Line data Source code
1 : #include "fd_snapct_tile.h"
2 : #include "utils/fd_ssping.h"
3 : #include "utils/fd_ssctrl.h"
4 : #include "utils/fd_ssarchive.h"
5 : #include "utils/fd_http_resolver.h"
6 : #include "utils/fd_ssmsg.h"
7 :
8 : #include "../../disco/topo/fd_topo.h"
9 : #include "../../disco/metrics/fd_metrics.h"
10 : #include "../../flamenco/gossip/fd_gossip_types.h"
11 : #include "../../waltz/openssl/fd_openssl_tile.h"
12 :
13 : #include <errno.h>
14 : #include <stdio.h>
15 : #include <fcntl.h>
16 : #include <unistd.h>
17 : #include <sys/stat.h>
18 : #include <netinet/tcp.h>
19 : #include <netinet/in.h>
20 :
21 : #include "generated/fd_snapct_tile_seccomp.h"
22 :
23 : #define NAME "snapct"
24 :
25 : /* FIXME: Implement full_effective_age_cancel_threshold */
26 : /* FIXME: Implement min_speed_mib and other download health logic */
27 : /* FIXME: Implement max_retry_abort and retry logic in general */
28 : /* FIXME: Add more timeout config options and have consistent behavior */
29 : /* FIXME: Do a finishing pass over the default.toml config options / comments */
30 : /* FIXME: Improve behavior when using incremental_snapshots = false */
31 : /* FIXME: Handle cases where no explicitly allowed peers advertise RPC */
32 : /* FIXME: Make the code more strict about duplicate IP:port's */
33 : /* FIXME: Handle cases where the slot number we start downloading differs from advertised */
34 :
35 0 : #define GOSSIP_PEERS_MAX (FD_CONTACT_INFO_TABLE_SIZE)
36 0 : #define SERVER_PEERS_MAX (FD_TOPO_SNAPSHOTS_SERVERS_MAX_RESOLVED)
37 0 : #define TOTAL_PEERS_MAX (GOSSIP_PEERS_MAX + SERVER_PEERS_MAX)
38 :
39 0 : #define IN_KIND_ACK (0)
40 0 : #define IN_KIND_SNAPLD (1)
41 0 : #define IN_KIND_GOSSIP (2)
42 : #define MAX_IN_LINKS (3)
43 :
44 0 : #define TEMP_FULL_SNAP_NAME ".snapshot.tar.bz2-partial"
45 0 : #define TEMP_INCR_SNAP_NAME ".incremental-snapshot.tar.bz2-partial"
46 :
47 : struct fd_snapct_out_link {
48 : ulong idx;
49 : fd_wksp_t * mem;
50 : ulong chunk0;
51 : ulong wmark;
52 : ulong chunk;
53 : ulong mtu;
54 : };
55 : typedef struct fd_snapct_out_link fd_snapct_out_link_t;
56 :
57 : #define FD_SNAPCT_GOSSIP_FRESH_DEADLINE_NANOS (10L*1000L*1000L*1000L) /* gossip contact info is pushed every ~7.5 seconds */
58 0 : #define FD_SNAPCT_GOSSIP_SATURATION_CHECK_INTERVAL ( 10L*1000L*1000L)
59 0 : #define FD_SNAPCT_GOSSIP_SATURATION_THRESHOLD (0.05) /* 5% fresh peers */
60 :
61 0 : #define FD_SNAPCT_COLLECTING_PEERS_TIMEOUT (2L*60L*1000L*1000L*1000L) /* 2 minutes */
62 0 : #define FD_SNAPCT_WAITING_FOR_PEERS_TIMEOUT (2L*60L*1000L*1000L*1000L) /* 2 minutes */
63 :
64 : struct gossip_ci_entry {
65 : fd_pubkey_t pubkey;
66 : int allowed;
67 : fd_ip4_port_t rpc_addr;
68 : long added_nanos;
69 : ulong map_next;
70 : };
71 : typedef struct gossip_ci_entry gossip_ci_entry_t;
72 :
73 : #define MAP_NAME gossip_ci_map
74 0 : #define MAP_KEY pubkey
75 : #define MAP_ELE_T gossip_ci_entry_t
76 : #define MAP_KEY_T fd_pubkey_t
77 0 : #define MAP_NEXT map_next
78 0 : #define MAP_KEY_EQ(k0,k1) fd_pubkey_eq( k0, k1 )
79 0 : #define MAP_KEY_HASH(key,seed) fd_hash( seed, key, sizeof(fd_pubkey_t) )
80 : #include "../../util/tmpl/fd_map_chain.c"
81 :
82 : struct fd_snapct_tile {
83 : struct fd_topo_tile_snapct config;
84 : int gossip_enabled;
85 : int download_enabled;
86 :
87 : fd_ssping_t * ssping;
88 : fd_http_resolver_t * ssresolver;
89 : fd_sspeer_selector_t * selector;
90 :
91 : int state;
92 : int malformed;
93 : long deadline_nanos;
94 : int flush_ack;
95 : fd_ip4_port_t addr;
96 :
97 : struct {
98 : int dir_fd;
99 : int full_snapshot_fd;
100 : int incremental_snapshot_fd;
101 : } local_out;
102 :
103 : char http_full_snapshot_name[ PATH_MAX ];
104 : char http_incr_snapshot_name[ PATH_MAX ];
105 :
106 : fd_wksp_t const * gossip_in_mem;
107 : fd_wksp_t const * snapld_in_mem;
108 : uchar in_kind[ MAX_IN_LINKS ];
109 :
110 : struct {
111 : ulong full_slot;
112 : ulong slot;
113 : int dirty;
114 : } predicted_incremental;
115 :
116 : struct {
117 : ulong full_snapshot_slot;
118 : char full_snapshot_path[ PATH_MAX ];
119 : ulong full_snapshot_size;
120 : int full_snapshot_zstd;
121 :
122 : ulong incremental_snapshot_slot;
123 : char incremental_snapshot_path[ PATH_MAX ];
124 : ulong incremental_snapshot_size;
125 : int incremental_snapshot_zstd;
126 : } local_in;
127 :
128 : struct {
129 : struct {
130 : ulong bytes_read;
131 : ulong bytes_written;
132 : ulong bytes_total;
133 : uint num_retries;
134 : } full;
135 :
136 : struct {
137 : ulong bytes_read;
138 : ulong bytes_written;
139 : ulong bytes_total;
140 : uint num_retries;
141 : } incremental;
142 : } metrics;
143 :
144 : struct {
145 : gossip_ci_entry_t * ci_table; /* flat array of all gossip entries, allowed or not */
146 : gossip_ci_map_t * ci_map; /* map from pubkey to only allowed gossip entries */
147 : ulong fresh_cnt;
148 : ulong total_cnt;
149 : int saturated;
150 : long next_saturated_check;
151 : } gossip;
152 :
153 : fd_snapct_out_link_t out_ld;
154 : fd_snapct_out_link_t out_gui;
155 : fd_snapct_out_link_t out_rp;
156 : };
157 : typedef struct fd_snapct_tile fd_snapct_tile_t;
158 :
159 : static int
160 0 : gossip_enabled( fd_topo_tile_t const * tile ) {
161 0 : return tile->snapct.sources.gossip.allow_any || tile->snapct.sources.gossip.allow_list_cnt>0UL;
162 0 : }
163 :
164 : static int
165 0 : download_enabled( fd_topo_tile_t const * tile ) {
166 0 : return gossip_enabled( tile ) || tile->snapct.sources.servers_cnt>0UL;
167 0 : }
168 :
169 : FD_FN_CONST static inline ulong
170 0 : loose_footprint( fd_topo_tile_t const * tile ) {
171 0 : (void)tile;
172 : /* Leftover space for OpenSSL allocations */
173 0 : return 1<<26UL; /* 64 MiB */
174 0 : }
175 :
176 : static ulong
177 0 : scratch_align( void ) {
178 0 : return fd_ulong_max( alignof(fd_snapct_tile_t),
179 0 : fd_ulong_max( fd_ssping_align(),
180 0 : fd_ulong_max( alignof(gossip_ci_entry_t),
181 0 : fd_ulong_max( gossip_ci_map_align(),
182 0 : fd_ulong_max( fd_http_resolver_align(),
183 0 : fd_sspeer_selector_align() ) ) ) ) );
184 0 : }
185 :
186 : static ulong
187 0 : scratch_footprint( fd_topo_tile_t const * tile FD_PARAM_UNUSED ) {
188 0 : ulong l = FD_LAYOUT_INIT;
189 0 : l = FD_LAYOUT_APPEND( l, alignof(fd_snapct_tile_t), sizeof(fd_snapct_tile_t) );
190 0 : l = FD_LAYOUT_APPEND( l, fd_ssping_align(), fd_ssping_footprint( TOTAL_PEERS_MAX ) );
191 0 : l = FD_LAYOUT_APPEND( l, alignof(gossip_ci_entry_t), sizeof(gossip_ci_entry_t) * GOSSIP_PEERS_MAX );
192 0 : l = FD_LAYOUT_APPEND( l, gossip_ci_map_align(), gossip_ci_map_footprint( gossip_ci_map_chain_cnt_est( GOSSIP_PEERS_MAX ) ) );
193 0 : l = FD_LAYOUT_APPEND( l, fd_http_resolver_align(), fd_http_resolver_footprint( SERVER_PEERS_MAX ) );
194 0 : l = FD_LAYOUT_APPEND( l, fd_sspeer_selector_align(), fd_sspeer_selector_footprint( TOTAL_PEERS_MAX ) );
195 0 : l = FD_LAYOUT_APPEND( l, fd_alloc_align(), fd_alloc_footprint() );
196 0 : return FD_LAYOUT_FINI( l, scratch_align() );
197 0 : }
198 :
199 : static inline int
200 0 : should_shutdown( fd_snapct_tile_t * ctx ) {
201 0 : return ctx->state==FD_SNAPCT_STATE_SHUTDOWN;
202 0 : }
203 :
204 : static void
205 0 : during_housekeeping( fd_snapct_tile_t * ctx ) {
206 0 : long now = fd_log_wallclock();
207 :
208 0 : if( FD_UNLIKELY( !ctx->gossip.saturated && now>ctx->gossip.next_saturated_check ) ) {
209 0 : ctx->gossip.next_saturated_check = now + FD_SNAPCT_GOSSIP_SATURATION_CHECK_INTERVAL;
210 :
211 0 : ulong fresh_cnt = 0UL;
212 0 : ulong total_cnt = 0UL;
213 0 : for( gossip_ci_map_iter_t iter = gossip_ci_map_iter_init( ctx->gossip.ci_map, ctx->gossip.ci_table );
214 0 : !gossip_ci_map_iter_done( iter, ctx->gossip.ci_map, ctx->gossip.ci_table );
215 0 : iter = gossip_ci_map_iter_next( iter, ctx->gossip.ci_map, ctx->gossip.ci_table ) ) {
216 0 : gossip_ci_entry_t const * ci_entry = gossip_ci_map_iter_ele_const( iter, ctx->gossip.ci_map, ctx->gossip.ci_table );
217 0 : if( FD_UNLIKELY( ci_entry->added_nanos>(now-FD_SNAPCT_GOSSIP_FRESH_DEADLINE_NANOS) ) ) fresh_cnt++;
218 0 : total_cnt++;
219 0 : }
220 0 : ctx->gossip.fresh_cnt = fresh_cnt;
221 0 : ctx->gossip.total_cnt = total_cnt;
222 :
223 0 : if( total_cnt!=0UL && total_cnt==ctx->config.sources.gossip.allow_list_cnt ) ctx->gossip.saturated = 1;
224 0 : else {
225 0 : double fresh = total_cnt ? (double)fresh_cnt/(double)total_cnt : 1.0;
