diff options
Diffstat (limited to 'src/feature/stats')
| -rw-r--r-- | src/feature/stats/geoip.c | 1889 | ||||
| -rw-r--r-- | src/feature/stats/geoip.h | 157 | ||||
| -rw-r--r-- | src/feature/stats/rephist.c | 3227 | ||||
| -rw-r--r-- | src/feature/stats/rephist.h | 142 |
4 files changed, 5415 insertions, 0 deletions
diff --git a/src/feature/stats/geoip.c b/src/feature/stats/geoip.c new file mode 100644 index 0000000000..019aa7a283 --- /dev/null +++ b/src/feature/stats/geoip.c @@ -0,0 +1,1889 @@ +/* Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file geoip.c + * \brief Functions related to maintaining an IP-to-country database; + * to summarizing client connections by country to entry guards, bridges, + * and directory servers; and for statistics on answering network status + * requests. + * + * There are two main kinds of functions in this module: geoip functions, + * which map groups of IPv4 and IPv6 addresses to country codes, and + * statistical functions, which collect statistics about different kinds of + * per-country usage. + * + * The geoip lookup tables are implemented as sorted lists of disjoint address + * ranges, each mapping to a singleton geoip_country_t. These country objects + * are also indexed by their names in a hashtable. + * + * The tables are populated from disk at startup by the geoip_load_file() + * function. For more information on the file format they read, see that + * function. See the scripts and the README file in src/config for more + * information about how those files are generated. + * + * Tor uses GeoIP information in order to implement user requests (such as + * ExcludeNodes {cc}), and to keep track of how much usage relays are getting + * for each country. + */ + +#define GEOIP_PRIVATE +#include "core/or/or.h" +#include "ht.h" +#include "lib/container/buffers.h" +#include "app/config/config.h" +#include "feature/control/control.h" +#include "feature/client/dnsserv.h" +#include "core/or/dos.h" +#include "feature/stats/geoip.h" +#include "feature/nodelist/routerlist.h" + +#include "lib/container/order.h" +#include "lib/time/tvdiff.h" + +static void init_geoip_countries(void); + +/** An entry from the GeoIP IPv4 file: maps an IPv4 range to a country. */ +typedef struct geoip_ipv4_entry_t { + uint32_t ip_low; /**< The lowest IP in the range, in host order */ + uint32_t ip_high; /**< The highest IP in the range, in host order */ + intptr_t country; /**< An index into geoip_countries */ +} geoip_ipv4_entry_t; + +/** An entry from the GeoIP IPv6 file: maps an IPv6 range to a country. */ +typedef struct geoip_ipv6_entry_t { + struct in6_addr ip_low; /**< The lowest IP in the range, in host order */ + struct in6_addr ip_high; /**< The highest IP in the range, in host order */ + intptr_t country; /**< An index into geoip_countries */ +} geoip_ipv6_entry_t; + +/** A per-country record for GeoIP request history. */ +typedef struct geoip_country_t { + char countrycode[3]; + uint32_t n_v3_ns_requests; +} geoip_country_t; + +/** A list of geoip_country_t */ +static smartlist_t *geoip_countries = NULL; +/** A map from lowercased country codes to their position in geoip_countries. + * The index is encoded in the pointer, and 1 is added so that NULL can mean + * not found. */ +static strmap_t *country_idxplus1_by_lc_code = NULL; +/** Lists of all known geoip_ipv4_entry_t and geoip_ipv6_entry_t, sorted + * by their respective ip_low. */ +static smartlist_t *geoip_ipv4_entries = NULL, *geoip_ipv6_entries = NULL; + +/** SHA1 digest of the GeoIP files to include in extra-info descriptors. */ +static char geoip_digest[DIGEST_LEN]; +static char geoip6_digest[DIGEST_LEN]; + +/* Total size in bytes of the geoip client history cache. Used by the OOM + * handler. */ +static size_t geoip_client_history_cache_size; + +/* Increment the geoip client history cache size counter with the given bytes. + * This prevents an overflow and set it to its maximum in that case. */ +static inline void +geoip_increment_client_history_cache_size(size_t bytes) +{ + /* This is shockingly high, lets log it so it can be reported. */ + IF_BUG_ONCE(geoip_client_history_cache_size > (SIZE_MAX - bytes)) { + geoip_client_history_cache_size = SIZE_MAX; + return; + } + geoip_client_history_cache_size += bytes; +} + +/* Decrement the geoip client history cache size counter with the given bytes. + * This prevents an underflow and set it to 0 in that case. */ +static inline void +geoip_decrement_client_history_cache_size(size_t bytes) +{ + /* Going below 0 means that we either allocated an entry without + * incrementing the counter or we have different sizes when allocating and + * freeing. It shouldn't happened so log it. */ + IF_BUG_ONCE(geoip_client_history_cache_size < bytes) { + geoip_client_history_cache_size = 0; + return; + } + geoip_client_history_cache_size -= bytes; +} + +/** Return the index of the <b>country</b>'s entry in the GeoIP + * country list if it is a valid 2-letter country code, otherwise + * return -1. */ +MOCK_IMPL(country_t, +geoip_get_country,(const char *country)) +{ + void *idxplus1_; + intptr_t idx; + + idxplus1_ = strmap_get_lc(country_idxplus1_by_lc_code, country); + if (!idxplus1_) + return -1; + + idx = ((uintptr_t)idxplus1_)-1; + return (country_t)idx; +} + +/** Add an entry to a GeoIP table, mapping all IP addresses between <b>low</b> + * and <b>high</b>, inclusive, to the 2-letter country code <b>country</b>. */ +static void +geoip_add_entry(const tor_addr_t *low, const tor_addr_t *high, + const char *country) +{ + intptr_t idx; + void *idxplus1_; + + IF_BUG_ONCE(tor_addr_family(low) != tor_addr_family(high)) + return; + IF_BUG_ONCE(tor_addr_compare(high, low, CMP_EXACT) < 0) + return; + + idxplus1_ = strmap_get_lc(country_idxplus1_by_lc_code, country); + + if (!idxplus1_) { + geoip_country_t *c = tor_malloc_zero(sizeof(geoip_country_t)); + strlcpy(c->countrycode, country, sizeof(c->countrycode)); + tor_strlower(c->countrycode); + smartlist_add(geoip_countries, c); + idx = smartlist_len(geoip_countries) - 1; + strmap_set_lc(country_idxplus1_by_lc_code, country, (void*)(idx+1)); + } else { + idx = ((uintptr_t)idxplus1_)-1; + } + { + geoip_country_t *c = smartlist_get(geoip_countries, (int)idx); + tor_assert(!strcasecmp(c->countrycode, country)); + } + + if (tor_addr_family(low) == AF_INET) { + geoip_ipv4_entry_t *ent = tor_malloc_zero(sizeof(geoip_ipv4_entry_t)); + ent->ip_low = tor_addr_to_ipv4h(low); + ent->ip_high = tor_addr_to_ipv4h(high); + ent->country = idx; + smartlist_add(geoip_ipv4_entries, ent); + } else if (tor_addr_family(low) == AF_INET6) { + geoip_ipv6_entry_t *ent = tor_malloc_zero(sizeof(geoip_ipv6_entry_t)); + ent->ip_low = *tor_addr_to_in6_assert(low); + ent->ip_high = *tor_addr_to_in6_assert(high); + ent->country = idx; + smartlist_add(geoip_ipv6_entries, ent); + } +} + +/** Add an entry to the GeoIP table indicated by <b>family</b>, + * parsing it from <b>line</b>. The format is as for geoip_load_file(). */ +STATIC int +geoip_parse_entry(const char *line, sa_family_t family) +{ + tor_addr_t low_addr, high_addr; + char c[3]; + char *country = NULL; + + if (!geoip_countries) + init_geoip_countries(); + if (family == AF_INET) { + if (!geoip_ipv4_entries) + geoip_ipv4_entries = smartlist_new(); + } else if (family == AF_INET6) { + if (!geoip_ipv6_entries) + geoip_ipv6_entries = smartlist_new(); + } else { + log_warn(LD_GENERAL, "Unsupported family: %d", family); + return -1; + } + + while (TOR_ISSPACE(*line)) + ++line; + if (*line == '#') + return 0; + + char buf[512]; + if (family == AF_INET) { + unsigned int low, high; + if (tor_sscanf(line,"%u,%u,%2s", &low, &high, c) == 3 || + tor_sscanf(line,"\"%u\",\"%u\",\"%2s\",", &low, &high, c) == 3) { + tor_addr_from_ipv4h(&low_addr, low); + tor_addr_from_ipv4h(&high_addr, high); + } else + goto fail; + country = c; + } else { /* AF_INET6 */ + char *low_str, *high_str; + struct in6_addr low, high; + char *strtok_state; + strlcpy(buf, line, sizeof(buf)); + low_str = tor_strtok_r(buf, ",", &strtok_state); + if (!low_str) + goto fail; + high_str = tor_strtok_r(NULL, ",", &strtok_state); + if (!high_str) + goto fail; + country = tor_strtok_r(NULL, "\n", &strtok_state); + if (!country) + goto fail; + if (strlen(country) != 2) + goto fail; + if (tor_inet_pton(AF_INET6, low_str, &low) <= 0) + goto fail; + tor_addr_from_in6(&low_addr, &low); + if (tor_inet_pton(AF_INET6, high_str, &high) <= 0) + goto fail; + tor_addr_from_in6(&high_addr, &high); + } + geoip_add_entry(&low_addr, &high_addr, country); + return 0; + + fail: + log_warn(LD_GENERAL, "Unable to parse line from GEOIP %s file: %s", + family == AF_INET ? "IPv4" : "IPv6", escaped(line)); + return -1; +} + +/** Sorting helper: return -1, 1, or 0 based on comparison of two + * geoip_ipv4_entry_t */ +static int +geoip_ipv4_compare_entries_(const void **_a, const void **_b) +{ + const geoip_ipv4_entry_t *a = *_a, *b = *_b; + if (a->ip_low < b->ip_low) + return -1; + else if (a->ip_low > b->ip_low) + return 1; + else + return 0; +} + +/** bsearch helper: return -1, 1, or 0 based on comparison of an IP (a pointer + * to a uint32_t in host order) to a geoip_ipv4_entry_t */ +static int +geoip_ipv4_compare_key_to_entry_(const void *_key, const void **_member) +{ + /* No alignment issue here, since _key really is a pointer to uint32_t */ + const uint32_t addr = *(uint32_t *)_key; + const geoip_ipv4_entry_t *entry = *_member; + if (addr < entry->ip_low) + return -1; + else if (addr > entry->ip_high) + return 1; + else + return 0; +} + +/** Sorting helper: return -1, 1, or 0 based on comparison of two + * geoip_ipv6_entry_t */ +static int +geoip_ipv6_compare_entries_(const void **_a, const void **_b) +{ + const geoip_ipv6_entry_t *a = *_a, *b = *_b; + return fast_memcmp(a->ip_low.s6_addr, b->ip_low.s6_addr, + sizeof(struct in6_addr)); +} + +/** bsearch helper: return -1, 1, or 0 based on comparison of an IPv6 + * (a pointer to a in6_addr) to a geoip_ipv6_entry_t */ +static int +geoip_ipv6_compare_key_to_entry_(const void *_key, const void **_member) +{ + const struct in6_addr *addr = (struct in6_addr *)_key; + const geoip_ipv6_entry_t *entry = *_member; + + if (fast_memcmp(addr->s6_addr, entry->ip_low.s6_addr, + sizeof(struct in6_addr)) < 0) + return -1; + else if (fast_memcmp(addr->s6_addr, entry->ip_high.s6_addr, + sizeof(struct in6_addr)) > 0) + return 1; + else + return 0; +} + +/** Return 1 if we should collect geoip stats on bridge users, and + * include them in our extrainfo descriptor. Else return 0. */ +int +should_record_bridge_info(const or_options_t *options) +{ + return options->BridgeRelay && options->BridgeRecordUsageByCountry; +} + +/** Set up a new list of geoip countries with no countries (yet) set in it, + * except for the unknown country. + */ +static void +init_geoip_countries(void) +{ + geoip_country_t *geoip_unresolved; + geoip_countries = smartlist_new(); + /* Add a geoip_country_t for requests that could not be resolved to a + * country as first element (index 0) to geoip_countries. */ + geoip_unresolved = tor_malloc_zero(sizeof(geoip_country_t)); + strlcpy(geoip_unresolved->countrycode, "??", + sizeof(geoip_unresolved->countrycode)); + smartlist_add(geoip_countries, geoip_unresolved); + country_idxplus1_by_lc_code = strmap_new(); + strmap_set_lc(country_idxplus1_by_lc_code, "??", (void*)(1)); +} + +/** Clear appropriate GeoIP database, based on <b>family</b>, and + * reload it from the file <b>filename</b>. Return 0 on success, -1 on + * failure. + * + * Recognized line formats for IPv4 are: + * INTIPLOW,INTIPHIGH,CC + * and + * "INTIPLOW","INTIPHIGH","CC","CC3","COUNTRY NAME" + * where INTIPLOW and INTIPHIGH are IPv4 addresses encoded as 4-byte unsigned + * integers, and CC is a country code. + * + * Recognized line format for IPv6 is: + * IPV6LOW,IPV6HIGH,CC + * where IPV6LOW and IPV6HIGH are IPv6 addresses and CC is a country code. + * + * It also recognizes, and skips over, blank lines and lines that start + * with '#' (comments). + */ +int +geoip_load_file(sa_family_t family, const char *filename) +{ + FILE *f; + const char *msg = ""; + const or_options_t *options = get_options(); + int severity = options_need_geoip_info(options, &msg) ? LOG_WARN : LOG_INFO; + crypto_digest_t *geoip_digest_env = NULL; + + tor_assert(family == AF_INET || family == AF_INET6); + + if (!(f = tor_fopen_cloexec(filename, "r"))) { + log_fn(severity, LD_GENERAL, "Failed to open GEOIP file %s. %s", + filename, msg); + return -1; + } + if (!geoip_countries) + init_geoip_countries(); + + if (family == AF_INET) { + if (geoip_ipv4_entries) { + SMARTLIST_FOREACH(geoip_ipv4_entries, geoip_ipv4_entry_t *, e, + tor_free(e)); + smartlist_free(geoip_ipv4_entries); + } + geoip_ipv4_entries = smartlist_new(); + } else { /* AF_INET6 */ + if (geoip_ipv6_entries) { + SMARTLIST_FOREACH(geoip_ipv6_entries, geoip_ipv6_entry_t *, e, + tor_free(e)); + smartlist_free(geoip_ipv6_entries); + } + geoip_ipv6_entries = smartlist_new(); + } + geoip_digest_env = crypto_digest_new(); + + log_notice(LD_GENERAL, "Parsing GEOIP %s file %s.", + (family == AF_INET) ? "IPv4" : "IPv6", filename); + while (!feof(f)) { + char buf[512]; + if (fgets(buf, (int)sizeof(buf), f) == NULL) + break; + crypto_digest_add_bytes(geoip_digest_env, buf, strlen(buf)); + /* FFFF track full country name. */ + geoip_parse_entry(buf, family); + } + /*XXXX abort and return -1 if no entries/illformed?*/ + fclose(f); + + /* Sort list and remember file digests so that we can include it in + * our extra-info descriptors. */ + if (family == AF_INET) { + smartlist_sort(geoip_ipv4_entries, geoip_ipv4_compare_entries_); + /* Okay, now we need to maybe change our mind about what is in + * which country. We do this for IPv4 only since that's what we + * store in node->country. */ + refresh_all_country_info(); + crypto_digest_get_digest(geoip_digest_env, geoip_digest, DIGEST_LEN); + } else { + /* AF_INET6 */ + smartlist_sort(geoip_ipv6_entries, geoip_ipv6_compare_entries_); + crypto_digest_get_digest(geoip_digest_env, geoip6_digest, DIGEST_LEN); + } + crypto_digest_free(geoip_digest_env); + + return 0; +} + +/** Given an IP address in host order, return a number representing the + * country to which that address belongs, -1 for "No geoip information + * available", or 0 for the 'unknown country'. The return value will always + * be less than geoip_get_n_countries(). To decode it, call + * geoip_get_country_name(). + */ +STATIC int +geoip_get_country_by_ipv4(uint32_t ipaddr) +{ + geoip_ipv4_entry_t *ent; + if (!geoip_ipv4_entries) + return -1; + ent = smartlist_bsearch(geoip_ipv4_entries, &ipaddr, + geoip_ipv4_compare_key_to_entry_); + return ent ? (int)ent->country : 0; +} + +/** Given an IPv6 address, return a number representing the country to + * which that address belongs, -1 for "No geoip information available", or + * 0 for the 'unknown country'. The return value will always be less than + * geoip_get_n_countries(). To decode it, call geoip_get_country_name(). + */ +STATIC int +geoip_get_country_by_ipv6(const struct in6_addr *addr) +{ + geoip_ipv6_entry_t *ent; + + if (!geoip_ipv6_entries) + return -1; + ent = smartlist_bsearch(geoip_ipv6_entries, addr, + geoip_ipv6_compare_key_to_entry_); + return ent ? (int)ent->country : 0; +} + +/** Given an IP address, return a number representing the country to which + * that address belongs, -1 for "No geoip information available", or 0 for + * the 'unknown country'. The return value will always be less than + * geoip_get_n_countries(). To decode it, call geoip_get_country_name(). + */ +MOCK_IMPL(int, +geoip_get_country_by_addr,(const tor_addr_t *addr)) +{ + if (tor_addr_family(addr) == AF_INET) { + return geoip_get_country_by_ipv4(tor_addr_to_ipv4h(addr)); + } else if (tor_addr_family(addr) == AF_INET6) { + return geoip_get_country_by_ipv6(tor_addr_to_in6(addr)); + } else { + return -1; + } +} + +/** Return the number of countries recognized by the GeoIP country list. */ +MOCK_IMPL(int, +geoip_get_n_countries,(void)) +{ + if (!geoip_countries) + init_geoip_countries(); + return (int) smartlist_len(geoip_countries); +} + +/** Return the two-letter country code associated with the number <b>num</b>, + * or "??" for an unknown value. */ +const char * +geoip_get_country_name(country_t num) +{ + if (geoip_countries && num >= 0 && num < smartlist_len(geoip_countries)) { + geoip_country_t *c = smartlist_get(geoip_countries, num); + return c->countrycode; + } else + return "??"; +} + +/** Return true iff we have loaded a GeoIP database.*/ +MOCK_IMPL(int, +geoip_is_loaded,(sa_family_t family)) +{ + tor_assert(family == AF_INET || family == AF_INET6); + if (geoip_countries == NULL) + return 0; + if (family == AF_INET) + return geoip_ipv4_entries != NULL; + else /* AF_INET6 */ + return geoip_ipv6_entries != NULL; +} + +/** Return the hex-encoded SHA1 digest of the loaded GeoIP file. The + * result does not need to be deallocated, but will be overwritten by the + * next call of hex_str(). */ +const char * +geoip_db_digest(sa_family_t family) +{ + tor_assert(family == AF_INET || family == AF_INET6); + if (family == AF_INET) + return hex_str(geoip_digest, DIGEST_LEN); + else /* AF_INET6 */ + return hex_str(geoip6_digest, DIGEST_LEN); +} + +/** Largest allowable value for last_seen_in_minutes. (It's a 30-bit field, + * so it can hold up to (1u<<30)-1, or 0x3fffffffu. + */ +#define MAX_LAST_SEEN_IN_MINUTES 0X3FFFFFFFu + +/** Map from client IP address to last time seen. */ +static HT_HEAD(clientmap, clientmap_entry_t) client_history = + HT_INITIALIZER(); + +/** Hashtable helper: compute a hash of a clientmap_entry_t. */ +static inline unsigned +clientmap_entry_hash(const clientmap_entry_t *a) +{ + unsigned h = (unsigned) tor_addr_hash(&a->addr); + + if (a->transport_name) + h += (unsigned) siphash24g(a->transport_name, strlen(a->transport_name)); + + return h; +} +/** Hashtable helper: compare two clientmap_entry_t values for equality. */ +static inline int +clientmap_entries_eq(const clientmap_entry_t *a, const clientmap_entry_t *b) +{ + if (strcmp_opt(a->transport_name, b->transport_name)) + return 0; + + return !tor_addr_compare(&a->addr, &b->addr, CMP_EXACT) && + a->action == b->action; +} + +HT_PROTOTYPE(clientmap, clientmap_entry_t, node, clientmap_entry_hash, + clientmap_entries_eq) +HT_GENERATE2(clientmap, clientmap_entry_t, node, clientmap_entry_hash, + clientmap_entries_eq, 0.6, tor_reallocarray_, tor_free_) + +#define clientmap_entry_free(ent) \ + FREE_AND_NULL(clientmap_entry_t, clientmap_entry_free_, ent) + +/** Return the size of a client map entry. */ +static inline size_t +clientmap_entry_size(const clientmap_entry_t *ent) +{ + tor_assert(ent); + return (sizeof(clientmap_entry_t) + + (ent->transport_name ? strlen(ent->transport_name) : 0)); +} + +/** Free all storage held by <b>ent</b>. */ +static void +clientmap_entry_free_(clientmap_entry_t *ent) +{ + if (!ent) + return; + + /* This entry is about to be freed so pass it to the DoS subsystem to see if + * any actions can be taken about it. */ + dos_geoip_entry_about_to_free(ent); + geoip_decrement_client_history_cache_size(clientmap_entry_size(ent)); + + tor_free(ent->transport_name); + tor_free(ent); +} + +/* Return a newly allocated clientmap entry with the given action and address + * that are mandatory. The transport_name can be optional. This can't fail. */ +static clientmap_entry_t * +clientmap_entry_new(geoip_client_action_t action, const tor_addr_t *addr, + const char *transport_name) +{ + clientmap_entry_t *entry; + + tor_assert(action == GEOIP_CLIENT_CONNECT || + action == GEOIP_CLIENT_NETWORKSTATUS); + tor_assert(addr); + + entry = tor_malloc_zero(sizeof(clientmap_entry_t)); + entry->action = action; + tor_addr_copy(&entry->addr, addr); + if (transport_name) { + entry->transport_name = tor_strdup(transport_name); + } + + /* Allocated and initialized, note down its size for the OOM handler. */ + geoip_increment_client_history_cache_size(clientmap_entry_size(entry)); + + return entry; +} + +/** Clear history of connecting clients used by entry and bridge stats. */ +static void +client_history_clear(void) +{ + clientmap_entry_t **ent, **next, *this; + for (ent = HT_START(clientmap, &client_history); ent != NULL; + ent = next) { + if ((*ent)->action == GEOIP_CLIENT_CONNECT) { + this = *ent; + next = HT_NEXT_RMV(clientmap, &client_history, ent); + clientmap_entry_free(this); + } else { + next = HT_NEXT(clientmap, &client_history, ent); + } + } +} + +/** Note that we've seen a client connect from the IP <b>addr</b> + * at time <b>now</b>. Ignored by all but bridges and directories if + * configured accordingly. */ +void +geoip_note_client_seen(geoip_client_action_t action, + const tor_addr_t *addr, + const char *transport_name, + time_t now) +{ + const or_options_t *options = get_options(); + clientmap_entry_t *ent; + + if (action == GEOIP_CLIENT_CONNECT) { + /* Only remember statistics if the DoS mitigation subsystem is enabled. If + * not, only if as entry guard or as bridge. */ + if (!dos_enabled()) { + if (!options->EntryStatistics && !should_record_bridge_info(options)) { + return; + } + } + } else { + /* Only gather directory-request statistics if configured, and + * forcibly disable them on bridge authorities. */ + if (!options->DirReqStatistics || options->BridgeAuthoritativeDir) + return; + } + + log_debug(LD_GENERAL, "Seen client from '%s' with transport '%s'.", + safe_str_client(fmt_addr((addr))), + transport_name ? transport_name : "<no transport>"); + + ent = geoip_lookup_client(addr, transport_name, action); + if (! ent) { + ent = clientmap_entry_new(action, addr, transport_name); + HT_INSERT(clientmap, &client_history, ent); + } + if (now / 60 <= (int)MAX_LAST_SEEN_IN_MINUTES && now >= 0) + ent->last_seen_in_minutes = (unsigned)(now/60); + else + ent->last_seen_in_minutes = 0; + + if (action == GEOIP_CLIENT_NETWORKSTATUS) { + int country_idx = geoip_get_country_by_addr(addr); + if (country_idx < 0) + country_idx = 0; /** unresolved requests are stored at index 0. */ + if (country_idx >= 0 && country_idx < smartlist_len(geoip_countries)) { + geoip_country_t *country = smartlist_get(geoip_countries, country_idx); + ++country->n_v3_ns_requests; + } + } +} + +/** HT_FOREACH helper: remove a clientmap_entry_t from the hashtable if it's + * older than a certain time. */ +static int +remove_old_client_helper_(struct clientmap_entry_t *ent, void *_cutoff) +{ + time_t cutoff = *(time_t*)_cutoff / 60; + if (ent->last_seen_in_minutes < cutoff) { + clientmap_entry_free(ent); + return 1; + } else { + return 0; + } +} + +/** Forget about all clients that haven't connected since <b>cutoff</b>. */ +void +geoip_remove_old_clients(time_t cutoff) +{ + clientmap_HT_FOREACH_FN(&client_history, + remove_old_client_helper_, + &cutoff); +} + +/* Return a client entry object matching the given address, transport name and + * geoip action from the clientmap. NULL if not found. The transport_name can + * be NULL. */ +clientmap_entry_t * +geoip_lookup_client(const tor_addr_t *addr, const char *transport_name, + geoip_client_action_t action) +{ + clientmap_entry_t lookup; + + tor_assert(addr); + + /* We always look for a client connection with no transport. */ + tor_addr_copy(&lookup.addr, addr); + lookup.action = action; + lookup.transport_name = (char *) transport_name; + + return HT_FIND(clientmap, &client_history, &lookup); +} + +/* Cleanup client entries older than the cutoff. Used for the OOM. Return the + * number of bytes freed. If 0 is returned, nothing was freed. */ +static size_t +oom_clean_client_entries(time_t cutoff) +{ + size_t bytes = 0; + clientmap_entry_t **ent, **ent_next; + + for (ent = HT_START(clientmap, &client_history); ent; ent = ent_next) { + clientmap_entry_t *entry = *ent; + if (entry->last_seen_in_minutes < (cutoff / 60)) { + ent_next = HT_NEXT_RMV(clientmap, &client_history, ent); + bytes += clientmap_entry_size(entry); + clientmap_entry_free(entry); + } else { + ent_next = HT_NEXT(clientmap, &client_history, ent); + } + } + return bytes; +} + +/* Below this minimum lifetime, the OOM won't cleanup any entries. */ +#define GEOIP_CLIENT_CACHE_OOM_MIN_CUTOFF (4 * 60 * 60) +/* The OOM moves the cutoff by that much every run. */ +#define GEOIP_CLIENT_CACHE_OOM_STEP (15 * 50) + +/* Cleanup the geoip client history cache called from the OOM handler. Return + * the amount of bytes removed. This can return a value below or above + * min_remove_bytes but will stop as oon as the min_remove_bytes has been + * reached. */ +size_t +geoip_client_cache_handle_oom(time_t now, size_t min_remove_bytes) +{ + time_t k; + size_t bytes_removed = 0; + + /* Our OOM handler called with 0 bytes to remove is a code flow error. */ + tor_assert(min_remove_bytes != 0); + + /* Set k to the initial cutoff of an entry. We then going to move it by step + * to try to remove as much as we can. */ + k = WRITE_STATS_INTERVAL; + + do { + time_t cutoff; + + /* If k has reached the minimum lifetime, we have to stop else we might + * remove every single entries which would be pretty bad for the DoS + * mitigation subsystem if by just filling the geoip cache, it was enough + * to trigger the OOM and clean every single entries. */ + if (k <= GEOIP_CLIENT_CACHE_OOM_MIN_CUTOFF) { + break; + } + + cutoff = now - k; + bytes_removed += oom_clean_client_entries(cutoff); + k -= GEOIP_CLIENT_CACHE_OOM_STEP; + } while (bytes_removed < min_remove_bytes); + + return bytes_removed; +} + +/* Return the total size in bytes of the client history cache. */ +size_t +geoip_client_cache_total_allocation(void) +{ + return geoip_client_history_cache_size; +} + +/** How many responses are we giving to clients requesting v3 network + * statuses? */ +static uint32_t ns_v3_responses[GEOIP_NS_RESPONSE_NUM]; + +/** Note that we've rejected a client's request for a v3 network status + * for reason <b>reason</b> at time <b>now</b>. */ +void +geoip_note_ns_response(geoip_ns_response_t response) +{ + static int arrays_initialized = 0; + if (!get_options()->DirReqStatistics) + return; + if (!arrays_initialized) { + memset(ns_v3_responses, 0, sizeof(ns_v3_responses)); + arrays_initialized = 1; + } + tor_assert(response < GEOIP_NS_RESPONSE_NUM); + ns_v3_responses[response]++; +} + +/** Do not mention any country from which fewer than this number of IPs have + * connected. This conceivably avoids reporting information that could + * deanonymize users, though analysis is lacking. */ +#define MIN_IPS_TO_NOTE_COUNTRY 1 +/** Do not report any geoip data at all if we have fewer than this number of + * IPs to report about. */ +#define MIN_IPS_TO_NOTE_ANYTHING 1 +/** When reporting geoip data about countries, round up to the nearest + * multiple of this value. */ +#define IP_GRANULARITY 8 + +/** Helper type: used to sort per-country totals by value. */ +typedef struct c_hist_t { + char country[3]; /**< Two-letter country code. */ + unsigned total; /**< Total IP addresses seen in this country. */ +} c_hist_t; + +/** Sorting helper: return -1, 1, or 0 based on comparison of two + * geoip_ipv4_entry_t. Sort in descending order of total, and then by country + * code. */ +static int +c_hist_compare_(const void **_a, const void **_b) +{ + const c_hist_t *a = *_a, *b = *_b; + if (a->total > b->total) + return -1; + else if (a->total < b->total) + return 1; + else + return strcmp(a->country, b->country); +} + +/** When there are incomplete directory requests at the end of a 24-hour + * period, consider those requests running for longer than this timeout as + * failed, the others as still running. */ +#define DIRREQ_TIMEOUT (10*60) + +/** Entry in a map from either chan->global_identifier for direct requests + * or a unique circuit identifier for tunneled requests to request time, + * response size, and completion time of a network status request. Used to + * measure download times of requests to derive average client + * bandwidths. */ +typedef struct dirreq_map_entry_t { + HT_ENTRY(dirreq_map_entry_t) node; + /** Unique identifier for this network status request; this is either the + * chan->global_identifier of the dir channel (direct request) or a new + * locally unique identifier of a circuit (tunneled request). This ID is + * only unique among other direct or tunneled requests, respectively. */ + uint64_t dirreq_id; + unsigned int state:3; /**< State of this directory request. */ + unsigned int type:1; /**< Is this a direct or a tunneled request? */ + unsigned int completed:1; /**< Is this request complete? */ + /** When did we receive the request and started sending the response? */ + struct timeval request_time; + size_t response_size; /**< What is the size of the response in bytes? */ + struct timeval completion_time; /**< When did the request succeed? */ +} dirreq_map_entry_t; + +/** Map of all directory requests asking for v2 or v3 network statuses in + * the current geoip-stats interval. Values are + * of type *<b>dirreq_map_entry_t</b>. */ +static HT_HEAD(dirreqmap, dirreq_map_entry_t) dirreq_map = + HT_INITIALIZER(); + +static int +dirreq_map_ent_eq(const dirreq_map_entry_t *a, + const dirreq_map_entry_t *b) +{ + return a->dirreq_id == b->dirreq_id && a->type == b->type; +} + +/* DOCDOC dirreq_map_ent_hash */ +static unsigned +dirreq_map_ent_hash(const dirreq_map_entry_t *entry) +{ + unsigned u = (unsigned) entry->dirreq_id; + u += entry->type << 20; + return u; +} + +HT_PROTOTYPE(dirreqmap, dirreq_map_entry_t, node, dirreq_map_ent_hash, + dirreq_map_ent_eq) +HT_GENERATE2(dirreqmap, dirreq_map_entry_t, node, dirreq_map_ent_hash, + dirreq_map_ent_eq, 0.6, tor_reallocarray_, tor_free_) + +/** Helper: Put <b>entry</b> into map of directory requests using + * <b>type</b> and <b>dirreq_id</b> as key parts. If there is + * already an entry for that key, print out a BUG warning and return. */ +static void +dirreq_map_put_(dirreq_map_entry_t *entry, dirreq_type_t type, + uint64_t dirreq_id) +{ + dirreq_map_entry_t *old_ent; + tor_assert(entry->type == type); + tor_assert(entry->dirreq_id == dirreq_id); + + /* XXXX we could switch this to HT_INSERT some time, since it seems that + * this bug doesn't happen. But since this function doesn't seem to be + * critical-path, it's sane to leave it alone. */ + old_ent = HT_REPLACE(dirreqmap, &dirreq_map, entry); + if (old_ent && old_ent != entry) { + log_warn(LD_BUG, "Error when putting directory request into local " + "map. There was already an entry for the same identifier."); + return; + } +} + +/** Helper: Look up and return an entry in the map of directory requests + * using <b>type</b> and <b>dirreq_id</b> as key parts. If there + * is no such entry, return NULL. */ +static dirreq_map_entry_t * +dirreq_map_get_(dirreq_type_t type, uint64_t dirreq_id) +{ + dirreq_map_entry_t lookup; + lookup.type = type; + lookup.dirreq_id = dirreq_id; + return HT_FIND(dirreqmap, &dirreq_map, &lookup); +} + +/** Note that an either direct or tunneled (see <b>type</b>) directory + * request for a v3 network status with unique ID <b>dirreq_id</b> of size + * <b>response_size</b> has started. */ +void +geoip_start_dirreq(uint64_t dirreq_id, size_t response_size, + dirreq_type_t type) +{ + dirreq_map_entry_t *ent; + if (!get_options()->DirReqStatistics) + return; + ent = tor_malloc_zero(sizeof(dirreq_map_entry_t)); + ent->dirreq_id = dirreq_id; + tor_gettimeofday(&ent->request_time); + ent->response_size = response_size; + ent->type = type; + dirreq_map_put_(ent, type, dirreq_id); +} + +/** Change the state of the either direct or tunneled (see <b>type</b>) + * directory request with <b>dirreq_id</b> to <b>new_state</b> and + * possibly mark it as completed. If no entry can be found for the given + * key parts (e.g., if this is a directory request that we are not + * measuring, or one that was started in the previous measurement period), + * or if the state cannot be advanced to <b>new_state</b>, do nothing. */ +void +geoip_change_dirreq_state(uint64_t dirreq_id, dirreq_type_t type, + dirreq_state_t new_state) +{ + dirreq_map_entry_t *ent; + if (!get_options()->DirReqStatistics) + return; + ent = dirreq_map_get_(type, dirreq_id); + if (!ent) + return; + if (new_state == DIRREQ_IS_FOR_NETWORK_STATUS) + return; + if (new_state - 1 != ent->state) + return; + ent->state = new_state; + if ((type == DIRREQ_DIRECT && + new_state == DIRREQ_FLUSHING_DIR_CONN_FINISHED) || + (type == DIRREQ_TUNNELED && + new_state == DIRREQ_CHANNEL_BUFFER_FLUSHED)) { + tor_gettimeofday(&ent->completion_time); + ent->completed = 1; + } +} + +/** Return the bridge-ip-transports string that should be inserted in + * our extra-info descriptor. Return NULL if the bridge-ip-transports + * line should be empty. */ +char * +geoip_get_transport_history(void) +{ + unsigned granularity = IP_GRANULARITY; + /** String hash table (name of transport) -> (number of users). */ + strmap_t *transport_counts = strmap_new(); + + /** Smartlist that contains copies of the names of the transports + that have been used. */ + smartlist_t *transports_used = smartlist_new(); + + /* Special string to signify that no transport was used for this + connection. Pluggable transport names can't have symbols in their + names, so this string will never collide with a real transport. */ + static const char* no_transport_str = "<OR>"; + + clientmap_entry_t **ent; + smartlist_t *string_chunks = smartlist_new(); + char *the_string = NULL; + + /* If we haven't seen any clients yet, return NULL. */ + if (HT_EMPTY(&client_history)) + goto done; + + /** We do the following steps to form the transport history string: + * a) Foreach client that uses a pluggable transport, we increase the + * times that transport was used by one. If the client did not use + * a transport, we increase the number of times someone connected + * without obfuscation. + * b) Foreach transport we observed, we write its transport history + * string and push it to string_chunks. So, for example, if we've + * seen 665 obfs2 clients, we write "obfs2=665". + * c) We concatenate string_chunks to form the final string. + */ + + log_debug(LD_GENERAL,"Starting iteration for transport history. %d clients.", + HT_SIZE(&client_history)); + + /* Loop through all clients. */ + HT_FOREACH(ent, clientmap, &client_history) { + uintptr_t val; + void *ptr; + const char *transport_name = (*ent)->transport_name; + if (!transport_name) + transport_name = no_transport_str; + + /* Increase the count for this transport name. */ + ptr = strmap_get(transport_counts, transport_name); + val = (uintptr_t)ptr; + val++; + ptr = (void*)val; + strmap_set(transport_counts, transport_name, ptr); + + /* If it's the first time we see this transport, note it. */ + if (val == 1) + smartlist_add_strdup(transports_used, transport_name); + + log_debug(LD_GENERAL, "Client from '%s' with transport '%s'. " + "I've now seen %d clients.", + safe_str_client(fmt_addr(&(*ent)->addr)), + transport_name ? transport_name : "<no transport>", + (int)val); + } + + /* Sort the transport names (helps with unit testing). */ + smartlist_sort_strings(transports_used); + + /* Loop through all seen transports. */ + SMARTLIST_FOREACH_BEGIN(transports_used, const char *, transport_name) { + void *transport_count_ptr = strmap_get(transport_counts, transport_name); + uintptr_t transport_count = (uintptr_t) transport_count_ptr; + + log_debug(LD_GENERAL, "We got %"PRIu64" clients with transport '%s'.", + ((uint64_t)transport_count), transport_name); + + smartlist_add_asprintf(string_chunks, "%s=%"PRIu64, + transport_name, + (round_uint64_to_next_multiple_of( + (uint64_t)transport_count, + granularity))); + } SMARTLIST_FOREACH_END(transport_name); + + the_string = smartlist_join_strings(string_chunks, ",", 0, NULL); + + log_debug(LD_GENERAL, "Final bridge-ip-transports string: '%s'", the_string); + + done: + strmap_free(transport_counts, NULL); + SMARTLIST_FOREACH(transports_used, char *, s, tor_free(s)); + smartlist_free(transports_used); + SMARTLIST_FOREACH(string_chunks, char *, s, tor_free(s)); + smartlist_free(string_chunks); + + return the_string; +} + +/** Return a newly allocated comma-separated string containing statistics + * on network status downloads. The string contains the number of completed + * requests, timeouts, and still running requests as well as the download + * times by deciles and quartiles. Return NULL if we have not observed + * requests for long enough. */ +static char * +geoip_get_dirreq_history(dirreq_type_t type) +{ + char *result = NULL; + buf_t *buf = NULL; + smartlist_t *dirreq_completed = NULL; + uint32_t complete = 0, timeouts = 0, running = 0; + dirreq_map_entry_t **ptr, **next; + struct timeval now; + + tor_gettimeofday(&now); + dirreq_completed = smartlist_new(); + for (ptr = HT_START(dirreqmap, &dirreq_map); ptr; ptr = next) { + dirreq_map_entry_t *ent = *ptr; + if (ent->type != type) { + next = HT_NEXT(dirreqmap, &dirreq_map, ptr); + continue; + } else { + if (ent->completed) { + smartlist_add(dirreq_completed, ent); + complete++; + next = HT_NEXT_RMV(dirreqmap, &dirreq_map, ptr); + } else { + if (tv_mdiff(&ent->request_time, &now) / 1000 > DIRREQ_TIMEOUT) + timeouts++; + else + running++; + next = HT_NEXT_RMV(dirreqmap, &dirreq_map, ptr); + tor_free(ent); + } + } + } +#define DIR_REQ_GRANULARITY 4 + complete = round_uint32_to_next_multiple_of(complete, + DIR_REQ_GRANULARITY); + timeouts = round_uint32_to_next_multiple_of(timeouts, + DIR_REQ_GRANULARITY); + running = round_uint32_to_next_multiple_of(running, + DIR_REQ_GRANULARITY); + buf = buf_new_with_capacity(1024); + buf_add_printf(buf, "complete=%u,timeout=%u," + "running=%u", complete, timeouts, running); + +#define MIN_DIR_REQ_RESPONSES 16 + if (complete >= MIN_DIR_REQ_RESPONSES) { + uint32_t *dltimes; + /* We may have rounded 'completed' up. Here we want to use the + * real value. */ + complete = smartlist_len(dirreq_completed); + dltimes = tor_calloc(complete, sizeof(uint32_t)); + SMARTLIST_FOREACH_BEGIN(dirreq_completed, dirreq_map_entry_t *, ent) { + uint32_t bytes_per_second; + uint32_t time_diff_ = (uint32_t) tv_mdiff(&ent->request_time, + &ent->completion_time); + if (time_diff_ == 0) + time_diff_ = 1; /* Avoid DIV/0; "instant" answers are impossible + * by law of nature or something, but a millisecond + * is a bit greater than "instantly" */ + bytes_per_second = (uint32_t)(1000 * ent->response_size / time_diff_); + dltimes[ent_sl_idx] = bytes_per_second; + } SMARTLIST_FOREACH_END(ent); + median_uint32(dltimes, complete); /* sorts as a side effect. */ + buf_add_printf(buf, + ",min=%u,d1=%u,d2=%u,q1=%u,d3=%u,d4=%u,md=%u," + "d6=%u,d7=%u,q3=%u,d8=%u,d9=%u,max=%u", + dltimes[0], + dltimes[1*complete/10-1], + dltimes[2*complete/10-1], + dltimes[1*complete/4-1], + dltimes[3*complete/10-1], + dltimes[4*complete/10-1], + dltimes[5*complete/10-1], + dltimes[6*complete/10-1], + dltimes[7*complete/10-1], + dltimes[3*complete/4-1], + dltimes[8*complete/10-1], + dltimes[9*complete/10-1], + dltimes[complete-1]); + tor_free(dltimes); + } + + result = buf_extract(buf, NULL); + + SMARTLIST_FOREACH(dirreq_completed, dirreq_map_entry_t *, ent, + tor_free(ent)); + smartlist_free(dirreq_completed); + buf_free(buf); + return result; +} + +/** Store a newly allocated comma-separated string in + * *<a>country_str</a> containing entries for all the countries from + * which we've seen enough clients connect as a bridge, directory + * server, or entry guard. The entry format is cc=num where num is the + * number of IPs we've seen connecting from that country, and cc is a + * lowercased country code. *<a>country_str</a> is set to NULL if + * we're not ready to export per country data yet. + * + * Store a newly allocated comma-separated string in <a>ipver_str</a> + * containing entries for clients connecting over IPv4 and IPv6. The + * format is family=num where num is the nubmer of IPs we've seen + * connecting over that protocol family, and family is 'v4' or 'v6'. + * + * Return 0 on success and -1 if we're missing geoip data. */ +int +geoip_get_client_history(geoip_client_action_t action, + char **country_str, char **ipver_str) +{ + unsigned granularity = IP_GRANULARITY; + smartlist_t *entries = NULL; + int n_countries = geoip_get_n_countries(); + int i; + clientmap_entry_t **cm_ent; + unsigned *counts = NULL; + unsigned total = 0; + unsigned ipv4_count = 0, ipv6_count = 0; + + if (!geoip_is_loaded(AF_INET) && !geoip_is_loaded(AF_INET6)) + return -1; + + counts = tor_calloc(n_countries, sizeof(unsigned)); + HT_FOREACH(cm_ent, clientmap, &client_history) { + int country; + if ((*cm_ent)->action != (int)action) + continue; + country = geoip_get_country_by_addr(&(*cm_ent)->addr); + if (country < 0) + country = 0; /** unresolved requests are stored at index 0. */ + tor_assert(0 <= country && country < n_countries); + ++counts[country]; + ++total; + switch (tor_addr_family(&(*cm_ent)->addr)) { + case AF_INET: + ipv4_count++; + break; + case AF_INET6: + ipv6_count++; + break; + } + } + if (ipver_str) { + smartlist_t *chunks = smartlist_new(); + smartlist_add_asprintf(chunks, "v4=%u", + round_to_next_multiple_of(ipv4_count, granularity)); + smartlist_add_asprintf(chunks, "v6=%u", + round_to_next_multiple_of(ipv6_count, granularity)); + *ipver_str = smartlist_join_strings(chunks, ",", 0, NULL); + SMARTLIST_FOREACH(chunks, char *, c, tor_free(c)); + smartlist_free(chunks); + } + + /* Don't record per country data if we haven't seen enough IPs. */ + if (total < MIN_IPS_TO_NOTE_ANYTHING) { + tor_free(counts); + if (country_str) + *country_str = NULL; + return 0; + } + + /* Make a list of c_hist_t */ + entries = smartlist_new(); + for (i = 0; i < n_countries; ++i) { + unsigned c = counts[i]; + const char *countrycode; + c_hist_t *ent; + /* Only report a country if it has a minimum number of IPs. */ + if (c >= MIN_IPS_TO_NOTE_COUNTRY) { + c = round_to_next_multiple_of(c, granularity); + countrycode = geoip_get_country_name(i); + ent = tor_malloc(sizeof(c_hist_t)); + strlcpy(ent->country, countrycode, sizeof(ent->country)); + ent->total = c; + smartlist_add(entries, ent); + } + } + /* Sort entries. Note that we must do this _AFTER_ rounding, or else + * the sort order could leak info. */ + smartlist_sort(entries, c_hist_compare_); + + if (country_str) { + smartlist_t *chunks = smartlist_new(); + SMARTLIST_FOREACH(entries, c_hist_t *, ch, { + smartlist_add_asprintf(chunks, "%s=%u", ch->country, ch->total); + }); + *country_str = smartlist_join_strings(chunks, ",", 0, NULL); + SMARTLIST_FOREACH(chunks, char *, c, tor_free(c)); + smartlist_free(chunks); + } + + SMARTLIST_FOREACH(entries, c_hist_t *, c, tor_free(c)); + smartlist_free(entries); + tor_free(counts); + + return 0; +} + +/** Return a newly allocated string holding the per-country request history + * for v3 network statuses in a format suitable for an extra-info document, + * or NULL on failure. */ +char * +geoip_get_request_history(void) +{ + smartlist_t *entries, *strings; + char *result; + unsigned granularity = IP_GRANULARITY; + + if (!geoip_countries) + return NULL; + + entries = smartlist_new(); + SMARTLIST_FOREACH_BEGIN(geoip_countries, geoip_country_t *, c) { + uint32_t tot = 0; + c_hist_t *ent; + tot = c->n_v3_ns_requests; + if (!tot) + continue; + ent = tor_malloc_zero(sizeof(c_hist_t)); + strlcpy(ent->country, c->countrycode, sizeof(ent->country)); + ent->total = round_to_next_multiple_of(tot, granularity); + smartlist_add(entries, ent); + } SMARTLIST_FOREACH_END(c); + smartlist_sort(entries, c_hist_compare_); + + strings = smartlist_new(); + SMARTLIST_FOREACH(entries, c_hist_t *, ent, { + smartlist_add_asprintf(strings, "%s=%u", ent->country, ent->total); + }); + result = smartlist_join_strings(strings, ",", 0, NULL); + SMARTLIST_FOREACH(strings, char *, cp, tor_free(cp)); + SMARTLIST_FOREACH(entries, c_hist_t *, ent, tor_free(ent)); + smartlist_free(strings); + smartlist_free(entries); + return result; +} + +/** Start time of directory request stats or 0 if we're not collecting + * directory request statistics. */ +static time_t start_of_dirreq_stats_interval; + +/** Initialize directory request stats. */ +void +geoip_dirreq_stats_init(time_t now) +{ + start_of_dirreq_stats_interval = now; +} + +/** Reset counters for dirreq stats. */ +void +geoip_reset_dirreq_stats(time_t now) +{ + SMARTLIST_FOREACH(geoip_countries, geoip_country_t *, c, { + c->n_v3_ns_requests = 0; + }); + { + clientmap_entry_t **ent, **next, *this; + for (ent = HT_START(clientmap, &client_history); ent != NULL; + ent = next) { + if ((*ent)->action == GEOIP_CLIENT_NETWORKSTATUS) { + this = *ent; + next = HT_NEXT_RMV(clientmap, &client_history, ent); + clientmap_entry_free(this); + } else { + next = HT_NEXT(clientmap, &client_history, ent); + } + } + } + memset(ns_v3_responses, 0, sizeof(ns_v3_responses)); + { + dirreq_map_entry_t **ent, **next, *this; + for (ent = HT_START(dirreqmap, &dirreq_map); ent != NULL; ent = next) { + this = *ent; + next = HT_NEXT_RMV(dirreqmap, &dirreq_map, ent); + tor_free(this); + } + } + start_of_dirreq_stats_interval = now; +} + +/** Stop collecting directory request stats in a way that we can re-start + * doing so in geoip_dirreq_stats_init(). */ +void +geoip_dirreq_stats_term(void) +{ + geoip_reset_dirreq_stats(0); +} + +/** Return a newly allocated string containing the dirreq statistics + * until <b>now</b>, or NULL if we're not collecting dirreq stats. Caller + * must ensure start_of_dirreq_stats_interval is in the past. */ +char * +geoip_format_dirreq_stats(time_t now) +{ + char t[ISO_TIME_LEN+1]; + int i; + char *v3_ips_string = NULL, *v3_reqs_string = NULL, + *v3_direct_dl_string = NULL, *v3_tunneled_dl_string = NULL; + char *result = NULL; + + if (!start_of_dirreq_stats_interval) + return NULL; /* Not initialized. */ + + tor_assert(now >= start_of_dirreq_stats_interval); + + format_iso_time(t, now); + geoip_get_client_history(GEOIP_CLIENT_NETWORKSTATUS, &v3_ips_string, NULL); + v3_reqs_string = geoip_get_request_history(); + +#define RESPONSE_GRANULARITY 8 + for (i = 0; i < GEOIP_NS_RESPONSE_NUM; i++) { + ns_v3_responses[i] = round_uint32_to_next_multiple_of( + ns_v3_responses[i], RESPONSE_GRANULARITY); + } +#undef RESPONSE_GRANULARITY + + v3_direct_dl_string = geoip_get_dirreq_history(DIRREQ_DIRECT); + v3_tunneled_dl_string = geoip_get_dirreq_history(DIRREQ_TUNNELED); + + /* Put everything together into a single string. */ + tor_asprintf(&result, "dirreq-stats-end %s (%d s)\n" + "dirreq-v3-ips %s\n" + "dirreq-v3-reqs %s\n" + "dirreq-v3-resp ok=%u,not-enough-sigs=%u,unavailable=%u," + "not-found=%u,not-modified=%u,busy=%u\n" + "dirreq-v3-direct-dl %s\n" + "dirreq-v3-tunneled-dl %s\n", + t, + (unsigned) (now - start_of_dirreq_stats_interval), + v3_ips_string ? v3_ips_string : "", + v3_reqs_string ? v3_reqs_string : "", + ns_v3_responses[GEOIP_SUCCESS], + ns_v3_responses[GEOIP_REJECT_NOT_ENOUGH_SIGS], + ns_v3_responses[GEOIP_REJECT_UNAVAILABLE], + ns_v3_responses[GEOIP_REJECT_NOT_FOUND], + ns_v3_responses[GEOIP_REJECT_NOT_MODIFIED], + ns_v3_responses[GEOIP_REJECT_BUSY], + v3_direct_dl_string ? v3_direct_dl_string : "", + v3_tunneled_dl_string ? v3_tunneled_dl_string : ""); + + /* Free partial strings. */ + tor_free(v3_ips_string); + tor_free(v3_reqs_string); + tor_free(v3_direct_dl_string); + tor_free(v3_tunneled_dl_string); + + return result; +} + +/** If 24 hours have passed since the beginning of the current dirreq + * stats period, write dirreq stats to $DATADIR/stats/dirreq-stats + * (possibly overwriting an existing file) and reset counters. Return + * when we would next want to write dirreq stats or 0 if we never want to + * write. */ +time_t +geoip_dirreq_stats_write(time_t now) +{ + char *str = NULL; + + if (!start_of_dirreq_stats_interval) + return 0; /* Not initialized. */ + if (start_of_dirreq_stats_interval + WRITE_STATS_INTERVAL > now) + goto done; /* Not ready to write. */ + + /* Discard all items in the client history that are too old. */ + geoip_remove_old_clients(start_of_dirreq_stats_interval); + + /* Generate history string .*/ + str = geoip_format_dirreq_stats(now); + if (! str) + goto done; + + /* Write dirreq-stats string to disk. */ + if (!check_or_create_data_subdir("stats")) { + write_to_data_subdir("stats", "dirreq-stats", str, "dirreq statistics"); + /* Reset measurement interval start. */ + geoip_reset_dirreq_stats(now); + } + + done: + tor_free(str); + return start_of_dirreq_stats_interval + WRITE_STATS_INTERVAL; +} + +/** Start time of bridge stats or 0 if we're not collecting bridge + * statistics. */ +static time_t start_of_bridge_stats_interval; + +/** Initialize bridge stats. */ +void +geoip_bridge_stats_init(time_t now) +{ + start_of_bridge_stats_interval = now; +} + +/** Stop collecting bridge stats in a way that we can re-start doing so in + * geoip_bridge_stats_init(). */ +void +geoip_bridge_stats_term(void) +{ + client_history_clear(); + start_of_bridge_stats_interval = 0; +} + +/** Validate a bridge statistics string as it would be written to a + * current extra-info descriptor. Return 1 if the string is valid and + * recent enough, or 0 otherwise. */ +static int +validate_bridge_stats(const char *stats_str, time_t now) +{ + char stats_end_str[ISO_TIME_LEN+1], stats_start_str[ISO_TIME_LEN+1], + *eos; + + const char *BRIDGE_STATS_END = "bridge-stats-end "; + const char *BRIDGE_IPS = "bridge-ips "; + const char *BRIDGE_IPS_EMPTY_LINE = "bridge-ips\n"; + const char *BRIDGE_TRANSPORTS = "bridge-ip-transports "; + const char *BRIDGE_TRANSPORTS_EMPTY_LINE = "bridge-ip-transports\n"; + const char *tmp; + time_t stats_end_time; + int seconds; + tor_assert(stats_str); + + /* Parse timestamp and number of seconds from + "bridge-stats-end YYYY-MM-DD HH:MM:SS (N s)" */ + tmp = find_str_at_start_of_line(stats_str, BRIDGE_STATS_END); + if (!tmp) + return 0; + tmp += strlen(BRIDGE_STATS_END); + + if (strlen(tmp) < ISO_TIME_LEN + 6) + return 0; + strlcpy(stats_end_str, tmp, sizeof(stats_end_str)); + if (parse_iso_time(stats_end_str, &stats_end_time) < 0) + return 0; + if (stats_end_time < now - (25*60*60) || + stats_end_time > now + (1*60*60)) + return 0; + seconds = (int)strtol(tmp + ISO_TIME_LEN + 2, &eos, 10); + if (!eos || seconds < 23*60*60) + return 0; + format_iso_time(stats_start_str, stats_end_time - seconds); + + /* Parse: "bridge-ips CC=N,CC=N,..." */ + tmp = find_str_at_start_of_line(stats_str, BRIDGE_IPS); + if (!tmp) { + /* Look if there is an empty "bridge-ips" line */ + tmp = find_str_at_start_of_line(stats_str, BRIDGE_IPS_EMPTY_LINE); + if (!tmp) + return 0; + } + + /* Parse: "bridge-ip-transports PT=N,PT=N,..." */ + tmp = find_str_at_start_of_line(stats_str, BRIDGE_TRANSPORTS); + if (!tmp) { + /* Look if there is an empty "bridge-ip-transports" line */ + tmp = find_str_at_start_of_line(stats_str, BRIDGE_TRANSPORTS_EMPTY_LINE); + if (!tmp) + return 0; + } + + return 1; +} + +/** Most recent bridge statistics formatted to be written to extra-info + * descriptors. */ +static char *bridge_stats_extrainfo = NULL; + +/** Return a newly allocated string holding our bridge usage stats by country + * in a format suitable for inclusion in an extrainfo document. Return NULL on + * failure. */ +char * +geoip_format_bridge_stats(time_t now) +{ + char *out = NULL; + char *country_data = NULL, *ipver_data = NULL, *transport_data = NULL; + long duration = now - start_of_bridge_stats_interval; + char written[ISO_TIME_LEN+1]; + + if (duration < 0) + return NULL; + if (!start_of_bridge_stats_interval) + return NULL; /* Not initialized. */ + + format_iso_time(written, now); + geoip_get_client_history(GEOIP_CLIENT_CONNECT, &country_data, &ipver_data); + transport_data = geoip_get_transport_history(); + + tor_asprintf(&out, + "bridge-stats-end %s (%ld s)\n" + "bridge-ips %s\n" + "bridge-ip-versions %s\n" + "bridge-ip-transports %s\n", + written, duration, + country_data ? country_data : "", + ipver_data ? ipver_data : "", + transport_data ? transport_data : ""); + tor_free(country_data); + tor_free(ipver_data); + tor_free(transport_data); + + return out; +} + +/** Return a newly allocated string holding our bridge usage stats by country + * in a format suitable for the answer to a controller request. Return NULL on + * failure. */ +static char * +format_bridge_stats_controller(time_t now) +{ + char *out = NULL, *country_data = NULL, *ipver_data = NULL; + char started[ISO_TIME_LEN+1]; + (void) now; + + format_iso_time(started, start_of_bridge_stats_interval); + geoip_get_client_history(GEOIP_CLIENT_CONNECT, &country_data, &ipver_data); + + tor_asprintf(&out, + "TimeStarted=\"%s\" CountrySummary=%s IPVersions=%s", + started, + country_data ? country_data : "", + ipver_data ? ipver_data : ""); + tor_free(country_data); + tor_free(ipver_data); + return out; +} + +/** Return a newly allocated string holding our bridge usage stats by + * country in a format suitable for inclusion in our heartbeat + * message. Return NULL on failure. */ +char * +format_client_stats_heartbeat(time_t now) +{ + const int n_hours = 6; + char *out = NULL; + int n_clients = 0; + clientmap_entry_t **ent; + unsigned cutoff = (unsigned)( (now-n_hours*3600)/60 ); + + if (!start_of_bridge_stats_interval) + return NULL; /* Not initialized. */ + + /* count unique IPs */ + HT_FOREACH(ent, clientmap, &client_history) { + /* only count directly connecting clients */ + if ((*ent)->action != GEOIP_CLIENT_CONNECT) + continue; + if ((*ent)->last_seen_in_minutes < cutoff) + continue; + n_clients++; + } + + tor_asprintf(&out, "Heartbeat: " + "In the last %d hours, I have seen %d unique clients.", + n_hours, + n_clients); + + return out; +} + +/** Write bridge statistics to $DATADIR/stats/bridge-stats and return + * when we should next try to write statistics. */ +time_t +geoip_bridge_stats_write(time_t now) +{ + char *val = NULL; + + /* Check if 24 hours have passed since starting measurements. */ + if (now < start_of_bridge_stats_interval + WRITE_STATS_INTERVAL) + return start_of_bridge_stats_interval + WRITE_STATS_INTERVAL; + + /* Discard all items in the client history that are too old. */ + geoip_remove_old_clients(start_of_bridge_stats_interval); + + /* Generate formatted string */ + val = geoip_format_bridge_stats(now); + if (val == NULL) + goto done; + + /* Update the stored value. */ + tor_free(bridge_stats_extrainfo); + bridge_stats_extrainfo = val; + start_of_bridge_stats_interval = now; + + /* Write it to disk. */ + if (!check_or_create_data_subdir("stats")) { + write_to_data_subdir("stats", "bridge-stats", + bridge_stats_extrainfo, "bridge statistics"); + + /* Tell the controller, "hey, there are clients!" */ + { + char *controller_str = format_bridge_stats_controller(now); + if (controller_str) + control_event_clients_seen(controller_str); + tor_free(controller_str); + } + } + + done: + return start_of_bridge_stats_interval + WRITE_STATS_INTERVAL; +} + +/** Try to load the most recent bridge statistics from disk, unless we + * have finished a measurement interval lately, and check whether they + * are still recent enough. */ +static void +load_bridge_stats(time_t now) +{ + char *fname, *contents; + if (bridge_stats_extrainfo) + return; + + fname = get_datadir_fname2("stats", "bridge-stats"); + contents = read_file_to_str(fname, RFTS_IGNORE_MISSING, NULL); + if (contents && validate_bridge_stats(contents, now)) { + bridge_stats_extrainfo = contents; + } else { + tor_free(contents); + } + + tor_free(fname); +} + +/** Return most recent bridge statistics for inclusion in extra-info + * descriptors, or NULL if we don't have recent bridge statistics. */ +const char * +geoip_get_bridge_stats_extrainfo(time_t now) +{ + load_bridge_stats(now); + return bridge_stats_extrainfo; +} + +/** Return a new string containing the recent bridge statistics to be returned + * to controller clients, or NULL if we don't have any bridge statistics. */ +char * +geoip_get_bridge_stats_controller(time_t now) +{ + return format_bridge_stats_controller(now); +} + +/** Start time of entry stats or 0 if we're not collecting entry + * statistics. */ +static time_t start_of_entry_stats_interval; + +/** Initialize entry stats. */ +void +geoip_entry_stats_init(time_t now) +{ + start_of_entry_stats_interval = now; +} + +/** Reset counters for entry stats. */ +void +geoip_reset_entry_stats(time_t now) +{ + client_history_clear(); + start_of_entry_stats_interval = now; +} + +/** Stop collecting entry stats in a way that we can re-start doing so in + * geoip_entry_stats_init(). */ +void +geoip_entry_stats_term(void) +{ + geoip_reset_entry_stats(0); +} + +/** Return a newly allocated string containing the entry statistics + * until <b>now</b>, or NULL if we're not collecting entry stats. Caller + * must ensure start_of_entry_stats_interval lies in the past. */ +char * +geoip_format_entry_stats(time_t now) +{ + char t[ISO_TIME_LEN+1]; + char *data = NULL; + char *result; + + if (!start_of_entry_stats_interval) + return NULL; /* Not initialized. */ + + tor_assert(now >= start_of_entry_stats_interval); + + geoip_get_client_history(GEOIP_CLIENT_CONNECT, &data, NULL); + format_iso_time(t, now); + tor_asprintf(&result, + "entry-stats-end %s (%u s)\n" + "entry-ips %s\n", + t, (unsigned) (now - start_of_entry_stats_interval), + data ? data : ""); + tor_free(data); + return result; +} + +/** If 24 hours have passed since the beginning of the current entry stats + * period, write entry stats to $DATADIR/stats/entry-stats (possibly + * overwriting an existing file) and reset counters. Return when we would + * next want to write entry stats or 0 if we never want to write. */ +time_t +geoip_entry_stats_write(time_t now) +{ + char *str = NULL; + + if (!start_of_entry_stats_interval) + return 0; /* Not initialized. */ + if (start_of_entry_stats_interval + WRITE_STATS_INTERVAL > now) + goto done; /* Not ready to write. */ + + /* Discard all items in the client history that are too old. */ + geoip_remove_old_clients(start_of_entry_stats_interval); + + /* Generate history string .*/ + str = geoip_format_entry_stats(now); + + /* Write entry-stats string to disk. */ + if (!check_or_create_data_subdir("stats")) { + write_to_data_subdir("stats", "entry-stats", str, "entry statistics"); + + /* Reset measurement interval start. */ + geoip_reset_entry_stats(now); + } + + done: + tor_free(str); + return start_of_entry_stats_interval + WRITE_STATS_INTERVAL; +} + +/** Helper used to implement GETINFO ip-to-country/... controller command. */ +int +getinfo_helper_geoip(control_connection_t *control_conn, + const char *question, char **answer, + const char **errmsg) +{ + (void)control_conn; + if (!strcmpstart(question, "ip-to-country/")) { + int c; + sa_family_t family; + tor_addr_t addr; + question += strlen("ip-to-country/"); + + if (!strcmp(question, "ipv4-available") || + !strcmp(question, "ipv6-available")) { + family = !strcmp(question, "ipv4-available") ? AF_INET : AF_INET6; + const int available = geoip_is_loaded(family); + tor_asprintf(answer, "%d", !! available); + return 0; + } + + family = tor_addr_parse(&addr, question); + if (family != AF_INET && family != AF_INET6) { + *errmsg = "Invalid address family"; + return -1; + } + if (!geoip_is_loaded(family)) { + *errmsg = "GeoIP data not loaded"; + return -1; + } + if (family == AF_INET) + c = geoip_get_country_by_ipv4(tor_addr_to_ipv4h(&addr)); + else /* AF_INET6 */ + c = geoip_get_country_by_ipv6(tor_addr_to_in6(&addr)); + *answer = tor_strdup(geoip_get_country_name(c)); + } + return 0; +} + +/** Release all storage held by the GeoIP databases and country list. */ +STATIC void +clear_geoip_db(void) +{ + if (geoip_countries) { + SMARTLIST_FOREACH(geoip_countries, geoip_country_t *, c, tor_free(c)); + smartlist_free(geoip_countries); + } + + strmap_free(country_idxplus1_by_lc_code, NULL); + if (geoip_ipv4_entries) { + SMARTLIST_FOREACH(geoip_ipv4_entries, geoip_ipv4_entry_t *, ent, + tor_free(ent)); + smartlist_free(geoip_ipv4_entries); + } + if (geoip_ipv6_entries) { + SMARTLIST_FOREACH(geoip_ipv6_entries, geoip_ipv6_entry_t *, ent, + tor_free(ent)); + smartlist_free(geoip_ipv6_entries); + } + geoip_countries = NULL; + country_idxplus1_by_lc_code = NULL; + geoip_ipv4_entries = NULL; + geoip_ipv6_entries = NULL; +} + +/** Release all storage held in this file. */ +void +geoip_free_all(void) +{ + { + clientmap_entry_t **ent, **next, *this; + for (ent = HT_START(clientmap, &client_history); ent != NULL; ent = next) { + this = *ent; + next = HT_NEXT_RMV(clientmap, &client_history, ent); + clientmap_entry_free(this); + } + HT_CLEAR(clientmap, &client_history); + } + { + dirreq_map_entry_t **ent, **next, *this; + for (ent = HT_START(dirreqmap, &dirreq_map); ent != NULL; ent = next) { + this = *ent; + next = HT_NEXT_RMV(dirreqmap, &dirreq_map, ent); + tor_free(this); + } + HT_CLEAR(dirreqmap, &dirreq_map); + } + + clear_geoip_db(); + tor_free(bridge_stats_extrainfo); + + memset(geoip_digest, 0, sizeof(geoip_digest)); + memset(geoip6_digest, 0, sizeof(geoip6_digest)); +} diff --git a/src/feature/stats/geoip.h b/src/feature/stats/geoip.h new file mode 100644 index 0000000000..a2175c01e5 --- /dev/null +++ b/src/feature/stats/geoip.h @@ -0,0 +1,157 @@ +/* Copyright (c) 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file geoip.h + * \brief Header file for geoip.c. + **/ + +#ifndef TOR_GEOIP_H +#define TOR_GEOIP_H + +#include "lib/testsupport/testsupport.h" +#include "core/or/dos.h" + +/** Indicates an action that we might be noting geoip statistics on. + * Note that if we're noticing CONNECT, we're a bridge, and if we're noticing + * the others, we're not. + */ +typedef enum { + /** We've noticed a connection as a bridge relay or entry guard. */ + GEOIP_CLIENT_CONNECT = 0, + /** We've served a networkstatus consensus as a directory server. */ + GEOIP_CLIENT_NETWORKSTATUS = 1, +} geoip_client_action_t; +/** Indicates either a positive reply or a reason for rejectng a network + * status request that will be included in geoip statistics. */ +typedef enum { + /** Request is answered successfully. */ + GEOIP_SUCCESS = 0, + /** V3 network status is not signed by a sufficient number of requested + * authorities. */ + GEOIP_REJECT_NOT_ENOUGH_SIGS = 1, + /** Requested network status object is unavailable. */ + GEOIP_REJECT_UNAVAILABLE = 2, + /** Requested network status not found. */ + GEOIP_REJECT_NOT_FOUND = 3, + /** Network status has not been modified since If-Modified-Since time. */ + GEOIP_REJECT_NOT_MODIFIED = 4, + /** Directory is busy. */ + GEOIP_REJECT_BUSY = 5, +} geoip_ns_response_t; +#define GEOIP_NS_RESPONSE_NUM 6 + +/** Directory requests that we are measuring can be either direct or + * tunneled. */ +typedef enum { + DIRREQ_DIRECT = 0, + DIRREQ_TUNNELED = 1, +} dirreq_type_t; + +/** Possible states for either direct or tunneled directory requests that + * are relevant for determining network status download times. */ +typedef enum { + /** Found that the client requests a network status; applies to both + * direct and tunneled requests; initial state of a request that we are + * measuring. */ + DIRREQ_IS_FOR_NETWORK_STATUS = 0, + /** Finished writing a network status to the directory connection; + * applies to both direct and tunneled requests; completes a direct + * request. */ + DIRREQ_FLUSHING_DIR_CONN_FINISHED = 1, + /** END cell sent to circuit that initiated a tunneled request. */ + DIRREQ_END_CELL_SENT = 2, + /** Flushed last cell from queue of the circuit that initiated a + * tunneled request to the outbuf of the OR connection. */ + DIRREQ_CIRC_QUEUE_FLUSHED = 3, + /** Flushed last byte from buffer of the channel belonging to the + * circuit that initiated a tunneled request; completes a tunneled + * request. */ + DIRREQ_CHANNEL_BUFFER_FLUSHED = 4 +} dirreq_state_t; + +#ifdef GEOIP_PRIVATE +STATIC int geoip_parse_entry(const char *line, sa_family_t family); +STATIC int geoip_get_country_by_ipv4(uint32_t ipaddr); +STATIC int geoip_get_country_by_ipv6(const struct in6_addr *addr); +STATIC void clear_geoip_db(void); +#endif /* defined(GEOIP_PRIVATE) */ + +/** Entry in a map from IP address to the last time we've seen an incoming + * connection from that IP address. Used by bridges only to track which + * countries have them blocked, or the DoS mitigation subsystem if enabled. */ +typedef struct clientmap_entry_t { + HT_ENTRY(clientmap_entry_t) node; + tor_addr_t addr; + /* Name of pluggable transport used by this client. NULL if no + pluggable transport was used. */ + char *transport_name; + + /** Time when we last saw this IP address, in MINUTES since the epoch. + * + * (This will run out of space around 4011 CE. If Tor is still in use around + * 4000 CE, please remember to add more bits to last_seen_in_minutes.) */ + unsigned int last_seen_in_minutes:30; + unsigned int action:2; + + /* This object is used to keep some statistics per client address for the + * DoS mitigation subsystem. */ + dos_client_stats_t dos_stats; +} clientmap_entry_t; + +int should_record_bridge_info(const or_options_t *options); +int geoip_load_file(sa_family_t family, const char *filename); +MOCK_DECL(int, geoip_get_country_by_addr, (const tor_addr_t *addr)); +MOCK_DECL(int, geoip_get_n_countries, (void)); +const char *geoip_get_country_name(country_t num); +MOCK_DECL(int, geoip_is_loaded, (sa_family_t family)); +const char *geoip_db_digest(sa_family_t