1 /* Copyright (c) 2017-2021, The Tor Project, Inc. */
2 /* See LICENSE for licensing information */
7 * \brief consensus diff manager functions
9 * This module is run by directory authorities and caches in order
10 * to remember a number of past consensus documents, and to generate
11 * and serve the diffs from those documents to the latest consensus.
14 #define CONSDIFFMGR_PRIVATE
16 #include "core/or/or.h"
17 #include "app/config/config.h"
18 #include "feature/dircache/conscache.h"
19 #include "feature/dircommon/consdiff.h"
20 #include "feature/dircache/consdiffmgr.h"
21 #include "core/mainloop/cpuworker.h"
22 #include "feature/nodelist/networkstatus.h"
23 #include "feature/dirparse/ns_parse.h"
24 #include "lib/evloop/compat_libevent.h"
25 #include "lib/evloop/workqueue.h"
26 #include "lib/compress/compress.h"
27 #include "lib/encoding/confline.h"
29 #include "feature/nodelist/networkstatus_st.h"
30 #include "feature/nodelist/networkstatus_voter_info_st.h"
33 * Labels to apply to items in the conscache object.
37 /* One of DOCTYPE_CONSENSUS or DOCTYPE_CONSENSUS_DIFF */
38 #define LABEL_DOCTYPE "document-type"
39 /* The valid-after time for a consensus (or for the target consensus of a
40 * diff), encoded as ISO UTC. */
41 #define LABEL_VALID_AFTER "consensus-valid-after"
42 /* The fresh-until time for a consensus (or for the target consensus of a
43 * diff), encoded as ISO UTC. */
44 #define LABEL_FRESH_UNTIL "consensus-fresh-until"
45 /* The valid-until time for a consensus (or for the target consensus of a
46 * diff), encoded as ISO UTC. */
47 #define LABEL_VALID_UNTIL "consensus-valid-until"
48 /* Comma-separated list of hex-encoded identity digests for the voting
50 #define LABEL_SIGNATORIES "consensus-signatories"
51 /* A hex encoded SHA3 digest of the object, as compressed (if any) */
52 #define LABEL_SHA3_DIGEST "sha3-digest"
53 /* A hex encoded SHA3 digest of the object before compression. */
54 #define LABEL_SHA3_DIGEST_UNCOMPRESSED "sha3-digest-uncompressed"
55 /* A hex encoded SHA3 digest-as-signed of a consensus */
56 #define LABEL_SHA3_DIGEST_AS_SIGNED "sha3-digest-as-signed"
57 /* The flavor of the consensus or consensuses diff */
58 #define LABEL_FLAVOR "consensus-flavor"
59 /* Diff only: the SHA3 digest-as-signed of the source consensus. */
60 #define LABEL_FROM_SHA3_DIGEST "from-sha3-digest"
61 /* Diff only: the SHA3 digest-in-full of the target consensus. */
62 #define LABEL_TARGET_SHA3_DIGEST "target-sha3-digest"
63 /* Diff only: the valid-after date of the source consensus. */
64 #define LABEL_FROM_VALID_AFTER "from-valid-after"
65 /* What kind of compression was used? */
66 #define LABEL_COMPRESSION_TYPE "compression"
69 #define DOCTYPE_CONSENSUS "consensus"
70 #define DOCTYPE_CONSENSUS_DIFF "consensus-diff"
73 * Underlying directory that stores consensuses and consensus diffs. Don't
74 * use this directly: use cdm_cache_get() instead.
76 static consensus_cache_t
*cons_diff_cache
= NULL
;
78 * If true, we have learned at least one new consensus since the
79 * consensus cache was last up-to-date.
81 static int cdm_cache_dirty
= 0;
83 * If true, we have scanned the cache to update our hashtable of diffs.
85 static int cdm_cache_loaded
= 0;
88 * Possible status values for cdm_diff_t.cdm_diff_status
90 typedef enum cdm_diff_status_t
{
92 CDM_DIFF_IN_PROGRESS
=2,
96 /** Which methods do we use for precompressing diffs? */
97 static const compress_method_t compress_diffs_with
[] = {
109 * Event for rescanning the cache.
111 static mainloop_event_t
*consdiffmgr_rescan_ev
= NULL
;
113 static void consdiffmgr_rescan_cb(mainloop_event_t
*ev
, void *arg
);
114 static void mark_cdm_cache_dirty(void);
116 /** How many different methods will we try to use for diff compression? */
118 n_diff_compression_methods(void)
120 return ARRAY_LENGTH(compress_diffs_with
);
123 /** Which methods do we use for precompressing consensuses? */
124 static const compress_method_t compress_consensus_with
[] = {
134 /** How many different methods will we try to use for diff compression? */
136 n_consensus_compression_methods(void)
138 return ARRAY_LENGTH(compress_consensus_with
);
141 /** For which compression method do we retain old consensuses? There's no
142 * need to keep all of them, since we won't be serving them. We'll
143 * go with ZLIB_METHOD because it's pretty fast and everyone has it.
145 #define RETAIN_CONSENSUS_COMPRESSED_WITH_METHOD ZLIB_METHOD
147 /** Handles pointing to the latest consensus entries as compressed and
149 static consensus_cache_entry_handle_t
*
150 latest_consensus
[N_CONSENSUS_FLAVORS
]
151 [ARRAY_LENGTH(compress_consensus_with
)];
153 /** Hashtable node used to remember the current status of the diff
154 * from a given sha3 digest to the current consensus. */
155 typedef struct cdm_diff_t
{
156 HT_ENTRY(cdm_diff_t
) node
;
158 /** Consensus flavor for this diff (part of ht key) */
159 consensus_flavor_t flavor
;
160 /** SHA3-256 digest of the consensus that this diff is _from_. (part of the
162 uint8_t from_sha3
[DIGEST256_LEN
];
163 /** Method by which the diff is compressed. (part of the ht key */
164 compress_method_t compress_method
;
166 /** One of the CDM_DIFF_* values, depending on whether this diff
167 * is available, in progress, or impossible to compute. */
168 cdm_diff_status_t cdm_diff_status
;
169 /** SHA3-256 digest of the consensus that this diff is _to. */
170 uint8_t target_sha3
[DIGEST256_LEN
];
172 /** Handle to the cache entry for this diff, if any. We use a handle here
173 * to avoid thinking too hard about cache entry lifetime issues. */
174 consensus_cache_entry_handle_t
*entry
;
177 /** Hashtable mapping flavor and source consensus digest to status. */
178 static HT_HEAD(cdm_diff_ht
, cdm_diff_t
) cdm_diff_ht
= HT_INITIALIZER();
181 // XXX(ahf): For tor#24857, a contributor suggested that on Windows, the CPU
182 // begins to spike at 100% once the number of files handled by the consensus
183 // diff manager becomes larger than 64. To see if the issue goes away, we
184 // hardcode this value to 64 now while we investigate a better solution.
185 # define CACHE_MAX_NUM 64
186 #else /* !defined(_WIN32) */
187 # define CACHE_MAX_NUM 128
188 #endif /* defined(_WIN32) */
191 * Configuration for this module
193 static consdiff_cfg_t consdiff_cfg
= {
194 // XXXX I'd like to make this number bigger, but it interferes with the
195 // XXXX seccomp2 syscall filter, which tops out at BPF_MAXINS (4096)
197 /* .cache_max_num = */ CACHE_MAX_NUM
200 static int consdiffmgr_ensure_space_for_files(int n
);
201 static int consensus_queue_compression_work(const char *consensus
,
202 size_t consensus_len
,
203 const networkstatus_t
*as_parsed
);
204 static int consensus_diff_queue_diff_work(consensus_cache_entry_t
*diff_from
,
205 consensus_cache_entry_t
*diff_to
);
206 static void consdiffmgr_set_cache_flags(void);
212 /** Helper: hash the key of a cdm_diff_t. */
214 cdm_diff_hash(const cdm_diff_t
*diff
)
216 uint8_t tmp
[DIGEST256_LEN
+ 2];
217 memcpy(tmp
, diff
->from_sha3
, DIGEST256_LEN
);
218 tmp
[DIGEST256_LEN
] = (uint8_t) diff
->flavor
;
219 tmp
[DIGEST256_LEN
+1] = (uint8_t) diff
->compress_method
;
220 return (unsigned) siphash24g(tmp
, sizeof(tmp
));
222 /** Helper: compare two cdm_diff_t objects for key equality */
224 cdm_diff_eq(const cdm_diff_t
*diff1
, const cdm_diff_t
*diff2
)
226 return fast_memeq(diff1
->from_sha3
, diff2
->from_sha3
, DIGEST256_LEN
) &&
227 diff1
->flavor
== diff2
->flavor
&&
228 diff1
->compress_method
== diff2
->compress_method
;
231 HT_PROTOTYPE(cdm_diff_ht
, cdm_diff_t
, node
, cdm_diff_hash
, cdm_diff_eq
);
232 HT_GENERATE2(cdm_diff_ht
, cdm_diff_t
, node
, cdm_diff_hash
, cdm_diff_eq
,
233 0.6, tor_reallocarray
, tor_free_
);
235 #define cdm_diff_free(diff) \
236 FREE_AND_NULL(cdm_diff_t, cdm_diff_free_, (diff))
238 /** Release all storage held in <b>diff</b>. */
240 cdm_diff_free_(cdm_diff_t
*diff
)
244 consensus_cache_entry_handle_free(diff
->entry
);
248 /** Create and return a new cdm_diff_t with the given values. Does not
249 * add it to the hashtable. */
251 cdm_diff_new(consensus_flavor_t flav
,
252 const uint8_t *from_sha3
,
253 const uint8_t *target_sha3
,
254 compress_method_t method
)
257 ent
= tor_malloc_zero(sizeof(cdm_diff_t
));
259 memcpy(ent
->from_sha3
, from_sha3
, DIGEST256_LEN
);
260 memcpy(ent
->target_sha3
, target_sha3
, DIGEST256_LEN
);
261 ent
->compress_method
= method
;
266 * Examine the diff hashtable to see whether we know anything about computing
267 * a diff of type <b>flav</b> between consensuses with the two provided
268 * SHA3-256 digests. If a computation is in progress, or if the computation
269 * has already been tried and failed, return 1. Otherwise, note the
270 * computation as "in progress" so that we don't reattempt it later, and
274 cdm_diff_ht_check_and_note_pending(consensus_flavor_t flav
,
275 const uint8_t *from_sha3
,
276 const uint8_t *target_sha3
)
278 struct cdm_diff_t search
, *ent
;
281 for (u
= 0; u
< n_diff_compression_methods(); ++u
) {
282 compress_method_t method
= compress_diffs_with
[u
];
283 memset(&search
, 0, sizeof(cdm_diff_t
));
284 search
.flavor
= flav
;
285 search
.compress_method
= method
;
286 memcpy(search
.from_sha3
, from_sha3
, DIGEST256_LEN
);
287 ent
= HT_FIND(cdm_diff_ht
, &cdm_diff_ht
, &search
);
289 tor_assert_nonfatal(ent
->cdm_diff_status
!= CDM_DIFF_PRESENT
);
293 ent
= cdm_diff_new(flav
, from_sha3
, target_sha3
, method
);
294 ent
->cdm_diff_status
= CDM_DIFF_IN_PROGRESS
;
295 HT_INSERT(cdm_diff_ht
, &cdm_diff_ht
, ent
);
301 * Update the status of the diff of type <b>flav</b> between consensuses with
302 * the two provided SHA3-256 digests, so that its status becomes
303 * <b>status</b>, and its value becomes the <b>handle</b>. If <b>handle</b>
304 * is NULL, then the old handle (if any) is freed, and replaced with NULL.
