tor/src/or/routerparse.c

4933 lines
166 KiB
C

/* Copyright (c) 2001 Matej Pfajfar.
* Copyright (c) 2001-2004, Roger Dingledine.
* Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
* Copyright (c) 2007-2013, The Tor Project, Inc. */
/* See LICENSE for licensing information */
/**
* \file routerparse.c
* \brief Code to parse and validate router descriptors and directories.
**/
#include "or.h"
#include "config.h"
#include "circuitstats.h"
#include "dirserv.h"
#include "dirvote.h"
#include "policies.h"
#include "rendcommon.h"
#include "router.h"
#include "routerlist.h"
#include "memarea.h"
#include "microdesc.h"
#include "networkstatus.h"
#include "rephist.h"
#include "routerparse.h"
#undef log
#include <math.h>
/****************************************************************************/
/** Enumeration of possible token types. The ones starting with K_ correspond
* to directory 'keywords'. ERR_ is an error in the tokenizing process, EOF_
* is an end-of-file marker, and NIL_ is used to encode not-a-token.
*/
typedef enum {
K_ACCEPT = 0,
K_ACCEPT6,
K_DIRECTORY_SIGNATURE,
K_RECOMMENDED_SOFTWARE,
K_REJECT,
K_REJECT6,
K_ROUTER,
K_SIGNED_DIRECTORY,
K_SIGNING_KEY,
K_ONION_KEY,
K_ONION_KEY_NTOR,
K_ROUTER_SIGNATURE,
K_PUBLISHED,
K_RUNNING_ROUTERS,
K_ROUTER_STATUS,
K_PLATFORM,
K_OPT,
K_BANDWIDTH,
K_CONTACT,
K_NETWORK_STATUS,
K_UPTIME,
K_DIR_SIGNING_KEY,
K_FAMILY,
K_FINGERPRINT,
K_HIBERNATING,
K_READ_HISTORY,
K_WRITE_HISTORY,
K_NETWORK_STATUS_VERSION,
K_DIR_SOURCE,
K_DIR_OPTIONS,
K_CLIENT_VERSIONS,
K_SERVER_VERSIONS,
K_OR_ADDRESS,
K_P,
K_P6,
K_R,
K_A,
K_S,
K_V,
K_W,
K_M,
K_EXTRA_INFO,
K_EXTRA_INFO_DIGEST,
K_CACHES_EXTRA_INFO,
K_HIDDEN_SERVICE_DIR,
K_ALLOW_SINGLE_HOP_EXITS,
K_IPV6_POLICY,
K_DIRREQ_END,
K_DIRREQ_V2_IPS,
K_DIRREQ_V3_IPS,
K_DIRREQ_V2_REQS,
K_DIRREQ_V3_REQS,
K_DIRREQ_V2_SHARE,
K_DIRREQ_V3_SHARE,
K_DIRREQ_V2_RESP,
K_DIRREQ_V3_RESP,
K_DIRREQ_V2_DIR,
K_DIRREQ_V3_DIR,
K_DIRREQ_V2_TUN,
K_DIRREQ_V3_TUN,
K_ENTRY_END,
K_ENTRY_IPS,
K_CELL_END,
K_CELL_PROCESSED,
K_CELL_QUEUED,
K_CELL_TIME,
K_CELL_CIRCS,
K_EXIT_END,
K_EXIT_WRITTEN,
K_EXIT_READ,
K_EXIT_OPENED,
K_DIR_KEY_CERTIFICATE_VERSION,
K_DIR_IDENTITY_KEY,
K_DIR_KEY_PUBLISHED,
K_DIR_KEY_EXPIRES,
K_DIR_KEY_CERTIFICATION,
K_DIR_KEY_CROSSCERT,
K_DIR_ADDRESS,
K_VOTE_STATUS,
K_VALID_AFTER,
K_FRESH_UNTIL,
K_VALID_UNTIL,
K_VOTING_DELAY,
K_KNOWN_FLAGS,
K_PARAMS,
K_BW_WEIGHTS,
K_VOTE_DIGEST,
K_CONSENSUS_DIGEST,
K_ADDITIONAL_DIGEST,
K_ADDITIONAL_SIGNATURE,
K_CONSENSUS_METHODS,
K_CONSENSUS_METHOD,
K_LEGACY_DIR_KEY,
K_DIRECTORY_FOOTER,
A_PURPOSE,
A_LAST_LISTED,
A_UNKNOWN_,
R_RENDEZVOUS_SERVICE_DESCRIPTOR,
R_VERSION,
R_PERMANENT_KEY,
R_SECRET_ID_PART,
R_PUBLICATION_TIME,
R_PROTOCOL_VERSIONS,
R_INTRODUCTION_POINTS,
R_SIGNATURE,
R_IPO_IDENTIFIER,
R_IPO_IP_ADDRESS,
R_IPO_ONION_PORT,
R_IPO_ONION_KEY,
R_IPO_SERVICE_KEY,
C_CLIENT_NAME,
C_DESCRIPTOR_COOKIE,
C_CLIENT_KEY,
ERR_,
EOF_,
NIL_
} directory_keyword;
#define MIN_ANNOTATION A_PURPOSE
#define MAX_ANNOTATION A_UNKNOWN_
/** Structure to hold a single directory token.
*
* We parse a directory by breaking it into "tokens", each consisting
* of a keyword, a line full of arguments, and a binary object. The
* arguments and object are both optional, depending on the keyword
* type.
*
* This structure is only allocated in memareas; do not allocate it on
* the heap, or token_clear() won't work.
*/
typedef struct directory_token_t {
directory_keyword tp; /**< Type of the token. */
int n_args:30; /**< Number of elements in args */
char **args; /**< Array of arguments from keyword line. */
char *object_type; /**< -----BEGIN [object_type]-----*/
size_t object_size; /**< Bytes in object_body */
char *object_body; /**< Contents of object, base64-decoded. */
crypto_pk_t *key; /**< For public keys only. Heap-allocated. */
char *error; /**< For ERR_ tokens only. */
} directory_token_t;
/* ********************************************************************** */
/** We use a table of rules to decide how to parse each token type. */
/** Rules for whether the keyword needs an object. */
typedef enum {
NO_OBJ, /**< No object, ever. */
NEED_OBJ, /**< Object is required. */
NEED_SKEY_1024,/**< Object is required, and must be a 1024 bit private key */
NEED_KEY_1024, /**< Object is required, and must be a 1024 bit public key */
NEED_KEY, /**< Object is required, and must be a public key. */
OBJ_OK, /**< Object is optional. */
} obj_syntax;
#define AT_START 1
#define AT_END 2
/** Determines the parsing rules for a single token type. */
typedef struct token_rule_t {
/** The string value of the keyword identifying the type of item. */
const char *t;
/** The corresponding directory_keyword enum. */
directory_keyword v;
/** Minimum number of arguments for this item */
int min_args;
/** Maximum number of arguments for this item */
int max_args;
/** If true, we concatenate all arguments for this item into a single
* string. */
int concat_args;
/** Requirements on object syntax for this item. */
obj_syntax os;
/** Lowest number of times this item may appear in a document. */
int min_cnt;
/** Highest number of times this item may appear in a document. */
int max_cnt;
/** One or more of AT_START/AT_END to limit where the item may appear in a
* document. */
int pos;
/** True iff this token is an annotation. */
int is_annotation;
} token_rule_t;
/*
* Helper macros to define token tables. 's' is a string, 't' is a
* directory_keyword, 'a' is a trio of argument multiplicities, and 'o' is an
* object syntax.
