Line data Source code
1 : #ifndef HEADER_fd_src_disco_topo_fd_topo_h
2 : #define HEADER_fd_src_disco_topo_fd_topo_h
3 :
4 : #include "../stem/fd_stem.h"
5 : #include "../../tango/fd_tango.h"
6 : #include "../../waltz/xdp/fd_xdp1.h"
7 : #include "../../ballet/base58/fd_base58.h"
8 : #include "../../util/net/fd_net_headers.h"
9 :
10 : /* Maximum number of workspaces that may be present in a topology. */
11 : #define FD_TOPO_MAX_WKSPS (256UL)
12 : /* Maximum number of links that may be present in a topology. */
13 : #define FD_TOPO_MAX_LINKS (256UL)
14 : /* Maximum number of tiles that may be present in a topology. */
15 0 : #define FD_TOPO_MAX_TILES (256UL)
16 : /* Maximum number of objects that may be present in a topology. */
17 : #define FD_TOPO_MAX_OBJS (4096UL)
18 : /* Maximum number of links that may go into any one tile in the
19 : topology. */
20 : #define FD_TOPO_MAX_TILE_IN_LINKS ( 128UL)
21 : /* Maximum number of links that a tile may write to. */
22 : #define FD_TOPO_MAX_TILE_OUT_LINKS ( 32UL)
23 : /* Maximum number of objects that a tile can use. */
24 : #define FD_TOPO_MAX_TILE_OBJS ( 256UL)
25 :
26 : /* Maximum number of additional ip addresses */
27 : #define FD_NET_MAX_SRC_ADDR 4
28 :
29 : /* Maximum number of additional destinations for leader shreds and for retransmitted shreds */
30 : #define FD_TOPO_ADTL_DESTS_MAX ( 32UL)
31 :
32 0 : #define FD_TOPO_CORE_DUMP_LEVEL_DISABLED (0)
33 0 : #define FD_TOPO_CORE_DUMP_LEVEL_MINIMAL (1)
34 6 : #define FD_TOPO_CORE_DUMP_LEVEL_REGULAR (2)
35 0 : #define FD_TOPO_CORE_DUMP_LEVEL_FULL (3)
36 0 : #define FD_TOPO_CORE_DUMP_LEVEL_NEVER (4)
37 :
38 : /* A workspace is a Firedancer specific memory management structure that
39 : sits on top of 1 or more memory mapped gigantic or huge pages mounted
40 : to the hugetlbfs. */
41 : typedef struct {
42 : ulong id; /* The ID of this workspace. Indexed from [0, wksp_cnt). When placed in a topology, the ID must be the index of the workspace in the workspaces list. */
43 : char name[ 13UL ]; /* The name of this workspace, like "pack". There can be at most one of each workspace name in a topology. */
44 :
45 : ulong numa_idx; /* The index of the NUMA node on the system that this workspace should be allocated from. */
46 :
47 : ulong min_part_max; /* Artificially raise part_max */
48 : ulong min_loose_sz; /* Artificially raise loose footprint */
49 :
50 : /* Computed fields. These are not supplied as configuration but calculated as needed. */
51 : struct {
52 : ulong page_sz; /* The size of the pages that this workspace is backed by. One of FD_PAGE_SIZE_*. */
53 : ulong page_cnt; /* The number of pages that must be mapped to this workspace to store all the data needed by consumers. */
54 : ulong part_max; /* The maximum number of partitions in the underlying workspace. There can only be this many allocations made at any one time. */
55 :
56 : int core_dump_level; /* The core dump level required to be set in the application configuration to have this workspace appear in core dumps. */
57 :
58 : fd_wksp_t * wksp; /* The workspace memory in the local process. */
59 : ulong known_footprint; /* Total size in bytes of all data in Firedancer that will be stored in this workspace at startup. */
60 : ulong total_footprint; /* Total size in bytes of all data in Firedancer that could be stored in this workspace, includes known data and loose data. */
61 : };
62 : } fd_topo_wksp_t;
63 :
64 : /* A link is an mcache in a workspace that has one producer and one or
65 : more consumers. A link may optionally also have a dcache, that holds
66 : fragments referred to by the mcache entries.
67 :
68 : A link belongs to exactly one workspace. A link has exactly one
69 : producer, and 1 or more consumers. Each consumer is either reliable
70 : or not reliable. A link has a depth and a MTU, which correspond to
71 : the depth and MTU of the mcache and dcache respectively. A MTU of
72 : zero means no dcache is needed, as there is no data. */
73 : typedef struct {
74 : ulong id; /* The ID of this link. Indexed from [0, link_cnt). When placed in a topology, the ID must be the index of the link in the links list. */
75 : char name[ 13UL ]; /* The name of this link, like "pack_bank". There can be multiple of each link name in a topology. */
76 : ulong kind_id; /* The ID of this link within its name. If there are N links of a particular name, they have IDs [0, N). The pair (name, kind_id) uniquely identifies a link, as does "id" on its own. */
77 :
78 : ulong depth; /* The depth of the mcache representing the link. */
79 : ulong mtu; /* The MTU of data fragments in the mcache. A value of 0 means there is no dcache. */
80 : ulong burst; /* The max amount of MTU sized data fragments that might be bursted to the dcache. */
81 :
82 : ulong mcache_obj_id;
83 : ulong dcache_obj_id;
84 :
85 : /* Computed fields. These are not supplied as configuration but calculated as needed. */
86 : struct {
87 : fd_frag_meta_t * mcache; /* The mcache of this link. */
88 : void * dcache; /* The dcache of this link, if it has one. */
89 : };
90 :
91 : uint permit_no_consumers : 1; /* Permit a topology where this link has no consumers */
92 : uint permit_no_producers : 1; /* Permit a topology where this link has no producers */
93 : } fd_topo_link_t;
94 :
95 : /* Be careful: ip and host are in different byte order */
96 : typedef struct {
97 : uint ip; /* in network byte order */
98 : ushort port; /* in host byte order */
99 : } fd_topo_ip_port_t;
100 :
101 : struct fd_topo_net_tile {
102 : ulong umem_dcache_obj_id; /* dcache for XDP UMEM frames */
103 : uint bind_address;
104 :
105 : ushort shred_listen_port;
106 : ushort quic_transaction_listen_port;
107 : ushort legacy_transaction_listen_port;
108 : ushort gossip_listen_port;
109 : ushort repair_intake_listen_port;
110 : ushort repair_serve_listen_port;
111 : ushort send_src_port;
112 : };
113 : typedef struct fd_topo_net_tile fd_topo_net_tile_t;
114 :
115 : /* A tile is a unique process that is spawned by Firedancer to represent
116 : one thread of execution. Firedancer sandboxes all tiles to their own
117 : process for security reasons.