226 0 : ctx->gossip.saturated = fresh<FD_SNAPCT_GOSSIP_SATURATION_THRESHOLD;
227 0 : }
228 0 : }
229 0 : }
230 :
231 : static void
232 0 : metrics_write( fd_snapct_tile_t * ctx ) {
233 : /* FIXME: Track/report FULL_NUM_RETRIES & INCREMENTAL_NUM_RETRIES */
234 :
235 0 : FD_MGAUGE_SET( SNAPCT, FULL_BYTES_READ, ctx->metrics.full.bytes_read );
236 0 : FD_MGAUGE_SET( SNAPCT, FULL_BYTES_WRITTEN, ctx->metrics.full.bytes_written );
237 0 : FD_MGAUGE_SET( SNAPCT, FULL_BYTES_TOTAL, ctx->metrics.full.bytes_total );
238 0 : FD_MGAUGE_SET( SNAPCT, FULL_DOWNLOAD_RETRIES, ctx->metrics.full.num_retries );
239 :
240 0 : FD_MGAUGE_SET( SNAPCT, INCREMENTAL_BYTES_READ, ctx->metrics.incremental.bytes_read );
241 0 : FD_MGAUGE_SET( SNAPCT, INCREMENTAL_BYTES_WRITTEN, ctx->metrics.incremental.bytes_written );
242 0 : FD_MGAUGE_SET( SNAPCT, INCREMENTAL_BYTES_TOTAL, ctx->metrics.incremental.bytes_total );
243 0 : FD_MGAUGE_SET( SNAPCT, INCREMENTAL_DOWNLOAD_RETRIES, ctx->metrics.incremental.num_retries );
244 :
245 0 : FD_MGAUGE_SET( SNAPCT, GOSSIP_FRESH_COUNT, ctx->gossip.fresh_cnt );
246 0 : FD_MGAUGE_SET( SNAPCT, GOSSIP_TOTAL_COUNT, ctx->gossip.total_cnt );
247 :
248 0 : FD_MGAUGE_SET( SNAPCT, PREDICTED_SLOT, ctx->predicted_incremental.slot );
249 :
250 0 : #if FD_HAS_OPENSSL
251 0 : FD_MCNT_SET( SNAPCT, SSL_ALLOC_ERRORS, fd_ossl_alloc_errors );
252 0 : #endif
253 :
254 0 : FD_MGAUGE_SET( SNAPCT, STATE, (ulong)ctx->state );
255 0 : }
256 :
257 : static void
258 : snapshot_path_gui_publish( fd_snapct_tile_t * ctx,
259 : fd_stem_context_t * stem,
260 : char const * path,
261 0 : int is_full ) {
262 : /* FIXME: Consider whether we can get everything we need from metrics
263 : rather than creating an entire link for this rare message */
264 0 : fd_snapct_update_t * out = fd_chunk_to_laddr( ctx->out_gui.mem, ctx->out_gui.chunk );
265 0 : FD_TEST( fd_cstr_printf_check( out->read_path, PATH_MAX, NULL, "%s", path ) );
266 0 : out->is_download = 0;
267 0 : out->type = fd_int_if( is_full, FD_SNAPCT_SNAPSHOT_TYPE_FULL, FD_SNAPCT_SNAPSHOT_TYPE_INCREMENTAL );
268 0 : fd_stem_publish( stem, ctx->out_gui.idx, 0UL, ctx->out_gui.chunk, sizeof(fd_snapct_update_t) , 0UL, 0UL, 0UL );
269 0 : ctx->out_gui.chunk = fd_dcache_compact_next( ctx->out_gui.chunk, sizeof(fd_snapct_update_t), ctx->out_gui.chunk0, ctx->out_gui.wmark );
270 0 : }
271 :
272 : static void
273 0 : predict_incremental( fd_snapct_tile_t * ctx ) {
274 0 : if( FD_UNLIKELY( !ctx->config.incremental_snapshots ) ) return;
275 0 : if( FD_UNLIKELY( ctx->predicted_incremental.full_slot==ULONG_MAX ) ) return;
276 :
277 0 : fd_sspeer_t best = fd_sspeer_selector_best( ctx->selector, 1, ctx->predicted_incremental.full_slot );
278 :
279 0 : if( FD_LIKELY( best.addr.l ) ) {
280 0 : if( FD_UNLIKELY( ctx->predicted_incremental.slot!=best.ssinfo.incremental.slot ) ) {
281 0 : ctx->predicted_incremental.slot = best.ssinfo.incremental.slot;
282 0 : ctx->predicted_incremental.dirty = 1;
283 0 : }
284 0 : }
285 0 : }
286 :
287 : static void
288 : on_resolve( void * _ctx,
289 : fd_ip4_port_t addr,
290 0 : fd_ssinfo_t const * ssinfo ) {
291 0 : fd_snapct_tile_t * ctx = (fd_snapct_tile_t *)_ctx;
292 :
293 0 : fd_sspeer_selector_add( ctx->selector, addr, ULONG_MAX, ssinfo );
294 0 : fd_sspeer_selector_process_cluster_slot( ctx->selector, ssinfo->full.slot, ssinfo->incremental.slot );
295 0 : predict_incremental( ctx );
296 0 : }
297 :
298 : static void
299 : on_ping( void * _ctx,
300 : fd_ip4_port_t addr,
301 0 : ulong latency ) {
302 0 : fd_snapct_tile_t * ctx = (fd_snapct_tile_t *)_ctx;
303 :
304 0 : fd_sspeer_selector_add( ctx->selector, addr, latency, NULL );
305 0 : predict_incremental( ctx );
306 0 : }
307 :
308 : static void
309 : on_snapshot_hash( fd_snapct_tile_t * ctx,
310 : fd_ip4_port_t addr,
311 0 : fd_gossip_update_message_t const * msg ) {
312 0 : ulong full_slot = msg->snapshot_hashes.full->slot;
313 0 : ulong incr_slot = 0UL;
314 :
315 0 : for( ulong i=0UL; i<msg->snapshot_hashes.incremental_len; i++ ) {
316 0 : if( FD_LIKELY( msg->snapshot_hashes.incremental[ i ].slot>incr_slot ) ) {
317 0 : incr_slot = msg->snapshot_hashes.incremental[ i ].slot;
318 0 : }
319 0 : }
320 :
321 0 : fd_ssinfo_t ssinfo = { .full = { .slot = msg->snapshot_hashes.full->slot },
322 0 : .incremental = { .slot = incr_slot, .base_slot = full_slot } };
323 :
324 0 : fd_sspeer_selector_add( ctx->selector, addr, ULONG_MAX, &ssinfo );
325 0 : fd_sspeer_selector_process_cluster_slot( ctx->selector, full_slot, incr_slot );
326 0 : predict_incremental( ctx );
327 0 : }
328 :
329 : static void
330 : send_expected_slot( fd_snapct_tile_t * ctx,
331 : fd_stem_context_t * stem,
332 0 : ulong slot ) {
333 0 : uint tsorig; uint tspub;
334 0 : fd_ssmsg_slot_to_frag( slot, &tsorig, &tspub );
335 0 : fd_stem_publish( stem, ctx->out_rp.idx, FD_SSMSG_EXPECTED_SLOT, 0UL, 0UL, 0UL, tsorig, tspub );
336 0 : }
337 :
338 : static void
339 0 : rename_snapshots( fd_snapct_tile_t * ctx ) {
340 0 : FD_TEST( -1!=ctx->local_out.dir_fd );
341 :
342 : /* FIXME: We should rename the full snapshot earlier as soon as the
343 : download is complete. That way, if the validator crashes during the
344 : incremental load, we can still use the snapshot on the next run. */
345 :
346 0 : if( FD_LIKELY( -1!=ctx->local_out.full_snapshot_fd && ctx->http_full_snapshot_name[ 0 ]!='\0' ) ) {
347 0 : if( FD_UNLIKELY( -1==renameat( ctx->local_out.dir_fd, TEMP_FULL_SNAP_NAME, ctx->local_out.dir_fd, ctx->http_full_snapshot_name ) ) )
348 0 : FD_LOG_ERR(( "renameat() failed (%i-%s)", errno, fd_io_strerror( errno ) ));
349 0 : }
350 0 : if( FD_LIKELY( -1!=ctx->local_out.incremental_snapshot_fd && ctx->http_incr_snapshot_name[ 0 ]!='\0' ) ) {
351 0 : if( FD_UNLIKELY( -1==renameat( ctx->local_out.dir_fd, TEMP_INCR_SNAP_NAME, ctx->local_out.dir_fd, ctx->http_incr_snapshot_name ) ) )
352 0 : FD_LOG_ERR(( "renameat() failed (%i-%s)", errno, fd_io_strerror( errno ) ));
353 0 : }
354 0 : }
355 :
356 : static ulong
357 : rlimit_file_cnt( fd_topo_t const * topo FD_PARAM_UNUSED,
358 0 : fd_topo_tile_t const * tile ) {
359 0 : ulong cnt = 1UL + /* stderr */
360 0 : 1UL; /* logfile */
361 0 : if( download_enabled( tile ) ) {
362 0 : cnt += 1UL + /* ssping socket */
363 0 : 2UL + /* dirfd + full snapshot download temp fd */
364 0 : tile->snapct.sources.servers_cnt; /* http resolver peer full sockets */
365 0 : if( tile->snapct.incremental_snapshots ) {
366 0 : cnt += 1UL + /* incr snapshot download temp fd */
367 0 : tile->snapct.sources.servers_cnt; /* http resolver peer incr sockets */
368 0 : }
369 0 : }
370 0 : return cnt;
371 0 : }
372 :
373 : static ulong
374 : populate_allowed_seccomp( fd_topo_t const * topo,
375 : fd_topo_tile_t const * tile,
376 : ulong out_cnt,
377 0 : struct sock_filter * out ) {
378 :
379 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
380 :
381 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
382 0 : fd_snapct_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_snapct_tile_t), sizeof(fd_snapct_tile_t) );
383 :
384 0 : int ping_fd = download_enabled( tile ) ? fd_ssping_get_sockfd( ctx->ssping ) : -1;
385 0 : populate_sock_filter_policy_fd_snapct_tile( out_cnt, out, (uint)fd_log_private_logfile_fd(), (uint)ctx->local_out.dir_fd, (uint)ctx->local_out.full_snapshot_fd, (uint)ctx->local_out.incremental_snapshot_fd, (uint)ping_fd );
386 0 : return sock_filter_policy_fd_snapct_tile_instr_cnt;
387 0 : }
388 :
389 : static ulong
390 : populate_allowed_fds( fd_topo_t const * topo,
391 : fd_topo_tile_t const * tile,
392 : ulong out_fds_cnt,
393 0 : int * out_fds ) {
394 0 : if( FD_UNLIKELY( out_fds_cnt<6UL ) ) FD_LOG_ERR(( "out_fds_cnt %lu", out_fds_cnt ));
395 :
396 0 : ulong out_cnt = 0;
397 0 : out_fds[ out_cnt++ ] = 2UL; /* stderr */
398 0 : if( FD_LIKELY( -1!=fd_log_private_logfile_fd() ) ) {
399 0 : out_fds[ out_cnt++ ] = fd_log_private_logfile_fd(); /* logfile */
400 0 : }
401 :
402 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
403 :
404 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
405 0 : fd_snapct_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_snapct_tile_t), sizeof(fd_snapct_tile_t) );
406 0 : if( FD_LIKELY( -1!=ctx->local_out.dir_fd ) ) out_fds[ out_cnt++ ] = ctx->local_out.dir_fd;
407 0 : if( FD_LIKELY( -1!=ctx->local_out.full_snapshot_fd ) ) out_fds[ out_cnt++ ] = ctx->local_out.full_snapshot_fd;