307 cdm_diff_ht_set_status(consensus_flavor_t flav
,
308 const uint8_t *from_sha3
,
309 const uint8_t *to_sha3
,
310 compress_method_t method
,
312 consensus_cache_entry_handle_t
*handle
)
314 if (handle
== NULL
) {
315 tor_assert_nonfatal(status
!= CDM_DIFF_PRESENT
);
318 struct cdm_diff_t search
, *ent
;
319 memset(&search
, 0, sizeof(cdm_diff_t
));
320 search
.flavor
= flav
;
321 search
.compress_method
= method
,
322 memcpy(search
.from_sha3
, from_sha3
, DIGEST256_LEN
);
323 ent
= HT_FIND(cdm_diff_ht
, &cdm_diff_ht
, &search
);
325 ent
= cdm_diff_new(flav
, from_sha3
, to_sha3
, method
);
326 ent
->cdm_diff_status
= CDM_DIFF_IN_PROGRESS
;
327 HT_INSERT(cdm_diff_ht
, &cdm_diff_ht
, ent
);
328 } else if (fast_memneq(ent
->target_sha3
, to_sha3
, DIGEST256_LEN
)) {
329 // This can happen under certain really pathological conditions
330 // if we decide we don't care about a diff before it is actually
335 tor_assert_nonfatal(ent
->cdm_diff_status
== CDM_DIFF_IN_PROGRESS
);
337 ent
->cdm_diff_status
= status
;
338 consensus_cache_entry_handle_free(ent
->entry
);
343 * Helper: Remove from the hash table every present (actually computed) diff
344 * of type <b>flav</b> whose target digest does not match
345 * <b>unless_target_sha3_matches</b>.
347 * This function is used for the hash table to throw away references to diffs
348 * that do not lead to the most given consensus of a given flavor.
351 cdm_diff_ht_purge(consensus_flavor_t flav
,
352 const uint8_t *unless_target_sha3_matches
)
354 cdm_diff_t
**diff
, **next
;
355 for (diff
= HT_START(cdm_diff_ht
, &cdm_diff_ht
); diff
; diff
= next
) {
356 cdm_diff_t
*this = *diff
;
358 if ((*diff
)->cdm_diff_status
== CDM_DIFF_PRESENT
&&
359 flav
== (*diff
)->flavor
) {
361 if (BUG((*diff
)->entry
== NULL
) ||
362 consensus_cache_entry_handle_get((*diff
)->entry
) == NULL
) {
363 /* the underlying entry has gone away; drop this. */
364 next
= HT_NEXT_RMV(cdm_diff_ht
, &cdm_diff_ht
, diff
);
369 if (unless_target_sha3_matches
&&
370 fast_memneq(unless_target_sha3_matches
, (*diff
)->target_sha3
,
372 /* target hash doesn't match; drop this. */
373 next
= HT_NEXT_RMV(cdm_diff_ht
, &cdm_diff_ht
, diff
);
378 next
= HT_NEXT(cdm_diff_ht
, &cdm_diff_ht
, diff
);
383 * Helper: initialize <b>cons_diff_cache</b>.
388 unsigned n_entries
= consdiff_cfg
.cache_max_num
* 2;
390 tor_assert(cons_diff_cache
== NULL
);
391 cons_diff_cache
= consensus_cache_open("diff-cache", n_entries
);
392 if (cons_diff_cache
== NULL
) {
394 log_err(LD_FS
, "Error: Couldn't open storage for consensus diffs.");
395 tor_assert_unreached();
398 consdiffmgr_set_cache_flags();
400 consdiffmgr_rescan_ev
=
401 mainloop_event_postloop_new(consdiffmgr_rescan_cb
, NULL
);
402 mark_cdm_cache_dirty();
403 cdm_cache_loaded
= 0;
407 * Helper: return the consensus_cache_t * that backs this manager,
408 * initializing it if needed.
410 STATIC consensus_cache_t
*
413 if (PREDICT_UNLIKELY(cons_diff_cache
== NULL
)) {
416 return cons_diff_cache
;
420 * Helper: given a list of labels, prepend the hex-encoded SHA3 digest
421 * of the <b>bodylen</b>-byte object at <b>body</b> to those labels,
422 * with <b>label</b> as its label.
425 cdm_labels_prepend_sha3(config_line_t
**labels
,
430 uint8_t sha3_digest
[DIGEST256_LEN
];
431 char hexdigest
[HEX_DIGEST256_LEN
+1];
432 crypto_digest256((char *)sha3_digest
,
433 (const char *)body
, bodylen
, DIGEST_SHA3_256
);
434 base16_encode(hexdigest
, sizeof(hexdigest
),
435 (const char *)sha3_digest
, sizeof(sha3_digest
));
437 config_line_prepend(labels
, label
, hexdigest
);
440 /** Helper: if there is a sha3-256 hex-encoded digest in <b>ent</b> with the
441 * given label, set <b>digest_out</b> to that value (decoded), and return 0.
443 * Return -1 if there is no such label, and -2 if it is badly formatted. */
445 cdm_entry_get_sha3_value(uint8_t *digest_out
,
446 consensus_cache_entry_t
*ent
,
452 const char *hex
= consensus_cache_entry_get_value(ent
, label
);
456 int n
= base16_decode((char*)digest_out
, DIGEST256_LEN
, hex
, strlen(hex
));
457 if (n
!= DIGEST256_LEN
)
464 * Helper: look for a consensus with the given <b>flavor</b> and
465 * <b>valid_after</b> time in the cache. Return that consensus if it's
466 * present, or NULL if it's missing.
468 STATIC consensus_cache_entry_t
*
469 cdm_cache_lookup_consensus(consensus_flavor_t flavor
, time_t valid_after
)
471 char formatted_time
[ISO_TIME_LEN
+1];
472 format_iso_time_nospace(formatted_time
, valid_after
);
473 const char *flavname
= networkstatus_get_flavor_name(flavor
);
475 /* We'll filter by valid-after time first, since that should
476 * match the fewest documents. */
477 /* We could add an extra hashtable here, but since we only do this scan
478 * when adding a new consensus, it probably doesn't matter much. */
479 smartlist_t
*matches
= smartlist_new();
480 consensus_cache_find_all(matches
, cdm_cache_get(),
481 LABEL_VALID_AFTER
, formatted_time
);
482 consensus_cache_filter_list(matches
, LABEL_FLAVOR
, flavname
);
483 consensus_cache_filter_list(matches
, LABEL_DOCTYPE
, DOCTYPE_CONSENSUS
);
485 consensus_cache_entry_t
*result
= NULL
;
486 if (smartlist_len(matches
)) {
487 result
= smartlist_get(matches
, 0);
489 smartlist_free(matches
);
494 /** Return the maximum age (in seconds) of consensuses that we should consider
495 * storing. The available space in the directory may impose additional limits
496 * on how much we store. */
498 get_max_age_to_cache(void)
500 const int32_t DEFAULT_MAX_AGE_TO_CACHE
= 8192;
501 const int32_t MIN_MAX_AGE_TO_CACHE
= 0;
502 const int32_t MAX_MAX_AGE_TO_CACHE
= 8192;
503 const char MAX_AGE_TO_CACHE_NAME
[] = "max-consensus-age-to-cache-for-diff";
505 const or_options_t
*options
= get_options();
507 if (options
->MaxConsensusAgeForDiffs
) {
508 const int v
= options
->MaxConsensusAgeForDiffs
;
509 if (v
>= MAX_MAX_AGE_TO_CACHE
* 3600)
510 return MAX_MAX_AGE_TO_CACHE
;
515 /* The parameter is in hours, so we multiply */
516 return 3600 * networkstatus_get_param(NULL
,
517 MAX_AGE_TO_CACHE_NAME
,
518 DEFAULT_MAX_AGE_TO_CACHE
,
519 MIN_MAX_AGE_TO_CACHE
,
520 MAX_MAX_AGE_TO_CACHE
);
523 #ifdef TOR_UNIT_TESTS
524 /** As consdiffmgr_add_consensus, but requires a nul-terminated input. For
527 consdiffmgr_add_consensus_nulterm(const char *consensus
,
528 const networkstatus_t
*as_parsed
)
530 size_t len
= strlen(consensus
);
531 /* make a non-nul-terminated copy so that we can have a better chance
532 * of catching errors. */
533 char *ctmp
= tor_memdup(consensus
, len
);
534 int r
= consdiffmgr_add_consensus(ctmp
, len
, as_parsed
);
538 #endif /* defined(TOR_UNIT_TESTS) */
541 * Given a buffer containing a networkstatus consensus, and the results of
542 * having parsed that consensus, add that consensus to the cache if it is not
543 * already present and not too old. Create new consensus diffs from or to
544 * that consensus as appropriate.