*
*/
/** Appears to indicate the end of a table. */
#define END_OF_TABLE { NULL, NIL_, 0,0,0, NO_OBJ, 0, INT_MAX, 0, 0 }
/** An item with no restrictions: used for obsolete document types */
#define T(s,t,a,o) { s, t, a, o, 0, INT_MAX, 0, 0 }
/** An item with no restrictions on multiplicity or location. */
#define T0N(s,t,a,o) { s, t, a, o, 0, INT_MAX, 0, 0 }
/** An item that must appear exactly once */
#define T1(s,t,a,o) { s, t, a, o, 1, 1, 0, 0 }
/** An item that must appear exactly once, at the start of the document */
#define T1_START(s,t,a,o) { s, t, a, o, 1, 1, AT_START, 0 }
/** An item that must appear exactly once, at the end of the document */
#define T1_END(s,t,a,o) { s, t, a, o, 1, 1, AT_END, 0 }
/** An item that must appear one or more times */
#define T1N(s,t,a,o) { s, t, a, o, 1, INT_MAX, 0, 0 }
/** An item that must appear no more than once */
#define T01(s,t,a,o) { s, t, a, o, 0, 1, 0, 0 }
/** An annotation that must appear no more than once */
#define A01(s,t,a,o) { s, t, a, o, 0, 1, 0, 1 }
/* Argument multiplicity: any number of arguments. */
#define ARGS 0,INT_MAX,0
/* Argument multiplicity: no arguments. */
#define NO_ARGS 0,0,0
/* Argument multiplicity: concatenate all arguments. */
#define CONCAT_ARGS 1,1,1
/* Argument multiplicity: at least <b>n</b> arguments. */
#define GE(n) n,INT_MAX,0
/* Argument multiplicity: exactly <b>n</b> arguments. */
#define EQ(n) n,n,0
/** List of tokens recognized in router descriptors */
static token_rule_t routerdesc_token_table[] = {
T0N("reject", K_REJECT, ARGS, NO_OBJ ),
T0N("accept", K_ACCEPT, ARGS, NO_OBJ ),
T0N("reject6", K_REJECT6, ARGS, NO_OBJ ),
T0N("accept6", K_ACCEPT6, ARGS, NO_OBJ ),
T1_START( "router", K_ROUTER, GE(5), NO_OBJ ),
T01("ipv6-policy", K_IPV6_POLICY, CONCAT_ARGS, NO_OBJ),
T1( "signing-key", K_SIGNING_KEY, NO_ARGS, NEED_KEY_1024 ),
T1( "onion-key", K_ONION_KEY, NO_ARGS, NEED_KEY_1024 ),
T01("ntor-onion-key", K_ONION_KEY_NTOR, GE(1), NO_OBJ ),
T1_END( "router-signature", K_ROUTER_SIGNATURE, NO_ARGS, NEED_OBJ ),
T1( "published", K_PUBLISHED, CONCAT_ARGS, NO_OBJ ),
T01("uptime", K_UPTIME, GE(1), NO_OBJ ),
T01("fingerprint", K_FINGERPRINT, CONCAT_ARGS, NO_OBJ ),
T01("hibernating", K_HIBERNATING, GE(1), NO_OBJ ),
T01("platform", K_PLATFORM, CONCAT_ARGS, NO_OBJ ),
T01("contact", K_CONTACT, CONCAT_ARGS, NO_OBJ ),
T01("read-history", K_READ_HISTORY, ARGS, NO_OBJ ),
T01("write-history", K_WRITE_HISTORY, ARGS, NO_OBJ ),
T01("extra-info-digest", K_EXTRA_INFO_DIGEST, GE(1), NO_OBJ ),
T01("hidden-service-dir", K_HIDDEN_SERVICE_DIR, NO_ARGS, NO_OBJ ),
T01("allow-single-hop-exits",K_ALLOW_SINGLE_HOP_EXITS, NO_ARGS, NO_OBJ ),
T01("family", K_FAMILY, ARGS, NO_OBJ ),
T01("caches-extra-info", K_CACHES_EXTRA_INFO, NO_ARGS, NO_OBJ ),
T0N("or-address", K_OR_ADDRESS, GE(1), NO_OBJ ),
T0N("opt", K_OPT, CONCAT_ARGS, OBJ_OK ),
T1( "bandwidth", K_BANDWIDTH, GE(3), NO_OBJ ),
A01("@purpose", A_PURPOSE, GE(1), NO_OBJ ),
END_OF_TABLE
};
/** List of tokens recognized in extra-info documents. */
static token_rule_t extrainfo_token_table[] = {
T1_END( "router-signature", K_ROUTER_SIGNATURE, NO_ARGS, NEED_OBJ ),
T1( "published", K_PUBLISHED, CONCAT_ARGS, NO_OBJ ),
T0N("opt", K_OPT, CONCAT_ARGS, OBJ_OK ),
T01("read-history", K_READ_HISTORY, ARGS, NO_OBJ ),
T01("write-history", K_WRITE_HISTORY, ARGS, NO_OBJ ),
T01("dirreq-stats-end", K_DIRREQ_END, ARGS, NO_OBJ ),
T01("dirreq-v2-ips", K_DIRREQ_V2_IPS, ARGS, NO_OBJ ),
T01("dirreq-v3-ips", K_DIRREQ_V3_IPS, ARGS, NO_OBJ ),
T01("dirreq-v2-reqs", K_DIRREQ_V2_REQS, ARGS, NO_OBJ ),
T01("dirreq-v3-reqs", K_DIRREQ_V3_REQS, ARGS, NO_OBJ ),
T01("dirreq-v2-share", K_DIRREQ_V2_SHARE, ARGS, NO_OBJ ),
T01("dirreq-v3-share", K_DIRREQ_V3_SHARE, ARGS, NO_OBJ ),
T01("dirreq-v2-resp", K_DIRREQ_V2_RESP, ARGS, NO_OBJ ),
T01("dirreq-v3-resp", K_DIRREQ_V3_RESP, ARGS, NO_OBJ ),
T01("dirreq-v2-direct-dl", K_DIRREQ_V2_DIR, ARGS, NO_OBJ ),
T01("dirreq-v3-direct-dl", K_DIRREQ_V3_DIR, ARGS, NO_OBJ ),
T01("dirreq-v2-tunneled-dl", K_DIRREQ_V2_TUN, ARGS, NO_OBJ ),
T01("dirreq-v3-tunneled-dl", K_DIRREQ_V3_TUN, ARGS, NO_OBJ ),
T01("entry-stats-end", K_ENTRY_END, ARGS, NO_OBJ ),
T01("entry-ips", K_ENTRY_IPS, ARGS, NO_OBJ ),
T01("cell-stats-end", K_CELL_END, ARGS, NO_OBJ ),
T01("cell-processed-cells", K_CELL_PROCESSED, ARGS, NO_OBJ ),
T01("cell-queued-cells", K_CELL_QUEUED, ARGS, NO_OBJ ),
T01("cell-time-in-queue", K_CELL_TIME, ARGS, NO_OBJ ),
T01("cell-circuits-per-decile", K_CELL_CIRCS, ARGS, NO_OBJ ),
T01("exit-stats-end", K_EXIT_END, ARGS, NO_OBJ ),
T01("exit-kibibytes-written", K_EXIT_WRITTEN, ARGS, NO_OBJ ),
T01("exit-kibibytes-read", K_EXIT_READ, ARGS, NO_OBJ ),
T01("exit-streams-opened", K_EXIT_OPENED, ARGS, NO_OBJ ),
T1_START( "extra-info", K_EXTRA_INFO, GE(2), NO_OBJ ),
END_OF_TABLE
};
/** List of tokens recognized in the body part of v3 networkstatus
* documents. */
static token_rule_t rtrstatus_token_table[] = {
T01("p", K_P, CONCAT_ARGS, NO_OBJ ),
T1( "r", K_R, GE(7), NO_OBJ ),
T0N("a", K_A, GE(1), NO_OBJ ),
T1( "s", K_S, ARGS, NO_OBJ ),
T01("v", K_V, CONCAT_ARGS, NO_OBJ ),
T01("w", K_W, ARGS, NO_OBJ ),
T0N("m", K_M, CONCAT_ARGS, NO_OBJ ),
T0N("opt", K_OPT, CONCAT_ARGS, OBJ_OK ),
END_OF_TABLE
};
/** List of tokens common to V3 authority certificates and V3 consensuses. */
#define CERTIFICATE_MEMBERS \
T1("dir-key-certificate-version", K_DIR_KEY_CERTIFICATE_VERSION, \
GE(1), NO_OBJ ), \
T1("dir-identity-key", K_DIR_IDENTITY_KEY, NO_ARGS, NEED_KEY ),\
T1("dir-key-published",K_DIR_KEY_PUBLISHED, CONCAT_ARGS, NO_OBJ), \
T1("dir-key-expires", K_DIR_KEY_EXPIRES, CONCAT_ARGS, NO_OBJ), \
T1("dir-signing-key", K_DIR_SIGNING_KEY, NO_ARGS, NEED_KEY ),\
T1("dir-key-crosscert", K_DIR_KEY_CROSSCERT, NO_ARGS, NEED_OBJ ),\
T1("dir-key-certification", K_DIR_KEY_CERTIFICATION, \
NO_ARGS, NEED_OBJ), \
T01("dir-address", K_DIR_ADDRESS, GE(1), NO_OBJ),
/** List of tokens recognized in V3 authority certificates. */
static token_rule_t dir_key_certificate_table[] = {
CERTIFICATE_MEMBERS
T1("fingerprint", K_FINGERPRINT, CONCAT_ARGS, NO_OBJ ),
END_OF_TABLE
};