118 :
119 : A tile belongs to exactly one workspace. A tile is a consumer of 0
120 : or more links, it's inputs. A tile is a producer of 0 or more output
121 : links.
122 :
123 : All input links will be automatically polled by the tile
124 : infrastructure, and output links will automatically source and manage
125 : credits from consumers. */
126 : struct fd_topo_tile {
127 : ulong id; /* The ID of this tile. Indexed from [0, tile_cnt). When placed in a topology, the ID must be the index of the tile in the tiles list. */
128 : char name[ 7UL ]; /* The name of this tile. There can be multiple of each tile name in a topology. */
129 : char metrics_name[ 10UL ]; /* The name of this tile for looking up metrics. This is used so tiles can share a name but report different metrics, for Frankendancer and Firedancer. */
130 : ulong kind_id; /* The ID of this tile within its name. If there are n tile of a particular name, they have IDs [0, N). The pair (name, kind_id) uniquely identifies a tile, as does "id" on its own. */
131 : int is_agave; /* If the tile needs to run in the Agave (Anza) address space or not. */
132 : int allow_shutdown; /* If the tile is allowed to shutdown gracefully. If false, when the tile exits it will tear down the entire application. */
133 :
134 : ulong cpu_idx; /* The CPU index to pin the tile on. A value of ULONG_MAX or more indicates the tile should be floating and not pinned to a core. */
135 :
136 : ulong in_cnt; /* The number of links that this tile reads from. */
137 : ulong in_link_id[ FD_TOPO_MAX_TILE_IN_LINKS ]; /* The link_id of each link that this tile reads from, indexed in [0, in_cnt). */
138 : int in_link_reliable[ FD_TOPO_MAX_TILE_IN_LINKS ]; /* If each link that this tile reads from is a reliable or unreliable consumer, indexed in [0, in_cnt). */
139 : int in_link_poll[ FD_TOPO_MAX_TILE_IN_LINKS ]; /* If each link that this tile reads from should be polled by the tile infrastructure, indexed in [0, in_cnt).
140 : If the link is not polled, the tile will not receive frags for it and the tile writer is responsible for
141 : reading from the link. The link must be marked as unreliable as it is not flow controlled. */
142 :
143 : ulong out_cnt; /* The number of links that this tile writes to. */
144 : ulong out_link_id[ FD_TOPO_MAX_TILE_OUT_LINKS ]; /* The link_id of each link that this tile writes to, indexed in [0, link_cnt). */
145 :
146 : ulong tile_obj_id;
147 : ulong metrics_obj_id;
148 : ulong keyswitch_obj_id;
149 : ulong in_link_fseq_obj_id[ FD_TOPO_MAX_TILE_IN_LINKS ];
150 :
151 : ulong uses_obj_cnt;
152 : ulong uses_obj_id[ FD_TOPO_MAX_TILE_OBJS ];
153 : int uses_obj_mode[ FD_TOPO_MAX_TILE_OBJS ];
154 :
155 : /* Computed fields. These are not supplied as configuration but calculated as needed. */
156 : struct {
157 : ulong * metrics; /* The shared memory for metrics that this tile should write. Consumer by monitoring and metrics writing tiles. */
158 :
159 : /* The fseq of each link that this tile reads from. Multiple fseqs
160 : may point to the link, if there are multiple consumers. An fseq
161 : can be uniquely identified via (link_id, tile_id), or (link_kind,
162 : link_kind_id, tile_kind, tile_kind_id) */
163 : ulong * in_link_fseq[ FD_TOPO_MAX_TILE_IN_LINKS ];
164 : };
165 :
166 : /* Configuration fields. These are required to be known by the topology so it can determine the
167 : total size of Firedancer in memory. */
168 : union {
169 : fd_topo_net_tile_t net;
170 :
171 : struct {
172 : fd_topo_net_tile_t net;
173 :
174 : char if_virt[ 16 ]; /* device name (virtual, for routing) */
175 : char if_phys[ 16 ]; /* device name (physical, for RX/TX) */
176 : uint if_queue; /* device queue index */
177 :
178 : /* xdp specific options */
179 : ulong xdp_rx_queue_size;
180 : ulong xdp_tx_queue_size;
181 : ulong free_ring_depth;
182 : long tx_flush_timeout_ns;
183 : char xdp_mode[8];
184 : int zero_copy;
185 :
186 : ulong netdev_dbl_buf_obj_id; /* dbl_buf containing netdev_tbl */
187 : ulong fib4_main_obj_id; /* fib4 containing main route table */
188 : ulong fib4_local_obj_id; /* fib4 containing local route table */
189 : ulong neigh4_obj_id; /* neigh4 hash map */
190 :
191 : int xsk_core_dump;
192 : } xdp;
193 :
194 : struct {
195 : fd_topo_net_tile_t net;
196 : /* sock specific options */
197 : int so_sndbuf;
198 : int so_rcvbuf;
199 : } sock;
200 :
201 : struct {
202 : ulong netdev_dbl_buf_obj_id; /* dbl_buf containing netdev_tbl */
203 : ulong fib4_main_obj_id; /* fib4 containing main route table */
204 : ulong fib4_local_obj_id; /* fib4 containing local route table */
205 : char neigh_if[ 16 ]; /* neigh4 interface name */
206 : ulong neigh4_obj_id; /* neigh4 hash map */
207 : } netlink;
208 :
209 0 : #define FD_TOPO_GOSSIP_ENTRYPOINTS_MAX 16UL
210 :
211 : struct {
212 : char identity_key_path[ PATH_MAX ];
213 :
214 : ulong entrypoints_cnt;
215 : fd_ip4_port_t entrypoints[ FD_TOPO_GOSSIP_ENTRYPOINTS_MAX ];
216 :
217 : long boot_timestamp_nanos;
218 :
219 : ulong tcache_depth;
220 :
221 : ushort shred_version;
222 : int allow_private_address;
223 : } gossvf;
224 :
225 : struct {
226 : char identity_key_path[ PATH_MAX ];
227 :
228 : ulong entrypoints_cnt;