408 0 : if( FD_LIKELY( -1!=ctx->local_out.incremental_snapshot_fd ) ) out_fds[ out_cnt++ ] = ctx->local_out.incremental_snapshot_fd;
409 0 : if( FD_LIKELY( download_enabled( tile ) ) ) out_fds[ out_cnt++ ] = fd_ssping_get_sockfd( ctx->ssping );
410 :
411 0 : return out_cnt;
412 0 : }
413 :
414 : static void
415 : init_load( fd_snapct_tile_t * ctx,
416 : fd_stem_context_t * stem,
417 : int full,
418 0 : int file ) {
419 0 : fd_ssctrl_init_t * out = fd_chunk_to_laddr( ctx->out_ld.mem, ctx->out_ld.chunk );
420 0 : out->file = file;
421 0 : out->zstd = !file || (full ? ctx->local_in.full_snapshot_zstd : ctx->local_in.incremental_snapshot_zstd);
422 0 : if( !file ) {
423 0 : out->addr = ctx->addr;
424 0 : for( ulong i=0UL; i<SERVER_PEERS_MAX; i++ ) {
425 0 : if( FD_UNLIKELY( ctx->addr.l==ctx->config.sources.servers[ i ].addr.l ) ) {
426 0 : fd_cstr_ncpy( out->hostname, ctx->config.sources.servers[ i ].hostname, sizeof(out->hostname) );
427 0 : out->is_https = ctx->config.sources.servers[ i ].is_https;
428 0 : break;
429 0 : }
430 0 : }
431 0 : }
432 0 : fd_stem_publish( stem, ctx->out_ld.idx, full ? FD_SNAPSHOT_MSG_CTRL_INIT_FULL : FD_SNAPSHOT_MSG_CTRL_INIT_INCR, ctx->out_ld.chunk, sizeof(fd_ssctrl_init_t), 0UL, 0UL, 0UL );
433 0 : ctx->out_ld.chunk = fd_dcache_compact_next( ctx->out_ld.chunk, sizeof(fd_ssctrl_init_t), ctx->out_ld.chunk0, ctx->out_ld.wmark );
434 0 : ctx->flush_ack = 0;
435 :
436 0 : if( file ) {
437 : /* When loading from a local file and not from HTTP, there is no
438 : future metadata message to initialize total size / filename, as
439 : these are already known immediately. */
440 0 : if( full ) {
441 0 : ctx->metrics.full.bytes_total = ctx->local_in.full_snapshot_size;
442 0 : fd_cstr_fini( ctx->http_full_snapshot_name );
443 0 : if( FD_LIKELY( !!ctx->out_gui.mem ) ) {
444 0 : snapshot_path_gui_publish( ctx, stem, ctx->local_in.full_snapshot_path, 1 );
445 0 : }
446 0 : } else {
447 0 : ctx->metrics.incremental.bytes_total = ctx->local_in.incremental_snapshot_size;
448 0 : fd_cstr_fini( ctx->http_incr_snapshot_name );
449 0 : if( FD_LIKELY( !!ctx->out_gui.mem ) ) {
450 0 : snapshot_path_gui_publish( ctx, stem, ctx->local_in.incremental_snapshot_path, 0 );
451 0 : }
452 0 : }
453 0 : }
454 0 : }
455 :
456 : static void
457 : log_download( fd_snapct_tile_t * ctx,
458 : int full,
459 : fd_ip4_port_t addr,
460 0 : ulong slot ) {
461 0 : for( gossip_ci_map_iter_t iter = gossip_ci_map_iter_init( ctx->gossip.ci_map, ctx->gossip.ci_table );
462 0 : !gossip_ci_map_iter_done( iter, ctx->gossip.ci_map, ctx->gossip.ci_table );
463 0 : iter = gossip_ci_map_iter_next( iter, ctx->gossip.ci_map, ctx->gossip.ci_table ) ) {
464 0 : gossip_ci_entry_t const * ci_entry = gossip_ci_map_iter_ele_const( iter, ctx->gossip.ci_map, ctx->gossip.ci_table );
465 0 : if( ci_entry->rpc_addr.l==addr.l ) {
466 0 : FD_TEST( ci_entry->allowed );
467 0 : FD_BASE58_ENCODE_32_BYTES( ci_entry->pubkey.uc, pubkey_b58 );
468 0 : FD_LOG_NOTICE(( "downloading %s snapshot at slot %lu from allowed gossip peer %s at http://" FD_IP4_ADDR_FMT ":%hu/%s",
469 0 : full ? "full" : "incremental", slot, pubkey_b58,
470 0 : FD_IP4_ADDR_FMT_ARGS( addr.addr ), fd_ushort_bswap( addr.port ),
471 0 : full ? "snapshot.tar.bz2" : "incremental-snapshot.tar.bz2" ));
472 0 : return;
473 0 : }
474 0 : }
475 :
476 0 : for( ulong i=0UL; i<ctx->config.sources.servers_cnt; i++ ) {
477 0 : if( addr.l==ctx->config.sources.servers[ i ].addr.l ) {
478 0 : if( ctx->config.sources.servers[ i ].is_https ) {
479 0 : FD_LOG_NOTICE(( "downloading %s snapshot at slot %lu from configured server with index %lu at https://%s:%hu/%s",
480 0 : full ? "full" : "incremental", slot, i,
481 0 : ctx->config.sources.servers[ i ].hostname, fd_ushort_bswap( addr.port ),
482 0 : full ? "snapshot.tar.bz2" : "incremental-snapshot.tar.bz2" ));
483 0 : } else {
484 0 : FD_LOG_NOTICE(( "downloading %s snapshot at slot %lu from configured server with index %lu at http://" FD_IP4_ADDR_FMT ":%hu/%s",
485 0 : full ? "full" : "incremental", slot, i,
486 0 : FD_IP4_ADDR_FMT_ARGS( addr.addr ), fd_ushort_bswap( addr.port ),
487 0 : full ? "snapshot.tar.bz2" : "incremental-snapshot.tar.bz2" ));
488 0 : }
489 0 : return;
490 0 : }
491 0 : }
492 :
493 0 : FD_TEST( 0 ); /* should not be possible */
494 0 : }
495 :
496 : static void
497 : after_credit( fd_snapct_tile_t * ctx,
498 : fd_stem_context_t * stem,
499 : int * opt_poll_in FD_PARAM_UNUSED,
500 0 : int * charge_busy FD_PARAM_UNUSED ) {
501 0 : long now = fd_log_wallclock();
502 :
503 0 : if( FD_LIKELY( ctx->ssping ) ) fd_ssping_advance( ctx->ssping, now, ctx->selector );
504 0 : if( FD_LIKELY( ctx->ssresolver ) ) fd_http_resolver_advance( ctx->ssresolver, now, ctx->selector );
505 :
506 : /* send an expected slot message as the predicted incremental
507 : could have changed as a result of the pinger, resolver, or from
508 : processing gossip frags in gossip_frag. */
509 0 : if( FD_LIKELY( ctx->predicted_incremental.dirty ) ) {
510 0 : send_expected_slot( ctx, stem, ctx->predicted_incremental.slot );
511 0 : ctx->predicted_incremental.dirty = 0;
512 0 : }
513 :
514 : /* Note: All state transitions should occur within this switch
515 : statement to make it easier to reason about the state management. */
516 :
517 : /* FIXME: Collapse WAITING_FOR_PEERS and COLLECTING_PEERS states for
518 : both full and incremental variants? */
519 : /* FIXME: Add INIT state so that we don't put the !download_enabled
520 : logic in waiting_for_peers, which is weird. */
521 :
522 0 : switch ( ctx->state ) {
523 :
524 : /* ============================================================== */
525 0 : case FD_SNAPCT_STATE_WAITING_FOR_PEERS: {
526 0 : if( FD_UNLIKELY( now>ctx->deadline_nanos ) ) FD_LOG_ERR(( "timed out waiting for peers." ));
527 :
528 0 : if( FD_UNLIKELY( !ctx->download_enabled ) ) {
529 0 : ulong local_slot = ctx->config.incremental_snapshots ? ctx->local_in.incremental_snapshot_slot : ctx->local_in.full_snapshot_slot;
530 0 : send_expected_slot( ctx, stem, local_slot );
531 0 : FD_LOG_NOTICE(( "reading full snapshot at slot %lu from local file `%s`", ctx->local_in.full_snapshot_slot, ctx->local_in.full_snapshot_path ));
532 0 : ctx->predicted_incremental.full_slot = ctx->local_in.full_snapshot_slot;
533 0 : ctx->state = FD_SNAPCT_STATE_READING_FULL_FILE;
534 0 : init_load( ctx, stem, 1, 1 );
535 0 : break;
536 0 : }
537 :
538 0 : fd_sspeer_t best = fd_sspeer_selector_best( ctx->selector, 0, ULONG_MAX );
539 0 : if( FD_LIKELY( best.addr.l ) ) {
540 0 : ctx->state = FD_SNAPCT_STATE_COLLECTING_PEERS;
541 0 : ctx->deadline_nanos = now+FD_SNAPCT_COLLECTING_PEERS_TIMEOUT;
542 0 : }
543 0 : break;
544 0 : }
545 :
546 : /* ============================================================== */
547 0 : case FD_SNAPCT_STATE_WAITING_FOR_PEERS_INCREMENTAL: {
548 : /* FIXME: Handle the case where we have no download peers enabled,
549 : boot off the local full snapshot but do not have a local incr. */
550 0 : fd_sspeer_t best = fd_sspeer_selector_best( ctx->selector, 0, ULONG_MAX );
551 0 : if( FD_LIKELY( best.addr.l ) ) {
552 0 : ctx->state = FD_SNAPCT_STATE_COLLECTING_PEERS_INCREMENTAL;
553 0 : ctx->deadline_nanos = now;
554 0 : }
555 0 : break;
556 0 : }
557 :
558 : /* ============================================================== */
559 0 : case FD_SNAPCT_STATE_COLLECTING_PEERS: {
560 0 : if( FD_UNLIKELY( !ctx->gossip.saturated && now<ctx->deadline_nanos ) ) break;
561 :
562 0 : fd_sspeer_t best = fd_sspeer_selector_best( ctx->selector, 0, ULONG_MAX );
563 0 : if( FD_UNLIKELY( !best.addr.l ) ) {
564 0 : ctx->state = FD_SNAPCT_STATE_WAITING_FOR_PEERS;
565 0 : break;
566 0 : }
567 :
568 0 : fd_sscluster_slot_t cluster = fd_sspeer_selector_cluster_slot( ctx->selector );
569 0 : if( FD_UNLIKELY( cluster.incremental==ULONG_MAX && ctx->config.incremental_snapshots ) ) {
570 : /* We must have a cluster full slot to be in this state. */
571 0 : FD_TEST( cluster.full!=ULONG_MAX );
572 : /* fall back to full snapshot only if the highest cluster slot
573 : is a full snapshot only */
574 0 : ctx->config.incremental_snapshots = 0;
575 0 : }
576 :
577 : /* FIXME: Revisit the local age logic with new effective age
578 : concept. Measure cluster slot based on snapshots we can
579 : download / trust. Reevaluate incremental age after the full
580 : snapshot download is completed. etc. etc. */