/** List of tokens recognized in rendezvous service descriptors */
static token_rule_t desc_token_table[] = {
T1_START("rendezvous-service-descriptor", R_RENDEZVOUS_SERVICE_DESCRIPTOR,
EQ(1), NO_OBJ),
T1("version", R_VERSION, EQ(1), NO_OBJ),
T1("permanent-key", R_PERMANENT_KEY, NO_ARGS, NEED_KEY_1024),
T1("secret-id-part", R_SECRET_ID_PART, EQ(1), NO_OBJ),
T1("publication-time", R_PUBLICATION_TIME, CONCAT_ARGS, NO_OBJ),
T1("protocol-versions", R_PROTOCOL_VERSIONS, EQ(1), NO_OBJ),
T01("introduction-points", R_INTRODUCTION_POINTS, NO_ARGS, NEED_OBJ),
T1_END("signature", R_SIGNATURE, NO_ARGS, NEED_OBJ),
END_OF_TABLE
};
/** List of tokens recognized in the (encrypted) list of introduction points of
* rendezvous service descriptors */
static token_rule_t ipo_token_table[] = {
T1_START("introduction-point", R_IPO_IDENTIFIER, EQ(1), NO_OBJ),
T1("ip-address", R_IPO_IP_ADDRESS, EQ(1), NO_OBJ),
T1("onion-port", R_IPO_ONION_PORT, EQ(1), NO_OBJ),
T1("onion-key", R_IPO_ONION_KEY, NO_ARGS, NEED_KEY_1024),
T1("service-key", R_IPO_SERVICE_KEY, NO_ARGS, NEED_KEY_1024),
END_OF_TABLE
};
/** List of tokens recognized in the (possibly encrypted) list of introduction
* points of rendezvous service descriptors */
static token_rule_t client_keys_token_table[] = {
T1_START("client-name", C_CLIENT_NAME, CONCAT_ARGS, NO_OBJ),
T1("descriptor-cookie", C_DESCRIPTOR_COOKIE, EQ(1), NO_OBJ),
T01("client-key", C_CLIENT_KEY, NO_ARGS, NEED_SKEY_1024),
END_OF_TABLE
};
/** List of tokens recognized in V3 networkstatus votes. */
static token_rule_t networkstatus_token_table[] = {
T1_START("network-status-version", K_NETWORK_STATUS_VERSION,
GE(1), NO_OBJ ),
T1("vote-status", K_VOTE_STATUS, GE(1), NO_OBJ ),
T1("published", K_PUBLISHED, CONCAT_ARGS, NO_OBJ ),
T1("valid-after", K_VALID_AFTER, CONCAT_ARGS, NO_OBJ ),
T1("fresh-until", K_FRESH_UNTIL, CONCAT_ARGS, NO_OBJ ),
T1("valid-until", K_VALID_UNTIL, CONCAT_ARGS, NO_OBJ ),
T1("voting-delay", K_VOTING_DELAY, GE(2), NO_OBJ ),
T1("known-flags", K_KNOWN_FLAGS, ARGS, NO_OBJ ),
T01("params", K_PARAMS, ARGS, NO_OBJ ),
T( "fingerprint", K_FINGERPRINT, CONCAT_ARGS, NO_OBJ ),
CERTIFICATE_MEMBERS
T0N("opt", K_OPT, CONCAT_ARGS, OBJ_OK ),
T1( "contact", K_CONTACT, CONCAT_ARGS, NO_OBJ ),
T1( "dir-source", K_DIR_SOURCE, GE(6), NO_OBJ ),
T01("legacy-dir-key", K_LEGACY_DIR_KEY, GE(1), NO_OBJ ),
T1( "known-flags", K_KNOWN_FLAGS, CONCAT_ARGS, NO_OBJ ),
T01("client-versions", K_CLIENT_VERSIONS, CONCAT_ARGS, NO_OBJ ),
T01("server-versions", K_SERVER_VERSIONS, CONCAT_ARGS, NO_OBJ ),
T1( "consensus-methods", K_CONSENSUS_METHODS, GE(1), NO_OBJ ),
END_OF_TABLE
};
/** List of tokens recognized in V3 networkstatus consensuses. */
static token_rule_t networkstatus_consensus_token_table[] = {
T1_START("network-status-version", K_NETWORK_STATUS_VERSION,
GE(1), NO_OBJ ),
T1("vote-status", K_VOTE_STATUS, GE(1), NO_OBJ ),
T1("valid-after", K_VALID_AFTER, CONCAT_ARGS, NO_OBJ ),
T1("fresh-until", K_FRESH_UNTIL, CONCAT_ARGS, NO_OBJ ),
T1("valid-until", K_VALID_UNTIL, CONCAT_ARGS, NO_OBJ ),
T1("voting-delay", K_VOTING_DELAY, GE(2), NO_OBJ ),
T0N("opt", K_OPT, CONCAT_ARGS, OBJ_OK ),
T1N("dir-source", K_DIR_SOURCE, GE(6), NO_OBJ ),
T1N("contact", K_CONTACT, CONCAT_ARGS, NO_OBJ ),
T1N("vote-digest", K_VOTE_DIGEST, GE(1), NO_OBJ ),
T1( "known-flags", K_KNOWN_FLAGS, CONCAT_ARGS, NO_OBJ ),
T01("client-versions", K_CLIENT_VERSIONS, CONCAT_ARGS, NO_OBJ ),
T01("server-versions", K_SERVER_VERSIONS, CONCAT_ARGS, NO_OBJ ),
T01("consensus-method", K_CONSENSUS_METHOD, EQ(1), NO_OBJ),
T01("params", K_PARAMS, ARGS, NO_OBJ ),
END_OF_TABLE
};
/** List of tokens recognized in the footer of v1 directory footers. */
static token_rule_t networkstatus_vote_footer_token_table[] = {
T01("directory-footer", K_DIRECTORY_FOOTER, NO_ARGS, NO_OBJ ),
T01("bandwidth-weights", K_BW_WEIGHTS, ARGS, NO_OBJ ),
T( "directory-signature", K_DIRECTORY_SIGNATURE, GE(2), NEED_OBJ ),
END_OF_TABLE
};
/** List of tokens recognized in detached networkstatus signature documents. */
static token_rule_t networkstatus_detached_signature_token_table[] = {
T1_START("consensus-digest", K_CONSENSUS_DIGEST, GE(1), NO_OBJ ),
T("additional-digest", K_ADDITIONAL_DIGEST,GE(3), NO_OBJ ),
T1("valid-after", K_VALID_AFTER, CONCAT_ARGS, NO_OBJ ),
T1("fresh-until", K_FRESH_UNTIL, CONCAT_ARGS, NO_OBJ ),
T1("valid-until", K_VALID_UNTIL, CONCAT_ARGS, NO_OBJ ),
T("additional-signature", K_ADDITIONAL_SIGNATURE, GE(4), NEED_OBJ ),
T1N("directory-signature", K_DIRECTORY_SIGNATURE, GE(2), NEED_OBJ ),
END_OF_TABLE
};
/** List of tokens recognized in microdescriptors */
static token_rule_t microdesc_token_table[] = {
T1_START("onion-key", K_ONION_KEY, NO_ARGS, NEED_KEY_1024),
T01("ntor-onion-key", K_ONION_KEY_NTOR, GE(1), NO_OBJ ),
T0N("a", K_A, GE(1), NO_OBJ ),
T01("family", K_FAMILY, ARGS, NO_OBJ ),
T01("p", K_P, CONCAT_ARGS, NO_OBJ ),
T01("p6", K_P6, CONCAT_ARGS, NO_OBJ ),
A01("@last-listed", A_LAST_LISTED, CONCAT_ARGS, NO_OBJ ),
END_OF_TABLE
};
#undef T
/* static function prototypes */
static int router_add_exit_policy(routerinfo_t *router,directory_token_t *tok);
static addr_policy_t *router_parse_addr_policy(directory_token_t *tok,
unsigned fmt_flags);
static addr_policy_t *router_parse_addr_policy_private(directory_token_t *tok);
static int router_get_hash_impl(const char *s, size_t s_len, char *digest,
const char *start_str, const char *end_str,
char end_char,
digest_algorithm_t alg);
static int router_get_hashes_impl(const char *s, size_t s_len,
digests_t *digests,
const char *start_str, const char *end_str,
char end_char);
static void token_clear(directory_token_t *tok);
static smartlist_t *find_all_by_keyword(smartlist_t *s, directory_keyword k);
static smartlist_t *find_all_exitpolicy(smartlist_t *s);
static directory_token_t *find_by_keyword_(smartlist_t *s,
directory_keyword keyword,
const char *keyword_str);
#define find_by_keyword(s, keyword) find_by_keyword_((s), (keyword), #keyword)
static directory_token_t *find_opt_by_keyword(smartlist_t *s,
directory_keyword keyword);
#define TS_ANNOTATIONS_OK 1
#define TS_NOCHECK 2
#define TS_NO_NEW_ANNOTATIONS 4
