843 0 : ctx->ipecho_shred_version = 0;
2844 0 : fd_memcpy( ctx->genesis_path, tile->replay.genesis_path, sizeof(ctx->genesis_path) );
2845 0 : ctx->has_genesis_hash = 0;
2846 0 : ctx->has_genesis_timestamp = 0;
2847 0 : ctx->has_expected_genesis_timestamp = 0;
2848 0 : ctx->cluster_type = FD_CLUSTER_UNKNOWN;
2849 0 : ctx->hard_forks_cnt = ULONG_MAX;
2850 :
2851 0 : if( FD_UNLIKELY( tile->replay.bundle.enabled ) ) {
2852 0 : ctx->bundle.enabled = 1;
2853 0 : if( FD_UNLIKELY( !fd_bundle_crank_gen_init( ctx->bundle.gen,
2854 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_distribution_program_addr,
2855 0 : (fd_acct_addr_t const *)tile->replay.bundle.tip_payment_program_addr,
2856 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc,
2857 0 : (fd_acct_addr_t const *)ctx->bundle.vote_account.uc, "NAN", 0UL ) ) ) {
2858 0 : FD_LOG_ERR(( "failed to initialize bundle crank gen" ));
2859 0 : }
2860 0 : } else {
2861 0 : ctx->bundle.enabled = 0;
2862 0 : }
2863 :
2864 0 : fd_features_t * features = &bank->f.features;
2865 0 : fd_features_enable_cleaned_up( features );
2866 :
2867 0 : char const * one_off_features[ 16UL ];
2868 0 : FD_TEST( tile->replay.enable_features_cnt<=sizeof(one_off_features)/sizeof(one_off_features[0]) );
2869 0 : for( ulong i=0UL; i<tile->replay.enable_features_cnt; i++ ) one_off_features[ i ] = tile->replay.enable_features[i];
2870 0 : fd_features_enable_one_offs( features, one_off_features, (uint)tile->replay.enable_features_cnt, 0UL );
2871 :
2872 0 : fd_topo_obj_t const * vinyl_data = fd_topo_find_tile_obj( topo, tile, "vinyl_data" );
2873 :
2874 0 : ulong progcache_obj_id; FD_TEST( (progcache_obj_id = fd_pod_query_ulong( topo->props, "progcache", ULONG_MAX ) )!=ULONG_MAX );
2875 0 : FD_TEST( fd_progcache_shmem_join( ctx->progcache, fd_topo_obj_laddr( topo, progcache_obj_id ) ) );
2876 :
2877 0 : fd_wksp_t * progcache_wksp = fd_wksp_containing( ctx->progcache->shmem );
2878 0 : FD_TEST( progcache_wksp );
2879 0 : fd_wksp_mon_init( ctx->progcache_wksp_mon, progcache_wksp, FD_WKSP_MON_DEFAULT_RATE, fd_tickcount() );
2880 :
2881 0 : ulong funk_obj_id; FD_TEST( (funk_obj_id = fd_pod_query_ulong( topo->props, "funk", ULONG_MAX ) )!=ULONG_MAX );
2882 0 : ulong funk_locks_obj_id; FD_TEST( (funk_locks_obj_id = fd_pod_query_ulong( topo->props, "funk_locks", ULONG_MAX ) )!=ULONG_MAX );
2883 0 : ulong max_depth = tile->replay.max_live_slots + tile->replay.write_delay_slots;
2884 0 : if( !vinyl_data ) {
2885 0 : FD_TEST( fd_accdb_admin_v1_init( ctx->accdb_admin,
2886 0 : fd_topo_obj_laddr( topo, funk_obj_id ),
2887 0 : fd_topo_obj_laddr( topo, funk_locks_obj_id ) ) );
2888 0 : } else {