581 :
582 0 : ulong cluster_slot = ctx->config.incremental_snapshots ? cluster.incremental : cluster.full;
583 0 : ulong local_slot = ctx->config.incremental_snapshots ? ctx->local_in.incremental_snapshot_slot : ctx->local_in.full_snapshot_slot;
584 0 : ulong local_slot_with_download = local_slot;
585 0 : int local_too_old = local_slot!=ULONG_MAX && local_slot<fd_ulong_sat_sub( cluster_slot, ctx->config.sources.max_local_incremental_age );
586 0 : int local_full_only = ctx->local_in.incremental_snapshot_slot==ULONG_MAX && ctx->local_in.full_snapshot_slot!=ULONG_MAX;
587 0 : if( FD_LIKELY( (ctx->config.incremental_snapshots && local_full_only) || local_too_old ) ) {
588 0 : fd_sspeer_t best_incremental = fd_sspeer_selector_best( ctx->selector, 1, ctx->local_in.full_snapshot_slot );
589 0 : if( FD_LIKELY( best_incremental.addr.l ) ) {
590 0 : ctx->predicted_incremental.slot = best_incremental.ssinfo.incremental.slot;
591 0 : local_slot_with_download = best_incremental.ssinfo.incremental.slot;
592 0 : ctx->local_in.incremental_snapshot_slot = ULONG_MAX; /* don't use the local incremental snapshot */
593 0 : }
594 0 : }
595 :
596 0 : int can_use_local_full = local_slot_with_download!=ULONG_MAX && local_slot_with_download>=fd_ulong_sat_sub( cluster_slot, ctx->config.sources.max_local_full_effective_age );
597 0 : if( FD_LIKELY( can_use_local_full ) ) {
598 0 : send_expected_slot( ctx, stem, local_slot );
599 :
600 0 : FD_LOG_NOTICE(( "reading full snapshot at slot %lu from local file `%s`", ctx->local_in.full_snapshot_slot, ctx->local_in.full_snapshot_path ));
601 0 : ctx->predicted_incremental.full_slot = ctx->local_in.full_snapshot_slot;
602 0 : ctx->state = FD_SNAPCT_STATE_READING_FULL_FILE;
603 0 : init_load( ctx, stem, 1, 1 );
604 0 : } else {
605 0 : if( FD_UNLIKELY( !ctx->config.incremental_snapshots ) ) send_expected_slot( ctx, stem, best.ssinfo.full.slot );
606 :
607 0 : fd_sspeer_t best_incremental = fd_sspeer_selector_best( ctx->selector, 1, best.ssinfo.full.slot );
608 0 : if( FD_LIKELY( best_incremental.addr.l ) ) {
609 0 : ctx->predicted_incremental.slot = best_incremental.ssinfo.incremental.slot;
610 0 : send_expected_slot( ctx, stem, best_incremental.ssinfo.incremental.slot );
611 0 : }
612 :
613 0 : ctx->addr = best.addr;
614 0 : ctx->state = FD_SNAPCT_STATE_READING_FULL_HTTP;
615 0 : ctx->predicted_incremental.full_slot = best.ssinfo.full.slot;
616 0 : init_load( ctx, stem, 1, 0 );
617 0 : log_download( ctx, 1, best.addr, best.ssinfo.full.slot );
618 0 : }
619 0 : break;
620 0 : }
621 :
622 : /* ============================================================== */
623 0 : case FD_SNAPCT_STATE_COLLECTING_PEERS_INCREMENTAL: {
624 0 : if( FD_UNLIKELY( now<ctx->deadline_nanos ) ) break;
625 :
626 0 : fd_sspeer_t best = fd_sspeer_selector_best( ctx->selector, 1, ctx->predicted_incremental.full_slot );
627 0 : if( FD_UNLIKELY( !best.addr.l ) ) {
628 0 : ctx->state = FD_SNAPCT_STATE_WAITING_FOR_PEERS_INCREMENTAL;
629 0 : break;
630 0 : }
631 :
632 : /* FIXME: predicted_incremental? */
633 :
634 0 : ctx->addr = best.addr;
635 0 : ctx->state = FD_SNAPCT_STATE_READING_INCREMENTAL_HTTP;
636 0 : init_load( ctx, stem, 0, 0 );
637 0 : log_download( ctx, 0, best.addr, best.ssinfo.incremental.slot );
638 0 : break;
639 0 : }
640 :
641 : /* ============================================================== */
642 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_FILE:
643 0 : if( !ctx->flush_ack ) break;
644 :
645 0 : if( FD_UNLIKELY( ctx->malformed ) ) {
646 0 : ctx->malformed = 0;
647 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_FAIL, 0UL, 0UL, 0UL, 0UL, 0UL );
648 0 : ctx->flush_ack = 0;
649 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_FILE_RESET;
650 0 : FD_LOG_WARNING(( "error reading snapshot from local file `%s`", ctx->local_in.incremental_snapshot_path ));
651 0 : break;
652 0 : }
653 :
654 0 : ctx->state = FD_SNAPCT_STATE_SHUTDOWN;
655 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_SHUTDOWN, 0UL, 0UL, 0UL, 0UL, 0UL );
656 0 : break;
657 :
658 : /* ============================================================== */
659 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP:
660 0 : if( !ctx->flush_ack ) break;
661 :
662 0 : if( FD_UNLIKELY( ctx->malformed ) ) {
663 0 : ctx->malformed = 0;
664 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_FAIL, 0UL, 0UL, 0UL, 0UL, 0UL );
665 0 : ctx->flush_ack = 0;
666 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP_RESET;
667 0 : FD_LOG_WARNING(( "error downloading snapshot from http://" FD_IP4_ADDR_FMT ":%hu/incremental-snapshot.tar.bz2",
668 0 : FD_IP4_ADDR_FMT_ARGS( ctx->addr.addr ), fd_ushort_bswap( ctx->addr.port ) ));
669 0 : fd_ssping_invalidate( ctx->ssping, ctx->addr, fd_log_wallclock() );
670 0 : fd_sspeer_selector_remove( ctx->selector, ctx->addr );
671 0 : break;
672 0 : }
673 :
674 0 : ctx->state = FD_SNAPCT_STATE_SHUTDOWN;
675 0 : rename_snapshots( ctx );
676 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_SHUTDOWN, 0UL, 0UL, 0UL, 0UL, 0UL );
677 0 : break;
678 :
679 : /* ============================================================== */
680 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_FILE:
681 0 : if( !ctx->flush_ack ) break;
682 :
683 0 : if( FD_UNLIKELY( ctx->malformed ) ) {
684 0 : ctx->malformed = 0;
685 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_FAIL, 0UL, 0UL, 0UL, 0UL, 0UL );
686 0 : ctx->flush_ack = 0;
687 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_FULL_FILE_RESET;
688 0 : FD_LOG_WARNING(( "error reading snapshot from local file `%s`", ctx->local_in.full_snapshot_path ));
689 0 : break;
690 0 : }
691 :
692 0 : if( FD_LIKELY( !ctx->config.incremental_snapshots ) ) {
693 0 : ctx->state = FD_SNAPCT_STATE_SHUTDOWN;
694 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_SHUTDOWN, 0UL, 0UL, 0UL, 0UL, 0UL );
695 0 : break;
696 0 : }
697 :
698 0 : if( FD_LIKELY( ctx->local_in.incremental_snapshot_slot==ULONG_MAX ) ) {
699 0 : ctx->state = FD_SNAPCT_STATE_COLLECTING_PEERS_INCREMENTAL;
700 0 : ctx->deadline_nanos = 0L;
701 0 : } else {
702 0 : FD_LOG_NOTICE(( "reading incremental snapshot at slot %lu from local file `%s`", ctx->local_in.incremental_snapshot_slot, ctx->local_in.incremental_snapshot_path ));
703 0 : ctx->state = FD_SNAPCT_STATE_READING_INCREMENTAL_FILE;
704 0 : init_load( ctx, stem, 0, 1 );
705 0 : }
706 0 : break;
707 :
708 : /* ============================================================== */
709 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_HTTP:
710 0 : if( !ctx->flush_ack ) break;
711 :
712 0 : if( FD_UNLIKELY( ctx->malformed ) ) {
713 0 : ctx->malformed = 0;
714 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_FAIL, 0UL, 0UL, 0UL, 0UL, 0UL );
715 0 : ctx->flush_ack = 0;
716 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_FULL_HTTP_RESET;
717 0 : FD_LOG_WARNING(( "error downloading snapshot from http://" FD_IP4_ADDR_FMT ":%hu/snapshot.tar.bz2",
718 0 : FD_IP4_ADDR_FMT_ARGS( ctx->addr.addr ), fd_ushort_bswap( ctx->addr.port ) ));
719 0 : fd_ssping_invalidate( ctx->ssping, ctx->addr, fd_log_wallclock() );
720 0 : fd_sspeer_selector_remove( ctx->selector, ctx->addr );
721 0 : break;
722 0 : }
723 :
724 0 : if( FD_LIKELY( !ctx->config.incremental_snapshots ) ) {
725 0 : ctx->state = FD_SNAPCT_STATE_SHUTDOWN;
726 0 : rename_snapshots( ctx );
727 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_SHUTDOWN, 0UL, 0UL, 0UL, 0UL, 0UL );
728 0 : break;
729 0 : }
730 :
731 : /* Get the best incremental peer to download from */
732 0 : fd_sspeer_t best = fd_sspeer_selector_best( ctx->selector, 1, ctx->predicted_incremental.full_slot );
733 0 : if( FD_UNLIKELY( !best.addr.l ) ) {
734 : /* FIXME: We should just transition to collecting_peers_incremental
735 : here rather than failing the full snapshot? */
736 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_FAIL, 0UL, 0UL, 0UL, 0UL, 0UL );
737 0 : ctx->flush_ack = 0;
738 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_FULL_HTTP_RESET;
739 0 : break;
740 0 : }
741 :
742 0 : if( FD_UNLIKELY( ctx->predicted_incremental.slot!=best.ssinfo.incremental.slot ) ) {
743 0 : ctx->predicted_incremental.slot = best.ssinfo.incremental.slot;
744 0 : send_expected_slot( ctx, stem, best.ssinfo.incremental.slot );
745 0 : }
746 :
747 0 : ctx->addr = best.addr;