static int tokenize_string(memarea_t *area,
const char *start, const char *end,
smartlist_t *out,
token_rule_t *table,
int flags);
static directory_token_t *get_next_token(memarea_t *area,
const char **s,
const char *eos,
token_rule_t *table);
#define CST_CHECK_AUTHORITY (1<<0)
#define CST_NO_CHECK_OBJTYPE (1<<1)
static int check_signature_token(const char *digest,
ssize_t digest_len,
directory_token_t *tok,
crypto_pk_t *pkey,
int flags,
const char *doctype);
#undef DEBUG_AREA_ALLOC
#ifdef DEBUG_AREA_ALLOC
#define DUMP_AREA(a,name) STMT_BEGIN \
size_t alloc=0, used=0; \
memarea_get_stats((a),&alloc,&used); \
log_debug(LD_MM, "Area for %s has %lu allocated; using %lu.", \
name, (unsigned long)alloc, (unsigned long)used); \
STMT_END
#else
#define DUMP_AREA(a,name) STMT_NIL
#endif
/** Last time we dumped a descriptor to disk. */
static time_t last_desc_dumped = 0;
/** For debugging purposes, dump unparseable descriptor *<b>desc</b> of
* type *<b>type</b> to file $DATADIR/unparseable-desc. Do not write more
* than one descriptor to disk per minute. If there is already such a
* file in the data directory, overwrite it. */
static void
dump_desc(const char *desc, const char *type)
{
time_t now = time(NULL);
tor_assert(desc);
tor_assert(type);
if (!last_desc_dumped || last_desc_dumped + 60 < now) {
char *debugfile = get_datadir_fname("unparseable-desc");
size_t filelen = 50 + strlen(type) + strlen(desc);
char *content = tor_malloc_zero(filelen);
tor_snprintf(content, filelen, "Unable to parse descriptor of type "
"%s:\n%s", type, desc);
write_str_to_file(debugfile, content, 1);
log_info(LD_DIR, "Unable to parse descriptor of type %s. See file "
"unparseable-desc in data directory for details.", type);
tor_free(content);
tor_free(debugfile);
last_desc_dumped = now;
}
}
/** Set <b>digest</b> to the SHA-1 digest of the hash of the directory in
* <b>s</b>. Return 0 on success, -1 on failure.
*/
int
router_get_dir_hash(const char *s, char *digest)
{
return router_get_hash_impl(s, strlen(s), digest,
"signed-directory","\ndirectory-signature",'\n',
DIGEST_SHA1);
}
/** Set <b>digest</b> to the SHA-1 digest of the hash of the first router in
* <b>s</b>. Return 0 on success, -1 on failure.
*/
int
router_get_router_hash(const char *s, size_t s_len, char *digest)
{
return router_get_hash_impl(s, s_len, digest,
"router ","\nrouter-signature", '\n',
DIGEST_SHA1);
}
/** Set <b>digests</b> to all the digests of the consensus document in
* <b>s</b> */
int
router_get_networkstatus_v3_hashes(const char *s, digests_t *digests)
{
return router_get_hashes_impl(s,strlen(s),digests,
"network-status-version",
"\ndirectory-signature",
' ');
}
/** Set <b>digest</b> to the SHA-1 digest of the hash of the <b>s_len</b>-byte
* extrainfo string at <b>s</b>. Return 0 on success, -1 on failure. */
int
router_get_extrainfo_hash(const char *s, size_t s_len, char *digest)
{
return router_get_hash_impl(s, s_len, digest, "extra-info",
"\nrouter-signature",'\n', DIGEST_SHA1);
}
/** Helper: used to generate signatures for routers, directories and
* network-status objects. Given a <b>digest_len</b>-byte digest in
* <b>digest</b> and a secret <b>private_key</b>, generate an PKCS1-padded
* signature, BASE64-encode it, surround it with -----BEGIN/END----- pairs,
* and return the new signature on success or NULL on failure.
*/
char *
router_get_dirobj_signature(const char *digest,
size_t digest_len,
crypto_pk_t *private_key)
{
char *signature;
size_t i, keysize;
int siglen;
char *buf = NULL;
size_t buf_len;
/* overestimate of BEGIN/END lines total len. */
#define BEGIN_END_OVERHEAD_LEN 64
keysize = crypto_pk_keysize(private_key);
signature = tor_malloc(keysize);
siglen = crypto_pk_private_sign(private_key, signature, keysize,
digest, digest_len);
if (siglen < 0) {
log_warn(LD_BUG,"Couldn't sign digest.");
goto err;
}
/* The *2 here is a ridiculous overestimate of base-64 overhead. */
buf_len = (siglen * 2) + BEGIN_END_OVERHEAD_LEN;
buf = tor_malloc(buf_len);
if (strlcpy(buf, "-----BEGIN SIGNATURE-----\n", buf_len) >= buf_len)
goto truncated;
i = strlen(buf);
if (base64_encode(buf+i, buf_len-i, signature, siglen) < 0) {
log_warn(LD_BUG,"couldn't base64-encode signature");
goto err;
}
if (strlcat(buf, "-----END SIGNATURE-----\n", buf_len) >= buf_len)
goto truncated;
tor_free(signature);
return buf;
truncated:
log_warn(LD_BUG,"tried to exceed string length.");
err:
tor_free(signature);
tor_free(buf);
return NULL;
}
/** Helper: used to generate signatures for routers, directories and
* network-status objects. Given a digest in <b>digest</b> and a secret
* <b>private_key</b>, generate a PKCS1-padded signature, BASE64-encode it,
* surround it with -----BEGIN/END----- pairs, and write it to the
* <b>buf_len</b>-byte buffer at <b>buf</b>. Return 0 on success, -1 on
* failure.
*/
int
router_append_dirobj_signature(char *buf, size_t buf_len, const char *digest,
size_t digest_len, crypto_pk_t *private_key)
{
size_t sig_len, s_len;
char *sig = router_get_dirobj_signature(digest, digest_len, private_key);
if (!sig) {
log_warn(LD_BUG, "No signature generated");
return -1;
}
sig_len = strlen(sig);
s_len = strlen(buf);
if (sig_len + s_len + 1 > buf_len) {
log_warn(LD_BUG, "Not enough room for signature");
tor_free(sig);
return -1;
}
memcpy(buf+s_len, sig, sig_len+1);
tor_free(sig);
return 0;
}
/** Return VS_RECOMMENDED if <b>myversion</b> is contained in
* <b>versionlist</b>. Else, return VS_EMPTY if versionlist has no
* entries. Else, return VS_OLD if every member of
* <b>versionlist</b> is newer than <b>myversion</b>. Else, return
* VS_NEW_IN_SERIES if there is at least one member of <b>versionlist</b> in
* the same series (major.minor.micro) as <b>myversion</b>, but no such member
* is newer than <b>myversion.</b>. Else, return VS_NEW if every member of
* <b>versionlist</b> is older than <b>myversion</b>. Else, return
* VS_UNRECOMMENDED.
*
* (versionlist is a comma-separated list of version strings,
* optionally prefixed with "Tor". Versions that can't be parsed are
* ignored.)