546 * Return 0 on success and -1 on failure.
549 consdiffmgr_add_consensus(const char *consensus
,
550 size_t consensus_len
,
551 const networkstatus_t
*as_parsed
)
553 if (BUG(consensus
== NULL
) || BUG(as_parsed
== NULL
))
554 return -1; // LCOV_EXCL_LINE
555 if (BUG(as_parsed
->type
!= NS_TYPE_CONSENSUS
))
556 return -1; // LCOV_EXCL_LINE
558 const consensus_flavor_t flavor
= as_parsed
->flavor
;
559 const time_t valid_after
= as_parsed
->valid_after
;
561 if (valid_after
< approx_time() - get_max_age_to_cache()) {
562 log_info(LD_DIRSERV
, "We don't care about this consensus document; it's "
567 /* Do we already have this one? */
568 consensus_cache_entry_t
*entry
=
569 cdm_cache_lookup_consensus(flavor
, valid_after
);
571 log_info(LD_DIRSERV
, "We already have a copy of that consensus");
575 /* We don't have it. Add it to the cache. */
576 return consensus_queue_compression_work(consensus
, consensus_len
, as_parsed
);
580 * Helper: used to sort two smartlists of consensus_cache_entry_t by their
581 * LABEL_VALID_AFTER labels.
584 compare_by_valid_after_(const void **a
, const void **b
)
586 const consensus_cache_entry_t
*e1
= *a
;
587 const consensus_cache_entry_t
*e2
= *b
;
588 /* We're in luck here: sorting UTC iso-encoded values lexically will work
589 * fine (until 9999). */
590 return strcmp_opt(consensus_cache_entry_get_value(e1
, LABEL_VALID_AFTER
),
591 consensus_cache_entry_get_value(e2
, LABEL_VALID_AFTER
));
595 * Helper: Sort <b>lst</b> by LABEL_VALID_AFTER and return the most recent
598 static consensus_cache_entry_t
*
599 sort_and_find_most_recent(smartlist_t
*lst
)
601 smartlist_sort(lst
, compare_by_valid_after_
);
602 if (smartlist_len(lst
)) {
603 return smartlist_get(lst
, smartlist_len(lst
) - 1);
609 /** Return i such that compress_consensus_with[i] == method. Return
610 * -1 if no such i exists. */
612 consensus_compression_method_pos(compress_method_t method
)
615 for (i
= 0; i
< n_consensus_compression_methods(); ++i
) {
616 if (compress_consensus_with
[i
] == method
) {
624 * If we know a consensus with the flavor <b>flavor</b> compressed with
625 * <b>method</b>, set *<b>entry_out</b> to that value. Return values are as
626 * for consdiffmgr_find_diff_from().
629 consdiffmgr_find_consensus(struct consensus_cache_entry_t
**entry_out
,
630 consensus_flavor_t flavor
,
631 compress_method_t method
)
633 tor_assert(entry_out
);
634 tor_assert((int)flavor
< N_CONSENSUS_FLAVORS
);
636 int pos
= consensus_compression_method_pos(method
);
638 // We don't compress consensuses with this method.
639 return CONSDIFF_NOT_FOUND
;
641 consensus_cache_entry_handle_t
*handle
= latest_consensus
[flavor
][pos
];
643 return CONSDIFF_NOT_FOUND
;
644 *entry_out
= consensus_cache_entry_handle_get(handle
);
646 return CONSDIFF_AVAILABLE
;
648 return CONSDIFF_NOT_FOUND
;
652 * Look up consensus_cache_entry_t for the consensus of type <b>flavor</b>,
653 * from the source consensus with the specified digest (which must be SHA3).
655 * If the diff is present, store it into *<b>entry_out</b> and return
656 * CONSDIFF_AVAILABLE. Otherwise return CONSDIFF_NOT_FOUND or
657 * CONSDIFF_IN_PROGRESS.
660 consdiffmgr_find_diff_from(consensus_cache_entry_t
**entry_out
,
661 consensus_flavor_t flavor
,
663 const uint8_t *digest
,
665 compress_method_t method
)
667 if (BUG(digest_type
!= DIGEST_SHA3_256
) ||
668 BUG(digestlen
!= DIGEST256_LEN
)) {
669 return CONSDIFF_NOT_FOUND
; // LCOV_EXCL_LINE
672 // Try to look up the entry in the hashtable.
673 cdm_diff_t search
, *ent
;
674 memset(&search
, 0, sizeof(search
));
675 search
.flavor
= flavor
;
676 search
.compress_method
= method
;
677 memcpy(search
.from_sha3
, digest
, DIGEST256_LEN
);
678 ent
= HT_FIND(cdm_diff_ht
, &cdm_diff_ht
, &search
);
681 ent
->cdm_diff_status
== CDM_DIFF_ERROR
) {
682 return CONSDIFF_NOT_FOUND
;
683 } else if (ent
->cdm_diff_status
== CDM_DIFF_IN_PROGRESS
) {
684 return CONSDIFF_IN_PROGRESS
;
685 } else if (BUG(ent
->cdm_diff_status
!= CDM_DIFF_PRESENT
)) {
686 return CONSDIFF_IN_PROGRESS
;
689 if (BUG(ent
->entry
== NULL
)) {
690 return CONSDIFF_NOT_FOUND
;
692 *entry_out
= consensus_cache_entry_handle_get(ent
->entry
);
693 return (*entry_out
) ? CONSDIFF_AVAILABLE
: CONSDIFF_NOT_FOUND
;
696 // XXXX Remove this. I'm keeping it around for now in case we need to
697 // XXXX debug issues in the hashtable.
698 char hex
[HEX_DIGEST256_LEN
+1];
699 base16_encode(hex
, sizeof(hex
), (const char *)digest
, digestlen
);
700 const char *flavname
= networkstatus_get_flavor_name(flavor
);
702 smartlist_t
*matches
= smartlist_new();
703 consensus_cache_find_all(matches
, cdm_cache_get(),
704 LABEL_FROM_SHA3_DIGEST
, hex
);
705 consensus_cache_filter_list(matches
, LABEL_FLAVOR
, flavname
);
706 consensus_cache_filter_list(matches
, LABEL_DOCTYPE
, DOCTYPE_CONSENSUS_DIFF
);
708 *entry_out
= sort_and_find_most_recent(matches
);
709 consdiff_status_t result
=
710 (*entry_out
) ? CONSDIFF_AVAILABLE
: CONSDIFF_NOT_FOUND
;
711 smartlist_free(matches
);
718 * Perform periodic cleanup tasks on the consensus diff cache. Return
719 * the number of objects marked for deletion.
722 consdiffmgr_cleanup(void)
724 smartlist_t
*objects
= smartlist_new();
725 smartlist_t
*consensuses
= smartlist_new();
726 smartlist_t
*diffs
= smartlist_new();
729 log_debug(LD_DIRSERV
, "Looking for consdiffmgr entries to remove");
731 // 1. Delete any consensus or diff or anything whose valid_after is too old.