229 : fd_ip4_port_t entrypoints[ FD_TOPO_GOSSIP_ENTRYPOINTS_MAX ];
230 :
231 : long boot_timestamp_nanos;
232 :
233 : uint ip_addr;
234 : ushort shred_version;
235 :
236 : ulong max_entries;
237 : ulong max_purged;
238 : ulong max_failed;
239 :
240 : struct {
241 : ushort gossip;
242 : ushort tvu;
243 : ushort tvu_quic;
244 : ushort tpu;
245 : ushort tpu_quic;
246 : ushort repair;
247 : } ports;
248 : } gossip;
249 :
250 : struct {
251 : uint out_depth;
252 : uint reasm_cnt;
253 : ulong max_concurrent_connections;
254 : ulong max_concurrent_handshakes;
255 : ushort quic_transaction_listen_port;
256 : long idle_timeout_millis;
257 : uint ack_delay_millis;
258 : int retry;
259 : char key_log_path[ PATH_MAX ];
260 : } quic;
261 :
262 : struct {
263 : ulong tcache_depth;
264 : } verify;
265 :
266 : struct {
267 : ulong tcache_depth;
268 : } dedup;
269 :
270 : struct {
271 : char url[ 256 ];
272 : ulong url_len;
273 : char sni[ 256 ];
274 : ulong sni_len;
275 : char identity_key_path[ PATH_MAX ];
276 : char key_log_path[ PATH_MAX ];
277 : ulong buf_sz;
278 : ulong ssl_heap_sz;
279 : ulong keepalive_interval_nanos;
280 : uchar tls_cert_verify : 1;
281 : } bundle;
282 :
283 : struct {
284 : ulong max_pending_transactions;
285 : ulong bank_tile_count;
286 : int larger_max_cost_per_block;
287 : int larger_shred_limits_per_block;
288 : int use_consumed_cus;
289 : int schedule_strategy;
290 : struct {
291 : int enabled;
292 : uchar tip_distribution_program_addr[ 32 ];
293 : uchar tip_payment_program_addr[ 32 ];
294 : uchar tip_distribution_authority[ 32 ];
295 : ulong commission_bps;
296 : char identity_key_path[ PATH_MAX ];
297 : char vote_account_path[ PATH_MAX ]; /* or pubkey is okay */
298 : } bundle;
299 : } pack;
300 :
301 : struct {
302 : int lagged_consecutive_leader_start;
303 : int plugins_enabled;
304 : ulong bank_cnt;
305 : char identity_key_path[ PATH_MAX ];
306 : struct {
307 : int enabled;
308 : uchar tip_payment_program_addr[ 32 ];
309 : uchar tip_distribution_program_addr[ 32 ];
310 : char vote_account_path[ PATH_MAX ];
311 : } bundle;
312 : } poh;
313 :
314 : struct {
315 : ulong depth;
316 : ulong fec_resolver_depth;
317 : char identity_key_path[ PATH_MAX ];
318 : ushort shred_listen_port;
319 : int larger_shred_limits_per_block;
320 : ushort expected_shred_version;
321 : ulong adtl_dests_retransmit_cnt;
322 : fd_topo_ip_port_t adtl_dests_retransmit[ FD_TOPO_ADTL_DESTS_MAX ];
323 : ulong adtl_dests_leader_cnt;
324 : fd_topo_ip_port_t adtl_dests_leader[ FD_TOPO_ADTL_DESTS_MAX ];
325 : } shred;
326 :
327 : struct {
328 : ulong disable_blockstore_from_slot;
329 : } store;
330 :
331 : struct {
332 : char identity_key_path[ PATH_MAX ];
333 : } sign;
334 :
335 : struct {
336 : uint listen_addr;
337 : ushort listen_port;
338 :
339 : int is_voting;
340 :
341 : char cluster[ 32 ];
342 : char identity_key_path[ PATH_MAX ];
343 : char vote_key_path[ PATH_MAX ];
344 :
345 : ulong max_http_connections;
346 : ulong max_websocket_connections;
347 : ulong max_http_request_length;
348 : ulong send_buffer_size_mb;
349 : int schedule_strategy;
350 :
351 : int websocket_compression;
352 : int frontend_release_channel;
353 : } gui;
354 :
355 : struct {
356 : uint listen_addr;
357 : ushort listen_port;
358 :
359 : ulong max_http_connections;
360 : ulong send_buffer_size_mb;
361 : ulong max_http_request_length;
362 :
363 : ulong max_live_slots;
364 :
365 : char identity_key_path[ PATH_MAX ];
366 : } rpc;
367 :
368 : struct {
369 : uint prometheus_listen_addr;
370 : ushort prometheus_listen_port;
371 : } metric;
372 :
373 : struct {
374 : ulong fec_max;
375 : ulong max_vote_accounts;
376 :
377 : ulong funk_obj_id;
378 :
379 : ulong txncache_obj_id;
380 : ulong progcache_obj_id;
381 :
382 : char shred_cap[ PATH_MAX ];
383 :
384 : char identity_key_path[ PATH_MAX ];
385 : uint ip_addr;
386 : char vote_account_path[ PATH_MAX ];
387 :
388 : ushort expected_shred_version;
389 :
390 : ulong heap_size_gib;
391 : ulong max_live_slots;
392 :
393 : /* not specified in TOML */
394 :
395 : ulong enable_features_cnt;
396 : char enable_features[ 16 ][ FD_BASE58_ENCODED_32_SZ ];
397 :
398 : char genesis_path[ PATH_MAX ];
399 :
400 : int larger_max_cost_per_block;
401 :
402 : ulong capture_start_slot;
403 : char solcap_capture[ PATH_MAX ];
404 : char dump_proto_dir[ PATH_MAX ];
405 : int dump_block_to_pb;
406 :
407 : struct {
408 : int enabled;
409 : uchar tip_payment_program_addr[ 32 ];
410 : uchar tip_distribution_program_addr[ 32 ];
411 : char vote_account_path[ PATH_MAX ];
412 : } bundle;
413 :
414 : } replay;
415 :
416 : struct {
417 : ulong funk_obj_id;
418 : ulong txncache_obj_id;
419 : ulong progcache_obj_id;
420 : ulong acc_pool_obj_id;
421 :
422 : ulong max_live_slots;
423 :
424 : ulong capture_start_slot;
425 : char solcap_capture[ PATH_MAX ];
426 : char dump_proto_dir[ PATH_MAX ];
427 : int dump_instr_to_pb;
428 : int dump_txn_to_pb;
429 : int dump_syscall_to_pb;
430 : int dump_elf_to_pb;
431 : } exec;
432 :
433 : struct {
434 : ushort send_to_port;
435 : uint send_to_ip_addr;
436 : ulong conn_cnt;
437 : int no_quic;
438 : } benchs;
439 :
440 : struct {
441 : ushort rpc_port;
442 : uint rpc_ip_addr;
443 : } bencho;