*/
version_status_t
tor_version_is_obsolete(const char *myversion, const char *versionlist)
{
tor_version_t mine, other;
int found_newer = 0, found_older = 0, found_newer_in_series = 0,
found_any_in_series = 0, r, same;
version_status_t ret = VS_UNRECOMMENDED;
smartlist_t *version_sl;
log_debug(LD_CONFIG,"Checking whether version '%s' is in '%s'",
myversion, versionlist);
if (tor_version_parse(myversion, &mine)) {
log_err(LD_BUG,"I couldn't parse my own version (%s)", myversion);
tor_assert(0);
}
version_sl = smartlist_new();
smartlist_split_string(version_sl, versionlist, ",", SPLIT_SKIP_SPACE, 0);
if (!strlen(versionlist)) { /* no authorities cared or agreed */
ret = VS_EMPTY;
goto done;
}
SMARTLIST_FOREACH_BEGIN(version_sl, const char *, cp) {
if (!strcmpstart(cp, "Tor "))
cp += 4;
if (tor_version_parse(cp, &other)) {
/* Couldn't parse other; it can't be a match. */
} else {
same = tor_version_same_series(&mine, &other);
if (same)
found_any_in_series = 1;
r = tor_version_compare(&mine, &other);
if (r==0) {
ret = VS_RECOMMENDED;
goto done;
} else if (r<0) {
found_newer = 1;
if (same)
found_newer_in_series = 1;
} else if (r>0) {
found_older = 1;
}
}
} SMARTLIST_FOREACH_END(cp);
/* We didn't find the listed version. Is it new or old? */
if (found_any_in_series && !found_newer_in_series && found_newer) {
ret = VS_NEW_IN_SERIES;
} else if (found_newer && !found_older) {
ret = VS_OLD;
} else if (found_older && !found_newer) {
ret = VS_NEW;
} else {
ret = VS_UNRECOMMENDED;
}
done:
SMARTLIST_FOREACH(version_sl, char *, version, tor_free(version));
smartlist_free(version_sl);
return ret;
}
/** Return true iff <b>key</b> is allowed to sign directories.
*/
static int
dir_signing_key_is_trusted(crypto_pk_t *key)
{
char digest[DIGEST_LEN];
if (!key) return 0;
if (crypto_pk_get_digest(key, digest) < 0) {
log_warn(LD_DIR, "Error computing dir-signing-key digest");
return 0;
}
if (!router_digest_is_trusted_dir(digest)) {
log_warn(LD_DIR, "Listed dir-signing-key is not trusted");
return 0;
}
return 1;
}
/** Check whether the object body of the token in <b>tok</b> has a good
* signature for <b>digest</b> using key <b>pkey</b>. If
* <b>CST_CHECK_AUTHORITY</b> is set, make sure that <b>pkey</b> is the key of
* a directory authority. If <b>CST_NO_CHECK_OBJTYPE</b> is set, do not check
* the object type of the signature object. Use <b>doctype</b> as the type of
* the document when generating log messages. Return 0 on success, negative
* on failure.
*/
static int
check_signature_token(const char *digest,
ssize_t digest_len,
directory_token_t *tok,
crypto_pk_t *pkey,
int flags,
const char *doctype)
{
char *signed_digest;
size_t keysize;
const int check_authority = (flags & CST_CHECK_AUTHORITY);
const int check_objtype = ! (flags & CST_NO_CHECK_OBJTYPE);
tor_assert(pkey);
tor_assert(tok);
tor_assert(digest);
tor_assert(doctype);
if (check_authority && !dir_signing_key_is_trusted(pkey)) {
log_warn(LD_DIR, "Key on %s did not come from an authority; rejecting",
doctype);
return -1;
}
if (check_objtype) {
if (strcmp(tok->object_type, "SIGNATURE")) {
log_warn(LD_DIR, "Bad object type on %s signature", doctype);
return -1;
}
}
keysize = crypto_pk_keysize(pkey);
signed_digest = tor_malloc(keysize);
if (crypto_pk_public_checksig(pkey, signed_digest, keysize,
tok->object_body, tok->object_size)
< digest_len) {
log_warn(LD_DIR, "Error reading %s: invalid signature.", doctype);
tor_free(signed_digest);
return -1;
}
// log_debug(LD_DIR,"Signed %s hash starts %s", doctype,
// hex_str(signed_digest,4));
if (tor_memneq(digest, signed_digest, digest_len)) {
log_warn(LD_DIR, "Error reading %s: signature does not match.", doctype);
tor_free(signed_digest);
return -1;
}
tor_free(signed_digest);
return 0;
}
/** Helper: move *<b>s_ptr</b> ahead to the next router, the next extra-info,
* or to the first of the annotations proceeding the next router or
* extra-info---whichever comes first. Set <b>is_extrainfo_out</b> to true if
* we found an extrainfo, or false if found a router. Do not scan beyond
* <b>eos</b>. Return -1 if we found nothing; 0 if we found something. */
static int
find_start_of_next_router_or_extrainfo(const char **s_ptr,
const char *eos,
int *is_extrainfo_out)
{
const char *annotations = NULL;
const char *s = *s_ptr;
s = eat_whitespace_eos(s, eos);
while (s < eos-32) { /* 32 gives enough room for a the first keyword. */
/* We're at the start of a line. */
tor_assert(*s != '\n');
if (*s == '@' && !annotations) {
annotations = s;
} else if (*s == 'r' && !strcmpstart(s, "router ")) {
*s_ptr = annotations ? annotations : s;
*is_extrainfo_out = 0;
return 0;
} else if (*s == 'e' && !strcmpstart(s, "extra-info ")) {
*s_ptr = annotations ? annotations : s;
*is_extrainfo_out = 1;
return 0;
}
if (!(s = memchr(s+1, '\n', eos-(s+1))))
break;
s = eat_whitespace_eos(s, eos);
}
return -1;
}
/** Given a string *<b>s</b> containing a concatenated sequence of router
* descriptors (or extra-info documents if <b>is_extrainfo</b> is set), parses
* them and stores the result in <b>dest</b>. All routers are marked running
* and valid. Advances *s to a point immediately following the last router
* entry. Ignore any trailing router entries that are not complete.
*
* If <b>saved_location</b> isn't SAVED_IN_CACHE, make a local copy of each
* descriptor in the signed_descriptor_body field of each routerinfo_t. If it
* isn't SAVED_NOWHERE, remember the offset of each descriptor.
*
* Returns 0 on success and -1 on failure.