732 const time_t valid_after_cutoff
= approx_time() - get_max_age_to_cache();
734 consensus_cache_find_all(objects
, cdm_cache_get(),
736 SMARTLIST_FOREACH_BEGIN(objects
, consensus_cache_entry_t
*, ent
) {
737 const char *lv_valid_after
=
738 consensus_cache_entry_get_value(ent
, LABEL_VALID_AFTER
);
739 if (! lv_valid_after
) {
740 log_debug(LD_DIRSERV
, "Ignoring entry because it had no %s label",
744 time_t valid_after
= 0;
745 if (parse_iso_time_nospace(lv_valid_after
, &valid_after
) < 0) {
746 log_debug(LD_DIRSERV
, "Ignoring entry because its %s value (%s) was "
747 "unparseable", LABEL_VALID_AFTER
, escaped(lv_valid_after
));
750 if (valid_after
< valid_after_cutoff
) {
751 log_debug(LD_DIRSERV
, "Deleting entry because its %s value (%s) was "
752 "too old", LABEL_VALID_AFTER
, lv_valid_after
);
753 consensus_cache_entry_mark_for_removal(ent
);
756 } SMARTLIST_FOREACH_END(ent
);
758 // 2. Delete all diffs that lead to a consensus whose valid-after is not the
760 for (int flav
= 0; flav
< N_CONSENSUS_FLAVORS
; ++flav
) {
761 const char *flavname
= networkstatus_get_flavor_name(flav
);
762 /* Determine the most recent consensus of this flavor */
763 consensus_cache_find_all(consensuses
, cdm_cache_get(),
764 LABEL_DOCTYPE
, DOCTYPE_CONSENSUS
);
765 consensus_cache_filter_list(consensuses
, LABEL_FLAVOR
, flavname
);
766 consensus_cache_entry_t
*most_recent
=
767 sort_and_find_most_recent(consensuses
);
768 if (most_recent
== NULL
)
770 const char *most_recent_sha3
=
771 consensus_cache_entry_get_value(most_recent
,
772 LABEL_SHA3_DIGEST_UNCOMPRESSED
);
773 if (BUG(most_recent_sha3
== NULL
))
774 continue; // LCOV_EXCL_LINE
776 /* consider all such-flavored diffs, and look to see if they match. */
777 consensus_cache_find_all(diffs
, cdm_cache_get(),
778 LABEL_DOCTYPE
, DOCTYPE_CONSENSUS_DIFF
);
779 consensus_cache_filter_list(diffs
, LABEL_FLAVOR
, flavname
);
780 SMARTLIST_FOREACH_BEGIN(diffs
, consensus_cache_entry_t
*, diff
) {
781 const char *this_diff_target_sha3
=
782 consensus_cache_entry_get_value(diff
, LABEL_TARGET_SHA3_DIGEST
);
783 if (!this_diff_target_sha3
)
785 if (strcmp(this_diff_target_sha3
, most_recent_sha3
)) {
786 consensus_cache_entry_mark_for_removal(diff
);
789 } SMARTLIST_FOREACH_END(diff
);
790 smartlist_clear(consensuses
);
791 smartlist_clear(diffs
);
794 // 3. Delete all consensuses except the most recent that are compressed with
795 // an un-preferred method.
796 for (int flav
= 0; flav
< N_CONSENSUS_FLAVORS
; ++flav
) {
797 const char *flavname
= networkstatus_get_flavor_name(flav
);
798 /* Determine the most recent consensus of this flavor */
799 consensus_cache_find_all(consensuses
, cdm_cache_get(),
800 LABEL_DOCTYPE
, DOCTYPE_CONSENSUS
);
801 consensus_cache_filter_list(consensuses
, LABEL_FLAVOR
, flavname
);
802 consensus_cache_entry_t
*most_recent
=
803 sort_and_find_most_recent(consensuses
);
804 if (most_recent
== NULL
)
806 const char *most_recent_sha3_uncompressed
=
807 consensus_cache_entry_get_value(most_recent
,
808 LABEL_SHA3_DIGEST_UNCOMPRESSED
);
809 const char *retain_methodname
= compression_method_get_name(
810 RETAIN_CONSENSUS_COMPRESSED_WITH_METHOD
);
812 if (BUG(most_recent_sha3_uncompressed
== NULL
))
814 SMARTLIST_FOREACH_BEGIN(consensuses
, consensus_cache_entry_t
*, ent
) {
815 const char *lv_sha3_uncompressed
=
816 consensus_cache_entry_get_value(ent
, LABEL_SHA3_DIGEST_UNCOMPRESSED
);
817 if (BUG(! lv_sha3_uncompressed
))
819 if (!strcmp(lv_sha3_uncompressed
, most_recent_sha3_uncompressed
))
820 continue; // This _is_ the most recent.
821 const char *lv_methodname
=
822 consensus_cache_entry_get_value(ent
, LABEL_COMPRESSION_TYPE
);
823 if (! lv_methodname
|| strcmp(lv_methodname
, retain_methodname
)) {
824 consensus_cache_entry_mark_for_removal(ent
);
827 } SMARTLIST_FOREACH_END(ent
);
830 smartlist_free(objects
);
831 smartlist_free(consensuses
);
832 smartlist_free(diffs
);
834 // Actually remove files, if they're not used.
835 consensus_cache_delete_pending(cdm_cache_get(), 0);
840 * Initialize the consensus diff manager and its cache, and configure
841 * its parameters based on the latest torrc and networkstatus parameters.
844 consdiffmgr_configure(const consdiff_cfg_t
*cfg
)
847 memcpy(&consdiff_cfg
, cfg
, sizeof(consdiff_cfg
));
849 (void) cdm_cache_get();
853 * Tell the sandbox (if any) configured by <b>cfg</b> to allow the
854 * operations that the consensus diff manager will need.
857 consdiffmgr_register_with_sandbox(struct sandbox_cfg_elem_t
**cfg
)
859 return consensus_cache_register_with_sandbox(cdm_cache_get(), cfg
);
863 * Scan the consensus diff manager's cache for any grossly malformed entries,
864 * and mark them as deletable. Return 0 if no problems were found; 1
865 * if problems were found and fixed.
868 consdiffmgr_validate(void)
870 /* Right now, we only check for entries that have bad sha3 values */
873 smartlist_t
*objects
= smartlist_new();
874 consensus_cache_find_all(objects
, cdm_cache_get(),
876 SMARTLIST_FOREACH_BEGIN(objects
, consensus_cache_entry_t
*, obj
) {
877 uint8_t sha3_expected
[DIGEST256_LEN
];
878 uint8_t sha3_received
[DIGEST256_LEN
];
879 int r
= cdm_entry_get_sha3_value(sha3_expected
, obj
, LABEL_SHA3_DIGEST
);
881 /* digest isn't there; that's allowed */
883 } else if (r
== -2) {
884 /* digest is malformed; that's not allowed */
886 consensus_cache_entry_mark_for_removal(obj
);
891 consensus_cache_entry_incref(obj
);
892 r
= consensus_cache_entry_get_body(obj
, &body
, &bodylen
);
894 crypto_digest256((char *)sha3_received
, (const char *)body
, bodylen
,
897 consensus_cache_entry_decref(obj
);
901 // Deconfuse coverity about the possibility of sha3_received being
905 if (fast_memneq(sha3_received
, sha3_expected
, DIGEST256_LEN
)) {
907 consensus_cache_entry_mark_for_removal(obj
);
911 } SMARTLIST_FOREACH_END(obj
);
912 smartlist_free(objects
);
917 * Helper: build new diffs of <b>flavor</b> as needed
920 consdiffmgr_rescan_flavor_(consensus_flavor_t flavor
)
922 smartlist_t
*matches
= NULL
;
923 smartlist_t
*diffs
= NULL
;
924 smartlist_t
*compute_diffs_from
= NULL
;
925 strmap_t
*have_diff_from
= NULL
;
927 // look for the most recent consensus, and for all previous in-range
928 // consensuses. Do they all have diffs to it?
929 const char *flavname
= networkstatus_get_flavor_name(flavor
);
931 // 1. find the most recent consensus, and the ones that we might want
933 const char *methodname
= compression_method_get_name(
934 RETAIN_CONSENSUS_COMPRESSED_WITH_METHOD
);
936 matches
= smartlist_new();
937 consensus_cache_find_all(matches
, cdm_cache_get(),
938 LABEL_FLAVOR
, flavname
);
939 consensus_cache_filter_list(matches
, LABEL_DOCTYPE
, DOCTYPE_CONSENSUS
);
940 consensus_cache_filter_list(matches
, LABEL_COMPRESSION_TYPE
, methodname
);
941 consensus_cache_entry_t
*most_recent
= sort_and_find_most_recent(matches
);
943 log_info(LD_DIRSERV
, "No 'most recent' %s consensus found; "
944 "not making diffs", flavname
);
947 tor_assert(smartlist_len(matches
));
948 smartlist_del(matches
, smartlist_len(matches
) - 1);
950 const char *most_recent_valid_after
=
951 consensus_cache_entry_get_value(most_recent
, LABEL_VALID_AFTER
);
952 if (BUG(most_recent_valid_after
== NULL
))
953 goto done
; //LCOV_EXCL_LINE
954 uint8_t most_recent_sha3
[DIGEST256_LEN
];
955 if (BUG(cdm_entry_get_sha3_value(most_recent_sha3
, most_recent
,
956 LABEL_SHA3_DIGEST_UNCOMPRESSED
) < 0))
957 goto done
; //LCOV_EXCL_LINE
959 // 2. Find all the relevant diffs _to_ this consensus. These are ones
960 // that we don't need to compute.
961 diffs
= smartlist_new();
962 consensus_cache_find_all(diffs
, cdm_cache_get(),
963 LABEL_VALID_AFTER
, most_recent_valid_after
);
964 consensus_cache_filter_list(diffs
, LABEL_DOCTYPE
, DOCTYPE_CONSENSUS_DIFF
);
965 consensus_cache_filter_list(diffs
, LABEL_FLAVOR
, flavname
);
966 have_diff_from
= strmap_new();
967 SMARTLIST_FOREACH_BEGIN(diffs
, consensus_cache_entry_t
*, diff
) {
968 const char *va
= consensus_cache_entry_get_value(diff
,
969 LABEL_FROM_VALID_AFTER
);
971 continue; // LCOV_EXCL_LINE
972 strmap_set(have_diff_from
, va
, diff
);
973 } SMARTLIST_FOREACH_END(diff
);
975 // 3. See which consensuses in 'matches' don't have diffs yet.