2889 0 : fd_topo_obj_t const * vinyl_rq = fd_topo_find_tile_obj( topo, tile, "vinyl_rq" );
2890 0 : fd_topo_obj_t const * vinyl_req_pool = fd_topo_find_tile_obj( topo, tile, "vinyl_rpool" );
2891 0 : FD_TEST( fd_accdb_admin_v2_init( ctx->accdb_admin,
2892 0 : fd_topo_obj_laddr( topo, funk_obj_id ),
2893 0 : fd_topo_obj_laddr( topo, funk_locks_obj_id ),
2894 0 : fd_topo_obj_laddr( topo, vinyl_rq->id ),
2895 0 : topo->workspaces[ vinyl_data->wksp_id ].wksp,
2896 0 : fd_topo_obj_laddr( topo, vinyl_req_pool->id ),
2897 0 : vinyl_rq->id,
2898 0 : max_depth ) );
2899 0 : fd_accdb_admin_v2_delay_set( ctx->accdb_admin, tile->replay.write_delay_slots );
2900 0 : }
2901 0 : fd_accdb_init_from_topo( ctx->accdb, topo, tile, max_depth );
2902 :
2903 0 : fd_wksp_t * funk_wksp = fd_wksp_containing( fd_topo_obj_laddr( topo, funk_obj_id ) );
2904 0 : FD_TEST( funk_wksp );
2905 0 : fd_wksp_mon_init( ctx->accdb_cache_wksp_mon, funk_wksp, FD_WKSP_MON_DEFAULT_RATE, fd_tickcount() );
2906 :
2907 0 : void * _txncache_shmem = fd_topo_obj_laddr( topo, tile->replay.txncache_obj_id );
2908 0 : fd_txncache_shmem_t * txncache_shmem = fd_txncache_shmem_join( _txncache_shmem );
2909 0 : FD_TEST( txncache_shmem );
2910 0 : ctx->txncache = fd_txncache_join( fd_txncache_new( _txncache, txncache_shmem ) );
2911 0 : FD_TEST( ctx->txncache );
2912 :
2913 0 : ctx->capture_ctx = NULL;
2914 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) ) ) {
2915 0 : ctx->capture_ctx = fd_capture_ctx_join( fd_capture_ctx_new( _capture_ctx ) );
2916 0 : ctx->capture_ctx->solcap_start_slot = tile->replay.capture_start_slot;
2917 0 : ctx->capture_ctx->capture_solcap = 1;
2918 0 : }
2919 :
2920 0 : ctx->dump_proto_ctx = NULL;
2921 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.dump_proto_dir ) ) ) {
2922 0 : ctx->dump_proto_ctx = dump_proto_ctx_mem;
2923 0 : ctx->dump_proto_ctx->dump_proto_output_dir = tile->replay.dump_proto_dir;
2924 0 : if( FD_LIKELY( tile->replay.dump_block_to_pb ) ) {
2925 0 : ctx->dump_proto_ctx->dump_block_to_pb = !!tile->replay.dump_block_to_pb;
2926 0 : }
2927 0 : }
2928 :
2929 0 : # if FD_HAS_FLATCC
2930 0 : if( FD_UNLIKELY( tile->replay.dump_block_to_pb ) ) {
2931 0 : ctx->block_dump_ctx = fd_block_dump_context_join( fd_block_dump_context_new( block_dump_ctx ) );
2932 0 : } else {
2933 0 : ctx->block_dump_ctx = NULL;
2934 0 : }
2935 0 : # endif
2936 :
2937 0 : ctx->is_booted = 0;
2938 :
2939 0 : ctx->larger_max_cost_per_block = tile->replay.larger_max_cost_per_block;
2940 :
2941 0 : FD_TEST( fd_rng_new( ctx->rng, ctx->rng_seed, 0UL ) );
2942 :
2943 0 : ctx->reasm = fd_reasm_join( fd_reasm_new( reasm_mem, tile->replay.fec_max, ctx->reasm_seed ) );