*/
int
router_parse_list_from_string(const char **s, const char *eos,
smartlist_t *dest,
saved_location_t saved_location,
int want_extrainfo,
int allow_annotations,
const char *prepend_annotations)
{
routerinfo_t *router;
extrainfo_t *extrainfo;
signed_descriptor_t *signed_desc = NULL;
void *elt;
const char *end, *start;
int have_extrainfo;
tor_assert(s);
tor_assert(*s);
tor_assert(dest);
start = *s;
if (!eos)
eos = *s + strlen(*s);
tor_assert(eos >= *s);
while (1) {
if (find_start_of_next_router_or_extrainfo(s, eos, &have_extrainfo) < 0)
break;
end = tor_memstr(*s, eos-*s, "\nrouter-signature");
if (end)
end = tor_memstr(end, eos-end, "\n-----END SIGNATURE-----\n");
if (end)
end += strlen("\n-----END SIGNATURE-----\n");
if (!end)
break;
elt = NULL;
if (have_extrainfo && want_extrainfo) {
routerlist_t *rl = router_get_routerlist();
extrainfo = extrainfo_parse_entry_from_string(*s, end,
saved_location != SAVED_IN_CACHE,
rl->identity_map);
if (extrainfo) {
signed_desc = &extrainfo->cache_info;
elt = extrainfo;
}
} else if (!have_extrainfo && !want_extrainfo) {
router = router_parse_entry_from_string(*s, end,
saved_location != SAVED_IN_CACHE,
allow_annotations,
prepend_annotations);
if (router) {
log_debug(LD_DIR, "Read router '%s', purpose '%s'",
router_describe(router),
router_purpose_to_string(router->purpose));
signed_desc = &router->cache_info;
elt = router;
}
}
if (!elt) {
*s = end;
continue;
}
if (saved_location != SAVED_NOWHERE) {
tor_assert(signed_desc);
signed_desc->saved_location = saved_location;
signed_desc->saved_offset = *s - start;
}
*s = end;
smartlist_add(dest, elt);
}
return 0;
}
/* For debugging: define to count every descriptor digest we've seen so we
* know if we need to try harder to avoid duplicate verifies. */
#undef COUNT_DISTINCT_DIGESTS
#ifdef COUNT_DISTINCT_DIGESTS
static digestmap_t *verified_digests = NULL;
#endif
/** Log the total count of the number of distinct router digests we've ever
* verified. When compared to the number of times we've verified routerdesc
* signatures <i>in toto</i>, this will tell us if we're doing too much
* multiple-verification. */
void
dump_distinct_digest_count(int severity)
{
#ifdef COUNT_DISTINCT_DIGESTS
if (!verified_digests)
verified_digests = digestmap_new();
tor_log(severity, LD_GENERAL, "%d *distinct* router digests verified",
digestmap_size(verified_digests));
#else
(void)severity; /* suppress "unused parameter" warning */
#endif
}
/** Try to find an IPv6 OR port in <b>list</b> of directory_token_t's
* with at least one argument (use GE(1) in setup). If found, store
* address and port number to <b>addr_out</b> and
* <b>port_out</b>. Return number of OR ports found. */
static int
find_single_ipv6_orport(const smartlist_t *list,
tor_addr_t *addr_out,
uint16_t *port_out)
{
int ret = 0;
tor_assert(list != NULL);
tor_assert(addr_out != NULL);
tor_assert(port_out != NULL);
SMARTLIST_FOREACH_BEGIN(list, directory_token_t *, t) {
tor_addr_t a;
maskbits_t bits;
uint16_t port_min, port_max;
tor_assert(t->n_args >= 1);
/* XXXX Prop186 the full spec allows much more than this. */
if (tor_addr_parse_mask_ports(t->args[0], 0,
&a, &bits, &port_min,
&port_max) == AF_INET6 &&
bits == 128 &&
port_min == port_max) {
/* Okay, this is one we can understand. Use it and ignore
any potential more addresses in list. */
tor_addr_copy(addr_out, &a);
*port_out = port_min;
ret = 1;
break;
}
} SMARTLIST_FOREACH_END(t);
return ret;
}
/** Helper function: reads a single router entry from *<b>s</b> ...
* *<b>end</b>. Mallocs a new router and returns it if all goes well, else
* returns NULL. If <b>cache_copy</b> is true, duplicate the contents of
* s through end into the signed_descriptor_body of the resulting
* routerinfo_t.
*
* If <b>end</b> is NULL, <b>s</b> must be properly NUL-terminated.
*
* If <b>allow_annotations</b>, it's okay to encounter annotations in <b>s</b>
* before the router; if it's false, reject the router if it's annotated. If
* <b>prepend_annotations</b> is set, it should contain some annotations:
* append them to the front of the router before parsing it, and keep them
* around when caching the router.
*
* Only one of allow_annotations and prepend_annotations may be set.
*/
routerinfo_t *
router_parse_entry_from_string(const char *s, const char *end,
int cache_copy, int allow_annotations,
const char *prepend_annotations)
{
routerinfo_t *router = NULL;
char digest[128];
smartlist_t *tokens = NULL, *exit_policy_tokens = NULL;
directory_token_t *tok;
struct in_addr in;
const char *start_of_annotations, *cp, *s_dup = s;
size_t prepend_len = prepend_annotations ? strlen(prepend_annotations) : 0;
int ok = 1;
memarea_t *area = NULL;
tor_assert(!allow_annotations || !prepend_annotations);
if (!end) {
end = s + strlen(s);
}
/* point 'end' to a point immediately after the final newline. */
while (end > s+2 && *(end-1) == '\n' && *(end-2) == '\n')
--end;
area = memarea_new();
tokens = smartlist_new();
if (prepend_annotations) {
if (tokenize_string(area,prepend_annotations,NULL,tokens,
routerdesc_token_table,TS_NOCHECK)) {
log_warn(LD_DIR, "Error tokenizing router descriptor (annotations).");
goto err;
}
}
start_of_annotations = s;
cp = tor_memstr(s, end-s, "\nrouter ");
if (!cp) {
if (end-s < 7 || strcmpstart(s, "router ")) {
log_warn(LD_DIR, "No router keyword found.");
goto err;
}
} else {
s = cp+1;
}
if (start_of_annotations != s) { /* We have annotations */
if (allow_annotations) {
if (tokenize_string(area,start_of_annotations,s,tokens,
routerdesc_token_table,TS_NOCHECK)) {
log_warn(LD_DIR, "Error tokenizing router descriptor (annotations).");
goto err;
}
} else {
log_warn(LD_DIR, "Found unexpected annotations on router descriptor not "
"loaded from disk. Dropping it.");
goto err;
}
}
if (router_get_router_hash(s, end - s, digest) < 0) {
log_warn(LD_DIR, "Couldn't compute router hash.");
goto err;
}
{
int flags = 0;
if (allow_annotations)
flags |= TS_ANNOTATIONS_OK;
if (prepend_annotations)
flags |= TS_ANNOTATIONS_OK|TS_NO_NEW_ANNOTATIONS;
if (tokenize_string(area,s,end,tokens,routerdesc_token_table, flags)) {
log_warn(LD_DIR, "Error tokenizing router descriptor.");
goto err;
}
}
if (smartlist_len(tokens) < 2) {
log_warn(LD_DIR, "Impossibly short router descriptor.");
goto err;
}
tok = find_by_keyword(tokens, K_ROUTER);
tor_assert(tok->n_args >= 5);
router = tor_malloc_zero(sizeof(routerinfo_t));
router->cache_info.routerlist_index = -1;
router->cache_info.annotations_len = s-start_of_annotations + prepend_len;
router->cache_info.signed_descriptor_len = end-s;
if (cache_copy) {
size_t len = router->cache_info.signed_descriptor_len +
router->cache_info.annotations_len;
char *cp =
router->cache_info.signed_descriptor_body = tor_malloc(len+1);
if (prepend_annotations) {
memcpy(cp, prepend_annotations, prepend_len);
cp += prepend_len;
}
/* This assertion will always succeed.