family); +MOCK_DECL(country_t, geoip_get_country, (const char *countrycode)); + +void geoip_note_client_seen(geoip_client_action_t action, + const tor_addr_t *addr, const char *transport_name, + time_t now); +void geoip_remove_old_clients(time_t cutoff); +clientmap_entry_t *geoip_lookup_client(const tor_addr_t *addr, + const char *transport_name, + geoip_client_action_t action); +size_t geoip_client_cache_total_allocation(void); +size_t geoip_client_cache_handle_oom(time_t now, size_t min_remove_bytes); + +void geoip_note_ns_response(geoip_ns_response_t response); +char *geoip_get_transport_history(void); +int geoip_get_client_history(geoip_client_action_t action, + char **country_str, char **ipver_str); +char *geoip_get_request_history(void); +int getinfo_helper_geoip(control_connection_t *control_conn, + const char *question, char **answer, + const char **errmsg); +void geoip_free_all(void); + +void geoip_start_dirreq(uint64_t dirreq_id, size_t response_size, + dirreq_type_t type); +void geoip_change_dirreq_state(uint64_t dirreq_id, dirreq_type_t type, + dirreq_state_t new_state); + +void geoip_dirreq_stats_init(time_t now); +void geoip_reset_dirreq_stats(time_t now); +char *geoip_format_dirreq_stats(time_t now); +time_t geoip_dirreq_stats_write(time_t now); +void geoip_dirreq_stats_term(void); +void geoip_entry_stats_init(time_t now); +time_t geoip_entry_stats_write(time_t now); +void geoip_entry_stats_term(void); +void geoip_reset_entry_stats(time_t now); +char *geoip_format_entry_stats(time_t now); +void geoip_bridge_stats_init(time_t now); +char *geoip_format_bridge_stats(time_t now); +time_t geoip_bridge_stats_write(time_t now); +void geoip_bridge_stats_term(void); +const char *geoip_get_bridge_stats_extrainfo(time_t); +char *geoip_get_bridge_stats_controller(time_t); +char *format_client_stats_heartbeat(time_t now); + +#endif /* !defined(TOR_GEOIP_H) */ diff --git a/src/feature/stats/rephist.c b/src/feature/stats/rephist.c new file mode 100644 index 0000000000..2588bb4c00 --- /dev/null +++ b/src/feature/stats/rephist.c @@ -0,0 +1,3227 @@ +/* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file rephist.c + * \brief Basic history and performance-tracking functionality. + * + * Basic history and performance-tracking functionality to remember + * which servers have worked in the past, how much bandwidth we've + * been using, which ports we tend to want, and so on; further, + * exit port statistics, cell statistics, and connection statistics. + * + * The history and information tracked in this module could sensibly be + * divided into several categories: + * + * <ul><li>Statistics used by authorities to remember the uptime and + * stability information about various relays, including "uptime", + * "weighted fractional uptime" and "mean time between failures". + * + * <li>Bandwidth usage history, used by relays to self-report how much + * bandwidth they've used for different purposes over last day or so, + * in order to generate the {dirreq-,}{read,write}-history lines in + * that they publish. + * + * <li>Predicted ports, used by clients to remember how long it's been + * since they opened an exit connection to each given target + * port. Clients use this information in order to try to keep circuits + * open to exit nodes that can connect to the ports that they care + * about. (The predicted ports mechanism also handles predicted circuit + * usage that _isn't_ port-specific, such as resolves, internal circuits, + * and so on.) + * + * <li>Public key operation counters, for tracking how many times we've + * done each public key operation. (This is unmaintained and we should + * remove it.) + * + * <li>Exit statistics by port, used by exits to keep track of the + * number of streams and bytes they've served at each exit port, so they + * can generate their exit-kibibytes-{read,written} and + * exit-streams-opened statistics. + * + * <li>Circuit stats, used by relays instances to tract circuit + * queue fullness and delay over time, and generate cell-processed-cells, + * cell-queued-cells, cell-time-in-queue, and cell-circuits-per-decile + * statistics. + * + * <li>Descriptor serving statistics, used by directory caches to track + * how many descriptors they've served. + * + * <li>Connection statistics, used by relays to track one-way and + * bidirectional connections. + * + * <li>Onion handshake statistics, used by relays to count how many + * TAP and ntor handshakes they've handled. + * + * <li>Hidden service statistics, used by relays to count rendezvous + * traffic and HSDir-stored descriptors. + * + * <li>Link protocol statistics, used by relays to count how many times + * each link protocol has been used. + * + * </ul> + * + * The entry points for this module are scattered throughout the + * codebase. Sending data, receiving data, connecting to a relay, + * losing a connection to a relay, and so on can all trigger a change in + * our current stats. Relays also invoke this module in order to + * extract their statistics when building routerinfo and extrainfo + * objects in router.c. + * + * TODO: This module should be broken up. + * + * (The "rephist" name originally stood for "reputation and history". ) + **/ + +#include "core/or/or.h" +#include "core/or/circuitlist.h" +#include "core/or/circuituse.h" +#include "app/config/config.h" +#include "lib/crypt_ops/crypto_rand.h" +#include "feature/nodelist/networkstatus.h" +#include "feature/nodelist/nodelist.h" +#include "feature/stats/rephist.h" +#include "feature/relay/router.h" +#include "feature/nodelist/routerlist.h" +#include "ht.h" +#include "core/or/channelpadding.h" +#include "core/or/connection_or.h" +#include "app/config/statefile.h" + +#include "feature/nodelist/networkstatus_st.h" +#include "core/or/or_circuit_st.h" +#include "app/config/or_state_st.h" + +#include "lib/container/bloomfilt.h" +#include "lib/container/order.h" +#include "lib/math/fp.h" +#include "lib/math/laplace.h" +#include "lib/time/tvdiff.h" + +#ifdef HAVE_FCNTL_H +#include <fcntl.h> +#endif + +static void bw_arrays_init(void); +static void predicted_ports_alloc(void); + +typedef struct bw_array_t bw_array_t; +STATIC uint64_t find_largest_max(bw_array_t *b); +STATIC void commit_max(bw_array_t *b); +STATIC void advance_obs(bw_array_t *b); + +/** Total number of bytes currently allocated in fields used by rephist.c. */ +uint64_t rephist_total_alloc=0; +/** Number of or_history_t objects currently allocated. */ +uint32_t rephist_total_num=0; + +/** If the total weighted run count of all runs for a router ever falls + * below this amount, the router can be treated as having 0 MTBF. */ +#define STABILITY_EPSILON 0.0001 +/** Value by which to discount all old intervals for MTBF purposes. This + * is compounded every STABILITY_INTERVAL. */ +#define STABILITY_ALPHA 0.95 +/** Interval at which to discount all old intervals for MTBF purposes. */ +#define STABILITY_INTERVAL (12*60*60) +/* (This combination of ALPHA, INTERVAL, and EPSILON makes it so that an + * interval that just ended counts twice as much as one that ended a week ago, + * 20X as much as one that ended a month ago, and routers that have had no + * uptime data for about half a year will get forgotten.) */ + +/** History of an OR. */ +typedef struct or_history_t { + /** When did we start tracking this OR? */ + time_t since; + /** When did we most recently note a change to this OR? */ + time_t changed; + + /** The address at which we most recently connected to this OR + * successfully. */ + tor_addr_t last_reached_addr; + + /** The port at which we most recently connected to this OR successfully */ + uint16_t last_reached_port; + + /* === For MTBF tracking: */ + /** Weighted sum total of all times that this router has been online. + */ + unsigned long weighted_run_length; + /** If the router is now online (according to stability-checking rules), + * when did it come online? */ + time_t start_of_run; + /** Sum of weights for runs in weighted_run_length. */ + double total_run_weights; + /* === For fractional uptime tracking: */ + time_t start_of_downtime; + unsigned long weighted_uptime; + unsigned long total_weighted_time; +} or_history_t; + +/** + * This structure holds accounting needed to calculate the padding overhead. + */ +typedef struct padding_counts_t { + /** Total number of cells we have received, including padding */ + uint64_t read_cell_count; + /** Total number of cells we have sent, including padding */ + uint64_t write_cell_count; + /** Total number of CELL_PADDING cells we have received */ + uint64_t read_pad_cell_count; + /** Total number of CELL_PADDING cells we have sent */ + uint64_t write_pad_cell_count; + /** Total number of read cells on padding-enabled conns */ + uint64_t enabled_read_cell_count; + /** Total number of sent cells on padding-enabled conns */ + uint64_t enabled_write_cell_count; + /** Total number of read CELL_PADDING cells on padding-enabled cons */ + uint64_t enabled_read_pad_cell_count; + /** Total number of sent CELL_PADDING cells on padding-enabled cons */ + uint64_t enabled_write_pad_cell_count; + /** Total number of RELAY_DROP cells we have received */ + uint64_t read_drop_cell_count; + /** Total number of RELAY_DROP cells we have sent */ + uint64_t write_drop_cell_count; + /** The maximum number of padding timers we've seen in 24 hours */ + uint64_t maximum_chanpad_timers; + /** When did we first copy padding_current into padding_published? */ + char first_published_at[ISO_TIME_LEN+1]; +} padding_counts_t; + +/** Holds the current values of our padding statistics. + * It is not published until it is transferred to padding_published. */ +static padding_counts_t padding_current; + +/** Remains fixed for a 24 hour period, and then is replaced + * by a redacted copy of padding_current */ +static padding_counts_t padding_published; + +/** When did we last multiply all routers' weighted_run_length and + * total_run_weights by STABILITY_ALPHA? */ +static time_t stability_last_downrated = 0; + +/** */ +static time_t started_tracking_stability = 0; + +/** Map from hex OR identity digest to or_history_t. */ +static digestmap_t *history_map = NULL; + +/** Return the or_history_t for the OR with identity digest <b>id</b>, + * creating it if necessary. */ +static or_history_t * +get_or_history(const char* id) +{ + or_history_t *hist; + + if (tor_digest_is_zero(id)) + return NULL; + + hist = digestmap_get(history_map, id); + if (!hist) { + hist = tor_malloc_zero(sizeof(or_history_t)); + rephist_total_alloc += sizeof(or_history_t); + rephist_total_num++; + hist->since = hist->changed = time(NULL); + tor_addr_make_unspec(&hist->last_reached_addr); + digestmap_set(history_map, id, hist); + } + return hist; +} + +/** Helper: free storage held by a single OR history entry. */ +static void +free_or_history(void *_hist) +{ + or_history_t *hist = _hist; + rephist_total_alloc -= sizeof(or_history_t); + rephist_total_num--; + tor_free(hist); +} + +/** Initialize the static data structures for tracking history. */ +void +rep_hist_init(void) +{ + history_map = digestmap_new(); + bw_arrays_init(); + predicted_ports_alloc(); +} + +/** We have just decided that this router with identity digest <b>id</b> is + * reachable, meaning we will give it a "Running" flag for the next while. */ +void +rep_hist_note_router_reachable(const char *id, const tor_addr_t *at_addr, + const uint16_t at_port, time_t when) +{ + or_history_t *hist = get_or_history(id); + int was_in_run = 1; + char tbuf[ISO_TIME_LEN+1]; + int addr_changed, port_changed; + + tor_assert(hist); + tor_assert((!at_addr && !at_port) || (at_addr && at_port)); + + addr_changed = at_addr && !tor_addr_is_null(&hist->last_reached_addr) && + tor_addr_compare(at_addr, &hist->last_reached_addr, CMP_EXACT) != 0; + port_changed = at_port && hist->last_reached_port && + at_port != hist->last_reached_port; + + if (!started_tracking_stability) + started_tracking_stability = time(NULL); + if (!hist->start_of_run) { + hist->start_of_run = when; + was_in_run = 0; + } + if (hist->start_of_downtime) { + long down_length; + + format_local_iso_time(tbuf, hist->start_of_downtime); + log_info(LD_HIST, "Router %s is now Running; it had been down since %s.", + hex_str(id, DIGEST_LEN), tbuf); + if (was_in_run) + log_info(LD_HIST, " (Paradoxically, it was already Running too.)"); + + down_length = when - hist->start_of_downtime; + hist->total_weighted_time += down_length; + hist->start_of_downtime = 0; + } else if (addr_changed || port_changed) { + /* If we're reachable, but the address changed, treat this as some + * downtime. */ + int penalty = get_options()->TestingTorNetwork ? 240 : 3600; + networkstatus_t *ns; + + if ((ns = networkstatus_get_latest_consensus())) { + int fresh_interval = (int)(ns->fresh_until - ns->valid_after); + int live_interval = (int)(ns->valid_until - ns->valid_after); + /* on average, a descriptor addr change takes .5 intervals to make it + * into a consensus, and half a liveness period to make it to + * clients. */ + penalty = (int)(fresh_interval + live_interval) / 2; + } + format_local_iso_time(tbuf, hist->start_of_run); + log_info(LD_HIST,"Router %s still seems Running, but its address appears " + "to have changed since the last time it was reachable. I'm " + "going to treat it as having been down for %d seconds", + hex_str(id, DIGEST_LEN), penalty); + rep_hist_note_router_unreachable(id, when-penalty); + rep_hist_note_router_reachable(id, NULL, 0, when); + } else { + format_local_iso_time(tbuf, hist->start_of_run); + if (was_in_run) + log_debug(LD_HIST, "Router %s is still Running; it has been Running " + "since %s", hex_str(id, DIGEST_LEN), tbuf); + else + log_info(LD_HIST,"Router %s is now Running; it was previously untracked", + hex_str(id, DIGEST_LEN)); + } + if (at_addr) + tor_addr_copy(&hist->last_reached_addr, at_addr); + if (at_port) + hist->last_reached_port = at_port; +} + +/** We have just decided that this router is unreachable, meaning + * we are taking away its "Running" flag. */ +void +rep_hist_note_router_unreachable(const char *id, time_t when) +{ + or_history_t *hist = get_or_history(id); + char tbuf[ISO_TIME_LEN+1]; + int was_running = 0; + if (!started_tracking_stability) + started_tracking_stability = time(NULL); + + tor_assert(hist); + if (hist->start_of_run) { + /*XXXX We could treat failed connections differently from failed + * connect attempts. */ + long run_length = when - hist->start_of_run; + format_local_iso_time(tbuf, hist->start_of_run); + + hist->total_run_weights += 1.0; + hist->start_of_run = 0; + if (run_length < 0) { + unsigned long penalty = -run_length; +#define SUBTRACT_CLAMPED(var, penalty) \ + do { (var) = (var) < (penalty) ? 0 : (var) - (penalty); } while (0) + + SUBTRACT_CLAMPED(hist->weighted_run_length, penalty); + SUBTRACT_CLAMPED(hist->weighted_uptime, penalty); + } else { + hist->weighted_run_length += run_length; + hist->weighted_uptime += run_length; + hist->total_weighted_time += run_length; + } + was_running = 1; + log_info(LD_HIST, "Router %s is now non-Running: it had previously been " + "Running since %s. Its total weighted uptime is %lu/%lu.", + hex_str(id, DIGEST_LEN), tbuf, hist->weighted_uptime, + hist->total_weighted_time); + } + if (!hist->start_of_downtime) { + hist->start_of_downtime = when; + + if (!was_running) + log_info(LD_HIST, "Router %s is now non-Running; it was previously " + "untracked.", hex_str(id, DIGEST_LEN)); + } else { + if (!was_running) { + format_local_iso_time(tbuf, hist->start_of_downtime); + + log_info(LD_HIST, "Router %s is still non-Running; it has been " + "non-Running since %s.", hex_str(id, DIGEST_LEN), tbuf); + } + } +} + +/** Mark a router with ID <b>id</b> as non-Running, and retroactively declare + * that it has never been running: give it no stability and no WFU. */ +void +rep_hist_make_router_pessimal(const char *id, time_t when) +{ + or_history_t *hist = get_or_history(id); + tor_assert(hist); + + rep_hist_note_router_unreachable(id, when); + + hist->weighted_run_length = 0; + hist->weighted_uptime = 0; +} + +/** Helper: Discount all old MTBF data, if it is time to do so. Return + * the time at which we should next discount MTBF data. */ +time_t +rep_hist_downrate_old_runs(time_t now) +{ + digestmap_iter_t *orhist_it; + const char *digest1; + or_history_t *hist; + void *hist_p; + double alpha = 1.0; + + if (!history_map) + history_map = digestmap_new(); + if (!stability_last_downrated) + stability_last_downrated = now; + if (stability_last_downrated + STABILITY_INTERVAL > now) + return stability_last_downrated + STABILITY_INTERVAL; + + /* Okay, we should downrate the data. By how much? */ + while (stability_last_downrated + STABILITY_INTERVAL < now) { + stability_last_downrated += STABILITY_INTERVAL; + alpha *= STABILITY_ALPHA; + } + + log_info(LD_HIST, "Discounting all old stability info by a factor of %f", + alpha); + + /* Multiply every w_r_l, t_r_w pair by alpha. */ + for (orhist_it = digestmap_iter_init(history_map); + !digestmap_iter_done(orhist_it); + orhist_it = digestmap_iter_next(history_map,orhist_it)) { + digestmap_iter_get(orhist_it, &digest1, &hist_p); + hist = hist_p; + + hist->weighted_run_length = + (unsigned long)(hist->weighted_run_length * alpha); + hist->total_run_weights *= alpha; + + hist->weighted_uptime = (unsigned long)(hist->weighted_uptime * alpha); + hist->total_weighted_time = (unsigned long) + (hist->total_weighted_time * alpha); + } + + return stability_last_downrated + STABILITY_INTERVAL; +} + +/** Helper: Return the weighted MTBF of the router with history <b>hist</b>. */ +static double +get_stability(or_history_t *hist, time_t when) +{ + long total = hist->weighted_run_length; + double total_weights = hist->total_run_weights; + + if (hist->start_of_run) { + /* We're currently in a run. Let total and total_weights hold the values + * they would hold if the current run were to end now. */ + total += (when-hist->start_of_run); + total_weights += 1.0; + } + if (total_weights < STABILITY_EPSILON) { + /* Round down to zero, and avoid divide-by-zero. */ + return 0.0; + } + + return total / total_weights; +} + +/** Return the total amount of time we've been observing, with each run of + * time downrated by the appropriate factor. */ +static long +get_total_weighted_time(or_history_t *hist, time_t when) +{ + long total = hist->total_weighted_time; + if (hist->start_of_run) { + total += (when - hist->start_of_run); + } else if (hist->start_of_downtime) { + total += (when - hist->start_of_downtime); + } + return total; +} + +/** Helper: Return the weighted percent-of-time-online of the router with + * history <b>hist</b>. */ +static double +get_weighted_fractional_uptime(or_history_t *hist, time_t when) +{ + long total = hist->total_weighted_time; + long up = hist->weighted_uptime; + + if (hist->start_of_run) { + long run_length = (when - hist->start_of_run); + up += run_length; + total += run_length; + } else if (hist->start_of_downtime) { + total += (when - hist->start_of_downtime); + } + + if (!total) { + /* Avoid calling anybody's uptime infinity (which should be impossible if + * the code is working), or NaN (which can happen for any router we haven't + * observed up or down yet). */ + return 0.0; + } + + return ((double) up) / total; +} + +/** Return how long the router whose identity digest is <b>id</b> has + * been reachable. Return 0 if the router is unknown or currently deemed + * unreachable. */ +long +rep_hist_get_uptime(const char *id, time_t when) +{ + or_history_t *hist = get_or_history(id); + if (!hist) + return 0; + if (!hist->start_of_run || when < hist->start_of_run) + return 0; + return when - hist->start_of_run; +} + +/** Return an estimated MTBF for the router whose identity digest is + * <b>id</b>. Return 0 if the router is unknown. */ +double +rep_hist_get_stability(const char *id, time_t when) +{ + or_history_t *hist = get_or_history(id); + if (!hist) + return 0.0; + + return get_stability(hist, when); +} + +/** Return an estimated percent-of-time-online for the router whose identity + * digest is <b>id</b>. Return 0 if the router is unknown. */ +double +rep_hist_get_weighted_fractional_uptime(const char *id, time_t when) +{ + or_history_t *hist = get_or_history(id); + if (!hist) + return 0.0; + + return get_weighted_fractional_uptime(hist, when); +} + +/** Return a number representing how long we've known about the router whose + * digest is <b>id</b>. Return 0 if the router is unknown. + * + * Be careful: this measure increases monotonically as we know the router for + * longer and longer, but it doesn't increase linearly. + */ +long +rep_hist_get_weighted_time_known(const char *id, time_t when) +{ + or_history_t *hist = get_or_history(id); + if (!hist) + return 0; + + return get_total_weighted_time(hist, when); +} + +/** Return true if we've been measuring MTBFs for long enough to + * pronounce on Stability. */ +int +rep_hist_have_measured_enough_stability(void) +{ + /* XXXX++ This doesn't do so well when we change our opinion + * as to whether we're tracking router stability. */ + return started_tracking_stability < time(NULL) - 4*60*60; +} + +/** Log all the reliability data we have remembered, with the chosen + * severity. + */ +void +rep_hist_dump_stats(time_t now, int severity) +{ + digestmap_iter_t *orhist_it; + const char *name1, *digest1; + char hexdigest1[HEX_DIGEST_LEN+1]; + or_history_t *or_history; + void *or_history_p; + const node_t *node; + + rep_history_clean(now - get_options()->RephistTrackTime); + + tor_log(severity, LD_HIST, "--------------- Dumping history information:"); + + for (orhist_it = digestmap_iter_init(history_map); + !digestmap_iter_done(orhist_it); + orhist_it = digestmap_iter_next(history_map,orhist_it)) { + double s; + long