748 0 : ctx->state = FD_SNAPCT_STATE_READING_INCREMENTAL_HTTP;
749 0 : init_load( ctx, stem, 0, 0 );
750 0 : log_download( ctx, 0, best.addr, best.ssinfo.incremental.slot );
751 0 : break;
752 :
753 : /* ============================================================== */
754 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_HTTP_RESET:
755 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_FILE_RESET:
756 0 : if( !ctx->flush_ack ) break;
757 :
758 0 : ctx->metrics.full.bytes_read = 0UL;
759 0 : ctx->metrics.full.bytes_written = 0UL;
760 0 : ctx->metrics.full.bytes_total = 0UL;
761 :
762 0 : ctx->metrics.incremental.bytes_read = 0UL;
763 0 : ctx->metrics.incremental.bytes_written = 0UL;
764 0 : ctx->metrics.incremental.bytes_total = 0UL;
765 :
766 0 : ctx->state = FD_SNAPCT_STATE_COLLECTING_PEERS;
767 0 : ctx->deadline_nanos = 0L;
768 0 : break;
769 :
770 : /* ============================================================== */
771 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_FILE_RESET:
772 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP_RESET:
773 0 : if( !ctx->flush_ack ) break;
774 :
775 0 : ctx->metrics.incremental.bytes_read = 0UL;
776 0 : ctx->metrics.incremental.bytes_written = 0UL;
777 0 : ctx->metrics.incremental.bytes_total = 0UL;
778 :
779 0 : ctx->state = FD_SNAPCT_STATE_COLLECTING_PEERS_INCREMENTAL;
780 0 : ctx->deadline_nanos = 0L;
781 0 : break;
782 :
783 : /* ============================================================== */
784 0 : case FD_SNAPCT_STATE_READING_FULL_FILE:
785 0 : if( FD_UNLIKELY( !ctx->flush_ack ) ) break;
786 0 : if( FD_UNLIKELY( ctx->malformed ) ) {
787 0 : ctx->malformed = 0;
788 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_FAIL, 0UL, 0UL, 0UL, 0UL, 0UL );
789 0 : ctx->flush_ack = 0;
790 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_FULL_FILE_RESET;
791 0 : FD_LOG_WARNING(( "error reading snapshot from local file `%s`", ctx->local_in.full_snapshot_path ));
792 0 : break;
793 0 : }
794 0 : FD_TEST( ctx->metrics.full.bytes_total!=0UL );
795 0 : if( FD_UNLIKELY( ctx->metrics.full.bytes_read == ctx->metrics.full.bytes_total ) ) {
796 0 : ulong sig = ctx->config.incremental_snapshots ? FD_SNAPSHOT_MSG_CTRL_NEXT : FD_SNAPSHOT_MSG_CTRL_DONE;
797 0 : fd_stem_publish( stem, ctx->out_ld.idx, sig, 0UL, 0UL, 0UL, 0UL, 0UL );
798 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_FULL_FILE;
799 0 : ctx->flush_ack = 0;
800 0 : }
801 0 : break;
802 :
803 : /* ============================================================== */
804 0 : case FD_SNAPCT_STATE_READING_INCREMENTAL_FILE:
805 0 : if( FD_UNLIKELY( !ctx->flush_ack ) ) break;
806 0 : if( FD_UNLIKELY( ctx->malformed ) ) {
807 0 : ctx->malformed = 0;
808 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_FAIL, 0UL, 0UL, 0UL, 0UL, 0UL );
809 0 : ctx->flush_ack = 0;
810 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_FILE_RESET;
811 0 : FD_LOG_WARNING(( "error reading snapshot from local file `%s`", ctx->local_in.incremental_snapshot_path ));
812 0 : break;
813 0 : }
814 0 : FD_TEST( ctx->metrics.incremental.bytes_total!=0UL );
815 0 : if ( FD_UNLIKELY( ctx->metrics.incremental.bytes_read == ctx->metrics.incremental.bytes_total ) ) {
816 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_DONE, 0UL, 0UL, 0UL, 0UL, 0UL );
817 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_FILE;
818 0 : ctx->flush_ack = 0;
819 0 : }
820 0 : break;
821 :
822 : /* ============================================================== */
823 0 : case FD_SNAPCT_STATE_READING_FULL_HTTP:
824 0 : if( FD_UNLIKELY( !ctx->flush_ack ) ) break;
825 0 : if( FD_UNLIKELY( ctx->malformed ) ) {
826 0 : ctx->malformed = 0;
827 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_FAIL, 0UL, 0UL, 0UL, 0UL, 0UL );
828 0 : ctx->flush_ack = 0;
829 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_FULL_HTTP_RESET;
830 0 : FD_LOG_WARNING(( "error downloading snapshot from http://" FD_IP4_ADDR_FMT ":%hu/snapshot.tar.bz2",
831 0 : FD_IP4_ADDR_FMT_ARGS( ctx->addr.addr ), fd_ushort_bswap( ctx->addr.port ) ));
832 0 : fd_ssping_invalidate( ctx->ssping, ctx->addr, fd_log_wallclock() );
833 0 : fd_sspeer_selector_remove( ctx->selector, ctx->addr );
834 0 : break;
835 0 : }
836 0 : if( FD_UNLIKELY( ctx->metrics.full.bytes_total!=0UL && ctx->metrics.full.bytes_read==ctx->metrics.full.bytes_total ) ) {
837 0 : ulong sig = ctx->config.incremental_snapshots ? FD_SNAPSHOT_MSG_CTRL_NEXT : FD_SNAPSHOT_MSG_CTRL_DONE;
838 0 : fd_stem_publish( stem, ctx->out_ld.idx, sig, 0UL, 0UL, 0UL, 0UL, 0UL );
839 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_FULL_HTTP;
840 0 : ctx->flush_ack = 0;
841 0 : }
842 0 : break;
843 :
844 : /* ============================================================== */
845 0 : case FD_SNAPCT_STATE_READING_INCREMENTAL_HTTP:
846 0 : if( FD_UNLIKELY( !ctx->flush_ack ) ) break;
847 0 : if( FD_UNLIKELY( ctx->malformed ) ) {
848 0 : ctx->malformed = 0;
849 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_FAIL, 0UL, 0UL, 0UL, 0UL, 0UL );
850 0 : ctx->flush_ack = 0;
851 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP_RESET;
852 0 : FD_LOG_WARNING(( "error downloading snapshot from http://" FD_IP4_ADDR_FMT ":%hu/incremental-snapshot.tar.bz2",
853 0 : FD_IP4_ADDR_FMT_ARGS( ctx->addr.addr ), fd_ushort_bswap( ctx->addr.port ) ));
854 0 : fd_ssping_invalidate( ctx->ssping, ctx->addr, fd_log_wallclock() );
855 0 : fd_sspeer_selector_remove( ctx->selector, ctx->addr );
856 0 : break;
857 0 : }
858 0 : if ( FD_UNLIKELY( ctx->metrics.incremental.bytes_total!=0UL && ctx->metrics.incremental.bytes_read==ctx->metrics.incremental.bytes_total ) ) {
859 0 : fd_stem_publish( stem, ctx->out_ld.idx, FD_SNAPSHOT_MSG_CTRL_DONE, 0UL, 0UL, 0UL, 0UL, 0UL );
860 0 : ctx->state = FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP;
861 0 : ctx->flush_ack = 0;
862 0 : }
863 0 : break;
864 :
865 : /* ============================================================== */
866 0 : case FD_SNAPCT_STATE_SHUTDOWN:
867 0 : break;
868 :
869 : /* ============================================================== */
870 0 : default: FD_LOG_ERR(( "unexpected state %d", ctx->state ));
871 0 : }
872 0 : }
873 :
874 : static void
875 : gossip_frag( fd_snapct_tile_t * ctx,
876 : ulong sig,
877 : ulong sz FD_PARAM_UNUSED,
878 0 : ulong chunk ) {
879 0 : FD_TEST( ctx->gossip_enabled );
880 :
881 0 : if( !( sig==FD_GOSSIP_UPDATE_TAG_CONTACT_INFO ||
882 0 : sig==FD_GOSSIP_UPDATE_TAG_CONTACT_INFO_REMOVE ||
883 0 : sig==FD_GOSSIP_UPDATE_TAG_SNAPSHOT_HASHES ) ) return;
884 :
885 0 : fd_gossip_update_message_t const * msg = fd_chunk_to_laddr_const( ctx->gossip_in_mem, chunk );
886 0 : fd_pubkey_t const * pubkey = (fd_pubkey_t const *)msg->origin_pubkey;
887 0 : switch( msg->tag ) {
888 0 : case FD_GOSSIP_UPDATE_TAG_CONTACT_INFO: {
889 0 : FD_TEST( msg->contact_info.idx<GOSSIP_PEERS_MAX );
890 0 : gossip_ci_entry_t * entry = ctx->gossip.ci_table + msg->contact_info.idx;
891 0 : if( FD_UNLIKELY( !fd_pubkey_eq( &entry->pubkey, pubkey ) ) ) {
892 : /* Initialize the new gossip entry, which may or may not be allowed */
893 0 : FD_TEST( fd_pubkey_check_zero( &entry->pubkey ) );
894 0 : entry->pubkey = *pubkey;
895 0 : entry->rpc_addr.l = 0UL;
896 0 : entry->added_nanos = fd_log_wallclock();
897 0 : if( ctx->config.sources.gossip.allow_any ) {
898 0 : entry->allowed = 1;
899 0 : for( ulong i=0UL; i<ctx->config.sources.gossip.block_list_cnt; i++ ) {
900 0 : if( fd_pubkey_eq( pubkey, &ctx->config.sources.gossip.block_list[ i ] ) ) {
901 0 : entry->allowed = 0;
902 0 : break;
903 0 : }
904 0 : }
905 0 : } else {
906 0 : entry->allowed = 0;
907 0 : for( ulong i=0UL; i<ctx->config.sources.gossip.allow_list_cnt; i++ ) {
908 0 : if( fd_pubkey_eq( pubkey, &ctx->config.sources.gossip.allow_list[ i ] ) ) {
909 0 : entry->allowed = 1;
910 0 : break;
911 0 : }
912 0 : }
913 0 : }
914 0 : FD_TEST( ULONG_MAX==gossip_ci_map_idx_query_const( ctx->gossip.ci_map, pubkey, ULONG_MAX, ctx->gossip.ci_table ) );
915 0 : if( entry->allowed ) gossip_ci_map_idx_insert( ctx->gossip.ci_map, msg->contact_info.idx, ctx->gossip.ci_table );
916 0 : }
917 0 : if( !entry->allowed ) break;
918 : /* Maybe update the RPC address of a new or existing allowed gossip peer */
919 0 : fd_ip4_port_t cur_addr = entry->rpc_addr;
920 0 : fd_ip4_port_t new_addr = msg->contact_info.contact_info->sockets[ FD_CONTACT_INFO_SOCKET_RPC ];