976 smartlist_reverse(matches
); // from newest to oldest.
977 compute_diffs_from
= smartlist_new();
978 SMARTLIST_FOREACH_BEGIN(matches
, consensus_cache_entry_t
*, ent
) {
979 const char *va
= consensus_cache_entry_get_value(ent
, LABEL_VALID_AFTER
);
981 continue; // LCOV_EXCL_LINE
982 if (strmap_get(have_diff_from
, va
) != NULL
)
983 continue; /* we already have this one. */
984 smartlist_add(compute_diffs_from
, ent
);
985 /* Since we are not going to serve this as the most recent consensus
986 * any more, we should stop keeping it mmap'd when it's not in use.
988 consensus_cache_entry_mark_for_aggressive_release(ent
);
989 } SMARTLIST_FOREACH_END(ent
);
992 "The most recent %s consensus is valid-after %s. We have diffs to "
993 "this consensus for %d/%d older %s consensuses. Generating diffs "
996 most_recent_valid_after
,
997 smartlist_len(matches
) - smartlist_len(compute_diffs_from
),
998 smartlist_len(matches
),
1000 smartlist_len(compute_diffs_from
));
1002 // 4. Update the hashtable; remove entries in this flavor to other
1003 // target consensuses.
1004 cdm_diff_ht_purge(flavor
, most_recent_sha3
);
1006 // 5. Actually launch the requests.
1007 SMARTLIST_FOREACH_BEGIN(compute_diffs_from
, consensus_cache_entry_t
*, c
) {
1008 if (BUG(c
== most_recent
))
1009 continue; // LCOV_EXCL_LINE
1011 uint8_t this_sha3
[DIGEST256_LEN
];
1012 if (cdm_entry_get_sha3_value(this_sha3
, c
,
1013 LABEL_SHA3_DIGEST_AS_SIGNED
)<0) {
1014 // Not actually a bug, since we might be running with a directory
1015 // with stale files from before the #22143 fixes.
1018 if (cdm_diff_ht_check_and_note_pending(flavor
,
1019 this_sha3
, most_recent_sha3
)) {
1020 // This is already pending, or we encountered an error.
1023 consensus_diff_queue_diff_work(c
, most_recent
);
1024 } SMARTLIST_FOREACH_END(c
);
1027 smartlist_free(matches
);
1028 smartlist_free(diffs
);
1029 smartlist_free(compute_diffs_from
);
1030 strmap_free(have_diff_from
, NULL
);
1034 * Scan the cache for the latest consensuses and add their handles to
1038 consdiffmgr_consensus_load(void)
1040 smartlist_t
*matches
= smartlist_new();
1041 for (int flav
= 0; flav
< N_CONSENSUS_FLAVORS
; ++flav
) {
1042 const char *flavname
= networkstatus_get_flavor_name(flav
);
1043 smartlist_clear(matches
);
1044 consensus_cache_find_all(matches
, cdm_cache_get(),
1045 LABEL_FLAVOR
, flavname
);
1046 consensus_cache_filter_list(matches
, LABEL_DOCTYPE
, DOCTYPE_CONSENSUS
);
1047 consensus_cache_entry_t
*most_recent
= sort_and_find_most_recent(matches
);
1049 continue; // no consensuses.
1050 const char *most_recent_sha3
=
1051 consensus_cache_entry_get_value(most_recent
,
1052 LABEL_SHA3_DIGEST_UNCOMPRESSED
);
1053 if (BUG(most_recent_sha3
== NULL
))
1054 continue; // LCOV_EXCL_LINE
1055 consensus_cache_filter_list(matches
, LABEL_SHA3_DIGEST_UNCOMPRESSED
,
1058 // Everything that remains matches the most recent consensus of this
1060 SMARTLIST_FOREACH_BEGIN(matches
, consensus_cache_entry_t
*, ent
) {
1061 const char *lv_compression
=
1062 consensus_cache_entry_get_value(ent
, LABEL_COMPRESSION_TYPE
);
1063 compress_method_t method
=
1064 compression_method_get_by_name(lv_compression
);
1065 int pos
= consensus_compression_method_pos(method
);
1068 consensus_cache_entry_handle_free(latest_consensus
[flav
][pos
]);
1069 latest_consensus
[flav
][pos
] = consensus_cache_entry_handle_new(ent
);
1070 } SMARTLIST_FOREACH_END(ent
);
1072 smartlist_free(matches
);
1076 * Scan the cache for diffs, and add them to the hashtable.
1079 consdiffmgr_diffs_load(void)
1081 smartlist_t
*diffs
= smartlist_new();
1082 consensus_cache_find_all(diffs
, cdm_cache_get(),
1083 LABEL_DOCTYPE
, DOCTYPE_CONSENSUS_DIFF
);
1084 SMARTLIST_FOREACH_BEGIN(diffs
, consensus_cache_entry_t
*, diff
) {
1085 const char *lv_flavor
=
1086 consensus_cache_entry_get_value(diff
, LABEL_FLAVOR
);
1089 int flavor
= networkstatus_parse_flavor_name(lv_flavor
);
1092 const char *lv_compression
=
1093 consensus_cache_entry_get_value(diff
, LABEL_COMPRESSION_TYPE
);
1094 compress_method_t method
= NO_METHOD
;
1095 if (lv_compression
) {
1096 method
= compression_method_get_by_name(lv_compression
);
1097 if (method
== UNKNOWN_METHOD
) {
1102 uint8_t from_sha3
[DIGEST256_LEN
];
1103 uint8_t to_sha3
[DIGEST256_LEN
];
1104 if (cdm_entry_get_sha3_value(from_sha3
, diff
, LABEL_FROM_SHA3_DIGEST
)<0)
1106 if (cdm_entry_get_sha3_value(to_sha3
, diff
, LABEL_TARGET_SHA3_DIGEST
)<0)
1109 cdm_diff_ht_set_status(flavor
, from_sha3
, to_sha3
,
1112 consensus_cache_entry_handle_new(diff
));
1113 } SMARTLIST_FOREACH_END(diff
);
1114 smartlist_free(diffs
);
1118 * Build new diffs as needed.
1121 consdiffmgr_rescan(void)
1123 if (cdm_cache_dirty
== 0)
1126 // Clean up here to make room for new diffs, and to ensure that older
1127 // consensuses do not have any entries.
1128 consdiffmgr_cleanup();
1130 if (cdm_cache_loaded
== 0) {
1131 consdiffmgr_diffs_load();
1132 consdiffmgr_consensus_load();
1133 cdm_cache_loaded
= 1;
1136 for (int flav
= 0; flav
< N_CONSENSUS_FLAVORS
; ++flav
) {
1137 consdiffmgr_rescan_flavor_((consensus_flavor_t
) flav
);
1140 cdm_cache_dirty
= 0;
1143 /** Callback wrapper for consdiffmgr_rescan */
1145 consdiffmgr_rescan_cb(mainloop_event_t
*ev
, void *arg
)
1149 consdiffmgr_rescan();
1152 /** Mark the cache as dirty, and schedule a rescan event. */
1154 mark_cdm_cache_dirty(void)
1156 cdm_cache_dirty
= 1;
1157 tor_assert(consdiffmgr_rescan_ev
);
1158 mainloop_event_activate(consdiffmgr_rescan_ev
);
1162 * Helper: compare two files by their from-valid-after and valid-after labels,
1163 * trying to sort in ascending order by from-valid-after (when present) and
1164 * valid-after (when not). Place everything that has neither label first in
1168 compare_by_staleness_(const void **a
, const void **b
)
1170 const consensus_cache_entry_t
*e1
= *a
;
1171 const consensus_cache_entry_t
*e2
= *b
;
1172 const char *va1
, *fva1
, *va2
, *fva2
;
1173 va1
= consensus_cache_entry_get_value(e1
, LABEL_VALID_AFTER
);
1174 va2
= consensus_cache_entry_get_value(e2
, LABEL_VALID_AFTER
);
1175 fva1
= consensus_cache_entry_get_value(e1
, LABEL_FROM_VALID_AFTER
);
1176 fva2
= consensus_cache_entry_get_value(e2
, LABEL_FROM_VALID_AFTER
);
1183 /* See note about iso-encoded values in compare_by_valid_after_. Also note
1184 * that missing dates will get placed first. */
1185 return strcmp_opt(va1
, va2
);
1188 /** If there are not enough unused filenames to store <b>n</b> files, then
1189 * delete old consensuses until there are. (We have to keep track of the
1190 * number of filenames because of the way that the seccomp2 cache works.)
1192 * Return 0 on success, -1 on failure.
1195 consdiffmgr_ensure_space_for_files(int n
)
1197 consensus_cache_t
*cache
= cdm_cache_get();
1198 if (consensus_cache_get_n_filenames_available(cache
) >= n
) {
1199 // there are already enough unused filenames.
1202 // Try a cheap deletion of stuff that's waiting to get deleted.
1203 consensus_cache_delete_pending(cache
, 0);
1204 if (consensus_cache_get_n_filenames_available(cache
) >= n
) {
1205 // okay, _that_ made enough filenames available.
1208 // Let's get more assertive: clean out unused stuff, and force-remove
1209 // the files that we can.