stability; + digestmap_iter_get(orhist_it, &digest1, &or_history_p); + or_history = (or_history_t*) or_history_p; + + if ((node = node_get_by_id(digest1)) && node_get_nickname(node)) + name1 = node_get_nickname(node); + else + name1 = "(unknown)"; + base16_encode(hexdigest1, sizeof(hexdigest1), digest1, DIGEST_LEN); + s = get_stability(or_history, now); + stability = (long)s; + tor_log(severity, LD_HIST, + "OR %s [%s]: wmtbf %lu:%02lu:%02lu", + name1, hexdigest1, + stability/3600, (stability/60)%60, stability%60); + } +} + +/** Remove history info for routers/links that haven't changed since + * <b>before</b>. + */ +void +rep_history_clean(time_t before) +{ + int authority = authdir_mode(get_options()); + or_history_t *or_history; + void *or_history_p; + digestmap_iter_t *orhist_it; + const char *d1; + + orhist_it = digestmap_iter_init(history_map); + while (!digestmap_iter_done(orhist_it)) { + int should_remove; + digestmap_iter_get(orhist_it, &d1, &or_history_p); + or_history = or_history_p; + + should_remove = authority ? + (or_history->total_run_weights < STABILITY_EPSILON && + !or_history->start_of_run) + : (or_history->changed < before); + if (should_remove) { + orhist_it = digestmap_iter_next_rmv(history_map, orhist_it); + free_or_history(or_history); + continue; + } + orhist_it = digestmap_iter_next(history_map, orhist_it); + } +} + +/** Write MTBF data to disk. Return 0 on success, negative on failure. + * + * If <b>missing_means_down</b>, then if we're about to write an entry + * that is still considered up but isn't in our routerlist, consider it + * to be down. */ +int +rep_hist_record_mtbf_data(time_t now, int missing_means_down) +{ + char time_buf[ISO_TIME_LEN+1]; + + digestmap_iter_t *orhist_it; + const char *digest; + void *or_history_p; + or_history_t *hist; + open_file_t *open_file = NULL; + FILE *f; + + { + char *filename = get_datadir_fname("router-stability"); + f = start_writing_to_stdio_file(filename, OPEN_FLAGS_REPLACE|O_TEXT, 0600, + &open_file); + tor_free(filename); + if (!f) + return -1; + } + + /* File format is: + * FormatLine *KeywordLine Data + * + * FormatLine = "format 1" NL + * KeywordLine = Keyword SP Arguments NL + * Data = "data" NL *RouterMTBFLine "." NL + * RouterMTBFLine = Fingerprint SP WeightedRunLen SP + * TotalRunWeights [SP S=StartRunTime] NL + */ +#define PUT(s) STMT_BEGIN if (fputs((s),f)<0) goto err; STMT_END +#define PRINTF(args) STMT_BEGIN if (fprintf args <0) goto err; STMT_END + + PUT("format 2\n"); + + format_iso_time(time_buf, time(NULL)); + PRINTF((f, "stored-at %s\n", time_buf)); + + if (started_tracking_stability) { + format_iso_time(time_buf, started_tracking_stability); + PRINTF((f, "tracked-since %s\n", time_buf)); + } + if (stability_last_downrated) { + format_iso_time(time_buf, stability_last_downrated); + PRINTF((f, "last-downrated %s\n", time_buf)); + } + + PUT("data\n"); + + /* XXX Nick: now bridge auths record this for all routers too. + * Should we make them record it only for bridge routers? -RD + * Not for 0.2.0. -NM */ + for (orhist_it = digestmap_iter_init(history_map); + !digestmap_iter_done(orhist_it); + orhist_it = digestmap_iter_next(history_map,orhist_it)) { + char dbuf[HEX_DIGEST_LEN+1]; + const char *t = NULL; + digestmap_iter_get(orhist_it, &digest, &or_history_p); + hist = (or_history_t*) or_history_p; + + base16_encode(dbuf, sizeof(dbuf), digest, DIGEST_LEN); + + if (missing_means_down && hist->start_of_run && + !connection_or_digest_is_known_relay(digest)) { + /* We think this relay is running, but it's not listed in our + * consensus. Somehow it fell out without telling us it went + * down. Complain and also correct it. */ + log_info(LD_HIST, + "Relay '%s' is listed as up in rephist, but it's not in " + "our routerlist. Correcting.", dbuf); + rep_hist_note_router_unreachable(digest, now); + } + + PRINTF((f, "R %s\n", dbuf)); + if (hist->start_of_run > 0) { + format_iso_time(time_buf, hist->start_of_run); + t = time_buf; + } + PRINTF((f, "+MTBF %lu %.5f%s%s\n", + hist->weighted_run_length, hist->total_run_weights, + t ? " S=" : "", t ? t : "")); + t = NULL; + if (hist->start_of_downtime > 0) { + format_iso_time(time_buf, hist->start_of_downtime); + t = time_buf; + } + PRINTF((f, "+WFU %lu %lu%s%s\n", + hist->weighted_uptime, hist->total_weighted_time, + t ? " S=" : "", t ? t : "")); + } + + PUT(".\n"); + +#undef PUT +#undef PRINTF + + return finish_writing_to_file(open_file); + err: + abort_writing_to_file(open_file); + return -1; +} + +/** Helper: return the first j >= i such that !strcmpstart(sl[j], prefix) and + * such that no line sl[k] with i <= k < j starts with "R ". Return -1 if no + * such line exists. */ +static int +find_next_with(smartlist_t *sl, int i, const char *prefix) +{ + for ( ; i < smartlist_len(sl); ++i) { + const char *line = smartlist_get(sl, i); + if (!strcmpstart(line, prefix)) + return i; + if (!strcmpstart(line, "R ")) + return -1; + } + return -1; +} + +/** How many bad times has parse_possibly_bad_iso_time() parsed? */ +static int n_bogus_times = 0; +/** Parse the ISO-formatted time in <b>s</b> into *<b>time_out</b>, but + * round any pre-1970 date to Jan 1, 1970. */ +static int +parse_possibly_bad_iso_time(const char *s, time_t *time_out) +{ + int year; + char b[5]; + strlcpy(b, s, sizeof(b)); + b[4] = '\0'; + year = (int)tor_parse_long(b, 10, 0, INT_MAX, NULL, NULL); + if (year < 1970) { + *time_out = 0; + ++n_bogus_times; + return 0; + } else + return parse_iso_time(s, time_out); +} + +/** We've read a time <b>t</b> from a file stored at <b>stored_at</b>, which + * says we started measuring at <b>started_measuring</b>. Return a new number + * that's about as much before <b>now</b> as <b>t</b> was before + * <b>stored_at</b>. + */ +static inline time_t +correct_time(time_t t, time_t now, time_t stored_at, time_t started_measuring) +{ + if (t < started_measuring - 24*60*60*365) + return 0; + else if (t < started_measuring) + return started_measuring; + else if (t > stored_at) + return 0; + else { + long run_length = stored_at - t; + t = (time_t)(now - run_length); + if (t < started_measuring) + t = started_measuring; + return t; + } +} + +/** Load MTBF data from disk. Returns 0 on success or recoverable error, -1 + * on failure. */ +int +rep_hist_load_mtbf_data(time_t now) +{ + /* XXXX won't handle being called while history is already populated. */ + smartlist_t *lines; + const char *line = NULL; + int r=0, i; + time_t last_downrated = 0, stored_at = 0, tracked_since = 0; + time_t latest_possible_start = now; + long format = -1; + + { + char *filename = get_datadir_fname("router-stability"); + char *d = read_file_to_str(filename, RFTS_IGNORE_MISSING, NULL); + tor_free(filename); + if (!d) + return -1; + lines = smartlist_new(); + smartlist_split_string(lines, d, "\n", SPLIT_SKIP_SPACE, 0); + tor_free(d); + } + + { + const char *firstline; + if (smartlist_len(lines)>4) { + firstline = smartlist_get(lines, 0); + if (!strcmpstart(firstline, "format ")) + format = tor_parse_long(firstline+strlen("format "), + 10, -1, LONG_MAX, NULL, NULL); + } + } + if (format != 1 && format != 2) { + log_warn(LD_HIST, + "Unrecognized format in mtbf history file. Skipping."); + goto err; + } + for (i = 1; i < smartlist_len(lines); ++i) { + line = smartlist_get(lines, i); + if (!strcmp(line, "data")) + break; + if (!strcmpstart(line, "last-downrated ")) { + if (parse_iso_time(line+strlen("last-downrated "), &last_downrated)<0) + log_warn(LD_HIST,"Couldn't parse downrate time in mtbf " + "history file."); + } + if (!strcmpstart(line, "stored-at ")) { + if (parse_iso_time(line+strlen("stored-at "), &stored_at)<0) + log_warn(LD_HIST,"Couldn't parse stored time in mtbf " + "history file."); + } + if (!strcmpstart(line, "tracked-since ")) { + if (parse_iso_time(line+strlen("tracked-since "), &tracked_since)<0) + log_warn(LD_HIST,"Couldn't parse started-tracking time in mtbf " + "history file."); + } + } + if (last_downrated > now) + last_downrated = now; + if (tracked_since > now) + tracked_since = now; + + if (!stored_at) { + log_warn(LD_HIST, "No stored time recorded."); + goto err; + } + + if (line && !strcmp(line, "data")) + ++i; + + n_bogus_times = 0; + + for (; i < smartlist_len(lines); ++i) { + char digest[DIGEST_LEN]; + char hexbuf[HEX_DIGEST_LEN+1]; + char mtbf_timebuf[ISO_TIME_LEN+1]; + char wfu_timebuf[ISO_TIME_LEN+1]; + time_t start_of_run = 0; + time_t start_of_downtime = 0; + int have_mtbf = 0, have_wfu = 0; + long wrl = 0; + double trw = 0; + long wt_uptime = 0, total_wt_time = 0; + int n; + or_history_t *hist; + line = smartlist_get(lines, i); + if (!strcmp(line, ".")) + break; + + mtbf_timebuf[0] = '\0'; + wfu_timebuf[0] = '\0'; + + if (format == 1) { + n = tor_sscanf(line, "%40s %ld %lf S=%10s %8s", + hexbuf, &wrl, &trw, mtbf_timebuf, mtbf_timebuf+11); + if (n != 3 && n != 5) { + log_warn(LD_HIST, "Couldn't scan line %s", escaped(line)); + continue; + } + have_mtbf = 1; + } else { + // format == 2. + int mtbf_idx, wfu_idx; + if (strcmpstart(line, "R ") || strlen(line) < 2+HEX_DIGEST_LEN) + continue; + strlcpy(hexbuf, line+2, sizeof(hexbuf)); + mtbf_idx = find_next_with(lines, i+1, "+MTBF "); + wfu_idx = find_next_with(lines, i+1, "+WFU "); + if (mtbf_idx >= 0) { + const char *mtbfline = smartlist_get(lines, mtbf_idx); + n = tor_sscanf(mtbfline, "+MTBF %lu %lf S=%10s %8s", + &wrl, &trw, mtbf_timebuf, mtbf_timebuf+11); + if (n == 2 || n == 4) { + have_mtbf = 1; + } else { + log_warn(LD_HIST, "Couldn't scan +MTBF line %s", + escaped(mtbfline)); + } + } + if (wfu_idx >= 0) { + const char *wfuline = smartlist_get(lines, wfu_idx); + n = tor_sscanf(wfuline, "+WFU %lu %lu S=%10s %8s", + &wt_uptime, &total_wt_time, + wfu_timebuf, wfu_timebuf+11); + if (n == 2 || n == 4) { + have_wfu = 1; + } else { + log_warn(LD_HIST, "Couldn't scan +WFU line %s", escaped(wfuline)); + } + } + if (wfu_idx > i) + i = wfu_idx; + if (mtbf_idx > i) + i = mtbf_idx; + } + if (base16_decode(digest, DIGEST_LEN, + hexbuf, HEX_DIGEST_LEN) != DIGEST_LEN) { + log_warn(LD_HIST, "Couldn't hex string %s", escaped(hexbuf)); + continue; + } + hist = get_or_history(digest); + if (!hist) + continue; + + if (have_mtbf) { + if (mtbf_timebuf[0]) { + mtbf_timebuf[10] = ' '; + if (parse_possibly_bad_iso_time(mtbf_timebuf, &start_of_run)<0) + log_warn(LD_HIST, "Couldn't parse time %s", + escaped(mtbf_timebuf)); + } + hist->start_of_run = correct_time(start_of_run, now, stored_at, + tracked_since); + if (hist->start_of_run < latest_possible_start + wrl) + latest_possible_start = (time_t)(hist->start_of_run - wrl); + + hist->weighted_run_length = wrl; + hist->total_run_weights = trw; + } + if (have_wfu) { + if (wfu_timebuf[0]) { + wfu_timebuf[10] = ' '; + if (parse_possibly_bad_iso_time(wfu_timebuf, &start_of_downtime)<0) + log_warn(LD_HIST, "Couldn't parse time %s", escaped(wfu_timebuf)); + } + } + hist->start_of_downtime = correct_time(start_of_downtime, now, stored_at, + tracked_since); + hist->weighted_uptime = wt_uptime; + hist->total_weighted_time = total_wt_time; + } + if (strcmp(line, ".")) + log_warn(LD_HIST, "Truncated MTBF file."); + + if (tracked_since < 86400*365) /* Recover from insanely early value. */ + tracked_since = latest_possible_start; + + stability_last_downrated = last_downrated; + started_tracking_stability = tracked_since; + + goto done; + err: + r = -1; + done: + SMARTLIST_FOREACH(lines, char *, cp, tor_free(cp)); + smartlist_free(lines); + return r; +} + +/** For how many seconds do we keep track of individual per-second bandwidth + * totals? */ +#define NUM_SECS_ROLLING_MEASURE 10 +/** How large are the intervals for which we track and report bandwidth use? */ +#define NUM_SECS_BW_SUM_INTERVAL (24*60*60) +/** How far in the past do we remember and publish bandwidth use? */ +#define NUM_SECS_BW_SUM_IS_VALID (5*24*60*60) +/** How many bandwidth usage intervals do we remember? (derived) */ +#define NUM_TOTALS (NUM_SECS_BW_SUM_IS_VALID/NUM_SECS_BW_SUM_INTERVAL) + +/** Structure to track bandwidth use, and remember the maxima for a given + * time period. + */ +struct bw_array_t { + /** Observation array: Total number of bytes transferred in each of the last + * NUM_SECS_ROLLING_MEASURE seconds. This is used as a circular array. */ + uint64_t obs[NUM_SECS_ROLLING_MEASURE]; + int cur_obs_idx; /**< Current position in obs. */ + time_t cur_obs_time; /**< Time represented in obs[cur_obs_idx] */ + uint64_t total_obs; /**< Total for all members of obs except + * obs[cur_obs_idx] */ + uint64_t max_total; /**< Largest value that total_obs has taken on in the + * current period. */ + uint64_t total_in_period; /**< Total bytes transferred in the current + * period. */ + + /** When does the next period begin? */ + time_t next_period; + /** Where in 'maxima' should the maximum bandwidth usage for the current + * period be stored? */ + int next_max_idx; + /** How many values in maxima/totals have been set ever? */ + int num_maxes_set; + /** Circular array of the maximum + * bandwidth-per-NUM_SECS_ROLLING_MEASURE usage for the last + * NUM_TOTALS periods */ + uint64_t maxima[NUM_TOTALS]; + /** Circular array of the total bandwidth usage for the last NUM_TOTALS + * periods */ + uint64_t totals[NUM_TOTALS]; +}; + +/** Shift the current period of b forward by one. */ +STATIC void +commit_max(bw_array_t *b) +{ + /* Store total from current period. */ + b->totals[b->next_max_idx] = b->total_in_period; + /* Store maximum from current period. */ + b->maxima[b->next_max_idx++] = b->max_total; + /* Advance next_period and next_max_idx */ + b->next_period += NUM_SECS_BW_SUM_INTERVAL; + if (b->next_max_idx == NUM_TOTALS) + b->next_max_idx = 0; + if (b->num_maxes_set < NUM_TOTALS) + ++b->num_maxes_set; + /* Reset max_total. */ + b->max_total = 0; + /* Reset total_in_period. */ + b->total_in_period = 0; +} + +/** Shift the current observation time of <b>b</b> forward by one second. */ +STATIC void +advance_obs(bw_array_t *b) +{ + int nextidx; + uint64_t total; + + /* Calculate the total bandwidth for the last NUM_SECS_ROLLING_MEASURE + * seconds; adjust max_total as needed.*/ + total = b->total_obs + b->obs[b->cur_obs_idx]; + if (total > b->max_total) + b->max_total = total; + + nextidx = b->cur_obs_idx+1; + if (nextidx == NUM_SECS_ROLLING_MEASURE) + nextidx = 0; + + b->total_obs = total - b->obs[nextidx]; + b->obs[nextidx]=0; + b->cur_obs_idx = nextidx; + + if (++b->cur_obs_time >= b->next_period) + commit_max(b); +} + +/** Add <b>n</b> bytes to the number of bytes in <b>b</b> for second + * <b>when</b>. */ +static inline void +add_obs(bw_array_t *b, time_t when, uint64_t n) +{ + if (when < b->cur_obs_time) + return; /* Don't record data in the past. */ + + /* If we're currently adding observations for an earlier second than + * 'when', advance b->cur_obs_time and b->cur_obs_idx by an + * appropriate number of seconds, and do all the other housekeeping. */ + while (when > b->cur_obs_time) { + /* Doing this one second at a time is potentially inefficient, if we start + with a state file that is very old. Fortunately, it doesn't seem to + show up in profiles, so we can just ignore it for now. */ + advance_obs(b); + } + + b->obs[b->cur_obs_idx] += n; + b->total_in_period += n; +} + +/** Allocate, initialize, and return a new bw_array. */ +static bw_array_t * +bw_array_new(void) +{ + bw_array_t *b; + time_t start; + b = tor_malloc_zero(sizeof(bw_array_t)); + rephist_total_alloc += sizeof(bw_array_t); + start = time(NULL); + b->cur_obs_time = start; + b->next_period = start + NUM_SECS_BW_SUM_INTERVAL; + return b; +} + +#define bw_array_free(val) \ + FREE_AND_NULL(bw_array_t, bw_array_free_, (val)) + +/** Free storage held by bandwidth array <b>b</b>. */ +static void +bw_array_free_(bw_array_t *b) +{ + if (!b) { + return; + } + + rephist_total_alloc -= sizeof(bw_array_t); + tor_free(b); +} + +/** Recent history of bandwidth observations for read operations. */ +static bw_array_t *read_array = NULL; +/** Recent history of bandwidth observations for write operations. */ +STATIC bw_array_t *write_array = NULL; +/** Recent history of bandwidth observations for read operations for the + directory protocol. */ +static bw_array_t *dir_read_array = NULL; +/** Recent history of bandwidth observations for write operations for the + directory protocol. */ +static bw_array_t *dir_write_array = NULL; + +/** Set up [dir-]read_array and [dir-]write_array, freeing them if they + * already exist. */ +static void +bw_arrays_init(void) +{ + bw_array_free(read_array); + bw_array_free(write_array); + bw_array_free(dir_read_array); + bw_array_free(dir_write_array); + + read_array = bw_array_new(); + write_array = bw_array_new(); + dir_read_array = bw_array_new(); + dir_write_array = bw_array_new(); +} + +/** Remember that we read <b>num_bytes</b> bytes in second <b>when</b>. + * + * Add num_bytes to the current running total for <b>when</b>. + * + * <b>when</b> can go back to time, but it's safe to ignore calls + * earlier than the latest <b>when</b> you've heard of. + */ +void +rep_hist_note_bytes_written(uint64_t num_bytes, time_t when) +{ +/* Maybe a circular array for recent seconds, and step to a new point + * every time a new second shows up. Or simpler is to just to have + * a normal array and push down each item every second; it's short. + */ +/* When a new second has rolled over, compute the sum of the bytes we've + * seen over when-1 to when-1-NUM_SECS_ROLLING_MEASURE, and stick it + * somewhere. See rep_hist_bandwidth_assess() below. + */ + add_obs(write_array, when, num_bytes); +} + +/** Remember that we wrote <b>num_bytes</b> bytes in second <b>when</b>. + * (like rep_hist_note_bytes_written() above) + */ +void +rep_hist_note_bytes_read(uint64_t num_bytes, time_t when) +{ +/* if we're smart, we can make this func and the one above share code */ + add_obs(read_array, when, num_bytes); +} + +/** Remember that we wrote <b>num_bytes</b> directory bytes in second + * <b>when</b>. (like rep_hist_note_bytes_written() above) + */ +void +rep_hist_note_dir_bytes_written(uint64_t num_bytes, time_t when) +{ + add_obs(dir_write_array, when, num_bytes); +} + +/** Remember that we read <b>num_bytes</b> directory bytes in second + * <b>when</b>. (like rep_hist_note_bytes_written() above) + */ +void +rep_hist_note_dir_bytes_read(uint64_t num_bytes, time_t when) +{ + add_obs(dir_read_array, when, num_bytes); +} + +/** Helper: Return the largest value in b->maxima. (This is equal to the + * most bandwidth used in any NUM_SECS_ROLLING_MEASURE period for the last + * NUM_SECS_BW_SUM_IS_VALID seconds.) + */ +STATIC uint64_t +find_largest_max(bw_array_t *b) +{ + int i; + uint64_t max; + max=0; + for (i=0; i<NUM_TOTALS; ++i) { + if (b->maxima[i]>max) + max = b->maxima[i]; + } + return max; +} + +/** Find the largest sums in the past NUM_SECS_BW_SUM_IS_VALID (roughly) + * seconds. Find one sum for reading and one for writing. They don't have + * to be at the same time. + * + * Return the smaller of these sums, divided by NUM_SECS_ROLLING_MEASURE. + */ +MOCK_IMPL(int, +rep_hist_bandwidth_assess,(void)) +{ + uint64_t w,r; + r = find_largest_max(read_array); + w = find_largest_max(write_array); + if (r>w) + return (int)(((double)w)/NUM_SECS_ROLLING_MEASURE); + else + return (int)(((double)r)/NUM_SECS_ROLLING_MEASURE); +} + +/** Print the bandwidth history of b (either [dir-]read_array or + * [dir-]write_array) into the buffer pointed to by buf. The format is + * simply comma separated numbers, from oldest to newest. + * + * It returns the number of bytes written. + */ +static size_t +rep_hist_fill_bandwidth_history(char *buf, size_t len, const bw_array_t *b) +{ + char *cp = buf; + int i, n; + const or_options_t *options = get_options(); + uint64_t cutoff; + + if (b->num_maxes_set <= b->next_max_idx) { + /* We haven't been through the circular array yet; time starts at i=0.*/ + i = 0; + } else { + /* We've been around the array at least once. The next i to be + overwritten is the oldest. */ + i = b->next_max_idx; + } + + if (options->RelayBandwidthRate) { + /* We don't want to report that we used more bandwidth than the max we're + * willing to relay; otherwise everybody will know how much traffic + * we used ourself. */ + cutoff = options->RelayBandwidthRate * NUM_SECS_BW_SUM_INTERVAL; + } else { + cutoff = UINT64_MAX; + } + + for (n=0; n<b->num_maxes_set; ++n,++i) { + uint64_t total; + if (i >= NUM_TOTALS) + i -= NUM_TOTALS; + tor_assert(i < NUM_TOTALS); + /* Round the bandwidth used down to the nearest 1k. */ + total = b->totals[i] & ~0x3ff; + if (total > cutoff) + total = cutoff; + + if (n==(b->num_maxes_set-1)) + tor_snprintf(cp, len-(cp-buf), "%"PRIu64, (total)); + else + tor_snprintf(cp, len-(cp-buf), "%"PRIu64",", (total)); + cp += strlen(cp); + } + return cp-buf; +} + +/** Allocate and return lines for representing this server's bandwidth + * history in its descriptor. We publish these lines in our extra-info + * descriptor. + */ +char * +rep_hist_get_bandwidth_lines(void) +{ + char *buf, *cp; + char t[ISO_TIME_LEN+1]; + int r; + bw_array_t *b = NULL; + const char *desc = NULL; + size_t len; + + /* [dirreq-](read|write)-history yyyy-mm-dd HH:MM:SS (n s) n,n,n... */ +/* The n,n,n part above. Largest representation of a uint64_t is 20 chars + * long, plus the comma. */ +#define MAX_HIST_VALUE_LEN (21*NUM_TOTALS) + len = (67+MAX_HIST_VALUE_LEN)*4; + buf = tor_malloc_zero(len); + cp = buf; + for (r=0;r<4;++r) { + char tmp[MAX_HIST_VALUE_LEN]; + size_t slen; + switch (r) { + case 0: + b = write_array; + desc = "write-history"; + break; + case 1: + b = read_array; + desc = "read-history"; + break; + case 2: + b = dir_write_array; + desc = "dirreq-write-history"; + break; + case 3: + b = dir_read_array; + desc = "dirreq-read-history"; + break; + } + tor_assert(b); + slen = rep_hist_fill_bandwidth_history(tmp, MAX_HIST_VALUE_LEN, b); + /* If we don't have anything to write, skip to the next entry. */ + if (slen == 0) + continue; + format_iso_time(t, b->next_period-NUM_SECS_BW_SUM_INTERVAL); + tor_snprintf(cp, len-(cp-buf), "%s %s (%d s) ", + desc, t, NUM_SECS_BW_SUM_INTERVAL); + cp += strlen(cp); + strlcat(cp, tmp, len-(cp-buf)); + cp += slen; + strlcat(cp, "\n", len-(cp-buf)); + ++cp; + } + return buf; +} + +/** Write a single bw_array_t into the Values, Ends, Interval, and Maximum + * entries of an or_state_t. Done before writing out a new state file. */ +static void +rep_hist_update_bwhist_state_section(or_state_t *state, + const bw_array_t *b, + smartlist_t **s_values, + smartlist_t **s_maxima, + time_t *s_begins, + int *s_interval) +{ + int i,j; + uint64_t maxval; + + if (*s_values) { + SMARTLIST_FOREACH(*s_values, char *, val, tor_free(val)); + smartlist_free(*s_values); + } + if (*s_maxima) { + SMARTLIST_FOREACH(*s_maxima, char *, val, tor_free(val)); + smartlist_free(*s_maxima); + } + if (! server_mode(get_options())) { + /* Clients don't need to store bandwidth history persistently; + * force these values to the defaults. */ + /* FFFF we should pull the default out of config.c's state table, + * so we don't have two defaults. */ + if (*s_begins != 0 || *s_interval != 900) { + time_t now = time(NULL); + time_t save_at = get_options()->AvoidDiskWrites ? now+3600 : now+600; + or_state_mark_dirty(state, save_at); + } + *s_begins = 0; + *s_interval = 900; + *s_values = smartlist_new(); + *s_maxima = smartlist_new(); + return; + } + *s_begins = b->next_period; + *s_interval = NUM_SECS_BW_SUM_INTERVAL; + + *s_values = smartlist_new(); + *s_maxima = smartlist_new(); + /* Set i to first position in circular array */ + i = (b->num_maxes_set <= b->next_max_idx) ? 