921 0 : if( FD_UNLIKELY( new_addr.l!=cur_addr.l ) ) {
922 0 : entry->rpc_addr = new_addr;
923 0 : if( FD_LIKELY( !!cur_addr.l ) ) {
924 0 : int removed = fd_ssping_remove( ctx->ssping, cur_addr );
925 0 : if( FD_LIKELY( removed ) ) fd_sspeer_selector_remove( ctx->selector, cur_addr );
926 0 : }
927 0 : if( FD_LIKELY( !!new_addr.l ) ) fd_ssping_add( ctx->ssping, new_addr );
928 0 : if( !ctx->config.sources.gossip.allow_any ) {
929 0 : FD_BASE58_ENCODE_32_BYTES( pubkey->uc, pubkey_b58 );
930 0 : if( FD_LIKELY( !!new_addr.l ) ) {
931 0 : FD_LOG_NOTICE(( "allowed gossip peer added with public key `%s` and RPC address `" FD_IP4_ADDR_FMT ":%hu`",
932 0 : pubkey_b58, FD_IP4_ADDR_FMT_ARGS( new_addr.addr ), fd_ushort_bswap( new_addr.port ) ));
933 0 : } else {
934 0 : FD_LOG_WARNING(( "allowed gossip peer with public key `%s` does not advertise an RPC address", pubkey_b58 ));
935 0 : }
936 0 : }
937 0 : }
938 0 : break;
939 0 : }
940 0 : case FD_GOSSIP_UPDATE_TAG_CONTACT_INFO_REMOVE: {
941 0 : FD_TEST( msg->contact_info.idx<GOSSIP_PEERS_MAX );
942 0 : gossip_ci_entry_t * entry = ctx->gossip.ci_table + msg->contact_info.idx;
943 0 : if( FD_UNLIKELY( !fd_pubkey_eq( &entry->pubkey, pubkey ) ) ) {
944 0 : FD_TEST( fd_pubkey_check_zero( &entry->pubkey ) );
945 0 : break;
946 0 : }
947 0 : ulong rem_idx = gossip_ci_map_idx_remove( ctx->gossip.ci_map, pubkey, ULONG_MAX, ctx->gossip.ci_table );
948 0 : if( rem_idx==ULONG_MAX ) break;
949 0 : FD_TEST( entry->allowed && rem_idx==msg->contact_info.idx );
950 0 : fd_ip4_port_t addr = entry->rpc_addr;
951 0 : if( FD_LIKELY( !!addr.l ) ) {
952 0 : int removed = fd_ssping_remove( ctx->ssping, addr );
953 0 : if( FD_LIKELY( removed ) ) fd_sspeer_selector_remove( ctx->selector, addr );
954 0 : }
955 0 : if( !ctx->config.sources.gossip.allow_any ) {
956 0 : FD_BASE58_ENCODE_32_BYTES( pubkey->uc, pubkey_b58 );
957 0 : FD_LOG_WARNING(( "allowed gossip peer removed with public key `%s` and RPC address `" FD_IP4_ADDR_FMT ":%hu`",
958 0 : pubkey_b58, FD_IP4_ADDR_FMT_ARGS( addr.addr ), fd_ushort_bswap( addr.port ) ));
959 0 : }
960 0 : fd_memset( entry, 0, sizeof(*entry) );
961 0 : break;
962 0 : }
963 0 : case FD_GOSSIP_UPDATE_TAG_SNAPSHOT_HASHES: {
964 0 : ulong idx = gossip_ci_map_idx_query_const( ctx->gossip.ci_map, pubkey, ULONG_MAX, ctx->gossip.ci_table );
965 0 : if( FD_LIKELY( idx!=ULONG_MAX ) ) {
966 0 : gossip_ci_entry_t * entry = ctx->gossip.ci_table + idx;
967 0 : FD_TEST( entry->allowed );
968 0 : on_snapshot_hash( ctx, entry->rpc_addr, msg );
969 0 : }
970 0 : break;
971 0 : }
972 0 : default:
973 0 : FD_LOG_ERR(( "snapct: unexpected gossip tag %u", (uint)msg->tag ));
974 0 : break;
975 0 : }
976 0 : }
977 :
978 : static void
979 : snapld_frag( fd_snapct_tile_t * ctx,
980 : ulong sig,
981 : ulong sz,
982 : ulong chunk,
983 0 : fd_stem_context_t * stem ) {
984 0 : if( FD_UNLIKELY( sig==FD_SNAPSHOT_MSG_META ) ) {
985 : /* Before snapld starts sending down data fragments, it first sends
986 : a metadata message containing the total size of the snapshot as
987 : well as the filename. This is only done for HTTP loading. */
988 0 : int full;
989 0 : switch( ctx->state ) {
990 0 : case FD_SNAPCT_STATE_READING_FULL_HTTP: full = 1; break;
991 0 : case FD_SNAPCT_STATE_READING_INCREMENTAL_HTTP: full = 0; break;
992 :
993 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_HTTP_RESET:
994 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP_RESET:
995 0 : return; /* Ignore */
996 0 : default: FD_LOG_ERR(( "invalid meta frag in state %d", ctx->state ));
997 0 : }
998 :
999 0 : FD_TEST( sz==sizeof(fd_ssctrl_meta_t) );
1000 0 : fd_ssctrl_meta_t const * meta = fd_chunk_to_laddr_const( ctx->snapld_in_mem, chunk );
1001 :
1002 0 : fd_memcpy( full ? ctx->http_full_snapshot_name : ctx->http_incr_snapshot_name, meta->name, PATH_MAX );
1003 :
1004 0 : if( FD_LIKELY( !!ctx->out_gui.mem ) ) {
1005 0 : char snapshot_path[ PATH_MAX+30UL ]; /* 30 is fd_cstr_nlen( "https://255.255.255.255:65536/", ULONG_MAX ) */
1006 0 : FD_TEST( fd_cstr_printf_check( snapshot_path, sizeof(snapshot_path), NULL, "http://" FD_IP4_ADDR_FMT ":%hu/%s", FD_IP4_ADDR_FMT_ARGS( ctx->addr.addr ), fd_ushort_bswap( ctx->addr.port ), meta->name ) );
1007 0 : snapshot_path_gui_publish( ctx, stem, snapshot_path, full );
1008 0 : }
1009 :
1010 0 : if( full ) ctx->metrics.full.bytes_total = meta->total_sz;
1011 0 : else ctx->metrics.incremental.bytes_total = meta->total_sz;
1012 :
1013 0 : return;
1014 0 : }
1015 0 : if( FD_UNLIKELY( sig!=FD_SNAPSHOT_MSG_DATA ) ) return;
1016 :
1017 0 : int full, file;
1018 0 : switch( ctx->state ) {
1019 : /* Expected cases, fall through below */
1020 0 : case FD_SNAPCT_STATE_READING_FULL_FILE: full = 1; file = 1; break;
1021 0 : case FD_SNAPCT_STATE_READING_INCREMENTAL_FILE: full = 0; file = 1; break;
1022 0 : case FD_SNAPCT_STATE_READING_FULL_HTTP: full = 1; file = 0; break;
1023 0 : case FD_SNAPCT_STATE_READING_INCREMENTAL_HTTP: full = 0; file = 0; break;
1024 :
1025 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_FILE_RESET:
1026 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_FILE_RESET:
1027 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_HTTP_RESET:
1028 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP_RESET:
1029 : /* We are waiting for a reset to fully propagate through the
1030 : pipeline, just throw away any trailing data frags. */
1031 0 : return;
1032 :
1033 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_FILE:
1034 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_FILE:
1035 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_HTTP:
1036 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP:
1037 : /* Based on previously received data frags, we expected that the
1038 : current full / incremental snapshot was finished, but then we
1039 : received additional data frags. Unsafe to continue so throw
1040 : away the whole snapshot. */
1041 0 : if( !ctx->malformed ) {
1042 0 : ctx->malformed = 1;
1043 0 : FD_LOG_WARNING(( "complete snapshot loaded but read %lu extra bytes", sz ));
1044 0 : }
1045 0 : return;
1046 :
1047 0 : case FD_SNAPCT_STATE_WAITING_FOR_PEERS:
1048 0 : case FD_SNAPCT_STATE_WAITING_FOR_PEERS_INCREMENTAL:
1049 0 : case FD_SNAPCT_STATE_COLLECTING_PEERS:
1050 0 : case FD_SNAPCT_STATE_COLLECTING_PEERS_INCREMENTAL:
1051 0 : case FD_SNAPCT_STATE_SHUTDOWN:
1052 0 : default:
1053 0 : FD_LOG_ERR(( "invalid data frag in state %d", ctx->state ));
1054 0 : return;
1055 0 : }
1056 :
1057 0 : if( full ) FD_TEST( ctx->metrics.full.bytes_total !=0UL );
1058 0 : else FD_TEST( ctx->metrics.incremental.bytes_total!=0UL );
1059 :
1060 0 : if( full ) ctx->metrics.full.bytes_read += sz;
1061 0 : else ctx->metrics.incremental.bytes_read += sz;
1062 :
1063 0 : if( !file && -1!=ctx->local_out.dir_fd ) {
1064 0 : uchar const * data = fd_chunk_to_laddr_const( ctx->snapld_in_mem, chunk );
1065 0 : int fd = full ? ctx->local_out.full_snapshot_fd : ctx->local_out.incremental_snapshot_fd;
1066 0 : long result = write( fd, data, sz );
1067 0 : if( FD_UNLIKELY( -1==result && errno==ENOSPC ) ) {
1068 0 : FD_LOG_ERR(( "Out of disk space when writing out snapshot data to `%s`", ctx->config.snapshots_path ));
1069 0 : } else if( FD_UNLIKELY( 0L>result ) ) {
1070 0 : FD_LOG_ERR(( "write() failed (%i-%s)", errno, fd_io_strerror( errno ) ));
1071 0 : } else if( FD_UNLIKELY( sz!=(ulong)result ) ) {
1072 0 : FD_LOG_ERR(( "paritial write(%lu)=%ld", sz, result ));
1073 0 : }
1074 0 : if( full ) ctx->metrics.full.bytes_written += sz;
1075 0 : else ctx->metrics.incremental.bytes_written += sz;
1076 0 : }
1077 :
1078 0 : if( FD_UNLIKELY( ( full && ctx->metrics.full.bytes_read > ctx->metrics.full.bytes_total ) ||
1079 0 : (!full && ctx->metrics.incremental.bytes_read > ctx->metrics.incremental.bytes_total ) ) ) {
1080 0 : if( !ctx->malformed ) {
1081 0 : ctx->malformed = 1;
1082 0 : FD_LOG_WARNING(( "expected %s snapshot size of %lu bytes but read %lu bytes",
1083 0 : full ? "full" : "incremental",
1084 0 : full ? ctx->metrics.full.bytes_total : ctx->metrics.incremental.bytes_total,
1085 0 : full ? ctx->metrics.full.bytes_read : ctx->metrics.incremental.bytes_read ));
1086 :
1087 0 : }
1088 0 : }
1089 0 : }
1090 :
1091 : static void
1092 : snapin_frag( fd_snapct_tile_t * ctx,
1093 0 : ulong sig ) {
1094 0 : switch( sig ) {