444 :
445 : struct {
446 : ulong accounts_cnt;
447 : int mode;
448 : float contending_fraction;
449 : float cu_price_spread;
450 : } benchg;
451 :
452 : struct {
453 : ushort repair_intake_listen_port;
454 : ushort repair_serve_listen_port;
455 : char identity_key_path[ PATH_MAX ];
456 : ulong max_pending_shred_sets;
457 : ulong slot_max;
458 :
459 : /* non-config */
460 :
461 : ulong repair_sign_depth;
462 : ulong repair_sign_cnt;
463 :
464 : ulong end_slot; /* repair profiler mode only */
465 : } repair;
466 :
467 : struct {
468 : ushort send_src_port;
469 :
470 : /* non-config */
471 :
472 : uint ip_addr;
473 : char identity_key_path[ PATH_MAX ];
474 : } send;
475 :
476 : struct {
477 : uint fake_dst_ip;
478 : } pktgen;
479 :
480 : struct {
481 : ulong end_slot;
482 : char rocksdb_path[ PATH_MAX ];
483 : char ingest_mode[ 32 ];
484 :
485 : /* Set internally by the archiver tile */
486 : int archive_fd;
487 : } archiver;
488 :
489 : struct {
490 : ulong end_slot;
491 : char rocksdb_path[ PATH_MAX ];
492 : char shredcap_path[ PATH_MAX ];
493 : } backtest;
494 :
495 : struct {
496 : int hard_fork_fatal;
497 : ulong max_live_slots;
498 : ulong max_vote_lookahead;
499 : char identity_key[ PATH_MAX ];
500 : char vote_account[ PATH_MAX ];
501 : char base_path[PATH_MAX];
502 : } tower;
503 :
504 : struct {
505 : char folder_path[ PATH_MAX ];
506 : ushort repair_intake_listen_port;
507 : ulong write_buffer_size; /* Size of the write buffer for the capture tile */
508 : int enable_publish_stake_weights;
509 : char manifest_path[ PATH_MAX ];
510 :
511 : /* Set internally by the capture tile */
512 : int shreds_fd;
513 : int requests_fd;
514 : int fecs_fd;
515 : int peers_fd;
516 : int bank_hashes_fd;
517 : int slices_fd;
518 : } shredcap;
519 :
520 : #define FD_TOPO_SNAPSHOTS_GOSSIP_LIST_MAX (32UL)
521 0 : #define FD_TOPO_SNAPSHOTS_SERVERS_MAX (16UL)
522 0 : #define FD_TOPO_MAX_RESOLVED_ADDRS ( 4UL)
523 0 : #define FD_TOPO_SNAPSHOTS_SERVERS_MAX_RESOLVED (FD_TOPO_MAX_RESOLVED_ADDRS*FD_TOPO_SNAPSHOTS_SERVERS_MAX)
524 :
525 : struct fd_topo_tile_snapct {
526 : char snapshots_path[ PATH_MAX ];
527 :
528 : struct {
529 : uint max_local_full_effective_age;
530 : uint max_local_incremental_age;
531 :
532 : struct {
533 : int allow_any;
534 : ulong allow_list_cnt;
535 : fd_pubkey_t allow_list[ FD_TOPO_SNAPSHOTS_GOSSIP_LIST_MAX ];
536 : ulong block_list_cnt;
537 : fd_pubkey_t block_list[ FD_TOPO_SNAPSHOTS_GOSSIP_LIST_MAX ];
538 : } gossip;
539 :
540 : ulong servers_cnt;
541 : struct {
542 : fd_ip4_port_t addr;
543 : char hostname[ 256UL ];
544 : int is_https;
545 : } servers[ FD_TOPO_SNAPSHOTS_SERVERS_MAX_RESOLVED ];
546 : } sources;
547 :
548 : int incremental_snapshots;
549 : uint max_full_snapshots_to_keep;
550 : uint max_incremental_snapshots_to_keep;
551 : uint full_effective_age_cancel_threshold;
552 : uint max_retry_abort;
553 : } snapct;
554 :
555 : struct {
556 : char snapshots_path[ PATH_MAX ];
557 : } snapld;
558 :
559 : struct {
560 : ulong max_live_slots;
561 : ulong funk_obj_id;
562 : ulong txncache_obj_id;
563 :
564 : uint lthash_disabled : 1;
565 : uint use_vinyl : 1;
566 : } snapin;
567 :
568 : struct {
569 : uint lthash_disabled : 1;
570 : ulong vinyl_meta_map_obj_id;
571 : ulong vinyl_meta_pool_obj_id;
572 : ulong snapwr_depth;
573 : char vinyl_path[ PATH_MAX ];
574 : } snapwm;
575 :
576 : struct {
577 : ulong dcache_obj_id;
578 : char vinyl_path[ PATH_MAX ];
579 : } snapwr;
580 :
581 : struct {
582 : ulong dcache_obj_id;
583 : int io_uring_enabled;
584 : char vinyl_path[ PATH_MAX ];
585 : } snaplh;
586 :
587 : struct {
588 :
589 : uint bind_address;
590 : ushort bind_port;
591 :
592 : ushort expected_shred_version;
593 : ulong entrypoints_cnt;
594 : fd_ip4_port_t entrypoints[ FD_TOPO_GOSSIP_ENTRYPOINTS_MAX ];
595 : } ipecho;
596 :
597 : struct {
598 : ulong max_live_slots;
599 : ulong txncache_obj_id;
600 : ulong funk_obj_id;
601 : ulong progcache_obj_id;
602 : ulong acc_pool_obj_id;
603 : } bank;
604 :
605 : struct {
606 : ulong funk_obj_id;
607 : } resolv;
608 :
609 : struct {
610 : ulong funk_obj_id;
611 :
612 : int allow_download;
613 :
614 : ushort expected_shred_version;
615 : ulong entrypoints_cnt;
616 : fd_ip4_port_t entrypoints[ FD_TOPO_GOSSIP_ENTRYPOINTS_MAX ];
617 :
618 : int has_expected_genesis_hash;
619 : uchar expected_genesis_hash[ 32UL ];
620 :
621 : char genesis_path[ PATH_MAX ];
622 :
623 : uint target_gid;
624 : uint target_uid;
625 : } genesi;
626 :
627 : struct {
628 : ulong vinyl_meta_map_obj_id;
629 : ulong vinyl_meta_pool_obj_id;
630 : ulong vinyl_line_max;
631 : ulong vinyl_data_obj_id;
632 : char vinyl_bstream_path[ PATH_MAX ];
633 :
634 : int io_type; /* FD_VINYL_IO_TYPE_* */
635 : uint uring_depth;
636 : } vinyl;
637 :
638 : struct {
639 : ulong capture_start_slot;
640 : char solcap_capture[ PATH_MAX ];
641 : int recent_only;
642 : ulong recent_slots_per_file;
643 : } solcap;
644 : };
645 : };
646 :
647 : typedef struct fd_topo_tile fd_topo_tile_t;
648 :
649 : typedef struct {
650 : ulong id;
651 : char name[ 13UL ]; /* object type */
652 : ulong wksp_id;
653 :
654 : /* Optional label for object */
655 : char label[ 13UL ]; /* object label */