2944 0 : FD_TEST( ctx->reasm );
2945 :
2946 0 : ctx->sched = fd_sched_join( fd_sched_new( sched_mem, ctx->rng, tile->replay.sched_depth, tile->replay.max_live_slots, fd_topo_tile_name_cnt( topo, "execrp" ) ) );
2947 0 : FD_TEST( ctx->sched );
2948 :
2949 0 : FD_TEST( fd_vinyl_req_pool_new( vinyl_req_pool_mem, 1UL, 1UL ) );
2950 :
2951 0 : ctx->vote_tracker = fd_vote_tracker_join( fd_vote_tracker_new( vote_tracker_mem, ctx->vote_tracker_seed ) );
2952 0 : FD_TEST( ctx->vote_tracker );
2953 :
2954 0 : ctx->identity_vote_rooted = 0;
2955 :
2956 0 : ctx->wait_for_vote_to_start_leader = tile->replay.wait_for_vote_to_start_leader;
2957 :
2958 0 : ctx->wfs_enabled = memcmp( tile->replay.wait_for_supermajority_with_bank_hash.uc, ((fd_pubkey_t){ 0 }).uc, sizeof(fd_pubkey_t) );
2959 0 : ctx->expected_bank_hash = tile->replay.wait_for_supermajority_with_bank_hash;
2960 0 : ctx->wfs_complete = !ctx->wfs_enabled;
2961 :
2962 0 : ctx->mleaders = fd_multi_epoch_leaders_join( fd_multi_epoch_leaders_new( ctx->mleaders_mem ) );
2963 0 : FD_TEST( ctx->mleaders );
2964 :
2965 0 : ctx->is_leader = 0;
2966 0 : ctx->supports_leader = fd_topo_find_tile( topo, "pack", 0UL )!=ULONG_MAX;
2967 0 : ctx->reset_slot = 0UL;
2968 0 : ctx->reset_bank = NULL;
2969 0 : ctx->reset_block_id = ctx->initial_block_id;
2970 0 : ctx->reset_timestamp_nanos = 0UL;
2971 0 : ctx->next_leader_slot = ULONG_MAX;
2972 0 : ctx->next_leader_tickcount = LONG_MAX;
2973 0 : ctx->highwater_leader_slot = ULONG_MAX;
2974 0 : ctx->slot_duration_nanos = 350L*1000L*1000L; /* TODO: Not fixed ... not always 350ms ... */
2975 0 : ctx->slot_duration_ticks = (double)ctx->slot_duration_nanos*fd_tempo_tick_per_ns( NULL );
2976 0 : ctx->leader_bank = NULL;
2977 :
2978 0 : ctx->block_id_len = tile->replay.max_live_slots;
2979 0 : ctx->block_id_arr = (fd_block_id_ele_t *)block_id_arr_mem;
2980 0 : ctx->block_id_map = fd_block_id_map_join( fd_block_id_map_new( block_id_map_mem, chain_cnt, ctx->block_id_map_seed ) );
2981 0 : FD_TEST( ctx->block_id_map );
2982 0 : for( ulong i=0UL; i<tile->replay.max_live_slots; i++ ) ctx->block_id_arr[ i ].block_id_seen = 0;
2983 :
2984 0 : ctx->resolv_tile_cnt = fd_topo_tile_name_cnt( topo, "resolv" );
2985 :
2986 0 : ctx->keyswitch = fd_keyswitch_join( fd_topo_obj_laddr( topo, tile->id_keyswitch_obj_id ) );
2987 0 : FD_TEST( ctx->keyswitch );
2988 0 : ctx->halt_leader = 0;
2989 :
2990 0 : FD_TEST( tile->in_cnt<=sizeof(ctx->in)/sizeof(ctx->in[0]) );
2991 0 : for( ulong i=0UL; i<tile->in_cnt; i++ ) {
2992 0 : fd_topo_link_t * link = &topo->links[ tile->in_link_id[ i ] ];
2993 0 : fd_topo_wksp_t * link_wksp = &topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ];