* len == signed_desc_len + annotations_len
* == end-s + s-start_of_annotations + prepend_len
* == end-start_of_annotations + prepend_len
* We already wrote prepend_len bytes into the buffer; now we're
* writing end-start_of_annotations -NM. */
tor_assert(cp+(end-start_of_annotations) ==
router->cache_info.signed_descriptor_body+len);
memcpy(cp, start_of_annotations, end-start_of_annotations);
router->cache_info.signed_descriptor_body[len] = '\0';
tor_assert(strlen(router->cache_info.signed_descriptor_body) == len);
}
memcpy(router->cache_info.signed_descriptor_digest, digest, DIGEST_LEN);
router->nickname = tor_strdup(tok->args[0]);
if (!is_legal_nickname(router->nickname)) {
log_warn(LD_DIR,"Router nickname is invalid");
goto err;
}
if (!tor_inet_aton(tok->args[1], &in)) {
log_warn(LD_DIR,"Router address is not an IP address.");
goto err;
}
router->addr = ntohl(in.s_addr);
router->or_port =
(uint16_t) tor_parse_long(tok->args[2],10,0,65535,&ok,NULL);
if (!ok) {
log_warn(LD_DIR,"Invalid OR port %s", escaped(tok->args[2]));
goto err;
}
router->dir_port =
(uint16_t) tor_parse_long(tok->args[4],10,0,65535,&ok,NULL);
if (!ok) {
log_warn(LD_DIR,"Invalid dir port %s", escaped(tok->args[4]));
goto err;
}
tok = find_by_keyword(tokens, K_BANDWIDTH);
tor_assert(tok->n_args >= 3);
router->bandwidthrate = (int)
tor_parse_long(tok->args[0],10,1,INT_MAX,&ok,NULL);
if (!ok) {
log_warn(LD_DIR, "bandwidthrate %s unreadable or 0. Failing.",
escaped(tok->args[0]));
goto err;
}
router->bandwidthburst =
(int) tor_parse_long(tok->args[1],10,0,INT_MAX,&ok,NULL);
if (!ok) {
log_warn(LD_DIR, "Invalid bandwidthburst %s", escaped(tok->args[1]));
goto err;
}
router->bandwidthcapacity = (int)
tor_parse_long(tok->args[2],10,0,INT_MAX,&ok,NULL);
if (!ok) {
log_warn(LD_DIR, "Invalid bandwidthcapacity %s", escaped(tok->args[1]));
goto err;
}
if ((tok = find_opt_by_keyword(tokens, A_PURPOSE))) {
tor_assert(tok->n_args);
router->purpose = router_purpose_from_string(tok->args[0]);
} else {
router->purpose = ROUTER_PURPOSE_GENERAL;
}
router->cache_info.send_unencrypted =
(router->purpose == ROUTER_PURPOSE_GENERAL) ? 1 : 0;
if ((tok = find_opt_by_keyword(tokens, K_UPTIME))) {
tor_assert(tok->n_args >= 1);
router->uptime = tor_parse_long(tok->args[0],10,0,LONG_MAX,&ok,NULL);
if (!ok) {
log_warn(LD_DIR, "Invalid uptime %s", escaped(tok->args[0]));
goto err;
}
}
if ((tok = find_opt_by_keyword(tokens, K_HIBERNATING))) {
tor_assert(tok->n_args >= 1);
router->is_hibernating
= (tor_parse_long(tok->args[0],10,0,LONG_MAX,NULL,NULL) != 0);
}
tok = find_by_keyword(tokens, K_PUBLISHED);
tor_assert(tok->n_args == 1);
if (parse_iso_time(tok->args[0], &router->cache_info.published_on) < 0)
goto err;
tok = find_by_keyword(tokens, K_ONION_KEY);
if (!crypto_pk_public_exponent_ok(tok->key)) {
log_warn(LD_DIR,
"Relay's onion key had invalid exponent.");
goto err;
}
router->onion_pkey = tok->key;
tok->key = NULL; /* Prevent free */
if ((tok = find_opt_by_keyword(tokens, K_ONION_KEY_NTOR))) {
curve25519_public_key_t k;
tor_assert(tok->n_args >= 1);
if (curve25519_public_from_base64(&k, tok->args[0]) < 0) {
log_warn(LD_DIR, "Bogus ntor-onion-key in routerinfo");
goto err;
}
router->onion_curve25519_pkey =
tor_memdup(&k, sizeof(curve25519_public_key_t));
}
tok = find_by_keyword(tokens, K_SIGNING_KEY);
router->identity_pkey = tok->key;
tok->key = NULL; /* Prevent free */
if (crypto_pk_get_digest(router->identity_pkey,
router->cache_info.identity_digest)) {
log_warn(LD_DIR, "Couldn't calculate key digest"); goto err;
}
if ((tok = find_opt_by_keyword(tokens, K_FINGERPRINT))) {
/* If there's a fingerprint line, it must match the identity digest. */
char d[DIGEST_LEN];
tor_assert(tok->n_args == 1);
tor_strstrip(tok->args[0], " ");
if (base16_decode(d, DIGEST_LEN, tok->args[0], strlen(tok->args[0]))) {
log_warn(LD_DIR, "Couldn't decode router fingerprint %s",
escaped(tok->args[0]));
goto err;
}
if (tor_memneq(d,router->cache_info.identity_digest, DIGEST_LEN)) {
log_warn(LD_DIR, "Fingerprint '%s' does not match identity digest.",
tok->args[0]);
goto err;
}
}
if ((tok = find_opt_by_keyword(tokens, K_PLATFORM))) {
router->platform = tor_strdup(tok->args[0]);
}
if ((tok = find_opt_by_keyword(tokens, K_CONTACT))) {
router->contact_info = tor_strdup(tok->args[0]);
}
if (find_opt_by_keyword(tokens, K_REJECT6) ||
find_opt_by_keyword(tokens, K_ACCEPT6)) {
log_warn(LD_DIR, "Rejecting router with reject6/accept6 line: they crash "
"older Tors.");
goto err;
}
{
smartlist_t *or_addresses = find_all_by_keyword(tokens, K_OR_ADDRESS);
if (or_addresses) {
find_single_ipv6_orport(or_addresses, &router->ipv6_addr,
&router->ipv6_orport);
smartlist_free(or_addresses);
}
}
exit_policy_tokens = find_all_exitpolicy(tokens);
if (!smartlist_len(exit_policy_tokens)) {
log_warn(LD_DIR, "No exit policy tokens in descriptor.");
goto err;
}
SMARTLIST_FOREACH(exit_policy_tokens, directory_token_t *, t,
if (router_add_exit_policy(router,t)<0) {
log_warn(LD_DIR,"Error in exit policy");
goto err;
});
policy_expand_private(&router->exit_policy);
if ((tok = find_opt_by_keyword(tokens, K_IPV6_POLICY)) && tok->n_args) {
router->ipv6_exit_policy = parse_short_policy(tok->args[0]);
if (! router->ipv6_exit_policy) {
log_warn(LD_DIR , "Error in ipv6-policy %s", escaped(tok->args[0]));
goto err;
}
}
if (policy_is_reject_star(router->exit_policy, AF_INET) &&
(!router->ipv6_exit_policy ||
short_policy_is_reject_star(router->ipv6_exit_policy)))
router->policy_is_reject_star = 1;
if ((tok = find_opt_by_keyword(tokens, K_FAMILY)) && tok->n_args) {
int i;
router->declared_family = smartlist_new();
for (i=0;i<tok->n_args;++i) {
if (!is_legal_nickname_or_hexdigest(tok->args[i])) {
log_warn(LD_DIR, "Illegal nickname %s in family line",
escaped(tok->args[i]));
goto err;
}
smartlist_add(router->declared_family, tor_strdup(tok->args[i]));
}
}
if (find_opt_by_keyword(tokens, K_CACHES_EXTRA_INFO))
router->caches_extra_info = 1;
if (find_opt_by_keyword(tokens, K_ALLOW_SINGLE_HOP_EXITS))
router->allow_single_hop_exits = 1;
if ((tok = find_opt_by_keyword(tokens, K_EXTRA_INFO_DIGEST))) {
tor_assert(tok->n_args >= 1);
if (strlen(tok->args[0]) == HEX_DIGEST_LEN) {
base16_decode(router->cache_info.extra_info_digest,
DIGEST_LEN, tok->args[0], HEX_DIGEST_LEN);
} else {
log_warn(LD_DIR, "Invalid extra info digest %s", escaped(tok->args[0]));
}
}
if (find_opt_by_keyword(tokens, K_HIDDEN_SERVICE_DIR)) {
router->wants_to_be_hs_dir = 1;
}
tok = find_by_keyword(tokens, K_ROUTER_SIGNATURE);
note_crypto_pk_op(VERIFY_RTR);
#ifdef COUNT_DISTINCT_DIGESTS
if (!verified_digests)
verified_digests = digestmap_new();
digestmap_set(verified_digests, signed_digest, (void*)(uintptr_t)1);
#endif
if (check_signature_token(digest, DIGEST_LEN, tok, router->identity_pkey, 0,
"router descriptor") < 0)
goto err;
if (!router->or_port) {
log_warn(LD_DIR,"or_port unreadable or 0. Failing.");
goto err;
}
if (!router->platform) {
router->platform = tor_strdup("<unknown>");
}
goto done;
err:
dump_desc(s_dup, "router descriptor");
routerinfo_free(router);
router = NULL;
done:
if (tokens) {
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_free(tokens);
}
smartlist_free(exit_policy_tokens);
if (area) {
DUMP_AREA(area, "routerinfo");
memarea_drop_all(area);
}
return router;
}
/** Parse a single extrainfo entry from the string <b>s</b>, ending at
* <b>end</b>. (If <b>end</b> is NULL, parse up to the end of <b>s</b>.) If
* <b>cache_copy</b> is true, make a copy of the extra-info document in the
* cache_info fields of the result. If <b>routermap</b> is provided, use it
* as a map from router identity to routerinfo_t when looking up signing keys.