656 : ulong label_idx; /* index of object for this label (ULONG_MAX if not labelled) */
657 :
658 : ulong offset;
659 : ulong footprint;
660 : } fd_topo_obj_t;
661 :
662 : /* An fd_topo_t represents the overall structure of a Firedancer
663 : configuration, describing all the workspaces, tiles, and links
664 : between them. */
665 : struct fd_topo {
666 : char app_name[ 256UL ];
667 : uchar props[ 16384UL ];
668 :
669 : ulong wksp_cnt;
670 : ulong link_cnt;
671 : ulong tile_cnt;
672 : ulong obj_cnt;
673 :
674 : fd_topo_wksp_t workspaces[ FD_TOPO_MAX_WKSPS ];
675 : fd_topo_link_t links[ FD_TOPO_MAX_LINKS ];
676 : fd_topo_tile_t tiles[ FD_TOPO_MAX_TILES ];
677 : fd_topo_obj_t objs[ FD_TOPO_MAX_OBJS ];
678 :
679 : ulong agave_affinity_cnt;
680 : ulong agave_affinity_cpu_idx[ FD_TILE_MAX ];
681 :
682 : ulong max_page_size; /* 2^21 or 2^30 */
683 : ulong gigantic_page_threshold; /* see [hugetlbfs.gigantic_page_threshold_mib]*/
684 : };
685 : typedef struct fd_topo fd_topo_t;
686 :
687 : typedef struct {
688 : char const * name;
689 :
690 : int keep_host_networking;
691 : int allow_connect;
692 : int allow_renameat;
693 : ulong rlimit_file_cnt;
694 : ulong rlimit_address_space;
695 : ulong rlimit_data;
696 : int for_tpool;
697 :
698 : ulong (*populate_allowed_seccomp)( fd_topo_t const * topo, fd_topo_tile_t const * tile, ulong out_cnt, struct sock_filter * out );
699 : ulong (*populate_allowed_fds )( fd_topo_t const * topo, fd_topo_tile_t const * tile, ulong out_fds_sz, int * out_fds );
700 : ulong (*scratch_align )( void );
701 : ulong (*scratch_footprint )( fd_topo_tile_t const * tile );
702 : ulong (*loose_footprint )( fd_topo_tile_t const * tile );
703 : void (*privileged_init )( fd_topo_t * topo, fd_topo_tile_t * tile );
704 : void (*unprivileged_init )( fd_topo_t * topo, fd_topo_tile_t * tile );
705 : void (*run )( fd_topo_t * topo, fd_topo_tile_t * tile );
706 : ulong (*rlimit_file_cnt_fn )( fd_topo_t const * topo, fd_topo_tile_t const * tile );
707 : } fd_topo_run_tile_t;
708 :
709 : struct fd_topo_obj_callbacks {
710 : char const * name;
711 : ulong (* footprint )( fd_topo_t const * topo, fd_topo_obj_t const * obj );
712 : ulong (* align )( fd_topo_t const * topo, fd_topo_obj_t const * obj );
713 : ulong (* loose )( fd_topo_t const * topo, fd_topo_obj_t const * obj );
714 : void (* new )( fd_topo_t const * topo, fd_topo_obj_t const * obj );
715 : };
716 :
717 : typedef struct fd_topo_obj_callbacks fd_topo_obj_callbacks_t;
718 :
719 : FD_PROTOTYPES_BEGIN
720 :
721 : FD_FN_CONST static inline ulong
722 0 : fd_topo_workspace_align( void ) {
723 : /* This needs to be the max( align ) of all the child members that
724 : could be aligned into this workspace, otherwise our footprint
725 : calculation will not be correct. For now just set to 4096 but this
726 : should probably be calculated dynamically, or we should reduce
727 : those child aligns if we can. */
728 0 : return 4096UL;
729 0 : }
730 :
731 : void *
732 : fd_topo_obj_laddr( fd_topo_t const * topo,
733 : ulong obj_id );
734 :
735 : /* Returns a pointer in the local address space to the base address of
736 : the workspace out of which the given object was allocated. */
737 :
738 : static inline void *
739 : fd_topo_obj_wksp_base( fd_topo_t const * topo,
740 0 : ulong obj_id ) {
741 0 : FD_TEST( obj_id<FD_TOPO_MAX_OBJS );
742 0 : fd_topo_obj_t const * obj = &topo->objs[ obj_id ];
743 0 : FD_TEST( obj->id == obj_id );
744 0 : ulong const wksp_id = obj->wksp_id;
745 :
746 0 : FD_TEST( wksp_id<FD_TOPO_MAX_WKSPS );
747 0 : fd_topo_wksp_t const * wksp = &topo->workspaces[ wksp_id ];
748 0 : FD_TEST( wksp->id == wksp_id );
749 0 : return wksp->wksp;
750 0 : }
751 :
752 : FD_FN_PURE static inline ulong
753 : fd_topo_tile_name_cnt( fd_topo_t const * topo,
754 3 : char const * name ) {
755 3 : ulong cnt = 0;
756 6 : for( ulong i=0; i<topo->tile_cnt; i++ ) {
757 3 : if( FD_UNLIKELY( !strcmp( topo->tiles[ i ].name, name ) ) ) cnt++;
758 3 : }
759 3 : return cnt;
760 3 : }
761 :
762 : /* Finds the workspace of a given name in the topology. Returns
763 : ULONG_MAX if there is no such workspace. There can be at most one
764 : workspace of a given name. */
765 :
766 : FD_FN_PURE static inline ulong
767 : fd_topo_find_wksp( fd_topo_t const * topo,
768 66 : char const * name ) {
769 66 : for( ulong i=0; i<topo->wksp_cnt; i++ ) {
770 66 : if( FD_UNLIKELY( !strcmp( topo->workspaces[ i ].name, name ) ) ) return i;
771 66 : }
772 0 : return ULONG_MAX;
773 66 : }
774 :
775 : /* Find the tile of a given name and kind_id in the topology, there will
776 : be at most one such tile, since kind_id is unique among the name.
777 : Returns ULONG_MAX if there is no such tile. */
778 :
779 : FD_FN_PURE static inline ulong
780 : fd_topo_find_tile( fd_topo_t const * topo,
781 : char const * name,
782 21 : ulong kind_id ) {
783 21 : for( ulong i=0; i<topo->tile_cnt; i++ ) {
784 21 : if( FD_UNLIKELY( !strcmp( topo->tiles[ i ].name, name ) ) && topo->tiles[ i ].kind_id == kind_id ) return i;
785 21 : }
786 0 : return ULONG_MAX;
787 21 : }
788 :
789 : /* Find the link of a given name and kind_id in the topology, there will
790 : be at most one such link, since kind_id is unique among the name.