2994 :
2995 0 : if( FD_LIKELY( link->dcache ) ) {
2996 0 : ctx->in[ i ].mem = link_wksp->wksp;
2997 0 : ctx->in[ i ].chunk0 = fd_dcache_compact_chunk0( ctx->in[ i ].mem, link->dcache );
2998 0 : ctx->in[ i ].wmark = fd_dcache_compact_wmark ( ctx->in[ i ].mem, link->dcache, link->mtu );
2999 0 : ctx->in[ i ].mtu = link->mtu;
3000 0 : }
3001 :
3002 0 : if( !strcmp( link->name, "genesi_out" ) ) ctx->in_kind[ i ] = IN_KIND_GENESIS;
3003 0 : else if( !strcmp( link->name, "ipecho_out" ) ) ctx->in_kind[ i ] = IN_KIND_IPECHO;
3004 0 : else if( !strcmp( link->name, "snapin_manif" ) ) ctx->in_kind[ i ] = IN_KIND_SNAP;
3005 0 : else if( !strcmp( link->name, "execrp_replay" ) ) ctx->in_kind[ i ] = IN_KIND_EXECRP;
3006 0 : else if( !strcmp( link->name, "tower_out" ) ) ctx->in_kind[ i ] = IN_KIND_TOWER;
3007 0 : else if( !strcmp( link->name, "poh_replay" ) ) ctx->in_kind[ i ] = IN_KIND_POH;
3008 0 : else if( !strcmp( link->name, "resolv_replay" ) ) ctx->in_kind[ i ] = IN_KIND_RESOLV;
3009 0 : else if( !strcmp( link->name, "repair_out" ) ) ctx->in_kind[ i ] = IN_KIND_REPAIR;
3010 0 : else if( !strcmp( link->name, "txsend_out" ) ) ctx->in_kind[ i ] = IN_KIND_TXSEND;
3011 0 : else if( !strcmp( link->name, "rpc_replay" ) ) ctx->in_kind[ i ] = IN_KIND_RPC;
3012 0 : else if( !strcmp( link->name, "gossip_out" ) ) ctx->in_kind[ i ] = IN_KIND_GOSSIP_OUT;
3013 0 : else FD_LOG_ERR(( "unexpected input link name %s", link->name ));
3014 0 : }
3015 :
3016 0 : *ctx->epoch_out = out1( topo, tile, "replay_epoch" ); FD_TEST( ctx->epoch_out->idx!=ULONG_MAX );
3017 0 : *ctx->replay_out = out1( topo, tile, "replay_out" ); FD_TEST( ctx->replay_out->idx!=ULONG_MAX );
3018 0 : *ctx->exec_out = out1( topo, tile, "replay_execrp" ); FD_TEST( ctx->exec_out->idx!=ULONG_MAX );
3019 :
3020 0 : ctx->rpc_enabled = fd_topo_find_tile( topo, "rpc", 0UL )!=ULONG_MAX;
3021 :
3022 0 : if( FD_UNLIKELY( strcmp( "", tile->replay.solcap_capture ) ) ) {
3023 0 : ulong idx = fd_topo_find_tile_out_link( topo, tile, "cap_repl", 0UL );
3024 0 : FD_TEST( idx!=ULONG_MAX );
3025 0 : fd_topo_link_t * link = &topo->links[ tile->out_link_id[ idx ] ];
3026 :
3027 :
3028 0 : fd_capture_link_buf_t * cap_repl_out = ctx->cap_repl_out;
3029 0 : cap_repl_out->base.vt = &fd_capture_link_buf_vt;
3030 0 : cap_repl_out->idx = idx;
3031 0 : cap_repl_out->mem = topo->workspaces[ topo->objs[ link->dcache_obj_id ].wksp_id ].wksp;
3032 0 : cap_repl_out->chunk0 = fd_dcache_compact_chunk0( cap_repl_out->mem, link->dcache );
3033 0 : cap_repl_out->wmark = fd_dcache_compact_wmark( cap_repl_out->mem, link->dcache, link->mtu );