0 : b->next_max_idx; + for (j=0; j < b->num_maxes_set; ++j,++i) { + if (i >= NUM_TOTALS) + i = 0; + smartlist_add_asprintf(*s_values, "%"PRIu64, + (b->totals[i] & ~0x3ff)); + maxval = b->maxima[i] / NUM_SECS_ROLLING_MEASURE; + smartlist_add_asprintf(*s_maxima, "%"PRIu64, + (maxval & ~0x3ff)); + } + smartlist_add_asprintf(*s_values, "%"PRIu64, + (b->total_in_period & ~0x3ff)); + maxval = b->max_total / NUM_SECS_ROLLING_MEASURE; + smartlist_add_asprintf(*s_maxima, "%"PRIu64, + (maxval & ~0x3ff)); +} + +/** Update <b>state</b> with the newest bandwidth history. Done before + * writing out a new state file. */ +void +rep_hist_update_state(or_state_t *state) +{ +#define UPDATE(arrname,st) \ + rep_hist_update_bwhist_state_section(state,\ + (arrname),\ + &state->BWHistory ## st ## Values, \ + &state->BWHistory ## st ## Maxima, \ + &state->BWHistory ## st ## Ends, \ + &state->BWHistory ## st ## Interval) + + UPDATE(write_array, Write); + UPDATE(read_array, Read); + UPDATE(dir_write_array, DirWrite); + UPDATE(dir_read_array, DirRead); + + if (server_mode(get_options())) { + or_state_mark_dirty(state, time(NULL)+(2*3600)); + } +#undef UPDATE +} + +/** Load a single bw_array_t from its Values, Ends, Maxima, and Interval + * entries in an or_state_t. Done while reading the state file. */ +static int +rep_hist_load_bwhist_state_section(bw_array_t *b, + const smartlist_t *s_values, + const smartlist_t *s_maxima, + const time_t s_begins, + const int s_interval) +{ + time_t now = time(NULL); + int retval = 0; + time_t start; + + uint64_t v, mv; + int i,ok,ok_m = 0; + int have_maxima = s_maxima && s_values && + (smartlist_len(s_values) == smartlist_len(s_maxima)); + + if (s_values && s_begins >= now - NUM_SECS_BW_SUM_INTERVAL*NUM_TOTALS) { + start = s_begins - s_interval*(smartlist_len(s_values)); + if (start > now) + return 0; + b->cur_obs_time = start; + b->next_period = start + NUM_SECS_BW_SUM_INTERVAL; + SMARTLIST_FOREACH_BEGIN(s_values, const char *, cp) { + const char *maxstr = NULL; + v = tor_parse_uint64(cp, 10, 0, UINT64_MAX, &ok, NULL); + if (have_maxima) { + maxstr = smartlist_get(s_maxima, cp_sl_idx); + mv = tor_parse_uint64(maxstr, 10, 0, UINT64_MAX, &ok_m, NULL); + mv *= NUM_SECS_ROLLING_MEASURE; + } else { + /* No maxima known; guess average rate to be conservative. */ + mv = (v / s_interval) * NUM_SECS_ROLLING_MEASURE; + } + if (!ok) { + retval = -1; + log_notice(LD_HIST, "Could not parse value '%s' into a number.'",cp); + } + if (maxstr && !ok_m) { + retval = -1; + log_notice(LD_HIST, "Could not parse maximum '%s' into a number.'", + maxstr); + } + + if (start < now) { + time_t cur_start = start; + time_t actual_interval_len = s_interval; + uint64_t cur_val = 0; + /* Calculate the average per second. This is the best we can do + * because our state file doesn't have per-second resolution. */ + if (start + s_interval > now) + actual_interval_len = now - start; + cur_val = v / actual_interval_len; + /* This is potentially inefficient, but since we don't do it very + * often it should be ok. */ + while (cur_start < start + actual_interval_len) { + add_obs(b, cur_start, cur_val); + ++cur_start; + } + b->max_total = mv; + /* This will result in some fairly choppy history if s_interval + * is not the same as NUM_SECS_BW_SUM_INTERVAL. XXXX */ + start += actual_interval_len; + } + } SMARTLIST_FOREACH_END(cp); + } + + /* Clean up maxima and observed */ + for (i=0; i<NUM_SECS_ROLLING_MEASURE; ++i) { + b->obs[i] = 0; + } + b->total_obs = 0; + + return retval; +} + +/** Set bandwidth history from the state file we just loaded. */ +int +rep_hist_load_state(or_state_t *state, char **err) +{ + int all_ok = 1; + + /* Assert they already have been malloced */ + tor_assert(read_array && write_array); + tor_assert(dir_read_array && dir_write_array); + +#define LOAD(arrname,st) \ + if (rep_hist_load_bwhist_state_section( \ + (arrname), \ + state->BWHistory ## st ## Values, \ + state->BWHistory ## st ## Maxima, \ + state->BWHistory ## st ## Ends, \ + state->BWHistory ## st ## Interval)<0) \ + all_ok = 0 + + LOAD(write_array, Write); + LOAD(read_array, Read); + LOAD(dir_write_array, DirWrite); + LOAD(dir_read_array, DirRead); + +#undef LOAD + if (!all_ok) { + *err = tor_strdup("Parsing of bandwidth history values failed"); + /* and create fresh arrays */ + bw_arrays_init(); + return -1; + } + return 0; +} + +/*********************************************************************/ + +/** A single predicted port: used to remember which ports we've made + * connections to, so that we can try to keep making circuits that can handle + * those ports. */ +typedef struct predicted_port_t { + /** The port we connected to */ + uint16_t port; + /** The time at which we last used it */ + time_t time; +} predicted_port_t; + +/** A list of port numbers that have been used recently. */ +static smartlist_t *predicted_ports_list=NULL; +/** How long do we keep predicting circuits? */ +static time_t prediction_timeout=0; +/** When was the last time we added a prediction entry (HS or port) */ +static time_t last_prediction_add_time=0; + +/** + * How much time left until we stop predicting circuits? + */ +int +predicted_ports_prediction_time_remaining(time_t now) +{ + time_t seconds_waited; + time_t seconds_left; + + /* Protect against overflow of return value. This can happen if the clock + * jumps backwards in time. Update the last prediction time (aka last + * active time) to prevent it. This update is preferable to using monotonic + * time because it prevents clock jumps into the past from simply causing + * very long idle timeouts while the monotonic time stands still. */ + seconds_waited = time_diff(last_prediction_add_time, now); + if (seconds_waited == TIME_MAX) { + last_prediction_add_time = now; + seconds_waited = 0; + } + + /* Protect against underflow of the return value. This can happen for very + * large periods of inactivity/system sleep. */ + if (seconds_waited > prediction_timeout) + return 0; + + seconds_left = time_diff(seconds_waited, prediction_timeout); + if (BUG(seconds_left == TIME_MAX)) + return INT_MAX; + + return (int)(seconds_left); +} + +/** We just got an application request for a connection with + * port <b>port</b>. Remember it for the future, so we can keep + * some circuits open that will exit to this port. + */ +static void +add_predicted_port(time_t now, uint16_t port) +{ + predicted_port_t *pp = tor_malloc(sizeof(predicted_port_t)); + + // If the list is empty, re-randomize predicted ports lifetime + if (!any_predicted_circuits(now)) { + prediction_timeout = + (time_t)channelpadding_get_circuits_available_timeout(); + } + + last_prediction_add_time = now; + + log_info(LD_CIRC, + "New port prediction added. Will continue predictive circ building " + "for %d more seconds.", + predicted_ports_prediction_time_remaining(now)); + + pp->port = port; + pp->time = now; + rephist_total_alloc += sizeof(*pp); + smartlist_add(predicted_ports_list, pp); +} + +/** + * Allocate whatever memory and structs are needed for predicting + * which ports will be used. Also seed it with port 80, so we'll build + * circuits on start-up. + */ +static void +predicted_ports_alloc(void) +{ + predicted_ports_list = smartlist_new(); +} + +void +predicted_ports_init(void) +{ + add_predicted_port(time(NULL), 443); // Add a port to get us started +} + +/** Free whatever memory is needed for predicting which ports will + * be used. + */ +static void +predicted_ports_free_all(void) +{ + rephist_total_alloc -= + smartlist_len(predicted_ports_list)*sizeof(predicted_port_t); + SMARTLIST_FOREACH(predicted_ports_list, predicted_port_t *, + pp, tor_free(pp)); + smartlist_free(predicted_ports_list); +} + +/** Remember that <b>port</b> has been asked for as of time <b>now</b>. + * This is used for predicting what sorts of streams we'll make in the + * future and making exit circuits to anticipate that. + */ +void +rep_hist_note_used_port(time_t now, uint16_t port) +{ + tor_assert(predicted_ports_list); + + if (!port) /* record nothing */ + return; + + SMARTLIST_FOREACH_BEGIN(predicted_ports_list, predicted_port_t *, pp) { + if (pp->port == port) { + pp->time = now; + + last_prediction_add_time = now; + log_info(LD_CIRC, + "New port prediction added. Will continue predictive circ " + "building for %d more seconds.", + predicted_ports_prediction_time_remaining(now)); + return; + } + } SMARTLIST_FOREACH_END(pp); + /* it's not there yet; we need to add it */ + add_predicted_port(now, port); +} + +/** Return a newly allocated pointer to a list of uint16_t * for ports that + * are likely to be asked for in the near future. + */ +smartlist_t * +rep_hist_get_predicted_ports(time_t now) +{ + int predicted_circs_relevance_time; + smartlist_t *out = smartlist_new(); + tor_assert(predicted_ports_list); + + predicted_circs_relevance_time = (int)prediction_timeout; + + /* clean out obsolete entries */ + SMARTLIST_FOREACH_BEGIN(predicted_ports_list, predicted_port_t *, pp) { + if (pp->time + predicted_circs_relevance_time < now) { + log_debug(LD_CIRC, "Expiring predicted port %d", pp->port); + + rephist_total_alloc -= sizeof(predicted_port_t); + tor_free(pp); + SMARTLIST_DEL_CURRENT(predicted_ports_list, pp); + } else { + smartlist_add(out, tor_memdup(&pp->port, sizeof(uint16_t))); + } + } SMARTLIST_FOREACH_END(pp); + return out; +} + +/** + * Take a list of uint16_t *, and remove every port in the list from the + * current list of predicted ports. + */ +void +rep_hist_remove_predicted_ports(const smartlist_t *rmv_ports) +{ + /* Let's do this on O(N), not O(N^2). */ + bitarray_t *remove_ports = bitarray_init_zero(UINT16_MAX); + SMARTLIST_FOREACH(rmv_ports, const uint16_t *, p, + bitarray_set(remove_ports, *p)); + SMARTLIST_FOREACH_BEGIN(predicted_ports_list, predicted_port_t *, pp) { + if (bitarray_is_set(remove_ports, pp->port)) { + tor_free(pp); + rephist_total_alloc -= sizeof(*pp); + SMARTLIST_DEL_CURRENT(predicted_ports_list, pp); + } + } SMARTLIST_FOREACH_END(pp); + bitarray_free(remove_ports); +} + +/** The user asked us to do a resolve. Rather than keeping track of + * timings and such of resolves, we fake it for now by treating + * it the same way as a connection to port 80. This way we will continue + * to have circuits lying around if the user only uses Tor for resolves. + */ +void +rep_hist_note_used_resolve(time_t now) +{ + rep_hist_note_used_port(now, 80); +} + +/** The last time at which we needed an internal circ. */ +static time_t predicted_internal_time = 0; +/** The last time we needed an internal circ with good uptime. */ +static time_t predicted_internal_uptime_time = 0; +/** The last time we needed an internal circ with good capacity. */ +static time_t predicted_internal_capacity_time = 0; + +/** Remember that we used an internal circ at time <b>now</b>. */ +void +rep_hist_note_used_internal(time_t now, int need_uptime, int need_capacity) +{ + // If the list is empty, re-randomize predicted ports lifetime + if (!any_predicted_circuits(now)) { + prediction_timeout = channelpadding_get_circuits_available_timeout(); + } + + last_prediction_add_time = now; + + log_info(LD_CIRC, + "New port prediction added. Will continue predictive circ building " + "for %d more seconds.", + predicted_ports_prediction_time_remaining(now)); + + predicted_internal_time = now; + if (need_uptime) + predicted_internal_uptime_time = now; + if (need_capacity) + predicted_internal_capacity_time = now; +} + +/** Return 1 if we've used an internal circ recently; else return 0. */ +int +rep_hist_get_predicted_internal(time_t now, int *need_uptime, + int *need_capacity) +{ + int predicted_circs_relevance_time; + + predicted_circs_relevance_time = (int)prediction_timeout; + + if (!predicted_internal_time) { /* initialize it */ + predicted_internal_time = now; + predicted_internal_uptime_time = now; + predicted_internal_capacity_time = now; + } + if (predicted_internal_time + predicted_circs_relevance_time < now) + return 0; /* too long ago */ + if (predicted_internal_uptime_time + predicted_circs_relevance_time >= now) + *need_uptime = 1; + // Always predict that we need capacity. + *need_capacity = 1; + return 1; +} + +/** Any ports used lately? These are pre-seeded if we just started + * up or if we're running a hidden service. */ +int +any_predicted_circuits(time_t now) +{ + int predicted_circs_relevance_time; + predicted_circs_relevance_time = (int)prediction_timeout; + + return smartlist_len(predicted_ports_list) || + predicted_internal_time + predicted_circs_relevance_time >= now; +} + +/** Return 1 if we have no need for circuits currently, else return 0. */ +int +rep_hist_circbuilding_dormant(time_t now) +{ + const or_options_t *options = get_options(); + + if (any_predicted_circuits(now)) + return 0; + + /* see if we'll still need to build testing circuits */ + if (server_mode(options) && + (!check_whether_orport_reachable(options) || + !circuit_enough_testing_circs())) + return 0; + if (!check_whether_dirport_reachable(options)) + return 0; + + return 1; +} + +/*** Exit port statistics ***/ + +/* Some constants */ +/** To what multiple should byte numbers be rounded up? */ +#define EXIT_STATS_ROUND_UP_BYTES 1024 +/** To what multiple should stream counts be rounded up? */ +#define EXIT_STATS_ROUND_UP_STREAMS 4 +/** Number of TCP ports */ +#define EXIT_STATS_NUM_PORTS 65536 +/** Top n ports that will be included in exit stats. */ +#define EXIT_STATS_TOP_N_PORTS 10 + +/* The following data structures are arrays and no fancy smartlists or maps, + * so that all write operations can be done in constant time. This comes at + * the price of some memory (1.25 MB) and linear complexity when writing + * stats for measuring relays. */ +/** Number of bytes read in current period by exit port */ +static uint64_t *exit_bytes_read = NULL; +/** Number of bytes written in current period by exit port */ +static uint64_t *exit_bytes_written = NULL; +/** Number of streams opened in current period by exit port */ +static uint32_t *exit_streams = NULL; + +/** Start time of exit stats or 0 if we're not collecting exit stats. */ +static time_t start_of_exit_stats_interval; + +/** Initialize exit port stats. */ +void +rep_hist_exit_stats_init(time_t now) +{ + start_of_exit_stats_interval = now; + exit_bytes_read = tor_calloc(EXIT_STATS_NUM_PORTS, sizeof(uint64_t)); + exit_bytes_written = tor_calloc(EXIT_STATS_NUM_PORTS, sizeof(uint64_t)); + exit_streams = tor_calloc(EXIT_STATS_NUM_PORTS, sizeof(uint32_t)); +} + +/** Reset counters for exit port statistics. */ +void +rep_hist_reset_exit_stats(time_t now) +{ + start_of_exit_stats_interval = now; + memset(exit_bytes_read, 0, EXIT_STATS_NUM_PORTS * sizeof(uint64_t)); + memset(exit_bytes_written, 0, EXIT_STATS_NUM_PORTS * sizeof(uint64_t)); + memset(exit_streams, 0, EXIT_STATS_NUM_PORTS * sizeof(uint32_t)); +} + +/** Stop collecting exit port stats in a way that we can re-start doing + * so in rep_hist_exit_stats_init(). */ +void +rep_hist_exit_stats_term(void) +{ + start_of_exit_stats_interval = 0; + tor_free(exit_bytes_read); + tor_free(exit_bytes_written); + tor_free(exit_streams); +} + +/** Helper for qsort: compare two ints. Does not handle overflow properly, + * but works fine for sorting an array of port numbers, which is what we use + * it for. */ +static int +compare_int_(const void *x, const void *y) +{ + return (*(int*)x - *(int*)y); +} + +/** Return a newly allocated string containing the exit port statistics + * until <b>now</b>, or NULL if we're not collecting exit stats. Caller + * must ensure start_of_exit_stats_interval is in the past. */ +char * +rep_hist_format_exit_stats(time_t now) +{ + int i, j, top_elements = 0, cur_min_idx = 0, cur_port; + uint64_t top_bytes[EXIT_STATS_TOP_N_PORTS]; + int top_ports[EXIT_STATS_TOP_N_PORTS]; + uint64_t cur_bytes = 0, other_read = 0, other_written = 0, + total_read = 0, total_written = 0; + uint32_t total_streams = 0, other_streams = 0; + smartlist_t *written_strings, *read_strings, *streams_strings; + char *written_string, *read_string, *streams_string; + char t[ISO_TIME_LEN+1]; + char *result; + + if (!start_of_exit_stats_interval) + return NULL; /* Not initialized. */ + + tor_assert(now >= start_of_exit_stats_interval); + + /* Go through all ports to find the n ports that saw most written and + * read bytes. + * + * Invariant: at the end of the loop for iteration i, + * total_read is the sum of all exit_bytes_read[0..i] + * total_written is the sum of all exit_bytes_written[0..i] + * total_stream is the sum of all exit_streams[0..i] + * + * top_elements = MAX(EXIT_STATS_TOP_N_PORTS, + * #{j | 0 <= j <= i && volume(i) > 0}) + * + * For all 0 <= j < top_elements, + * top_bytes[j] > 0 + * 0 <= top_ports[j] <= 65535 + * top_bytes[j] = volume(top_ports[j]) + * + * There is no j in 0..i and k in 0..top_elements such that: + * volume(j) > top_bytes[k] AND j is not in top_ports[0..top_elements] + * + * There is no j!