791 : Returns ULONG_MAX if there is no such link. */
792 :
793 : FD_FN_PURE static inline ulong
794 : fd_topo_find_link( fd_topo_t const * topo,
795 : char const * name,
796 18 : ulong kind_id ) {
797 39 : for( ulong i=0; i<topo->link_cnt; i++ ) {
798 39 : if( FD_UNLIKELY( !strcmp( topo->links[ i ].name, name ) ) && topo->links[ i ].kind_id == kind_id ) return i;
799 39 : }
800 0 : return ULONG_MAX;
801 18 : }
802 :
803 : FD_FN_PURE static inline ulong
804 : fd_topo_find_tile_in_link( fd_topo_t const * topo,
805 : fd_topo_tile_t const * tile,
806 : char const * name,
807 0 : ulong kind_id ) {
808 0 : for( ulong i=0; i<tile->in_cnt; i++ ) {
809 0 : if( FD_UNLIKELY( !strcmp( topo->links[ tile->in_link_id[ i ] ].name, name ) )
810 0 : && topo->links[ tile->in_link_id[ i ] ].kind_id == kind_id ) return i;
811 0 : }
812 0 : return ULONG_MAX;
813 0 : }
814 :
815 : FD_FN_PURE static inline ulong
816 : fd_topo_find_tile_out_link( fd_topo_t const * topo,
817 : fd_topo_tile_t const * tile,
818 : char const * name,
819 0 : ulong kind_id ) {
820 0 : for( ulong i=0; i<tile->out_cnt; i++ ) {
821 0 : if( FD_UNLIKELY( !strcmp( topo->links[ tile->out_link_id[ i ] ].name, name ) )
822 0 : && topo->links[ tile->out_link_id[ i ] ].kind_id == kind_id ) return i;
823 0 : }
824 0 : return ULONG_MAX;
825 0 : }
826 :
827 : /* Find the id of the tile which is a producer for the given link. If
828 : no tile is a producer for the link, returns ULONG_MAX. This should
829 : not be possible for a well formed and validated topology. */
830 : FD_FN_PURE static inline ulong
831 : fd_topo_find_link_producer( fd_topo_t const * topo,
832 0 : fd_topo_link_t const * link ) {
833 0 : for( ulong i=0; i<topo->tile_cnt; i++ ) {
834 0 : fd_topo_tile_t const * tile = &topo->tiles[ i ];
835 :
836 0 : for( ulong j=0; j<tile->out_cnt; j++ ) {
837 0 : if( FD_UNLIKELY( tile->out_link_id[ j ] == link->id ) ) return i;
838 0 : }
839 0 : }
840 0 : return ULONG_MAX;
841 0 : }
842 :
843 : /* Given a link, count the number of consumers of that link among all
844 : the tiles in the topology. */
845 : FD_FN_PURE static inline ulong
846 : fd_topo_link_consumer_cnt( fd_topo_t const * topo,
847 0 : fd_topo_link_t const * link ) {
848 0 : ulong cnt = 0;
849 0 : for( ulong i=0; i<topo->tile_cnt; i++ ) {
850 0 : fd_topo_tile_t const * tile = &topo->tiles[ i ];
851 0 : for( ulong j=0; j<tile->in_cnt; j++ ) {
852 0 : if( FD_UNLIKELY( tile->in_link_id[ j ] == link->id ) ) cnt++;
853 0 : }
854 0 : }
855 :
856 0 : return cnt;
857 0 : }
858 :
859 : /* Given a link, count the number of reliable consumers of that link
860 : among all the tiles in the topology. */
861 : FD_FN_PURE static inline ulong
862 : fd_topo_link_reliable_consumer_cnt( fd_topo_t const * topo,
863 0 : fd_topo_link_t const * link ) {
864 0 : ulong cnt = 0;
865 0 : for( ulong i=0; i<topo->tile_cnt; i++ ) {
866 0 : fd_topo_tile_t const * tile = &topo->tiles[ i ];
867 0 : for( ulong j=0; j<tile->in_cnt; j++ ) {
868 0 : if( FD_UNLIKELY( tile->in_link_id[ j ] == link->id && tile->in_link_reliable[ j ] ) ) cnt++;
869 0 : }
870 0 : }
871 :
872 0 : return cnt;
873 0 : }
874 :
875 : FD_FN_PURE static inline ulong
876 : fd_topo_tile_consumer_cnt( fd_topo_t const * topo,
877 0 : fd_topo_tile_t const * tile ) {
878 0 : (void)topo;
879 0 : return tile->out_cnt;
880 0 : }
881 :
882 : FD_FN_PURE static inline ulong
883 : fd_topo_tile_reliable_consumer_cnt( fd_topo_t const * topo,
884 0 : fd_topo_tile_t const * tile ) {
885 0 : ulong reliable_cons_cnt = 0UL;
886 0 : for( ulong i=0UL; i<topo->tile_cnt; i++ ) {
887 0 : fd_topo_tile_t const * consumer_tile = &topo->tiles[ i ];
888 0 : for( ulong j=0UL; j<consumer_tile->in_cnt; j++ ) {
889 0 : for( ulong k=0UL; k<tile->out_cnt; k++ ) {
890 0 : if( FD_UNLIKELY( consumer_tile->in_link_id[ j ]==tile->out_link_id[ k ] && consumer_tile->in_link_reliable[ j ] ) ) {
891 0 : reliable_cons_cnt++;
892 0 : }
893 0 : }
894 0 : }
895 0 : }
896 0 : return reliable_cons_cnt;
897 0 : }
898 :
899 : FD_FN_PURE static inline ulong
900 : fd_topo_tile_producer_cnt( fd_topo_t const * topo,
901 0 : fd_topo_tile_t const * tile ) {
902 0 : (void)topo;
903 0 : ulong in_cnt = 0UL;
904 0 : for( ulong i=0UL; i<tile->in_cnt; i++ ) {
905 0 : if( FD_UNLIKELY( !tile->in_link_poll[ i ] ) ) continue;
906 0 : in_cnt++;
907 0 : }
908 0 : return in_cnt;
909 0 : }
910 :
911 : FD_FN_PURE FD_FN_UNUSED static ulong
912 : fd_topo_obj_cnt( fd_topo_t const * topo,
913 : char const * obj_type,
914 0 : char const * label ) {
915 0 : ulong cnt = 0UL;
916 0 : for( ulong i=0UL; i<topo->obj_cnt; i++ ) {
917 0 : fd_topo_obj_t const * obj = &topo->objs[ i ];
918 0 : if( strncmp( obj->name, obj_type, sizeof(obj->name) ) ) continue;
919 0 : if( label &&
920 0 : strncmp( obj->label, label, sizeof(obj->label) ) ) continue;
921 0 : cnt++;
922 0 : }
923 0 : return cnt;
924 0 : }
925 :
926 : FD_FN_PURE FD_FN_UNUSED static fd_topo_obj_t const *
927 : fd_topo_find_obj( fd_topo_t const * topo,
928 : char const * obj_type,
929 : char const * label,
930 0 : ulong label_idx ) {
931 0 : for( ulong i=0UL; i<topo->obj_cnt; i++ ) {
932 0 : fd_topo_obj_t const * obj = &topo->objs[ i ];
933 0 : if( strncmp( obj->name, obj_type, sizeof(obj->name) ) ) continue;
934 0 : if( label &&
935 0 : strncmp( obj->label, label, sizeof(obj->label) ) ) continue;
936 0 : if( label_idx != ULONG_MAX && obj->label_idx != label_idx ) continue;
937 0 : return obj;
938 0 : }
939 0 : return NULL;
940 0 : }
941 :
942 : FD_FN_PURE FD_FN_UNUSED static fd_topo_obj_t const *
943 : fd_topo_find_tile_obj( fd_topo_t const * topo,
944 : fd_topo_tile_t const * tile,
945 0 : char const * obj_type ) {
946 0 : for( ulong i=0UL; i<(tile->uses_obj_cnt); i++ ) {
947 0 : fd_topo_obj_t const * obj = &topo->objs[ tile->uses_obj_id[ i ] ];
948 0 : if( strncmp( obj->name, obj_type, sizeof(obj->name) ) ) continue;
949 0 : return obj;
950 0 : }
951 0 : return NULL;
952 0 : }
953 :
954 : /* Join (map into the process) all shared memory (huge/gigantic pages)
955 : needed by the tile, in the given topology. All memory associated
956 : with the tile (aka. used by links that the tile either produces to or
957 : consumes from, or used by the tile itself for its cnc) will be
958 : attached (mapped into the process).