3034 0 : cap_repl_out->chunk = cap_repl_out->chunk0;
3035 0 : cap_repl_out->mcache = link->mcache;
3036 0 : cap_repl_out->depth = fd_mcache_depth( link->mcache );
3037 0 : cap_repl_out->seq = 0UL;
3038 :
3039 0 : ctx->capture_ctx->capctx_type.buf = cap_repl_out;
3040 0 : ctx->capture_ctx->capture_link = &cap_repl_out->base;
3041 0 : ctx->capture_ctx->current_txn_idx = 0UL;
3042 :
3043 :
3044 0 : ulong consumer_tile_idx = fd_topo_find_tile( topo, "solcap", 0UL );
3045 0 : fd_topo_tile_t * consumer_tile = &topo->tiles[ consumer_tile_idx ];
3046 0 : cap_repl_out->fseq = NULL;
3047 0 : for( ulong j = 0UL; j < consumer_tile->in_cnt; j++ ) {
3048 0 : if( FD_UNLIKELY( consumer_tile->in_link_id[ j ] == link->id ) ) {
3049 0 : cap_repl_out->fseq = fd_fseq_join( fd_topo_obj_laddr( topo, consumer_tile->in_link_fseq_obj_id[ j ] ) );
3050 0 : FD_TEST( cap_repl_out->fseq );
3051 0 : break;
3052 0 : }
3053 0 : }
3054 0 : }
3055 :
3056 0 : fd_memset( &ctx->metrics, 0, sizeof(ctx->metrics) );
3057 :
3058 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_query_wait, FD_MHIST_SECONDS_MIN( REPLAY, STORE_QUERY_WAIT ),
3059 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_QUERY_WAIT ) ) );
3060 0 : fd_histf_join( fd_histf_new( ctx->metrics.store_query_work, FD_MHIST_SECONDS_MIN( REPLAY, STORE_QUERY_WORK ),
3061 0 : FD_MHIST_SECONDS_MAX( REPLAY, STORE_QUERY_WORK ) ) );
3062 :
3063 0 : fd_histf_join( fd_histf_new( ctx->metrics.root_slot_dur, FD_MHIST_SECONDS_MIN( REPLAY, ROOT_SLOT_DURATION_SECONDS ),
3064 0 : FD_MHIST_SECONDS_MAX( REPLAY, ROOT_SLOT_DURATION_SECONDS ) ) );
3065 0 : fd_histf_join( fd_histf_new( ctx->metrics.root_account_dur, FD_MHIST_SECONDS_MIN( REPLAY, ROOT_ACCOUNT_DURATION_SECONDS ),
3066 0 : FD_MHIST_SECONDS_MAX( REPLAY, ROOT_ACCOUNT_DURATION_SECONDS ) ) );
3067 :
3068 : /* Ensure precompiles are available, crash fast otherwise */
3069 0 : fd_precompiles();
3070 :
3071 0 : ulong scratch_top = FD_SCRATCH_ALLOC_FINI( l, 1UL );
3072 0 : if( FD_UNLIKELY( scratch_top > (ulong)scratch + scratch_footprint( tile ) ) )
3073 0 : FD_LOG_ERR(( "scratch overflow %lu %lu %lu", scratch_top - (ulong)scratch - scratch_footprint( tile ), scratch_top, (ulong)scratch + scratch_footprint( tile ) ));
3074 0 : }
3075 :
3076 : static ulong
3077 : populate_allowed_seccomp( fd_topo_t const * topo FD_FN_UNUSED,
3078 : fd_topo_tile_t const * tile FD_FN_UNUSED,
3079 : ulong out_cnt,
3080 0 : struct sock_filter * out ) {
3081 :
3082 0 : populate_sock_filter_policy_fd_replay_tile( out_cnt, out, (uint)fd_log_private_logfile_fd() );
3083 0 : return sock_filter_policy_fd_replay_tile_instr_cnt;
3084 0 : }
3085 :
3086 : static ulong