1095 0 : case FD_SNAPSHOT_MSG_CTRL_INIT_FULL:
1096 0 : if( FD_LIKELY( ctx->state==FD_SNAPCT_STATE_READING_FULL_HTTP ||
1097 0 : ctx->state==FD_SNAPCT_STATE_READING_FULL_FILE ) ) {
1098 0 : FD_TEST( !ctx->flush_ack );
1099 0 : ctx->flush_ack = 1;
1100 0 : } else FD_LOG_ERR(( "invalid control frag %lu in state %d", sig, ctx->state ));
1101 0 : break;
1102 :
1103 0 : case FD_SNAPSHOT_MSG_CTRL_INIT_INCR:
1104 0 : if( FD_LIKELY( ctx->state==FD_SNAPCT_STATE_READING_INCREMENTAL_HTTP ||
1105 0 : ctx->state==FD_SNAPCT_STATE_READING_INCREMENTAL_FILE ) ) {
1106 0 : FD_TEST( !ctx->flush_ack );
1107 0 : ctx->flush_ack = 1;
1108 0 : } else FD_LOG_ERR(( "invalid control frag %lu in state %d", sig, ctx->state ));
1109 0 : break;
1110 :
1111 0 : case FD_SNAPSHOT_MSG_CTRL_NEXT:
1112 0 : if( FD_LIKELY( ctx->state==FD_SNAPCT_STATE_FLUSHING_FULL_HTTP ||
1113 0 : ctx->state==FD_SNAPCT_STATE_FLUSHING_FULL_FILE ) ) {
1114 0 : FD_TEST( !ctx->flush_ack );
1115 0 : ctx->flush_ack = 1;
1116 0 : } else FD_LOG_ERR(( "invalid control frag %lu in state %d", sig, ctx->state ));
1117 0 : break;
1118 :
1119 0 : case FD_SNAPSHOT_MSG_CTRL_DONE:
1120 0 : if( FD_LIKELY( ctx->state==FD_SNAPCT_STATE_FLUSHING_FULL_HTTP ||
1121 0 : ctx->state==FD_SNAPCT_STATE_FLUSHING_FULL_FILE ||
1122 0 : ctx->state==FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP ||
1123 0 : ctx->state==FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_FILE ) ) {
1124 0 : FD_TEST( !ctx->flush_ack );
1125 0 : ctx->flush_ack = 1;
1126 0 : } else FD_LOG_ERR(( "invalid control frag %lu in state %d", sig, ctx->state ));
1127 0 : break;
1128 :
1129 0 : case FD_SNAPSHOT_MSG_CTRL_FAIL:
1130 0 : if( FD_LIKELY( ctx->state==FD_SNAPCT_STATE_FLUSHING_FULL_HTTP_RESET ||
1131 0 : ctx->state==FD_SNAPCT_STATE_FLUSHING_FULL_FILE_RESET ||
1132 0 : ctx->state==FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP_RESET ||
1133 0 : ctx->state==FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_FILE_RESET ) ) {
1134 0 : FD_TEST( !ctx->flush_ack );
1135 0 : ctx->flush_ack = 1;
1136 0 : } else FD_LOG_ERR(( "invalid control frag %lu in state %d", sig, ctx->state ));
1137 0 : break;
1138 :
1139 0 : case FD_SNAPSHOT_MSG_CTRL_SHUTDOWN:
1140 0 : break;
1141 :
1142 0 : case FD_SNAPSHOT_MSG_CTRL_ERROR:
1143 0 : switch( ctx->state ) {
1144 0 : case FD_SNAPCT_STATE_READING_FULL_FILE:
1145 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_FILE:
1146 0 : case FD_SNAPCT_STATE_READING_INCREMENTAL_FILE:
1147 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_FILE:
1148 0 : case FD_SNAPCT_STATE_READING_FULL_HTTP:
1149 0 : case FD_SNAPCT_STATE_FLUSHING_FULL_HTTP:
1150 0 : case FD_SNAPCT_STATE_READING_INCREMENTAL_HTTP:
1151 0 : case FD_SNAPCT_STATE_FLUSHING_INCREMENTAL_HTTP:
1152 0 : ctx->malformed = 1;
1153 0 : break;
1154 0 : default:
1155 0 : break;
1156 0 : }
1157 0 : break;
1158 0 : }
1159 0 : }
1160 :
1161 : static int
1162 : returnable_frag( fd_snapct_tile_t * ctx,
1163 : ulong in_idx,
1164 : ulong seq FD_PARAM_UNUSED,
1165 : ulong sig,
1166 : ulong chunk,
1167 : ulong sz,
1168 : ulong ctl FD_PARAM_UNUSED,
1169 : ulong tsorig FD_PARAM_UNUSED,
1170 : ulong tspub FD_PARAM_UNUSED,
1171 0 : fd_stem_context_t * stem ) {
1172 0 : if( FD_LIKELY( ctx->in_kind[ in_idx ]==IN_KIND_GOSSIP ) ) {
1173 0 : gossip_frag( ctx, sig, sz, chunk );
1174 0 : } else if( ctx->in_kind[ in_idx ]==IN_KIND_SNAPLD ) {
1175 0 : snapld_frag( ctx, sig, sz, chunk, stem );
1176 0 : } else if( ctx->in_kind[ in_idx ]==IN_KIND_ACK ) {
1177 0 : snapin_frag( ctx, sig );
1178 0 : } else FD_LOG_ERR(( "invalid in_kind %lu %u", in_idx, (uint)ctx->in_kind[ in_idx ] ));
1179 0 : return 0;
1180 0 : }
1181 :
1182 : static void
1183 : privileged_init( fd_topo_t * topo,
1184 0 : fd_topo_tile_t * tile ) {
1185 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
1186 :
1187 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
1188 0 : fd_snapct_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_snapct_tile_t), sizeof(fd_snapct_tile_t) );
1189 0 : void * _ssping = FD_SCRATCH_ALLOC_APPEND( l, fd_ssping_align(), fd_ssping_footprint( TOTAL_PEERS_MAX ) );
1190 0 : FD_SCRATCH_ALLOC_APPEND( l, alignof(gossip_ci_entry_t), sizeof(gossip_ci_entry_t)*GOSSIP_PEERS_MAX );
1191 0 : FD_SCRATCH_ALLOC_APPEND( l, gossip_ci_map_align(), gossip_ci_map_footprint( gossip_ci_map_chain_cnt_est( GOSSIP_PEERS_MAX ) ) );
1192 0 : void * _ssresolver = FD_SCRATCH_ALLOC_APPEND( l, fd_http_resolver_align(), fd_http_resolver_footprint( SERVER_PEERS_MAX ) );
1193 0 : FD_SCRATCH_ALLOC_APPEND( l, fd_sspeer_selector_align(), fd_sspeer_selector_footprint( TOTAL_PEERS_MAX ) );
1194 :
1195 0 : #if FD_HAS_OPENSSL
1196 0 : void * _alloc = FD_SCRATCH_ALLOC_APPEND( l, fd_alloc_align(), fd_alloc_footprint() );
1197 0 : fd_alloc_t * alloc = fd_alloc_join( fd_alloc_new( _alloc, 1UL ), tile->kind_id );
1198 0 : fd_ossl_tile_init( alloc );
1199 0 : #endif
1200 :
1201 0 : ctx->ssping = NULL;
1202 0 : if( FD_LIKELY( download_enabled( tile ) ) ) ctx->ssping = fd_ssping_join( fd_ssping_new( _ssping, TOTAL_PEERS_MAX, 1UL, on_ping, ctx ) );
1203 0 : if( FD_LIKELY( tile->snapct.sources.servers_cnt ) ) ctx->ssresolver = fd_http_resolver_join( fd_http_resolver_new( _ssresolver, SERVER_PEERS_MAX, tile->snapct.incremental_snapshots, on_resolve, ctx ) );
1204 0 : else ctx->ssresolver = NULL;
1205 :
1206 : /* FIXME: We will keep too many snapshots if we have local snapshots
1207 : but elect not to use them due to their age. */
1208 0 : fd_ssarchive_remove_old_snapshots( tile->snapct.snapshots_path,
1209 0 : tile->snapct.max_full_snapshots_to_keep,
1210 0 : tile->snapct.max_incremental_snapshots_to_keep );
1211 :
1212 0 : ulong full_slot = ULONG_MAX;
1213 0 : ulong incremental_slot = ULONG_MAX;
1214 0 : int full_is_zstd = 0;
1215 0 : int incremental_is_zstd = 0;
1216 0 : char full_path[ PATH_MAX ] = {0};
1217 0 : char incremental_path[ PATH_MAX ] = {0};
1218 0 : if( FD_UNLIKELY( -1==fd_ssarchive_latest_pair( tile->snapct.snapshots_path,
1219 0 : tile->snapct.incremental_snapshots,
1220 0 : &full_slot,
1221 0 : &incremental_slot,
1222 0 : full_path,
1223 0 : incremental_path,
1224 0 : &full_is_zstd,
1225 0 : &incremental_is_zstd ) ) ) {
1226 0 : if( FD_UNLIKELY( !download_enabled( tile ) ) ) {
1227 0 : FD_LOG_ERR(( "No snapshots found in `%s` and no download sources are enabled. "
1228 0 : "Please enable downloading via [snapshots.sources] and restart.", tile->snapct.snapshots_path ));
1229 0 : }
1230 0 : ctx->local_in.full_snapshot_slot = ULONG_MAX;
1231 0 : ctx->local_in.incremental_snapshot_slot = ULONG_MAX;
1232 0 : ctx->local_in.full_snapshot_size = 0UL;
1233 0 : ctx->local_in.incremental_snapshot_size = 0UL;
1234 0 : ctx->local_in.full_snapshot_zstd = 0;
1235 0 : ctx->local_in.incremental_snapshot_zstd = 0;
1236 0 : fd_cstr_fini( ctx->local_in.full_snapshot_path );
1237 0 : fd_cstr_fini( ctx->local_in.incremental_snapshot_path );
1238 0 : } else {
1239 0 : FD_TEST( full_slot!=ULONG_MAX );
1240 :
1241 0 : ctx->local_in.full_snapshot_slot = full_slot;
1242 0 : ctx->local_in.incremental_snapshot_slot = incremental_slot;
1243 0 : ctx->local_in.full_snapshot_zstd = full_is_zstd;
1244 0 : ctx->local_in.incremental_snapshot_zstd = incremental_is_zstd;
1245 :
1246 0 : strncpy( ctx->local_in.full_snapshot_path, full_path, PATH_MAX );
1247 0 : struct stat full_stat;
1248 0 : if( FD_UNLIKELY( -1==stat( ctx->local_in.full_snapshot_path, &full_stat ) ) ) FD_LOG_ERR(( "stat() failed `%s` (%i-%s)", full_path, errno, fd_io_strerror( errno ) ));
1249 0 : if( FD_UNLIKELY( !S_ISREG( full_stat.st_mode ) ) ) FD_LOG_ERR(( "full snapshot path `%s` is not a regular file", full_path ));
1250 0 : ctx->local_in.full_snapshot_size = (ulong)full_stat.st_size;
1251 :
1252 0 : if( FD_LIKELY( incremental_slot!=ULONG_MAX ) ) {
1253 0 : strncpy( ctx->local_in.incremental_snapshot_path, incremental_path, PATH_MAX );
1254 0 : struct stat incremental_stat;
1255 0 : if( FD_UNLIKELY( -1==stat( ctx->local_in.incremental_snapshot_path, &incremental_stat ) ) ) FD_LOG_ERR(( "stat() failed `%s` (%i-%s)", incremental_path, errno, fd_io_strerror( errno ) ));
1256 0 : if( FD_UNLIKELY( !S_ISREG( incremental_stat.st_mode ) ) ) FD_LOG_ERR(( "incremental snapshot path `%s` is not a regular file", incremental_path ));
1257 0 : ctx->local_in.incremental_snapshot_size = (ulong)incremental_stat.st_size;
1258 0 : } else {