959 :
960 : This is needed to play nicely with the sandbox. Once a process is
961 : sandboxed we can no longer map any memory. */
962 : void
963 : fd_topo_join_tile_workspaces( fd_topo_t * topo,
964 : fd_topo_tile_t * tile,
965 : int core_dump_level );
966 :
967 : /* Join (map into the process) the shared memory (huge/gigantic pages)
968 : for the given workspace. Mode is one of
969 : FD_SHMEM_JOIN_MODE_READ_WRITE or FD_SHMEM_JOIN_MODE_READ_ONLY and
970 : determines the prot argument that will be passed to mmap when mapping
971 : the pages in (PROT_WRITE or PROT_READ respectively).
972 :
973 : Dump should be set to 1 if the workspace memory should be dumpable
974 : when the process crashes, or 0 if not. */
975 : void
976 : fd_topo_join_workspace( fd_topo_t * topo,
977 : fd_topo_wksp_t * wksp,
978 : int mode,
979 : int dump );
980 :
981 : /* Join (map into the process) all shared memory (huge/gigantic pages)
982 : needed by all tiles in the topology. Mode is one of
983 : FD_SHMEM_JOIN_MODE_READ_WRITE or FD_SHMEM_JOIN_MODE_READ_ONLY and
984 : determines the prot argument that will be passed to mmap when
985 : mapping the pages in (PROT_WRITE or PROT_READ respectively). */
986 : void
987 : fd_topo_join_workspaces( fd_topo_t * topo,
988 : int mode,
989 : int core_dump_level );
990 :
991 : /* Leave (unmap from the process) the shared memory needed for the
992 : given workspace in the topology, if it was previously mapped.
993 :
994 : topo and wksp are assumed non-NULL. It is OK if the workspace
995 : has not been previously joined, in which case this is a no-op. */
996 :
997 : void
998 : fd_topo_leave_workspace( fd_topo_t * topo,
999 : fd_topo_wksp_t * wksp );
1000 :
1001 : /* Leave (unmap from the process) all shared memory needed by all
1002 : tiles in the topology, if each of them was mapped.
1003 :
1004 : topo is assumed non-NULL. Only workspaces which were previously
1005 : joined are unmapped. */
1006 :
1007 : void
1008 : fd_topo_leave_workspaces( fd_topo_t * topo );
1009 :
1010 : /* Create the given workspace needed by the topology on the system.
1011 : This does not "join" the workspaces (map their memory into the
1012 : process), but only creates the .wksp file and formats it correctly
1013 : as a workspace.
1014 :
1015 : Returns 0 on success and -1 on failure, with errno set to the error.
1016 : The only reason for failure currently that will be returned is
1017 : ENOMEM, as other unexpected errors will cause the program to exit.
1018 :
1019 : If update_existing is 1, the workspace will not be created from
1020 : scratch but it will be assumed that it already exists from a prior
1021 : run and needs to be maybe resized and then have the header
1022 : structures reinitialized. This can save a very expensive operation
1023 : of zeroing all of the workspace pages. This is dangerous in
1024 : production because it can leave stray memory from prior runs around,
1025 : and should only be used in development environments. */
1026 :
1027 : int
1028 : fd_topo_create_workspace( fd_topo_t * topo,
1029 : fd_topo_wksp_t * wksp,
1030 : int update_existing );
1031 :
1032 : /* Join the standard IPC objects needed by the topology of this particular
1033 : tile */
1034 :
1035 : void
1036 : fd_topo_fill_tile( fd_topo_t * topo,
1037 : fd_topo_tile_t * tile );
1038 :
1039 : /* Same as fd_topo_fill_tile but fills in all the objects for a
1040 : particular workspace with the given mode. */
1041 : void
1042 : fd_topo_workspace_fill( fd_topo_t * topo,
1043 : fd_topo_wksp_t * wksp );
1044 :
1045 : /* Apply a new function to every object that is resident in the given
1046 : workspace in the topology. */
1047 :
1048 : void
1049 : fd_topo_wksp_new( fd_topo_t const * topo,
1050 : fd_topo_wksp_t const * wksp,
1051 : fd_topo_obj_callbacks_t ** callbacks );
1052 :
1053 : /* Same as fd_topo_fill_tile but fills in all tiles in the topology. */
1054 :
1055 : void
1056 : fd_topo_fill( fd_topo_t * topo );
1057 :
1058 : /* fd_topo_tile_stack_join joins a huge page optimized stack for the
1059 : provided tile. The stack is assumed to already exist at a known
1060 : path in the hugetlbfs mount. */
1061 :
1062 : void *
1063 : fd_topo_tile_stack_join( char const * app_name,
1064 : char const * tile_name,
1065 : ulong tile_kind_id );
1066 :
1067 : /* fd_topo_run_single_process runs all the tiles in a single process
1068 : (the calling process). This spawns a thread for each tile, switches
1069 : that thread to the given UID and GID and then runs the tile in it.