1210 consdiffmgr_cleanup();
1211 consensus_cache_delete_pending(cache
, 1);
1212 const int n_to_remove
= n
- consensus_cache_get_n_filenames_available(cache
);
1213 if (n_to_remove
<= 0) {
1218 // At this point, we're going to have to throw out objects that will be
1220 smartlist_t
*objects
= smartlist_new();
1221 consensus_cache_find_all(objects
, cache
, NULL
, NULL
);
1222 smartlist_sort(objects
, compare_by_staleness_
);
1224 SMARTLIST_FOREACH_BEGIN(objects
, consensus_cache_entry_t
*, ent
) {
1225 consensus_cache_entry_mark_for_removal(ent
);
1226 if (++n_marked
>= n_to_remove
)
1228 } SMARTLIST_FOREACH_END(ent
);
1229 smartlist_free(objects
);
1231 consensus_cache_delete_pending(cache
, 1);
1233 if (consensus_cache_may_overallocate(cache
)) {
1234 /* If we're allowed to throw extra files into the cache, let's do so
1235 * rather getting upset.
1240 if (BUG(n_marked
< n_to_remove
))
1247 * Set consensus cache flags on the objects in this consdiffmgr.
1250 consdiffmgr_set_cache_flags(void)
1252 /* Right now, we just mark the consensus objects for aggressive release,
1253 * so that they get mmapped for as little time as possible. */
1254 smartlist_t
*objects
= smartlist_new();
1255 consensus_cache_find_all(objects
, cdm_cache_get(), LABEL_DOCTYPE
,
1257 SMARTLIST_FOREACH_BEGIN(objects
, consensus_cache_entry_t
*, ent
) {
1258 consensus_cache_entry_mark_for_aggressive_release(ent
);
1259 } SMARTLIST_FOREACH_END(ent
);
1260 smartlist_free(objects
);
1264 * Called before shutdown: drop all storage held by the consdiffmgr.c module.
1267 consdiffmgr_free_all(void)
1269 cdm_diff_t
**diff
, **next
;
1270 for (diff
= HT_START(cdm_diff_ht
, &cdm_diff_ht
); diff
; diff
= next
) {
1271 cdm_diff_t
*this = *diff
;
1272 next
= HT_NEXT_RMV(cdm_diff_ht
, &cdm_diff_ht
, diff
);
1273 cdm_diff_free(this);
1277 for (i
= 0; i
< N_CONSENSUS_FLAVORS
; ++i
) {
1278 for (j
= 0; j
< n_consensus_compression_methods(); ++j
) {
1279 consensus_cache_entry_handle_free(latest_consensus
[i
][j
]);
1282 memset(latest_consensus
, 0, sizeof(latest_consensus
));
1283 consensus_cache_free(cons_diff_cache
);
1284 cons_diff_cache
= NULL
;
1285 mainloop_event_free(consdiffmgr_rescan_ev
);
1292 typedef struct compressed_result_t
{
1293 config_line_t
*labels
;
1295 * Output: Body of the diff, as compressed.
1299 * Output: length of body_out
1302 } compressed_result_t
;
1305 * Compress the bytestring <b>input</b> of length <b>len</b> using the
1306 * <b>n_methods</b> compression methods listed in the array <b>methods</b>.
1308 * For each successful compression, set the fields in the <b>results_out</b>
1309 * array in the position corresponding to the compression method. Use
1310 * <b>labels_in</b> as a basis for the labels of the result.
1312 * Return 0 if all compression succeeded; -1 if any failed.
1315 compress_multiple(compressed_result_t
*results_out
, int n_methods
,
1316 const compress_method_t
*methods
,
1317 const uint8_t *input
, size_t len
,
1318 const config_line_t
*labels_in
)
1322 for (i
= 0; i
< n_methods
; ++i
) {
1323 compress_method_t method
= methods
[i
];
1324 const char *methodname
= compression_method_get_name(method
);
1327 if (0 == tor_compress(&result
, &sz
, (const char*)input
, len
, method
)) {
1328 results_out
[i
].body
= (uint8_t*)result
;
1329 results_out
[i
].bodylen
= sz
;
1330 results_out
[i
].labels
= config_lines_dup(labels_in
);
1331 cdm_labels_prepend_sha3(&results_out
[i
].labels
, LABEL_SHA3_DIGEST
,
1332 results_out
[i
].body
,
1333 results_out
[i
].bodylen
);
1334 config_line_prepend(&results_out
[i
].labels
,
1335 LABEL_COMPRESSION_TYPE
,
1345 * Given an array of <b>n</b> compressed_result_t in <b>results</b>,
1346 * as produced by compress_multiple, store them all into the
1347 * consdiffmgr, and store handles to them in the <b>handles_out</b>
1350 * Return CDM_DIFF_PRESENT if any was stored, and CDM_DIFF_ERROR if none
1353 static cdm_diff_status_t
1354 store_multiple(consensus_cache_entry_handle_t
**handles_out
,
1356 const compress_method_t
*methods
,
1357 const compressed_result_t
*results
,
1358 const char *description
)
1360 cdm_diff_status_t status
= CDM_DIFF_ERROR
;
1361 consdiffmgr_ensure_space_for_files(n
);
1364 for (i
= 0; i
< n
; ++i
) {
1365 compress_method_t method
= methods
[i
];
1366 uint8_t *body_out
= results
[i
].body
;
1367 size_t bodylen_out
= results
[i
].bodylen
;
1368 config_line_t
*labels
= results
[i
].labels
;
1369 const char *methodname
= compression_method_get_name(method
);
1370 if (body_out
&& bodylen_out
&& labels
) {
1371 /* Success! Store the results */
1372 log_info(LD_DIRSERV
, "Adding %s, compressed with %s",
1373 description
, methodname
);
1375 consensus_cache_entry_t
*ent
=
1376 consensus_cache_add(cdm_cache_get(),
1381 static ratelim_t cant_store_ratelim
= RATELIM_INIT(5*60);
1382 log_fn_ratelim(&cant_store_ratelim
, LOG_WARN
, LD_FS
,
1383 "Unable to store object %s compressed with %s.",
1384 description
, methodname
);
1388 status
= CDM_DIFF_PRESENT
;
1389 handles_out
[i
] = consensus_cache_entry_handle_new(ent
);
1390 consensus_cache_entry_decref(ent
);
1397 * An object passed to a worker thread that will try to produce a consensus
1400 typedef struct consensus_diff_worker_job_t
{
1402 * Input: The consensus to compute the diff from. Holds a reference to the
1403 * cache entry, which must not be released until the job is passed back to
1404 * the main thread. The body must be mapped into memory in the main thread.
1406 consensus_cache_entry_t
*diff_from
;
1408 * Input: The consensus to compute the diff to. Holds a reference to the
1409 * cache entry, which must not be released until the job is passed back to
1410 * the main thread. The body must be mapped into memory in the main thread.
1412 consensus_cache_entry_t
*diff_to
;
1414 /** Output: labels and bodies */
1415 compressed_result_t out
[ARRAY_LENGTH(compress_diffs_with
)];
1416 } consensus_diff_worker_job_t
;
1418 /** Given a consensus_cache_entry_t, check whether it has a label claiming
1419 * that it was compressed. If so, uncompress its contents into *<b>out</b> and
1420 * set <b>outlen</b> to hold their size, and set *<b>owned_out</b> to a pointer
1421 * that the caller will need to free. If not, just set *<b>out</b> and
1422 * <b>outlen</b> to its extent in memory. Return 0 on success, -1 on failure.
1425 uncompress_or_set_ptr(const char **out
, size_t *outlen
,
1427 consensus_cache_entry_t
*ent
)
1429 const uint8_t *body
;
1434 if (consensus_cache_entry_get_body(ent
, &body
, &bodylen
) < 0)
1437 const char *lv_compression
=
1438 consensus_cache_entry_get_value(ent
, LABEL_COMPRESSION_TYPE
);
1439 compress_method_t method
= NO_METHOD
;
1442 method
= compression_method_get_by_name(lv_compression
);
1445 if (method
== NO_METHOD
) {
1446 *out
= (const char *)body
;
1450 rv
= tor_uncompress(owned_out
, outlen
, (const char *)body
, bodylen
,
1451 method
, 1, LOG_WARN
);
1458 * Worker function. This function runs inside a worker thread and receives
1459 * a consensus_diff_worker_job_t as its input.
1461 static workqueue_reply_t
1462 consensus_diff_worker_threadfn(void *state_
, void *work_
)
1465 consensus_diff_worker_job_t
*job
= work_
;
1466 const uint8_t *diff_from
, *diff_to
;
1467 size_t len_from
, len_to
;
1469 /* We need to have the body already mapped into RAM here.