1259 0 : ctx->local_in.incremental_snapshot_size = 0UL;
1260 0 : fd_cstr_fini( ctx->local_in.incremental_snapshot_path );
1261 0 : }
1262 0 : }
1263 :
1264 0 : ctx->local_out.dir_fd = -1;
1265 0 : ctx->local_out.full_snapshot_fd = -1;
1266 0 : ctx->local_out.incremental_snapshot_fd = -1;
1267 0 : if( FD_LIKELY( download_enabled( tile ) ) ) {
1268 0 : ctx->local_out.dir_fd = open( tile->snapct.snapshots_path, O_DIRECTORY|O_CLOEXEC );
1269 0 : if( FD_UNLIKELY( -1==ctx->local_out.dir_fd ) ) FD_LOG_ERR(( "open(%s) failed (%i-%s)", tile->snapct.snapshots_path, errno, fd_io_strerror( errno ) ));
1270 :
1271 0 : ctx->local_out.full_snapshot_fd = openat( ctx->local_out.dir_fd, TEMP_FULL_SNAP_NAME, O_WRONLY|O_CREAT|O_TRUNC|O_NONBLOCK, S_IRUSR|S_IWUSR );
1272 0 : if( FD_UNLIKELY( -1==ctx->local_out.full_snapshot_fd ) ) FD_LOG_ERR(( "open(%s/%s) failed (%i-%s)", tile->snapct.snapshots_path, TEMP_FULL_SNAP_NAME, errno, fd_io_strerror( errno ) ));
1273 :
1274 0 : if( FD_LIKELY( tile->snapct.incremental_snapshots ) ) {
1275 0 : ctx->local_out.incremental_snapshot_fd = openat( ctx->local_out.dir_fd, TEMP_INCR_SNAP_NAME, O_WRONLY|O_CREAT|O_TRUNC|O_NONBLOCK, S_IRUSR|S_IWUSR );
1276 0 : if( FD_UNLIKELY( -1==ctx->local_out.incremental_snapshot_fd ) ) FD_LOG_ERR(( "open(%s/%s) failed (%i-%s)", tile->snapct.snapshots_path, TEMP_INCR_SNAP_NAME, errno, fd_io_strerror( errno ) ));
1277 0 : }
1278 0 : }
1279 0 : }
1280 :
1281 : static inline fd_snapct_out_link_t
1282 : out1( fd_topo_t const * topo,
1283 : fd_topo_tile_t const * tile,
1284 0 : char const * name ) {
1285 0 : ulong idx = ULONG_MAX;
1286 :
1287 0 : for( ulong i=0UL; i<tile->out_cnt; i++ ) {
1288 0 : fd_topo_link_t const * link = &topo->links[ tile->out_link_id[ i ] ];
1289 0 : if( !strcmp( link->name, name ) ) {
1290 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 ));
1291 0 : idx = i;
1292 0 : }
1293 0 : }
1294 :
1295 0 : if( FD_UNLIKELY( idx==ULONG_MAX ) ) return (fd_snapct_out_link_t){ .idx = ULONG_MAX, .mem = NULL, .chunk0 = 0, .wmark = 0, .chunk = 0, .mtu = 0 };
1296 :
1297 0 : ulong mtu = topo->links[ tile->out_link_id[ idx ] ].mtu;
1298 0 : if( FD_UNLIKELY( mtu==0UL ) ) return (fd_snapct_out_link_t){ .idx = idx, .mem = NULL, .chunk0 = ULONG_MAX, .wmark = ULONG_MAX, .chunk = ULONG_MAX, .mtu = mtu };
1299 :
1300 0 : void * mem = topo->workspaces[ topo->objs[ topo->links[ tile->out_link_id[ idx ] ].dcache_obj_id ].wksp_id ].wksp;
1301 0 : ulong chunk0 = fd_dcache_compact_chunk0( mem, topo->links[ tile->out_link_id[ idx ] ].dcache );
1302 0 : ulong wmark = fd_dcache_compact_wmark ( mem, topo->links[ tile->out_link_id[ idx ] ].dcache, mtu );
1303 0 : return (fd_snapct_out_link_t){ .idx = idx, .mem = mem, .chunk0 = chunk0, .wmark = wmark, .chunk = chunk0, .mtu = mtu };
1304 0 : }
1305 :
1306 : static void
1307 : unprivileged_init( fd_topo_t * topo,
1308 0 : fd_topo_tile_t * tile ) {
1309 0 : void * scratch = fd_topo_obj_laddr( topo, tile->tile_obj_id );
1310 :
1311 0 : FD_SCRATCH_ALLOC_INIT( l, scratch );
1312 0 : fd_snapct_tile_t * ctx = FD_SCRATCH_ALLOC_APPEND( l, alignof(fd_snapct_tile_t), sizeof(fd_snapct_tile_t) );
1313 0 : FD_SCRATCH_ALLOC_APPEND( l, fd_ssping_align(), fd_ssping_footprint( TOTAL_PEERS_MAX ) );
1314 0 : void * _ci_table = FD_SCRATCH_ALLOC_APPEND( l, alignof(gossip_ci_entry_t), sizeof(gossip_ci_entry_t) * GOSSIP_PEERS_MAX );
1315 0 : void * _ci_map = FD_SCRATCH_ALLOC_APPEND( l, gossip_ci_map_align(), gossip_ci_map_footprint( gossip_ci_map_chain_cnt_est( GOSSIP_PEERS_MAX ) ) );
1316 0 : FD_SCRATCH_ALLOC_APPEND( l, fd_http_resolver_align(), fd_http_resolver_footprint( SERVER_PEERS_MAX ) );
1317 0 : void * _selector = FD_SCRATCH_ALLOC_APPEND( l, fd_sspeer_selector_align(), fd_sspeer_selector_footprint( TOTAL_PEERS_MAX ) );
1318 :
1319 0 : fd_memcpy( &ctx->config, &tile->snapct, sizeof(ctx->config) );
1320 0 : ctx->gossip_enabled = gossip_enabled( tile );
1321 0 : ctx->download_enabled = download_enabled( tile );
1322 :
1323 0 : if( ctx->config.sources.servers_cnt ) {
1324 0 : for( ulong i=0UL; i<tile->snapct.sources.servers_cnt; i++ ) {
1325 0 : fd_ssping_add ( ctx->ssping, tile->snapct.sources.servers[ i ].addr );
1326 0 : fd_http_resolver_add( ctx->ssresolver,
1327 0 : tile->snapct.sources.servers[ i ].addr,
1328 0 : tile->snapct.sources.servers[ i ].hostname,
1329 0 : tile->snapct.sources.servers[ i ].is_https );
1330 0 : }
1331 0 : }
1332 :
1333 0 : ctx->selector = fd_sspeer_selector_join( fd_sspeer_selector_new( _selector, TOTAL_PEERS_MAX, ctx->config.incremental_snapshots, 1UL ) );
1334 :
1335 0 : ctx->state = FD_SNAPCT_STATE_WAITING_FOR_PEERS;
1336 0 : ctx->malformed = 0;
1337 0 : ctx->deadline_nanos = fd_log_wallclock() + FD_SNAPCT_WAITING_FOR_PEERS_TIMEOUT;
1338 0 : ctx->flush_ack = 0;
1339 0 : ctx->addr.l = 0UL;
1340 :
1341 0 : fd_memset( ctx->http_full_snapshot_name, 0, PATH_MAX );
1342 0 : fd_memset( ctx->http_incr_snapshot_name, 0, PATH_MAX );
1343 :
1344 0 : ctx->gossip_in_mem = NULL;
1345 0 : int has_snapld_dc = 0, has_ack_loopback = 0;
1346 0 : FD_TEST( tile->in_cnt<=MAX_IN_LINKS );
1347 0 : for( ulong i=0UL; i<(tile->in_cnt); i++ ) {
1348 0 : fd_topo_link_t * in_link = &topo->links[ tile->in_link_id[ i ] ];
1349 0 : if( 0==strcmp( in_link->name, "gossip_out" ) ) {
1350 0 : ctx->in_kind[ i ] = IN_KIND_GOSSIP;
1351 0 : ctx->gossip_in_mem = topo->workspaces[ topo->objs[ in_link->dcache_obj_id ].wksp_id ].wksp;
1352 0 : } else if( 0==strcmp( in_link->name, "snapld_dc" ) ) {
1353 0 : ctx->in_kind[ i ] = IN_KIND_SNAPLD;
1354 0 : ctx->snapld_in_mem = topo->workspaces[ topo->objs[ in_link->dcache_obj_id ].wksp_id ].wksp;
1355 0 : FD_TEST( !has_snapld_dc );
1356 0 : has_snapld_dc = 1;
1357 0 : } else if( 0==strcmp( in_link->name, "snapin_ct" ) || 0==strcmp( in_link->name, "snapls_ct" ) ) {
1358 0 : ctx->in_kind[ i ] = IN_KIND_ACK;
1359 0 : FD_TEST( !has_ack_loopback );
1360 0 : has_ack_loopback = 1;
1361 0 : }
1362 0 : }
1363 0 : FD_TEST( has_snapld_dc && has_ack_loopback );
1364 0 : FD_TEST( ctx->gossip_enabled==(ctx->gossip_in_mem!=NULL) );
1365 :
1366 0 : ctx->predicted_incremental.full_slot = ULONG_MAX;
1367 0 : ctx->predicted_incremental.slot = ULONG_MAX;
1368 0 : ctx->predicted_incremental.dirty = 0;
1369 :
1370 0 : fd_memset( &ctx->metrics, 0, sizeof(ctx->metrics) );
1371 :
1372 0 : fd_memset( _ci_table, 0, sizeof(gossip_ci_entry_t) * GOSSIP_PEERS_MAX );
1373 0 : ctx->gossip.ci_table = _ci_table;
1374 0 : ctx->gossip.ci_map = gossip_ci_map_join( gossip_ci_map_new( _ci_map, gossip_ci_map_chain_cnt_est( GOSSIP_PEERS_MAX ), 0UL ) );
1375 0 : ctx->gossip.fresh_cnt = 0UL;
1376 0 : ctx->gossip.total_cnt = 0UL;
1377 0 : ctx->gossip.saturated = !ctx->gossip_enabled;
1378 0 : ctx->gossip.next_saturated_check = 0;
1379 :
1380 0 : if( FD_UNLIKELY( tile->out_cnt<2UL || tile->out_cnt>3UL ) ) FD_LOG_ERR(( "tile `" NAME "` has %lu outs, expected 2-3", tile->out_cnt ));
1381 0 : ctx->out_ld = out1( topo, tile, "snapct_ld" );
1382 0 : ctx->out_gui = out1( topo, tile, "snapct_gui" );
1383 0 : ctx->out_rp = out1( topo, tile, "snapct_repr" );
1384 0 : }
1385 :
1386 : /* after_credit can result in as many as 5 stem publishes in some code
1387 : paths, and returnable_frag can result in 1. */
1388 0 : #define STEM_BURST 6UL
1389 :
1390 0 : #define STEM_LAZY 1000L
1391 :
1392 0 : #define STEM_CALLBACK_CONTEXT_TYPE fd_snapct_tile_t
1393 0 : #define STEM_CALLBACK_CONTEXT_ALIGN alignof(fd_snapct_tile_t)
1394 :
1395 : #define STEM_CALLBACK_SHOULD_SHUTDOWN should_shutdown
1396 0 : #define STEM_CALLBACK_DURING_HOUSEKEEPING during_housekeeping
1397 0 : #define STEM_CALLBACK_METRICS_WRITE metrics_write
1398 0 : #define STEM_CALLBACK_AFTER_CREDIT after_credit
1399 0 : #define STEM_CALLBACK_RETURNABLE_FRAG returnable_frag
1400 :
1401 : #include "../../disco/stem/fd_stem.c"
1402 :
1403 : fd_topo_run_tile_t fd_tile_snapct = {
1404 : .name = NAME,
1405 : .rlimit_file_cnt_fn = rlimit_file_cnt,
1406 : .populate_allowed_seccomp = populate_allowed_seccomp,
1407 : .populate_allowed_fds = populate_allowed_fds,
1408 : .scratch_align = scratch_align,
1409 : .scratch_footprint = scratch_footprint,
1410 : .loose_footprint = loose_footprint,
1411 : .privileged_init = privileged_init,
1412 : .unprivileged_init = unprivileged_init,
1413 : .run = stem_run,
1414 : .keep_host_networking = 1,
1415 : .allow_connect = 1,
1416 : .allow_renameat = 1,
1417 : };
1418 :
1419 : #undef NAME
|