1070 : Each thread will never exit, as tiles are expected to run forever.
1071 : An error is logged and the application will exit if a tile exits.
1072 : The function itself does return after spawning all the threads.
1073 :
1074 : The threads will not be sandboxed in any way, except switching to the
1075 : provided UID and GID, so they will share the same address space, and
1076 : not have any seccomp restrictions or use any Linux namespaces. The
1077 : calling thread will also switch to the provided UID and GID before
1078 : it returns.
1079 :
1080 : In production, when running with an Agave child process this is
1081 : used for spawning certain tiles inside the Agave address space.
1082 : It's also useful for tooling and debugging, but is not how the main
1083 : production Firedancer process runs. For production, each tile is run
1084 : in its own address space with a separate process and full security
1085 : sandbox.
1086 :
1087 : The agave argument determines which tiles are started. If the
1088 : argument is 0 or 1, only non-agave (or only agave) tiles are started.
1089 : If the argument is any other value, all tiles in the topology are
1090 : started regardless of if they are Agave tiles or not. */
1091 :
1092 : void
1093 : fd_topo_run_single_process( fd_topo_t * topo,
1094 : int agave,
1095 : uint uid,
1096 : uint gid,
1097 : fd_topo_run_tile_t (* tile_run )( fd_topo_tile_t const * tile ) );
1098 :
1099 : /* fd_topo_run_tile runs the given tile directly within the current
1100 : process (and thread). The function will never return, as tiles are
1101 : expected to run forever. An error is logged and the application will
1102 : exit if the tile exits.
1103 :
1104 : The sandbox argument determines if the current process will be
1105 : sandboxed fully before starting the tile. The thread will switch to
1106 : the UID and GID provided before starting the tile, even if the thread
1107 : is not being sandboxed. Although POSIX specifies that all threads in
1108 : a process must share a UID and GID, this is not the case on Linux.
1109 : The thread will switch to the provided UID and GID without switching
1110 : the other threads in the process.
1111 :
1112 : If keep_controlling_terminal is set to 0, and the sandbox is enabled
1113 : the controlling terminal will be detached as an additional sandbox
1114 : measure, but you will not be able to send Ctrl+C or other signals
1115 : from the terminal. See fd_sandbox.h for more information.
1116 :
1117 : The allow_fd argument is only used if sandbox is true, and is a file
1118 : descriptor which will be allowed to exist in the process. Normally
1119 : the sandbox code rejects and aborts if there is an unexpected file
1120 : descriptor present on boot. This is helpful to allow a parent
1121 : process to be notified on termination of the tile by waiting for a
1122 : pipe file descriptor to get closed.
1123 :
1124 : wait and debugger are both used in debugging. If wait is non-NULL,
1125 : the runner will wait until the value pointed to by wait is non-zero
1126 : before launching the tile. Likewise, if debugger is non-NULL, the
1127 : runner will wait until a debugger is attached before setting the
1128 : value pointed to by debugger to non-zero. These are intended to be
1129 : used as a pair, where many tiles share a waiting reference, and then
1130 : one of the tiles (a tile you want to attach the debugger to) has the
1131 : same reference provided as the debugger, so all tiles will stop and
1132 : wait for the debugger to attach to it before proceeding. */
1133 :
1134 : void
1135 : fd_topo_run_tile( fd_topo_t * topo,
1136 : fd_topo_tile_t * tile,
1137 : int sandbox,
1138 : int keep_controlling_terminal,
1139 : int dumpable,
1140 : uint uid,
1141 : uint gid,
1142 : int allow_fd,
1143 : volatile int * wait,
1144 : volatile int * debugger,
1145 : fd_topo_run_tile_t * tile_run );
1146 :
1147 : /* This is for determining the value of RLIMIT_MLOCK that we need to
1148 : successfully run all tiles in separate processes. The value returned
1149 : is the maximum amount of memory that will be locked with mlock() by
1150 : any individual process in the tree. Specifically, if we have three
1151 : tile processes, and they each need to lock 5, 9, and 2 MiB of memory
1152 : respectively, RLIMIT_MLOCK needs to be 9 MiB to allow all three
1153 : process mlock() calls to succeed.
1154 :
1155 : Tiles lock memory in three ways. Any workspace they are using, they
1156 : lock the entire workspace. Then each tile uses huge pages for the
1157 : stack which are also locked, and finally some tiles use private
1158 : locked mmaps outside the workspace for storing key material. The
1159 : results here include all of this memory together.
1160 :
1161 : The result is not necessarily the amount of memory used by the tile
1162 : process, although it will be quite close. Tiles could potentially
1163 : allocate memory (eg, with brk) without needing to lock it, which
1164 : would not need to included, and some kernel memory that tiles cause
1165 : to be allocated (for example XSK buffers) is also not included. The
1166 : actual amount of memory used will not be less than this value. */
1167 : FD_FN_PURE ulong
1168 : fd_topo_mlock_max_tile( fd_topo_t const * topo );
1169 :
1170 : /* Same as fd_topo_mlock_max_tile, but for loading the entire topology
1171 : into one process, rather than a separate process per tile. This is
1172 : used, for example, by the configuration code when it creates all the
1173 : workspaces, or the monitor that maps the entire system into one
1174 : address space. */
1175 : FD_FN_PURE ulong
1176 : fd_topo_mlock( fd_topo_t const * topo );
1177 :
1178 : /* This returns the number of gigantic pages needed by the topology on
1179 : the provided numa node. It includes pages needed by the workspaces,
1180 : as well as additional allocations like huge pages for process stacks
1181 : and private key storage. */
1182 :
1183 : FD_FN_PURE ulong
1184 : fd_topo_gigantic_page_cnt( fd_topo_t const * topo,
1185 : ulong numa_idx );
1186 :
1187 : /* This returns the number of huge pages in the application needed by
1188 : the topology on the provided numa node. It includes pages needed by
1189 : things placed in the hugetlbfs (workspaces, process stacks). If
1190 : include_anonymous is true, it also includes anonymous hugepages which
1191 : are needed but are not placed in the hugetlbfs. */
1192 :
1193 : FD_FN_PURE ulong
1194 : fd_topo_huge_page_cnt( fd_topo_t const * topo,
1195 : ulong numa_idx,
1196 : int include_anonymous );
1197 :
1198 : /* Prints a message describing the topology to an output stream. If
1199 : stdout is true, will be written to stdout, otherwise will be written
1200 : as a NOTICE log message to the log file. */
1201 : void
1202 : fd_topo_print_log( int stdout,
1203 : fd_topo_t * topo );
1204 :
1205 : FD_PROTOTYPES_END
1206 :
1207 : #endif /* HEADER_fd_src_disco_topo_fd_topo_h */
|