3087 : populate_allowed_fds( fd_topo_t const * topo FD_FN_UNUSED,
3088 : fd_topo_tile_t const * tile FD_FN_UNUSED,
3089 : ulong out_fds_cnt,
3090 0 : int * out_fds ) {
3091 :
3092 0 : if( FD_UNLIKELY( out_fds_cnt<2UL ) ) FD_LOG_ERR(( "out_fds_cnt %lu", out_fds_cnt ));
3093 :
3094 0 : ulong out_cnt = 0UL;
3095 0 : out_fds[ out_cnt++ ] = 2; /* stderr */
3096 0 : if( FD_LIKELY( -1!=fd_log_private_logfile_fd() ) )
3097 0 : out_fds[ out_cnt++ ] = fd_log_private_logfile_fd(); /* logfile */
3098 0 : return out_cnt;
3099 0 : }
3100 :
3101 : static inline void
3102 0 : during_housekeeping( fd_replay_tile_t * ctx ) {
3103 0 : if( FD_UNLIKELY( fd_keyswitch_state_query( ctx->keyswitch )==FD_KEYSWITCH_STATE_UNHALT_PENDING ) ) {
3104 0 : FD_CRIT( ctx->halt_leader, "state machine corruption" );
3105 0 : FD_LOG_DEBUG(( "keyswitch: unhalting leader" ));
3106 0 : ctx->halt_leader = 0;
3107 0 : fd_keyswitch_state( ctx->keyswitch, FD_KEYSWITCH_STATE_COMPLETED );
3108 0 : }
3109 :
3110 0 : if( FD_UNLIKELY( fd_keyswitch_state_query( ctx->keyswitch )==FD_KEYSWITCH_STATE_SWITCH_PENDING ) ) {
3111 0 : FD_LOG_DEBUG(( "keyswitch: halting leader" ));
3112 0 : ctx->halt_leader = 1;
3113 0 : if( !ctx->is_leader ) maybe_switch_identity( ctx );
3114 0 : }
3115 0 : }
3116 :
3117 : #undef DEBUG_LOGGING
3118 :
3119 : /* counting carefully, after_credit can generate at most 7 frags and
3120 : returnable_frag boot_genesis can also generate at most 7 frags, so 14
3121 : is a conservative bound. */
3122 0 : #define STEM_BURST (14UL)
3123 :
3124 : /* fd_tempo_lazy_default( 16384 ) where 16384 is the minimum out-link
3125 : depth (i.e. cr_max) but excludes replay_epoch, which is so infrequent
3126 : credit availability is a non-issue. */
3127 0 : #define STEM_LAZY ((long)36865)
3128 :
3129 0 : #define STEM_CALLBACK_CONTEXT_TYPE fd_replay_tile_t
3130 0 : #define STEM_CALLBACK_CONTEXT_ALIGN alignof(fd_replay_tile_t)
3131 :
3132 0 : #define STEM_CALLBACK_METRICS_WRITE metrics_write
3133 0 : #define STEM_CALLBACK_AFTER_CREDIT after_credit
3134 0 : #define STEM_CALLBACK_BEFORE_FRAG before_frag
3135 0 : #define STEM_CALLBACK_RETURNABLE_FRAG returnable_frag
3136 0 : #define STEM_CALLBACK_DURING_HOUSEKEEPING during_housekeeping
3137 :
3138 : #include "../../disco/stem/fd_stem.c"
3139 :
3140 : fd_topo_run_tile_t fd_tile_replay = {
3141 : .name = "replay",
3142 : .populate_allowed_seccomp = populate_allowed_seccomp,
3143 : .populate_allowed_fds = populate_allowed_fds,
3144 : .scratch_align = scratch_align,
3145 : .scratch_footprint = scratch_footprint,
3146 : .privileged_init = privileged_init,
3147 : .unprivileged_init = unprivileged_init,
3148 : .run = stem_run,
3149 : };
|