=cur_min_idx in 0..top_elements such that: + * top_bytes[j] < top_bytes[cur_min_idx] + * + * where volume(x) == exit_bytes_read[x]+exit_bytes_written[x] + * + * Worst case: O(EXIT_STATS_NUM_PORTS * EXIT_STATS_TOP_N_PORTS) + */ + for (i = 1; i < EXIT_STATS_NUM_PORTS; i++) { + total_read += exit_bytes_read[i]; + total_written += exit_bytes_written[i]; + total_streams += exit_streams[i]; + cur_bytes = exit_bytes_read[i] + exit_bytes_written[i]; + if (cur_bytes == 0) { + continue; + } + if (top_elements < EXIT_STATS_TOP_N_PORTS) { + top_bytes[top_elements] = cur_bytes; + top_ports[top_elements++] = i; + } else if (cur_bytes > top_bytes[cur_min_idx]) { + top_bytes[cur_min_idx] = cur_bytes; + top_ports[cur_min_idx] = i; + } else { + continue; + } + cur_min_idx = 0; + for (j = 1; j < top_elements; j++) { + if (top_bytes[j] < top_bytes[cur_min_idx]) { + cur_min_idx = j; + } + } + } + + /* Add observations of top ports to smartlists. */ + written_strings = smartlist_new(); + read_strings = smartlist_new(); + streams_strings = smartlist_new(); + other_read = total_read; + other_written = total_written; + other_streams = total_streams; + /* Sort the ports; this puts them out of sync with top_bytes, but we + * won't be using top_bytes again anyway */ + qsort(top_ports, top_elements, sizeof(int), compare_int_); + for (j = 0; j < top_elements; j++) { + cur_port = top_ports[j]; + if (exit_bytes_written[cur_port] > 0) { + uint64_t num = round_uint64_to_next_multiple_of( + exit_bytes_written[cur_port], + EXIT_STATS_ROUND_UP_BYTES); + num /= 1024; + smartlist_add_asprintf(written_strings, "%d=%"PRIu64, + cur_port, (num)); + other_written -= exit_bytes_written[cur_port]; + } + if (exit_bytes_read[cur_port] > 0) { + uint64_t num = round_uint64_to_next_multiple_of( + exit_bytes_read[cur_port], + EXIT_STATS_ROUND_UP_BYTES); + num /= 1024; + smartlist_add_asprintf(read_strings, "%d=%"PRIu64, + cur_port, (num)); + other_read -= exit_bytes_read[cur_port]; + } + if (exit_streams[cur_port] > 0) { + uint32_t num = round_uint32_to_next_multiple_of( + exit_streams[cur_port], + EXIT_STATS_ROUND_UP_STREAMS); + smartlist_add_asprintf(streams_strings, "%d=%u", cur_port, num); + other_streams -= exit_streams[cur_port]; + } + } + + /* Add observations of other ports in a single element. */ + other_written = round_uint64_to_next_multiple_of(other_written, + EXIT_STATS_ROUND_UP_BYTES); + other_written /= 1024; + smartlist_add_asprintf(written_strings, "other=%"PRIu64, + (other_written)); + other_read = round_uint64_to_next_multiple_of(other_read, + EXIT_STATS_ROUND_UP_BYTES); + other_read /= 1024; + smartlist_add_asprintf(read_strings, "other=%"PRIu64, + (other_read)); + other_streams = round_uint32_to_next_multiple_of(other_streams, + EXIT_STATS_ROUND_UP_STREAMS); + smartlist_add_asprintf(streams_strings, "other=%u", other_streams); + + /* Join all observations in single strings. */ + written_string = smartlist_join_strings(written_strings, ",", 0, NULL); + read_string = smartlist_join_strings(read_strings, ",", 0, NULL); + streams_string = smartlist_join_strings(streams_strings, ",", 0, NULL); + SMARTLIST_FOREACH(written_strings, char *, cp, tor_free(cp)); + SMARTLIST_FOREACH(read_strings, char *, cp, tor_free(cp)); + SMARTLIST_FOREACH(streams_strings, char *, cp, tor_free(cp)); + smartlist_free(written_strings); + smartlist_free(read_strings); + smartlist_free(streams_strings); + + /* Put everything together. */ + format_iso_time(t, now); + tor_asprintf(&result, "exit-stats-end %s (%d s)\n" + "exit-kibibytes-written %s\n" + "exit-kibibytes-read %s\n" + "exit-streams-opened %s\n", + t, (unsigned) (now - start_of_exit_stats_interval), + written_string, + read_string, + streams_string); + tor_free(written_string); + tor_free(read_string); + tor_free(streams_string); + return result; +} + +/** If 24 hours have passed since the beginning of the current exit port + * stats period, write exit stats to $DATADIR/stats/exit-stats (possibly + * overwriting an existing file) and reset counters. Return when we would + * next want to write exit stats or 0 if we never want to write. */ +time_t +rep_hist_exit_stats_write(time_t now) +{ + char *str = NULL; + + if (!start_of_exit_stats_interval) + return 0; /* Not initialized. */ + if (start_of_exit_stats_interval + WRITE_STATS_INTERVAL > now) + goto done; /* Not ready to write. */ + + log_info(LD_HIST, "Writing exit port statistics to disk."); + + /* Generate history string. */ + str = rep_hist_format_exit_stats(now); + + /* Reset counters. */ + rep_hist_reset_exit_stats(now); + + /* Try to write to disk. */ + if (!check_or_create_data_subdir("stats")) { + write_to_data_subdir("stats", "exit-stats", str, "exit port statistics"); + } + + done: + tor_free(str); + return start_of_exit_stats_interval + WRITE_STATS_INTERVAL; +} + +/** Note that we wrote <b>num_written</b> bytes and read <b>num_read</b> + * bytes to/from an exit connection to <b>port</b>. */ +void +rep_hist_note_exit_bytes(uint16_t port, size_t num_written, + size_t num_read) +{ + if (!start_of_exit_stats_interval) + return; /* Not initialized. */ + exit_bytes_written[port] += num_written; + exit_bytes_read[port] += num_read; + log_debug(LD_HIST, "Written %lu bytes and read %lu bytes to/from an " + "exit connection to port %d.", + (unsigned long)num_written, (unsigned long)num_read, port); +} + +/** Note that we opened an exit stream to <b>port</b>. */ +void +rep_hist_note_exit_stream_opened(uint16_t port) +{ + if (!start_of_exit_stats_interval) + return; /* Not initialized. */ + exit_streams[port]++; + log_debug(LD_HIST, "Opened exit stream to port %d", port); +} + +/*** cell statistics ***/ + +/** Start of the current buffer stats interval or 0 if we're not + * collecting buffer statistics. */ +static time_t start_of_buffer_stats_interval; + +/** Initialize buffer stats. */ +void +rep_hist_buffer_stats_init(time_t now) +{ + start_of_buffer_stats_interval = now; +} + +/** Statistics from a single circuit. Collected when the circuit closes, or + * when we flush statistics to disk. */ +typedef struct circ_buffer_stats_t { + /** Average number of cells in the circuit's queue */ + double mean_num_cells_in_queue; + /** Average time a cell waits in the queue. */ + double mean_time_cells_in_queue; + /** Total number of cells sent over this circuit */ + uint32_t processed_cells; +} circ_buffer_stats_t; + +/** List of circ_buffer_stats_t. */ +static smartlist_t *circuits_for_buffer_stats = NULL; + +/** Remember cell statistics <b>mean_num_cells_in_queue</b>, + * <b>mean_time_cells_in_queue</b>, and <b>processed_cells</b> of a + * circuit. */ +void +rep_hist_add_buffer_stats(double mean_num_cells_in_queue, + double mean_time_cells_in_queue, uint32_t processed_cells) +{ + circ_buffer_stats_t *stats; + if (!start_of_buffer_stats_interval) + return; /* Not initialized. */ + stats = tor_malloc_zero(sizeof(circ_buffer_stats_t)); + stats->mean_num_cells_in_queue = mean_num_cells_in_queue; + stats->mean_time_cells_in_queue = mean_time_cells_in_queue; + stats->processed_cells = processed_cells; + if (!circuits_for_buffer_stats) + circuits_for_buffer_stats = smartlist_new(); + smartlist_add(circuits_for_buffer_stats, stats); +} + +/** Remember cell statistics for circuit <b>circ</b> at time + * <b>end_of_interval</b> and reset cell counters in case the circuit + * remains open in the next measurement interval. */ +void +rep_hist_buffer_stats_add_circ(circuit_t *circ, time_t end_of_interval) +{ + time_t start_of_interval; + int interval_length; + or_circuit_t *orcirc; + double mean_num_cells_in_queue, mean_time_cells_in_queue; + uint32_t processed_cells; + if (CIRCUIT_IS_ORIGIN(circ)) + return; + orcirc = TO_OR_CIRCUIT(circ); + if (!orcirc->processed_cells) + return; + start_of_interval = (circ->timestamp_created.tv_sec > + start_of_buffer_stats_interval) ? + (time_t)circ->timestamp_created.tv_sec : + start_of_buffer_stats_interval; + interval_length = (int) (end_of_interval - start_of_interval); + if (interval_length <= 0) + return; + processed_cells = orcirc->processed_cells; + /* 1000.0 for s -> ms; 2.0 because of app-ward and exit-ward queues */ + mean_num_cells_in_queue = (double) orcirc->total_cell_waiting_time / + (double) interval_length / 1000.0 / 2.0; + mean_time_cells_in_queue = + (double) orcirc->total_cell_waiting_time / + (double) orcirc->processed_cells; + orcirc->total_cell_waiting_time = 0; + orcirc->processed_cells = 0; + rep_hist_add_buffer_stats(mean_num_cells_in_queue, + mean_time_cells_in_queue, + processed_cells); +} + +/** Sorting helper: return -1, 1, or 0 based on comparison of two + * circ_buffer_stats_t */ +static int +buffer_stats_compare_entries_(const void **_a, const void **_b) +{ + const circ_buffer_stats_t *a = *_a, *b = *_b; + if (a->processed_cells < b->processed_cells) + return 1; + else if (a->processed_cells > b->processed_cells) + return -1; + else + return 0; +} + +/** Stop collecting cell stats in a way that we can re-start doing so in + * rep_hist_buffer_stats_init(). */ +void +rep_hist_buffer_stats_term(void) +{ + rep_hist_reset_buffer_stats(0); +} + +/** Clear history of circuit statistics and set the measurement interval + * start to <b>now</b>. */ +void +rep_hist_reset_buffer_stats(time_t now) +{ + if (!circuits_for_buffer_stats) + circuits_for_buffer_stats = smartlist_new(); + SMARTLIST_FOREACH(circuits_for_buffer_stats, circ_buffer_stats_t *, + stats, tor_free(stats)); + smartlist_clear(circuits_for_buffer_stats); + start_of_buffer_stats_interval = now; +} + +/** Return a newly allocated string containing the buffer statistics until + * <b>now</b>, or NULL if we're not collecting buffer stats. Caller must + * ensure start_of_buffer_stats_interval is in the past. */ +char * +rep_hist_format_buffer_stats(time_t now) +{ +#define SHARES 10 + uint64_t processed_cells[SHARES]; + uint32_t circs_in_share[SHARES]; + int number_of_circuits, i; + double queued_cells[SHARES], time_in_queue[SHARES]; + smartlist_t *processed_cells_strings, *queued_cells_strings, + *time_in_queue_strings; + char *processed_cells_string, *queued_cells_string, + *time_in_queue_string; + char t[ISO_TIME_LEN+1]; + char *result; + + if (!start_of_buffer_stats_interval) + return NULL; /* Not initialized. */ + + tor_assert(now >= start_of_buffer_stats_interval); + + /* Calculate deciles if we saw at least one circuit. */ + memset(processed_cells, 0, SHARES * sizeof(uint64_t)); + memset(circs_in_share, 0, SHARES * sizeof(uint32_t)); + memset(queued_cells, 0, SHARES * sizeof(double)); + memset(time_in_queue, 0, SHARES * sizeof(double)); + if (!circuits_for_buffer_stats) + circuits_for_buffer_stats = smartlist_new(); + number_of_circuits = smartlist_len(circuits_for_buffer_stats); + if (number_of_circuits > 0) { + smartlist_sort(circuits_for_buffer_stats, + buffer_stats_compare_entries_); + i = 0; + SMARTLIST_FOREACH_BEGIN(circuits_for_buffer_stats, + circ_buffer_stats_t *, stats) + { + int share = i++ * SHARES / number_of_circuits; + processed_cells[share] += stats->processed_cells; + queued_cells[share] += stats->mean_num_cells_in_queue; + time_in_queue[share] += stats->mean_time_cells_in_queue; + circs_in_share[share]++; + } + SMARTLIST_FOREACH_END(stats); + } + + /* Write deciles to strings. */ + processed_cells_strings = smartlist_new(); + queued_cells_strings = smartlist_new(); + time_in_queue_strings = smartlist_new(); + for (i = 0; i < SHARES; i++) { + smartlist_add_asprintf(processed_cells_strings, + "%"PRIu64, !circs_in_share[i] ? 0 : + (processed_cells[i] / + circs_in_share[i])); + } + for (i = 0; i < SHARES; i++) { + smartlist_add_asprintf(queued_cells_strings, "%.2f", + circs_in_share[i] == 0 ? 0.0 : + queued_cells[i] / (double) circs_in_share[i]); + } + for (i = 0; i < SHARES; i++) { + smartlist_add_asprintf(time_in_queue_strings, "%.0f", + circs_in_share[i] == 0 ? 0.0 : + time_in_queue[i] / (double) circs_in_share[i]); + } + + /* Join all observations in single strings. */ + processed_cells_string = smartlist_join_strings(processed_cells_strings, + ",", 0, NULL); + queued_cells_string = smartlist_join_strings(queued_cells_strings, + ",", 0, NULL); + time_in_queue_string = smartlist_join_strings(time_in_queue_strings, + ",", 0, NULL); + SMARTLIST_FOREACH(processed_cells_strings, char *, cp, tor_free(cp)); + SMARTLIST_FOREACH(queued_cells_strings, char *, cp, tor_free(cp)); + SMARTLIST_FOREACH(time_in_queue_strings, char *, cp, tor_free(cp)); + smartlist_free(processed_cells_strings); + smartlist_free(queued_cells_strings); + smartlist_free(time_in_queue_strings); + + /* Put everything together. */ + format_iso_time(t, now); + tor_asprintf(&result, "cell-stats-end %s (%d s)\n" + "cell-processed-cells %s\n" + "cell-queued-cells %s\n" + "cell-time-in-queue %s\n" + "cell-circuits-per-decile %d\n", + t, (unsigned) (now - start_of_buffer_stats_interval), + processed_cells_string, + queued_cells_string, + time_in_queue_string, + CEIL_DIV(number_of_circuits, SHARES)); + tor_free(processed_cells_string); + tor_free(queued_cells_string); + tor_free(time_in_queue_string); + return result; +#undef SHARES +} + +/** If 24 hours have passed since the beginning of the current buffer + * stats period, write buffer stats to $DATADIR/stats/buffer-stats + * (possibly overwriting an existing file) and reset counters. Return + * when we would next want to write buffer stats or 0 if we never want to + * write. */ +time_t +rep_hist_buffer_stats_write(time_t now) +{ + char *str = NULL; + + if (!start_of_buffer_stats_interval) + return 0; /* Not initialized. */ + if (start_of_buffer_stats_interval + WRITE_STATS_INTERVAL > now) + goto done; /* Not ready to write */ + + /* Add open circuits to the history. */ + SMARTLIST_FOREACH_BEGIN(circuit_get_global_list(), circuit_t *, circ) { + rep_hist_buffer_stats_add_circ(circ, now); + } + SMARTLIST_FOREACH_END(circ); + + /* Generate history string. */ + str = rep_hist_format_buffer_stats(now); + + /* Reset both buffer history and counters of open circuits. */ + rep_hist_reset_buffer_stats(now); + + /* Try to write to disk. */ + if (!check_or_create_data_subdir("stats")) { + write_to_data_subdir("stats", "buffer-stats", str, "buffer statistics"); + } + + done: + tor_free(str); + return start_of_buffer_stats_interval + WRITE_STATS_INTERVAL; +} + +/*** Descriptor serving statistics ***/ + +/** Digestmap to track which descriptors were downloaded this stats + * collection interval. It maps descriptor digest to pointers to 1, + * effectively turning this into a list. */ +static digestmap_t *served_descs = NULL; + +/** Number of how many descriptors were downloaded in total during this + * interval. */ +static unsigned long total_descriptor_downloads; + +/** Start time of served descs stats or 0 if we're not collecting those. */ +static time_t start_of_served_descs_stats_interval; + +/** Initialize descriptor stats. */ +void +rep_hist_desc_stats_init(time_t now) +{ + if (served_descs) { + log_warn(LD_BUG, "Called rep_hist_desc_stats_init() when desc stats were " + "already initialized. This is probably harmless."); + return; // Already initialized + } + served_descs = digestmap_new(); + total_descriptor_downloads = 0; + start_of_served_descs_stats_interval = now; +} + +/** Reset served descs stats to empty, starting a new interval <b>now</b>. */ +static void +rep_hist_reset_desc_stats(time_t now) +{ + rep_hist_desc_stats_term(); + rep_hist_desc_stats_init(now); +} + +/** Stop collecting served descs stats, so that rep_hist_desc_stats_init() is + * safe to be called again. */ +void +rep_hist_desc_stats_term(void) +{ + digestmap_free(served_descs, NULL); + served_descs = NULL; + start_of_served_descs_stats_interval = 0; + total_descriptor_downloads = 0; +} + +/** Helper for rep_hist_desc_stats_write(). Return a newly allocated string + * containing the served desc statistics until now, or NULL if we're not + * collecting served desc stats. Caller must ensure that now is not before + * start_of_served_descs_stats_interval. */ +static char * +rep_hist_format_desc_stats(time_t now) +{ + char t[ISO_TIME_LEN+1]; + char *result; + + digestmap_iter_t *iter; + const char *key; + void *val; + unsigned size; + int *vals, max = 0, q3 = 0, md = 0, q1 = 0, min = 0; + int n = 0; + + if (!start_of_served_descs_stats_interval) + return NULL; + + size = digestmap_size(served_descs); + if (size > 0) { + vals = tor_calloc(size, sizeof(int)); + for (iter = digestmap_iter_init(served_descs); + !digestmap_iter_done(iter); + iter = digestmap_iter_next(served_descs, iter)) { + uintptr_t count; + digestmap_iter_get(iter, &key, &val); + count = (uintptr_t)val; + vals[n++] = (int)count; + (void)key; + } + max = find_nth_int(vals, size, size-1); + q3 = find_nth_int(vals, size, (3*size-1)/4); + md = find_nth_int(vals, size, (size-1)/2); + q1 = find_nth_int(vals, size, (size-1)/4); + min = find_nth_int(vals, size, 0); + tor_free(vals); + } + + format_iso_time(t, now); + + tor_asprintf(&result, + "served-descs-stats-end %s (%d s) total=%lu unique=%u " + "max=%d q3=%d md=%d q1=%d min=%d\n", + t, + (unsigned) (now - start_of_served_descs_stats_interval), + total_descriptor_downloads, + size, max, q3, md, q1, min); + + return result; +} + +/** If WRITE_STATS_INTERVAL seconds have passed since the beginning of + * the current served desc stats interval, write the stats to + * $DATADIR/stats/served-desc-stats (possibly appending to an existing file) + * and reset the state for the next interval. Return when we would next want + * to write served desc stats or 0 if we won't want to write. */ +time_t +rep_hist_desc_stats_write(time_t now) +{ + char *filename = NULL, *str = NULL; + + if (!start_of_served_descs_stats_interval) + return 0; /* We're not collecting stats. */ + if (start_of_served_descs_stats_interval + WRITE_STATS_INTERVAL > now) + return start_of_served_descs_stats_interval + WRITE_STATS_INTERVAL; + + str = rep_hist_format_desc_stats(now); + tor_assert(str != NULL); + + if (check_or_create_data_subdir("stats") < 0) { + goto done; + } + filename = get_datadir_fname2("stats", "served-desc-stats"); + if (append_bytes_to_file(filename, str, strlen(str), 0) < 0) + log_warn(LD_HIST, "Unable to write served descs statistics to disk!"); + + rep_hist_reset_desc_stats(now); + + done: + tor_free(filename); + tor_free(str); + return start_of_served_descs_stats_interval + WRITE_STATS_INTERVAL; +} + +/** Called to note that we've served a given descriptor (by + * digest). Increments the count of descriptors served, and the number + * of times we've served this descriptor. */ +void +rep_hist_note_desc_served(const char * desc) +{ + void *val; + uintptr_t count; + if (!served_descs) + return; // We're not collecting stats + val = digestmap_get(served_descs, desc); + count = (uintptr_t)val; + if (count != INT_MAX) + ++count; + digestmap_set(served_descs, desc, (void*)count); + total_descriptor_downloads++; +} + +/*** Connection statistics ***/ + +/** Start of the current connection stats interval or 0 if we're not + * collecting connection statistics. */ +static time_t start_of_conn_stats_interval; + +/** Initialize connection stats. */ +void +rep_hist_conn_stats_init(time_t now) +{ + start_of_conn_stats_interval = now; +} + +/* Count connections that we read and wrote less than these many bytes + * from/to as below threshold. */ +#define BIDI_THRESHOLD 20480 + +/* Count connections that we read or wrote at least this factor as many + * bytes from/to than we wrote or read to/from as mostly reading or + * writing. */ +#define BIDI_FACTOR 10 + +/* Interval length in seconds for considering read and written bytes for + * connection stats. */ +#define BIDI_INTERVAL 10 + +/** Start of next BIDI_INTERVAL second interval. */ +static time_t bidi_next_interval = 0; + +/** Number of connections that we read and wrote less than BIDI_THRESHOLD + * bytes from/to in BIDI_INTERVAL seconds. */ +static uint32_t below_threshold = 0; + +/** Number of connections that we read at least BIDI_FACTOR times more + * bytes from than we wrote to in BIDI_INTERVAL seconds. */ +static uint32_t mostly_read = 0; + +/** Number of connections that we wrote at least BIDI_FACTOR times more + * bytes to than we read from in BIDI_INTERVAL seconds. */ +static uint32_t mostly_written = 0; + +/** Number of connections that we read and wrote at least BIDI_THRESHOLD + * bytes from/to, but not BIDI_FACTOR times more in either direction in + * BIDI_INTERVAL seconds. */ +static uint32_t both_read_and_written = 0; + +/** Entry in a map from connection ID to the number of read and written + * bytes on this connection in a BIDI_INTERVAL second interval. */ +typedef struct bidi_map_entry_t { + HT_ENTRY(bidi_map_entry_t) node; + uint64_t conn_id; /**< Connection ID */ + size_t read; /**< Number of read bytes */ + size_t written; /**< Number of written bytes */ +} bidi_map_entry_t; + +/** Map of OR connections together with the number of read and written + * bytes in the current BIDI_INTERVAL second interval. */ +static HT_HEAD(bidimap, bidi_map_entry_t) bidi_map = + HT_INITIALIZER(); + +static int +bidi_map_ent_eq(const bidi_map_entry_t *a, const bidi_map_entry_t *b) +{ + return a->conn_id == b->conn_id; +} + +/* DOCDOC bidi_map_ent_hash */ +static unsigned +bidi_map_ent_hash(const bidi_map_entry_t *entry) +{ + return (unsigned) entry->conn_id; +} + +HT_PROTOTYPE(bidimap, bidi_map_entry_t, node, bidi_map_ent_hash, + bidi_map_ent_eq) +HT_GENERATE2(bidimap, bidi_map_entry_t, node, bidi_map_ent_hash, + bidi_map_ent_eq, 0.6, tor_reallocarray_, tor_free_) + +/* DOCDOC bidi_map_free */ +static void +bidi_map_free_all(void) +{ + bidi_map_entry_t **ptr, **next, *ent; + for (ptr = HT_START(bidimap, &bidi_map); ptr; ptr = next) { + ent = *ptr; + next = HT_NEXT_RMV(bidimap, &bidi_map, ptr); + tor_free(ent); + } + HT_CLEAR(bidimap, &bidi_map); +} + +/** Reset counters for conn statistics. */ +void +rep_hist_reset_conn_stats(time_t now) +{ + start_of_conn_stats_interval = now; + below_threshold = 0; + mostly_read = 0; + mostly_written = 0; + both_read_and_written = 0; + bidi_map_free_all(); +} + +/** Stop collecting connection stats in a way that we can re-start doing + * so in rep_hist_conn_stats_init(). */ +void +rep_hist_conn_stats_term(void) +{ + rep_hist_reset_conn_stats(0); +} + +/** We read <b>num_read</b> bytes and wrote <b>num_written</b> from/to OR + * connection <b>conn_id</b> in second <b>when</b>. If this is the first + * observation in a new interval, sum up the last observations. Add bytes + * for this connection. */ +void +rep_hist_note_or_conn_bytes(uint64_t conn_id, size_t num_read, + size_t num_written, time_t when) +{ + if (!start_of_conn_stats_interval) + return; + /* Initialize */ + if (bidi_next_interval == 0) + bidi_next_interval = when + BIDI_INTERVAL; + /* Sum up last period's statistics */ + if (when >= bidi_next_interval) { + bidi_map_entry_t **ptr, **next, *ent; + for (ptr = HT_START(bidimap, &bidi_map); ptr; ptr = next) { + ent = *ptr; + if (ent->read + ent->written < BIDI_THRESHOLD) + below_threshold++; + else if (ent->read >= ent->written * BIDI_FACTOR) + mostly_read++; + else if (ent->written >= ent->read * BIDI_FACTOR) + mostly_written++; + else + both_read_and_written++; + next = HT_NEXT_RMV(bidimap, &bidi_map, ptr); + tor_free(ent); + } + while (when >= bidi_next_interval) + bidi_next_interval += BIDI_INTERVAL; + log_info(LD_GENERAL, "%d below threshold, %d mostly read, " + "%d mostly written, %d both read and written.", + below_threshold, mostly_read, mostly_written, + both_read_and_written); + } + /* Add this connection's bytes. */ + if (num_read > 0 || num_written > 0) { + bidi_map_entry_t *entry, lookup; + lookup.conn_id = conn_id; + entry = HT_FIND(bidimap, &bidi_map, &lookup); + if (entry) { + entry->written += num_written; + entry->read += num_read; + } else { + entry = tor_malloc_zero(sizeof(bidi_map_entry_t)); + entry->conn_id = conn_id; + entry->written = num_written; + entry->read = num_read; + HT_INSERT(bidimap, &bidi_map, entry); + } + } +} + +/** Return a newly allocated string containing the connection statistics + * until <b>now</b>, or NULL if we're not collecting conn stats. Caller must + * ensure start_of_conn_stats_interval is in the past. */ +char * +rep_hist_format_conn_stats(time_t now) +{ + char *result, written[ISO_TIME_LEN+1]; + + if (!start_of_conn_stats_interval) + return NULL; /* Not initialized. */ + + tor_assert(now >= start_of_conn_stats_interval); + + format_iso_time(written, now); + tor_asprintf(&result, "conn-bi-direct %s (%d s) %d,%d,%d,%d\n", + written, + (unsigned) (now - start_of_conn_stats_interval), + below_threshold, + mostly_read, + mostly_written, + both_read_and_written); + return result; +} + +/** If 24 hours have passed since the beginning of the current conn stats + * period, write conn