1471 r
= consensus_cache_entry_get_body(job
->diff_from
, &diff_from
, &len_from
);
1473 return WQ_RPL_REPLY
; // LCOV_EXCL_LINE
1474 r
= consensus_cache_entry_get_body(job
->diff_to
, &diff_to
, &len_to
);
1476 return WQ_RPL_REPLY
; // LCOV_EXCL_LINE
1478 const char *lv_to_valid_after
=
1479 consensus_cache_entry_get_value(job
->diff_to
, LABEL_VALID_AFTER
);
1480 const char *lv_to_fresh_until
=
1481 consensus_cache_entry_get_value(job
->diff_to
, LABEL_FRESH_UNTIL
);
1482 const char *lv_to_valid_until
=
1483 consensus_cache_entry_get_value(job
->diff_to
, LABEL_VALID_UNTIL
);
1484 const char *lv_to_signatories
=
1485 consensus_cache_entry_get_value(job
->diff_to
, LABEL_SIGNATORIES
);
1486 const char *lv_from_valid_after
=
1487 consensus_cache_entry_get_value(job
->diff_from
, LABEL_VALID_AFTER
);
1488 const char *lv_from_digest
=
1489 consensus_cache_entry_get_value(job
->diff_from
,
1490 LABEL_SHA3_DIGEST_AS_SIGNED
);
1491 const char *lv_from_flavor
=
1492 consensus_cache_entry_get_value(job
->diff_from
, LABEL_FLAVOR
);
1493 const char *lv_to_flavor
=
1494 consensus_cache_entry_get_value(job
->diff_to
, LABEL_FLAVOR
);
1495 const char *lv_to_digest
=
1496 consensus_cache_entry_get_value(job
->diff_to
,
1497 LABEL_SHA3_DIGEST_UNCOMPRESSED
);
1499 if (! lv_from_digest
) {
1500 /* This isn't a bug right now, since it can happen if you're migrating
1501 * from an older version of master to a newer one. The older ones didn't
1502 * annotate their stored consensus objects with sha3-digest-as-signed.
1504 return WQ_RPL_REPLY
; // LCOV_EXCL_LINE
1507 /* All these values are mandatory on the input */
1508 if (BUG(!lv_to_valid_after
) ||
1509 BUG(!lv_from_valid_after
) ||
1510 BUG(!lv_from_flavor
) ||
1511 BUG(!lv_to_flavor
)) {
1512 return WQ_RPL_REPLY
; // LCOV_EXCL_LINE
1514 /* The flavors need to match */
1515 if (BUG(strcmp(lv_from_flavor
, lv_to_flavor
))) {
1516 return WQ_RPL_REPLY
; // LCOV_EXCL_LINE
1519 char *consensus_diff
;
1521 const char *diff_from_nt
= NULL
, *diff_to_nt
= NULL
;
1522 char *owned1
= NULL
, *owned2
= NULL
;
1523 size_t diff_from_nt_len
, diff_to_nt_len
;
1525 if (uncompress_or_set_ptr(&diff_from_nt
, &diff_from_nt_len
, &owned1
,
1526 job
->diff_from
) < 0) {
1527 return WQ_RPL_REPLY
;
1529 if (uncompress_or_set_ptr(&diff_to_nt
, &diff_to_nt_len
, &owned2
,
1530 job
->diff_to
) < 0) {
1532 return WQ_RPL_REPLY
;
1534 tor_assert(diff_from_nt
);
1535 tor_assert(diff_to_nt
);
1537 // XXXX ugh; this is going to calculate the SHA3 of both its
1538 // XXXX inputs again, even though we already have that. Maybe it's time
1539 // XXXX to change the API here?
1540 consensus_diff
= consensus_diff_generate(diff_from_nt
,
1547 if (!consensus_diff
) {
1548 /* Couldn't generate consensus; we'll leave the reply blank. */
1549 return WQ_RPL_REPLY
;
1552 /* Compress the results and send the reply */
1553 tor_assert(compress_diffs_with
[0] == NO_METHOD
);
1554 size_t difflen
= strlen(consensus_diff
);
1555 job
->out
[0].body
= (uint8_t *) consensus_diff
;
1556 job
->out
[0].bodylen
= difflen
;
1558 config_line_t
*common_labels
= NULL
;
1559 if (lv_to_valid_until
)
1560 config_line_prepend(&common_labels
, LABEL_VALID_UNTIL
, lv_to_valid_until
);
1561 if (lv_to_fresh_until
)
1562 config_line_prepend(&common_labels
, LABEL_FRESH_UNTIL
, lv_to_fresh_until
);
1563 if (lv_to_signatories
)
1564 config_line_prepend(&common_labels
, LABEL_SIGNATORIES
, lv_to_signatories
);
1565 cdm_labels_prepend_sha3(&common_labels
,
1566 LABEL_SHA3_DIGEST_UNCOMPRESSED
,
1568 job
->out
[0].bodylen
);
1569 config_line_prepend(&common_labels
, LABEL_FROM_VALID_AFTER
,
1570 lv_from_valid_after
);
1571 config_line_prepend(&common_labels
, LABEL_VALID_AFTER
,
1573 config_line_prepend(&common_labels
, LABEL_FLAVOR
, lv_from_flavor
);
1574 config_line_prepend(&common_labels
, LABEL_FROM_SHA3_DIGEST
,
1576 config_line_prepend(&common_labels
, LABEL_TARGET_SHA3_DIGEST
,
1578 config_line_prepend(&common_labels
, LABEL_DOCTYPE
,
1579 DOCTYPE_CONSENSUS_DIFF
);
1581 job
->out
[0].labels
= config_lines_dup(common_labels
);
1582 cdm_labels_prepend_sha3(&job
->out
[0].labels
,
1585 job
->out
[0].bodylen
);
1587 compress_multiple(job
->out
+1,
1588 n_diff_compression_methods()-1,
1589 compress_diffs_with
+1,
1590 (const uint8_t*)consensus_diff
, difflen
, common_labels
);
1592 config_free_lines(common_labels
);
1593 return WQ_RPL_REPLY
;
1596 #define consensus_diff_worker_job_free(job) \
1597 FREE_AND_NULL(consensus_diff_worker_job_t, \
1598 consensus_diff_worker_job_free_, (job))
1601 * Helper: release all storage held in <b>job</b>.
1604 consensus_diff_worker_job_free_(consensus_diff_worker_job_t
*job
)
1609 for (u
= 0; u
< n_diff_compression_methods(); ++u
) {
1610 config_free_lines(job
->out
[u
].labels
);
1611 tor_free(job
->out
[u
].body
);
1613 consensus_cache_entry_decref(job
->diff_from
);
1614 consensus_cache_entry_decref(job
->diff_to
);
1619 * Worker function: This function runs in the main thread, and receives
1620 * a consensus_diff_worker_job_t that the worker thread has already
1624 consensus_diff_worker_replyfn(void *work_
)
1626 tor_assert(in_main_thread());
1629 consensus_diff_worker_job_t
*job
= work_
;
1631 const char *lv_from_digest
=
1632 consensus_cache_entry_get_value(job
->diff_from
,
1633 LABEL_SHA3_DIGEST_AS_SIGNED
);
1634 const char *lv_to_digest
=
1635 consensus_cache_entry_get_value(job
->diff_to
,
1636 LABEL_SHA3_DIGEST_UNCOMPRESSED
);
1637 const char *lv_flavor
=
1638 consensus_cache_entry_get_value(job
->diff_to
, LABEL_FLAVOR
);
1639 if (BUG(lv_from_digest
== NULL
))
1640 lv_from_digest
= "???"; // LCOV_EXCL_LINE
1641 if (BUG(lv_to_digest
== NULL
))
1642 lv_to_digest
= "???"; // LCOV_EXCL_LINE
1644 uint8_t from_sha3
[DIGEST256_LEN
];
1645 uint8_t to_sha3
[DIGEST256_LEN
];
1648 if (BUG(cdm_entry_get_sha3_value(from_sha3
, job
->diff_from
,
1649 LABEL_SHA3_DIGEST_AS_SIGNED
) < 0))
1651 if (BUG(cdm_entry_get_sha3_value(to_sha3
, job
->diff_to
,
1652 LABEL_SHA3_DIGEST_UNCOMPRESSED
) < 0))
1654 if (BUG(lv_flavor
== NULL
)) {
1656 } else if ((flav
= networkstatus_parse_flavor_name(lv_flavor
)) < 0) {
1660 consensus_cache_entry_handle_t
*handles
[ARRAY_LENGTH(compress_diffs_with
)];
1661 memset(handles
, 0, sizeof(handles
));
1663 char description
[128];
1664 tor_snprintf(description
, sizeof(description
),
1665 "consensus diff from %s to %s",
1666 lv_from_digest
, lv_to_digest
);
1668 int status
= store_multiple(handles
,
1669 n_diff_compression_methods(),
1670 compress_diffs_with
,
1674 if (status
!= CDM_DIFF_PRESENT
) {
1675 /* Failure! Nothing to do but complain */
1676 log_warn(LD_DIRSERV
,
1677 "Worker was unable to compute consensus diff "
1678 "from %s to %s", lv_from_digest
, lv_to_digest
);
1679 /* Cache this error so we don't try to compute this one again. */
1680 status
= CDM_DIFF_ERROR
;
1684 for (u
= 0; u
< ARRAY_LENGTH(handles
); ++u
) {
1685 compress_method_t method
= compress_diffs_with
[u
];
1687 consensus_cache_entry_handle_t
*h
= handles
[u
];
1688 int this_status
= status
;
1690 this_status
= CDM_DIFF_ERROR
;
1692 tor_assert_nonfatal(h
!= NULL
|| this_status
== CDM_DIFF_ERROR
);
1693 cdm_diff_ht_set_status(flav
, from_sha3
, to_sha3
, method
, this_status
, h
);
1695 consensus_cache_entry_handle_free(handles
[u
]);
1699 consensus_diff_worker_job_free(job
);
1703 * Queue the job of computing the diff from <b>diff_from</b> to <b>diff_to</b>
1704 * in a worker thread.