stats to $DATADIR/stats/conn-stats (possibly + * overwriting an existing file) and reset counters. Return when we would + * next want to write conn stats or 0 if we never want to write. */ +time_t +rep_hist_conn_stats_write(time_t now) +{ + char *str = NULL; + + if (!start_of_conn_stats_interval) + return 0; /* Not initialized. */ + if (start_of_conn_stats_interval + WRITE_STATS_INTERVAL > now) + goto done; /* Not ready to write */ + + /* Generate history string. */ + str = rep_hist_format_conn_stats(now); + + /* Reset counters. */ + rep_hist_reset_conn_stats(now); + + /* Try to write to disk. */ + if (!check_or_create_data_subdir("stats")) { + write_to_data_subdir("stats", "conn-stats", str, "connection statistics"); + } + + done: + tor_free(str); + return start_of_conn_stats_interval + WRITE_STATS_INTERVAL; +} + +/** Internal statistics to track how many requests of each type of + * handshake we've received, and how many we've assigned to cpuworkers. + * Useful for seeing trends in cpu load. + * @{ */ +STATIC int onion_handshakes_requested[MAX_ONION_HANDSHAKE_TYPE+1] = {0}; +STATIC int onion_handshakes_assigned[MAX_ONION_HANDSHAKE_TYPE+1] = {0}; +/**@}*/ + +/** A new onionskin (using the <b>type</b> handshake) has arrived. */ +void +rep_hist_note_circuit_handshake_requested(uint16_t type) +{ + if (type <= MAX_ONION_HANDSHAKE_TYPE) + onion_handshakes_requested[type]++; +} + +/** We've sent an onionskin (using the <b>type</b> handshake) to a + * cpuworker. */ +void +rep_hist_note_circuit_handshake_assigned(uint16_t type) +{ + if (type <= MAX_ONION_HANDSHAKE_TYPE) + onion_handshakes_assigned[type]++; +} + +/** Log our onionskin statistics since the last time we were called. */ +void +rep_hist_log_circuit_handshake_stats(time_t now) +{ + (void)now; + log_notice(LD_HEARTBEAT, "Circuit handshake stats since last time: " + "%d/%d TAP, %d/%d NTor.", + onion_handshakes_assigned[ONION_HANDSHAKE_TYPE_TAP], + onion_handshakes_requested[ONION_HANDSHAKE_TYPE_TAP], + onion_handshakes_assigned[ONION_HANDSHAKE_TYPE_NTOR], + onion_handshakes_requested[ONION_HANDSHAKE_TYPE_NTOR]); + memset(onion_handshakes_assigned, 0, sizeof(onion_handshakes_assigned)); + memset(onion_handshakes_requested, 0, sizeof(onion_handshakes_requested)); +} + +/* Hidden service statistics section */ + +/** Start of the current hidden service stats interval or 0 if we're + * not collecting hidden service statistics. */ +static time_t start_of_hs_stats_interval; + +/** Carries the various hidden service statistics, and any other + * information needed. */ +typedef struct hs_stats_t { + /** How many relay cells have we seen as rendezvous points? */ + uint64_t rp_relay_cells_seen; + + /** Set of unique public key digests we've seen this stat period + * (could also be implemented as sorted smartlist). */ + digestmap_t *onions_seen_this_period; +} hs_stats_t; + +/** Our statistics structure singleton. */ +static hs_stats_t *hs_stats = NULL; + +/** Allocate, initialize and return an hs_stats_t structure. */ +static hs_stats_t * +hs_stats_new(void) +{ + hs_stats_t *new_hs_stats = tor_malloc_zero(sizeof(hs_stats_t)); + new_hs_stats->onions_seen_this_period = digestmap_new(); + + return new_hs_stats; +} + +#define hs_stats_free(val) \ + FREE_AND_NULL(hs_stats_t, hs_stats_free_, (val)) + +/** Free an hs_stats_t structure. */ +static void +hs_stats_free_(hs_stats_t *victim_hs_stats) +{ + if (!victim_hs_stats) { + return; + } + + digestmap_free(victim_hs_stats->onions_seen_this_period, NULL); + tor_free(victim_hs_stats); +} + +/** Initialize hidden service statistics. */ +void +rep_hist_hs_stats_init(time_t now) +{ + if (!hs_stats) { + hs_stats = hs_stats_new(); + } + + start_of_hs_stats_interval = now; +} + +/** Clear history of hidden service statistics and set the measurement + * interval start to <b>now</b>. */ +static void +rep_hist_reset_hs_stats(time_t now) +{ + if (!hs_stats) { + hs_stats = hs_stats_new(); + } + + hs_stats->rp_relay_cells_seen = 0; + + digestmap_free(hs_stats->onions_seen_this_period, NULL); + hs_stats->onions_seen_this_period = digestmap_new(); + + start_of_hs_stats_interval = now; +} + +/** Stop collecting hidden service stats in a way that we can re-start + * doing so in rep_hist_buffer_stats_init(). */ +void +rep_hist_hs_stats_term(void) +{ + rep_hist_reset_hs_stats(0); +} + +/** We saw a new HS relay cell, Count it! */ +void +rep_hist_seen_new_rp_cell(void) +{ + if (!hs_stats) { + return; // We're not collecting stats + } + + hs_stats->rp_relay_cells_seen++; +} + +/** As HSDirs, we saw another hidden service with public key + * <b>pubkey</b>. Check whether we have counted it before, if not + * count it now! */ +void +rep_hist_stored_maybe_new_hs(const crypto_pk_t *pubkey) +{ + char pubkey_hash[DIGEST_LEN]; + + if (!hs_stats) { + return; // We're not collecting stats + } + + /* Get the digest of the pubkey which will be used to detect whether + we've seen this hidden service before or not. */ + if (crypto_pk_get_digest(pubkey, pubkey_hash) < 0) { + /* This fail should not happen; key has been validated by + descriptor parsing code first. */ + return; + } + + /* Check if this is the first time we've seen this hidden + service. If it is, count it as new. */ + if (!digestmap_get(hs_stats->onions_seen_this_period, + pubkey_hash)) { + digestmap_set(hs_stats->onions_seen_this_period, + pubkey_hash, (void*)(uintptr_t)1); + } +} + +/* The number of cells that are supposed to be hidden from the adversary + * by adding noise from the Laplace distribution. This value, divided by + * EPSILON, is Laplace parameter b. It must be greather than 0. */ +#define REND_CELLS_DELTA_F 2048 +/* Security parameter for obfuscating number of cells with a value between + * ]0.0, 1.0]. Smaller values obfuscate observations more, but at the same + * time make statistics less usable. */ +#define REND_CELLS_EPSILON 0.3 +/* The number of cells that are supposed to be hidden from the adversary + * by rounding up to the next multiple of this number. */ +#define REND_CELLS_BIN_SIZE 1024 +/* The number of service identities that are supposed to be hidden from the + * adversary by adding noise from the Laplace distribution. This value, + * divided by EPSILON, is Laplace parameter b. It must be greater than 0. */ +#define ONIONS_SEEN_DELTA_F 8 +/* Security parameter for obfuscating number of service identities with a + * value between ]0.0, 1.0]. Smaller values obfuscate observations more, but + * at the same time make statistics less usable. */ +#define ONIONS_SEEN_EPSILON 0.3 +/* The number of service identities that are supposed to be hidden from + * the adversary by rounding up to the next multiple of this number. */ +#define ONIONS_SEEN_BIN_SIZE 8 + +/** Allocate and return a string containing hidden service stats that + * are meant to be placed in the extra-info descriptor. */ +static char * +rep_hist_format_hs_stats(time_t now) +{ + char t[ISO_TIME_LEN+1]; + char *hs_stats_string; + int64_t obfuscated_cells_seen; + int64_t obfuscated_onions_seen; + + uint64_t rounded_cells_seen + = round_uint64_to_next_multiple_of(hs_stats->rp_relay_cells_seen, + REND_CELLS_BIN_SIZE); + rounded_cells_seen = MIN(rounded_cells_seen, INT64_MAX); + obfuscated_cells_seen = add_laplace_noise((int64_t)rounded_cells_seen, + crypto_rand_double(), + REND_CELLS_DELTA_F, REND_CELLS_EPSILON); + + uint64_t rounded_onions_seen = + round_uint64_to_next_multiple_of((size_t)digestmap_size( + hs_stats->onions_seen_this_period), + ONIONS_SEEN_BIN_SIZE); + rounded_onions_seen = MIN(rounded_onions_seen, INT64_MAX); + obfuscated_onions_seen = add_laplace_noise((int64_t)rounded_onions_seen, + crypto_rand_double(), ONIONS_SEEN_DELTA_F, + ONIONS_SEEN_EPSILON); + + format_iso_time(t, now); + tor_asprintf(&hs_stats_string, "hidserv-stats-end %s (%d s)\n" + "hidserv-rend-relayed-cells %"PRId64" delta_f=%d " + "epsilon=%.2f bin_size=%d\n" + "hidserv-dir-onions-seen %"PRId64" delta_f=%d " + "epsilon=%.2f bin_size=%d\n", + t, (unsigned) (now - start_of_hs_stats_interval), + (obfuscated_cells_seen), REND_CELLS_DELTA_F, + REND_CELLS_EPSILON, REND_CELLS_BIN_SIZE, + (obfuscated_onions_seen), + ONIONS_SEEN_DELTA_F, + ONIONS_SEEN_EPSILON, ONIONS_SEEN_BIN_SIZE); + + return hs_stats_string; +} + +/** If 24 hours have passed since the beginning of the current HS + * stats period, write buffer stats to $DATADIR/stats/hidserv-stats + * (possibly overwriting an existing file) and reset counters. Return + * when we would next want to write buffer stats or 0 if we never want to + * write. */ +time_t +rep_hist_hs_stats_write(time_t now) +{ + char *str = NULL; + + if (!start_of_hs_stats_interval) { + return 0; /* Not initialized. */ + } + + if (start_of_hs_stats_interval + WRITE_STATS_INTERVAL > now) { + goto done; /* Not ready to write */ + } + + /* Generate history string. */ + str = rep_hist_format_hs_stats(now); + + /* Reset HS history. */ + rep_hist_reset_hs_stats(now); + + /* Try to write to disk. */ + if (!check_or_create_data_subdir("stats")) { + write_to_data_subdir("stats", "hidserv-stats", str, + "hidden service stats"); + } + + done: + tor_free(str); + return start_of_hs_stats_interval + WRITE_STATS_INTERVAL; +} + +static uint64_t link_proto_count[MAX_LINK_PROTO+1][2]; + +/** Note that we negotiated link protocol version <b>link_proto</b>, on + * a connection that started here iff <b>started_here</b> is true. + */ +void +rep_hist_note_negotiated_link_proto(unsigned link_proto, int started_here) +{ + started_here = !!started_here; /* force to 0 or 1 */ + if (link_proto > MAX_LINK_PROTO) { + log_warn(LD_BUG, "Can't log link protocol %u", link_proto); + return; + } + + link_proto_count[link_proto][started_here]++; +} + +/** + * Update the maximum count of total pending channel padding timers + * in this period. + */ +void +rep_hist_padding_count_timers(uint64_t num_timers) +{ + if (num_timers > padding_current.maximum_chanpad_timers) { + padding_current.maximum_chanpad_timers = num_timers; + } +} + +/** + * Count a cell that we sent for padding overhead statistics. + * + * RELAY_COMMAND_DROP and CELL_PADDING are accounted separately. Both should be + * counted for PADDING_TYPE_TOTAL. + */ +void +rep_hist_padding_count_write(padding_type_t type) +{ + switch (type) { + case PADDING_TYPE_DROP: + padding_current.write_drop_cell_count++; + break; + case PADDING_TYPE_CELL: + padding_current.write_pad_cell_count++; + break; + case PADDING_TYPE_TOTAL: + padding_current.write_cell_count++; + break; + case PADDING_TYPE_ENABLED_TOTAL: + padding_current.enabled_write_cell_count++; + break; + case PADDING_TYPE_ENABLED_CELL: + padding_current.enabled_write_pad_cell_count++; + break; + } +} + +/** + * Count a cell that we've received for padding overhead statistics. + * + * RELAY_COMMAND_DROP and CELL_PADDING are accounted separately. Both should be + * counted for PADDING_TYPE_TOTAL. + */ +void +rep_hist_padding_count_read(padding_type_t type) +{ + switch (type) { + case PADDING_TYPE_DROP: + padding_current.read_drop_cell_count++; + break; + case PADDING_TYPE_CELL: + padding_current.read_pad_cell_count++; + break; + case PADDING_TYPE_TOTAL: + padding_current.read_cell_count++; + break; + case PADDING_TYPE_ENABLED_TOTAL: + padding_current.enabled_read_cell_count++; + break; + case PADDING_TYPE_ENABLED_CELL: + padding_current.enabled_read_pad_cell_count++; + break; + } +} + +/** + * Reset our current padding statistics. Called once every 24 hours. + */ +void +rep_hist_reset_padding_counts(void) +{ + memset(&padding_current, 0, sizeof(padding_current)); +} + +/** + * Copy our current cell counts into a structure for listing in our + * extra-info descriptor. Also perform appropriate rounding and redaction. + * + * This function is called once every 24 hours. + */ +#define MIN_CELL_COUNTS_TO_PUBLISH 1 +#define ROUND_CELL_COUNTS_TO 10000 +void +rep_hist_prep_published_padding_counts(time_t now) +{ + memcpy(&padding_published, &padding_current, sizeof(padding_published)); + + if (padding_published.read_cell_count < MIN_CELL_COUNTS_TO_PUBLISH || + padding_published.write_cell_count < MIN_CELL_COUNTS_TO_PUBLISH) { + memset(&padding_published, 0, sizeof(padding_published)); + return; + } + + format_iso_time(padding_published.first_published_at, now); +#define ROUND_AND_SET_COUNT(x) (x) = round_uint64_to_next_multiple_of((x), \ + ROUND_CELL_COUNTS_TO) + ROUND_AND_SET_COUNT(padding_published.read_pad_cell_count); + ROUND_AND_SET_COUNT(padding_published.write_pad_cell_count); + ROUND_AND_SET_COUNT(padding_published.read_drop_cell_count); + ROUND_AND_SET_COUNT(padding_published.write_drop_cell_count); + ROUND_AND_SET_COUNT(padding_published.write_cell_count); + ROUND_AND_SET_COUNT(padding_published.read_cell_count); + ROUND_AND_SET_COUNT(padding_published.enabled_read_cell_count); + ROUND_AND_SET_COUNT(padding_published.enabled_read_pad_cell_count); + ROUND_AND_SET_COUNT(padding_published.enabled_write_cell_count); + ROUND_AND_SET_COUNT(padding_published.enabled_write_pad_cell_count); +#undef ROUND_AND_SET_COUNT +} + +/** + * Returns an allocated string for extra-info documents for publishing + * padding statistics from the last 24 hour interval. + */ +char * +rep_hist_get_padding_count_lines(void) +{ + char *result = NULL; + + if (!padding_published.read_cell_count || + !padding_published.write_cell_count) { + return NULL; + } + + tor_asprintf(&result, "padding-counts %s (%d s)" + " bin-size=%"PRIu64 + " write-drop=%"PRIu64 + " write-pad=%"PRIu64 + " write-total=%"PRIu64 + " read-drop=%"PRIu64 + " read-pad=%"PRIu64 + " read-total=%"PRIu64 + " enabled-read-pad=%"PRIu64 + " enabled-read-total=%"PRIu64 + " enabled-write-pad=%"PRIu64 + " enabled-write-total=%"PRIu64 + " max-chanpad-timers=%"PRIu64 + "\n", + padding_published.first_published_at, + REPHIST_CELL_PADDING_COUNTS_INTERVAL, + (uint64_t)ROUND_CELL_COUNTS_TO, + (padding_published.write_drop_cell_count), + (padding_published.write_pad_cell_count), + (padding_published.write_cell_count), + (padding_published.read_drop_cell_count), + (padding_published.read_pad_cell_count), + (padding_published.read_cell_count), + (padding_published.enabled_read_pad_cell_count), + (padding_published.enabled_read_cell_count), + (padding_published.enabled_write_pad_cell_count), + (padding_published.enabled_write_cell_count), + (padding_published.maximum_chanpad_timers) + ); + + return result; +} + +/** Log a heartbeat message explaining how many connections of each link + * protocol version we have used. + */ +void +rep_hist_log_link_protocol_counts(void) +{ + log_notice(LD_HEARTBEAT, + "Since startup, we have initiated " + "%"PRIu64" v1 connections, " + "%"PRIu64" v2 connections, " + "%"PRIu64" v3 connections, and " + "%"PRIu64" v4 connections; and received " + "%"PRIu64" v1 connections, " + "%"PRIu64" v2 connections, " + "%"PRIu64" v3 connections, and " + "%"PRIu64" v4 connections.", + (link_proto_count[1][1]), + (link_proto_count[2][1]), + (link_proto_count[3][1]), + (link_proto_count[4][1]), + (link_proto_count[1][0]), + (link_proto_count[2][0]), + (link_proto_count[3][0]), + (link_proto_count[4][0])); +} + +/** Free all storage held by the OR/link history caches, by the + * bandwidth history arrays, by the port history, or by statistics . */ +void +rep_hist_free_all(void) +{ + hs_stats_free(hs_stats); + digestmap_free(history_map, free_or_history); + + bw_array_free(read_array); + read_array = NULL; + + bw_array_free(write_array); + write_array = NULL; + + bw_array_free(dir_read_array); + dir_read_array = NULL; + + bw_array_free(dir_write_array); + dir_write_array = NULL; + + tor_free(exit_bytes_read); + tor_free(exit_bytes_written); + tor_free(exit_streams); + predicted_ports_free_all(); + bidi_map_free_all(); + + if (circuits_for_buffer_stats) { + SMARTLIST_FOREACH(circuits_for_buffer_stats, circ_buffer_stats_t *, s, + tor_free(s)); + smartlist_free(circuits_for_buffer_stats); + circuits_for_buffer_stats = NULL; + } + rep_hist_desc_stats_term(); + total_descriptor_downloads = 0; + + tor_assert_nonfatal(rephist_total_alloc == 0); + tor_assert_nonfatal_once(rephist_total_num == 0); +} diff --git a/src/feature/stats/rephist.h b/src/feature/stats/rephist.h new file mode 100644 index 0000000000..e17a722489 --- /dev/null +++ b/src/feature/stats/rephist.h @@ -0,0 +1,142 @@ +/* Copyright (c) 2001 Matej Pfajfar. + * Copyright (c) 2001-2004, Roger Dingledine. + * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson. + * Copyright (c) 2007-2018, The Tor Project, Inc. */ +/* See LICENSE for licensing information */ + +/** + * \file rephist.h + * \brief Header file for rephist.c. + **/ + +#ifndef TOR_REPHIST_H +#define TOR_REPHIST_H + +void rep_hist_init(void); +void rep_hist_dump_stats(time_t now, int severity); +void rep_hist_note_bytes_read(uint64_t num_bytes, time_t when); +void rep_hist_note_bytes_written(uint64_t num_bytes, time_t when); + +void rep_hist_make_router_pessimal(const char *id, time_t when); + +void rep_hist_note_dir_bytes_read(uint64_t num_bytes, time_t when); +void rep_hist_note_dir_bytes_written(uint64_t num_bytes, time_t when); + +MOCK_DECL(int, rep_hist_bandwidth_assess, (void)); +char *rep_hist_get_bandwidth_lines(void); +void rep_hist_update_state(or_state_t *state); +int rep_hist_load_state(or_state_t *state, char **err); +void rep_history_clean(time_t before); + +void rep_hist_note_router_reachable(const char *id, const tor_addr_t *at_addr, + const uint16_t at_port, time_t when); +void rep_hist_note_router_unreachable(const char *id, time_t when); +int rep_hist_record_mtbf_data(time_t now, int missing_means_down); +int rep_hist_load_mtbf_data(time_t now); + +time_t rep_hist_downrate_old_runs(time_t now); +long rep_hist_get_uptime(const char *id, time_t when); +double rep_hist_get_stability(const char *id, time_t when); +double rep_hist_get_weighted_fractional_uptime(const char *id, time_t when); +long rep_hist_get_weighted_time_known(const char *id, time_t when); +int rep_hist_have_measured_enough_stability(void); + +void predicted_ports_init(void); +void rep_hist_note_used_port(time_t now, uint16_t port); +smartlist_t *rep_hist_get_predicted_ports(time_t now); +void rep_hist_remove_predicted_ports(const smartlist_t *rmv_ports); +void rep_hist_note_used_resolve(time_t now); +void rep_hist_note_used_internal(time_t now, int need_uptime, + int need_capacity); +int rep_hist_get_predicted_internal(time_t now, int *need_uptime, + int *need_capacity); + +int any_predicted_circuits(time_t now); +int rep_hist_circbuilding_dormant(time_t now); +int predicted_ports_prediction_time_remaining(time_t now); + +void rep_hist_exit_stats_init(time_t now); +void rep_hist_reset_exit_stats(time_t now); +void rep_hist_exit_stats_term(void); +char *rep_hist_format_exit_stats(time_t now); +time_t rep_hist_exit_stats_write(time_t now); +void rep_hist_note_exit_bytes(uint16_t port, size_t num_written, + size_t num_read); +void rep_hist_note_exit_stream_opened(uint16_t port); + +void rep_hist_buffer_stats_init(time_t now); +void rep_hist_buffer_stats_add_circ(circuit_t *circ, + time_t end_of_interval); +time_t rep_hist_buffer_stats_write(time_t now); +void rep_hist_buffer_stats_term(void); +void rep_hist_add_buffer_stats(double mean_num_cells_in_queue, + double mean_time_cells_in_queue, uint32_t processed_cells); +char *rep_hist_format_buffer_stats(time_t now); +void rep_hist_reset_buffer_stats(time_t now); + +void rep_hist_desc_stats_init(time_t now); +void rep_hist_note_desc_served(const char * desc); +void rep_hist_desc_stats_term(void); +time_t rep_hist_desc_stats_write(time_t now); + +void rep_hist_conn_stats_init(time_t now); +void rep_hist_note_or_conn_bytes(uint64_t conn_id, size_t num_read, + size_t num_written, time_t when); +void rep_hist_reset_conn_stats(time_t now); +char *rep_hist_format_conn_stats(time_t now); +time_t rep_hist_conn_stats_write(time_t now); +void rep_hist_conn_stats_term(void); + +void rep_hist_note_circuit_handshake_requested(uint16_t type); +void rep_hist_note_circuit_handshake_assigned(uint16_t type); +void rep_hist_log_circuit_handshake_stats(time_t now); + +void rep_hist_hs_stats_init(time_t now); +void rep_hist_hs_stats_term(void); +time_t rep_hist_hs_stats_write(time_t now); +char *rep_hist_get_hs_stats_string(void); +void rep_hist_seen_new_rp_cell(void); +void rep_hist_stored_maybe_new_hs(const crypto_pk_t *pubkey); + +void rep_hist_free_all(void); + +void rep_hist_note_negotiated_link_proto(unsigned link_proto, + int started_here); +void rep_hist_log_link_protocol_counts(void); + +extern uint64_t rephist_total_alloc; +extern uint32_t rephist_total_num; +#ifdef TOR_UNIT_TESTS +extern int onion_handshakes_requested[MAX_ONION_HANDSHAKE_TYPE+1]; +extern int onion_handshakes_assigned[MAX_ONION_HANDSHAKE_TYPE+1]; +typedef struct bw_array_t bw_array_t; +extern bw_array_t *write_array; +#endif + +/** + * Represents the type of a cell for padding accounting + */ +typedef enum padding_type_t { + /** A RELAY_DROP cell */ + PADDING_TYPE_DROP, + /** A CELL_PADDING cell */ + PADDING_TYPE_CELL, + /** Total counts of padding and non-padding together */ + PADDING_TYPE_TOTAL, + /** Total cell counts for all padding-enabled channels */ + PADDING_TYPE_ENABLED_TOTAL, + /** CELL_PADDING counts for all padding-enabled channels */ + PADDING_TYPE_ENABLED_CELL +} padding_type_t; + +/** The amount of time over which the padding cell counts were counted */ +#define REPHIST_CELL_PADDING_COUNTS_INTERVAL (24*60*60) +void rep_hist_padding_count_read(padding_type_t type); +void rep_hist_padding_count_write(padding_type_t type); +char *rep_hist_get_padding_count_lines(void); +void rep_hist_reset_padding_counts(void); +void rep_hist_prep_published_padding_counts(time_t now); +void rep_hist_padding_count_timers(uint64_t num_timers); + +#endif /* !defined(TOR_REPHIST_H) */ + |