1707 consensus_diff_queue_diff_work(consensus_cache_entry_t
*diff_from
,
1708 consensus_cache_entry_t
*diff_to
)
1710 tor_assert(in_main_thread());
1712 consensus_cache_entry_incref(diff_from
);
1713 consensus_cache_entry_incref(diff_to
);
1715 consensus_diff_worker_job_t
*job
= tor_malloc_zero(sizeof(*job
));
1716 job
->diff_from
= diff_from
;
1717 job
->diff_to
= diff_to
;
1719 /* Make sure body is mapped. */
1720 const uint8_t *body
;
1722 int r1
= consensus_cache_entry_get_body(diff_from
, &body
, &bodylen
);
1723 int r2
= consensus_cache_entry_get_body(diff_to
, &body
, &bodylen
);
1724 if (r1
< 0 || r2
< 0)
1727 workqueue_entry_t
*work
;
1728 work
= cpuworker_queue_work(WQ_PRI_LOW
,
1729 consensus_diff_worker_threadfn
,
1730 consensus_diff_worker_replyfn
,
1737 consensus_diff_worker_job_free(job
); // includes decrefs.
1742 * Holds requests and replies for consensus_compress_workers.
1744 typedef struct consensus_compress_worker_job_t
{
1746 size_t consensus_len
;
1747 consensus_flavor_t flavor
;
1748 config_line_t
*labels_in
;
1749 compressed_result_t out
[ARRAY_LENGTH(compress_consensus_with
)];
1750 } consensus_compress_worker_job_t
;
1752 #define consensus_compress_worker_job_free(job) \
1753 FREE_AND_NULL(consensus_compress_worker_job_t, \
1754 consensus_compress_worker_job_free_, (job))
1757 * Free all resources held in <b>job</b>
1760 consensus_compress_worker_job_free_(consensus_compress_worker_job_t
*job
)
1764 tor_free(job
->consensus
);
1765 config_free_lines(job
->labels_in
);
1767 for (u
= 0; u
< n_consensus_compression_methods(); ++u
) {
1768 config_free_lines(job
->out
[u
].labels
);
1769 tor_free(job
->out
[u
].body
);
1774 * Worker function. This function runs inside a worker thread and receives
1775 * a consensus_compress_worker_job_t as its input.
1777 static workqueue_reply_t
1778 consensus_compress_worker_threadfn(void *state_
, void *work_
)
1781 consensus_compress_worker_job_t
*job
= work_
;
1782 consensus_flavor_t flavor
= job
->flavor
;
1783 const char *consensus
= job
->consensus
;
1784 size_t bodylen
= job
->consensus_len
;
1786 config_line_t
*labels
= config_lines_dup(job
->labels_in
);
1787 const char *flavname
= networkstatus_get_flavor_name(flavor
);
1789 cdm_labels_prepend_sha3(&labels
, LABEL_SHA3_DIGEST_UNCOMPRESSED
,
1790 (const uint8_t *)consensus
, bodylen
);
1792 const char *start
, *end
;
1793 if (router_get_networkstatus_v3_signed_boundaries(consensus
, bodylen
,
1794 &start
, &end
) < 0) {
1796 end
= consensus
+bodylen
;
1798 cdm_labels_prepend_sha3(&labels
, LABEL_SHA3_DIGEST_AS_SIGNED
,
1799 (const uint8_t *)start
,
1802 config_line_prepend(&labels
, LABEL_FLAVOR
, flavname
);
1803 config_line_prepend(&labels
, LABEL_DOCTYPE
, DOCTYPE_CONSENSUS
);
1805 compress_multiple(job
->out
,
1806 n_consensus_compression_methods(),
1807 compress_consensus_with
,
1808 (const uint8_t*)consensus
, bodylen
, labels
);
1809 config_free_lines(labels
);
1810 return WQ_RPL_REPLY
;
1814 * Worker function: This function runs in the main thread, and receives
1815 * a consensus_diff_compress_job_t that the worker thread has already
1819 consensus_compress_worker_replyfn(void *work_
)
1821 consensus_compress_worker_job_t
*job
= work_
;
1823 consensus_cache_entry_handle_t
*handles
[
1824 ARRAY_LENGTH(compress_consensus_with
)];
1825 memset(handles
, 0, sizeof(handles
));
1827 store_multiple(handles
,
1828 n_consensus_compression_methods(),
1829 compress_consensus_with
,
1832 mark_cdm_cache_dirty();
1835 consensus_flavor_t f
= job
->flavor
;
1836 tor_assert((int)f
< N_CONSENSUS_FLAVORS
);
1837 for (u
= 0; u
< ARRAY_LENGTH(handles
); ++u
) {
1838 if (handles
[u
] == NULL
)
1840 consensus_cache_entry_handle_free(latest_consensus
[f
][u
]);
1841 latest_consensus
[f
][u
] = handles
[u
];
1844 consensus_compress_worker_job_free(job
);
1848 * If true, we compress in worker threads.
1850 static int background_compression
= 0;
1853 * Queue a job to compress <b>consensus</b> and store its compressed
1854 * text in the cache.
1857 consensus_queue_compression_work(const char *consensus
,
1858 size_t consensus_len
,
1859 const networkstatus_t
*as_parsed
)
1861 tor_assert(consensus
);
1862 tor_assert(as_parsed
);
1864 consensus_compress_worker_job_t
*job
= tor_malloc_zero(sizeof(*job
));
1865 job
->consensus
= tor_memdup_nulterm(consensus
, consensus_len
);
1866 job
->consensus_len
= strlen(job
->consensus
);
1867 job
->flavor
= as_parsed
->flavor
;
1869 char va_str
[ISO_TIME_LEN
+1];
1870 char vu_str
[ISO_TIME_LEN
+1];
1871 char fu_str
[ISO_TIME_LEN
+1];
1872 format_iso_time_nospace(va_str
, as_parsed
->valid_after
);
1873 format_iso_time_nospace(fu_str
, as_parsed
->fresh_until
);
1874 format_iso_time_nospace(vu_str
, as_parsed
->valid_until
);
1875 config_line_append(&job
->labels_in
, LABEL_VALID_AFTER
, va_str
);
1876 config_line_append(&job
->labels_in
, LABEL_FRESH_UNTIL
, fu_str
);
1877 config_line_append(&job
->labels_in
, LABEL_VALID_UNTIL
, vu_str
);
1878 if (as_parsed
->voters
) {
1879 smartlist_t
*hexvoters
= smartlist_new();
1880 SMARTLIST_FOREACH_BEGIN(as_parsed
->voters
,
1881 networkstatus_voter_info_t
*, vi
) {
1882 if (smartlist_len(vi
->sigs
) == 0)
1883 continue; // didn't sign.
1884 char d
[HEX_DIGEST_LEN
+1];
1885 base16_encode(d
, sizeof(d
), vi
->identity_digest
, DIGEST_LEN
);
1886 smartlist_add_strdup(hexvoters
, d
);
1887 } SMARTLIST_FOREACH_END(vi
);
1888 char *signers
= smartlist_join_strings(hexvoters
, ",", 0, NULL
);
1889 config_line_prepend(&job
->labels_in
, LABEL_SIGNATORIES
, signers
);
1891 SMARTLIST_FOREACH(hexvoters
, char *, cp
, tor_free(cp
));
1892 smartlist_free(hexvoters
);
1895 if (background_compression
) {
1896 workqueue_entry_t
*work
;
1897 work
= cpuworker_queue_work(WQ_PRI_LOW
,
1898 consensus_compress_worker_threadfn
,
1899 consensus_compress_worker_replyfn
,
1902 consensus_compress_worker_job_free(job
);
1908 consensus_compress_worker_threadfn(NULL
, job
);
1909 consensus_compress_worker_replyfn(job
);
1915 * Tell the consdiffmgr backend to compress consensuses in worker threads.
1918 consdiffmgr_enable_background_compression(void)
1920 // This isn't the default behavior because it would break unit tests.
1921 background_compression
= 1;
1924 /** Read the set of voters from the cached object <b>ent</b> into
1925 * <b>out</b>, as a list of hex-encoded digests. Return 0 on success,
1926 * -1 if no signatories were recorded. */
1928 consensus_cache_entry_get_voter_id_digests(const consensus_cache_entry_t
*ent
,
1934 s
= consensus_cache_entry_get_value(ent
, LABEL_SIGNATORIES
);
1937 smartlist_split_string(out
, s
, ",", SPLIT_SKIP_SPACE
|SPLIT_STRIP_SPACE
, 0);
1941 /** Read the fresh-until time of cached object <b>ent</b> into *<b>out</b>
1942 * and return 0, or return -1 if no such time was recorded. */
1944 consensus_cache_entry_get_fresh_until(const consensus_cache_entry_t
*ent
,
1950 s
= consensus_cache_entry_get_value(ent
, LABEL_FRESH_UNTIL
);
1951 if (s
== NULL
|| parse_iso_time_nospace(s
, out
) < 0)
1957 /** Read the valid until timestamp from the cached object <b>ent</b> into
1958 * *<b>out</b> and return 0, or return -1 if no such time was recorded. */
1960 consensus_cache_entry_get_valid_until(const consensus_cache_entry_t
*ent
,
1967 s
= consensus_cache_entry_get_value(ent
, LABEL_VALID_UNTIL
);
1968 if (s
== NULL
|| parse_iso_time_nospace(s
, out
) < 0)
1974 /** Read the valid after timestamp from the cached object <b>ent</b> into
1975 * *<b>out</b> and return 0, or return -1 if no such time was recorded. */
1977 consensus_cache_entry_get_valid_after(const consensus_cache_entry_t
*ent
,
1984 s
= consensus_cache_entry_get_value(ent
, LABEL_VALID_AFTER
);
1986 if (s
== NULL
|| parse_iso_time_nospace(s
, out
) < 0)