*/
extrainfo_t *
extrainfo_parse_entry_from_string(const char *s, const char *end,
int cache_copy, struct digest_ri_map_t *routermap)
{
extrainfo_t *extrainfo = NULL;
char digest[128];
smartlist_t *tokens = NULL;
directory_token_t *tok;
crypto_pk_t *key = NULL;
routerinfo_t *router = NULL;
memarea_t *area = NULL;
const char *s_dup = s;
if (!end) {
end = s + strlen(s);
}
/* point 'end' to a point immediately after the final newline. */
while (end > s+2 && *(end-1) == '\n' && *(end-2) == '\n')
--end;
if (router_get_extrainfo_hash(s, end-s, digest) < 0) {
log_warn(LD_DIR, "Couldn't compute router hash.");
goto err;
}
tokens = smartlist_new();
area = memarea_new();
if (tokenize_string(area,s,end,tokens,extrainfo_token_table,0)) {
log_warn(LD_DIR, "Error tokenizing extra-info document.");
goto err;
}
if (smartlist_len(tokens) < 2) {
log_warn(LD_DIR, "Impossibly short extra-info document.");
goto err;
}
tok = smartlist_get(tokens,0);
if (tok->tp != K_EXTRA_INFO) {
log_warn(LD_DIR,"Entry does not start with \"extra-info\"");
goto err;
}
extrainfo = tor_malloc_zero(sizeof(extrainfo_t));
extrainfo->cache_info.is_extrainfo = 1;
if (cache_copy)
extrainfo->cache_info.signed_descriptor_body = tor_memdup_nulterm(s,end-s);
extrainfo->cache_info.signed_descriptor_len = end-s;
memcpy(extrainfo->cache_info.signed_descriptor_digest, digest, DIGEST_LEN);
tor_assert(tok->n_args >= 2);
if (!is_legal_nickname(tok->args[0])) {
log_warn(LD_DIR,"Bad nickname %s on \"extra-info\"",escaped(tok->args[0]));
goto err;
}
strlcpy(extrainfo->nickname, tok->args[0], sizeof(extrainfo->nickname));
if (strlen(tok->args[1]) != HEX_DIGEST_LEN ||
base16_decode(extrainfo->cache_info.identity_digest, DIGEST_LEN,
tok->args[1], HEX_DIGEST_LEN)) {
log_warn(LD_DIR,"Invalid fingerprint %s on \"extra-info\"",
escaped(tok->args[1]));
goto err;
}
tok = find_by_keyword(tokens, K_PUBLISHED);
if (parse_iso_time(tok->args[0], &extrainfo->cache_info.published_on)) {
log_warn(LD_DIR,"Invalid published time %s on \"extra-info\"",
escaped(tok->args[0]));
goto err;
}
if (routermap &&
(router = digestmap_get((digestmap_t*)routermap,
extrainfo->cache_info.identity_digest))) {
key = router->identity_pkey;
}
tok = find_by_keyword(tokens, K_ROUTER_SIGNATURE);
if (strcmp(tok->object_type, "SIGNATURE") ||
tok->object_size < 128 || tok->object_size > 512) {
log_warn(LD_DIR, "Bad object type or length on extra-info signature");
goto err;
}
if (key) {
note_crypto_pk_op(VERIFY_RTR);
if (check_signature_token(digest, DIGEST_LEN, tok, key, 0,
"extra-info") < 0)
goto err;
if (router)
extrainfo->cache_info.send_unencrypted =
router->cache_info.send_unencrypted;
} else {
extrainfo->pending_sig = tor_memdup(tok->object_body,
tok->object_size);
extrainfo->pending_sig_len = tok->object_size;
}
goto done;
err:
dump_desc(s_dup, "extra-info descriptor");
extrainfo_free(extrainfo);
extrainfo = NULL;
done:
if (tokens) {
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_free(tokens);
}
if (area) {
DUMP_AREA(area, "extrainfo");
memarea_drop_all(area);
}
return extrainfo;
}
/** Parse a key certificate from <b>s</b>; point <b>end-of-string</b> to
* the first character after the certificate. */
authority_cert_t *
authority_cert_parse_from_string(const char *s, const char **end_of_string)
{
/** Reject any certificate at least this big; it is probably an overflow, an
* attack, a bug, or some other nonsense. */
#define MAX_CERT_SIZE (128*1024)
authority_cert_t *cert = NULL, *old_cert;
smartlist_t *tokens = NULL;
char digest[DIGEST_LEN];
directory_token_t *tok;
char fp_declared[DIGEST_LEN];
char *eos;
size_t len;
int found;
memarea_t *area = NULL;
const char *s_dup = s;
s = eat_whitespace(s);
eos = strstr(s, "\ndir-key-certification");
if (! eos) {
log_warn(LD_DIR, "No signature found on key certificate");
return NULL;
}
eos = strstr(eos, "\n-----END SIGNATURE-----\n");
if (! eos) {
log_warn(LD_DIR, "No end-of-signature found on key certificate");
return NULL;
}
eos = strchr(eos+2, '\n');
tor_assert(eos);
++eos;
len = eos - s;
if (len > MAX_CERT_SIZE) {
log_warn(LD_DIR, "Certificate is far too big (at %lu bytes long); "
"rejecting", (unsigned long)len);
return NULL;
}
tokens = smartlist_new();
area = memarea_new();
if (tokenize_string(area,s, eos, tokens, dir_key_certificate_table, 0) < 0) {
log_warn(LD_DIR, "Error tokenizing key certificate");
goto err;
}
if (router_get_hash_impl(s, strlen(s), digest, "dir-key-certificate-version",
"\ndir-key-certification", '\n', DIGEST_SHA1) < 0)
goto err;
tok = smartlist_get(tokens, 0);
if (tok->tp != K_DIR_KEY_CERTIFICATE_VERSION || strcmp(tok->args[0], "3")) {
log_warn(LD_DIR,
"Key certificate does not begin with a recognized version (3).");
goto err;
}
cert = tor_malloc_zero(sizeof(authority_cert_t));
memcpy(cert->cache_info.signed_descriptor_digest, digest, DIGEST_LEN);
tok = find_by_keyword(tokens, K_DIR_SIGNING_KEY);
tor_assert(tok->key);
cert->signing_key = tok->key;
tok->key = NULL;
if (crypto_pk_get_digest(cert->signing_key, cert->signing_key_digest))
goto err;
tok = find_by_keyword(tokens, K_DIR_IDENTITY_KEY);
tor_assert(tok->key);
cert->identity_key = tok->key;
tok->key = NULL;
tok = find_by_keyword(tokens, K_FINGERPRINT);
tor_assert(tok->n_args);
if (base16_decode(fp_declared, DIGEST_LEN, tok->args[0],
strlen(tok->args[0]))) {
log_warn(LD_DIR, "Couldn't decode key certificate fingerprint %s",
escaped(tok->args[0]));
goto err;
}
if (crypto_pk_get_digest(cert->identity_key,
cert->cache_info.identity_digest))
goto err;
if (tor_memneq(cert->cache_info.identity_digest, fp_declared, DIGEST_LEN)) {
log_warn(LD_DIR, "Digest of certificate key didn't match declared "
"fingerprint");
goto err;
}
tok = find_opt_by_keyword(tokens, K_DIR_ADDRESS);
if (tok) {
struct in_addr in;
char *address = NULL;
tor_assert(tok->n_args);
/* XXX024 use some tor_addr parse function below instead. -RD */
if (tor_addr_port_split(LOG_WARN, tok->args[0], &address,
&cert->dir_port) < 0 ||
tor_inet_aton(address, &in) == 0) {
log_warn(LD_DIR, "Couldn't parse dir-address in certificate");
tor_free(address);
goto err;
}
cert->addr = ntohl(in.s_addr);
tor_free(address);
}
tok = find_by_keyword(tokens, K_DIR_KEY_PUBLISHED);
if (parse_iso_time(tok->args[0], &cert->cache_info.published_on) < 0) {
goto err;
}
tok = find_by_keyword(tokens, K_DIR_KEY_EXPIRES);
if (parse_iso_time(tok->args[0], &cert->expires) < 0) {
goto err;
}
tok = smartlist_get(tokens, smartlist_len(tokens)-1);
if (tok->tp != K_DIR_KEY_CERTIFICATION) {
log_warn(LD_DIR, "Certificate didn't end with dir-key-certification.");
goto err;
}
/* If we already have this cert, don't bother checking the signature. */
old_cert = authority_cert_get_by_digests(
cert->cache_info.identity_digest,
cert->signing_key_digest);
found = 0;
if (old_cert) {
/* XXXX We could just compare signed_descriptor_digest, but that wouldn't
* buy us much. */
if (old_cert->cache_info.signed_descriptor_len == len &&
old_cert->cache_info.signed_descriptor_body &&
tor_memeq(s, old_cert->cache_info.signed_descriptor_body, len)) {
log_debug(LD_DIR, "We already checked the signature on this "
"certificate; no need to do so again.");
found = 1;
}
}
if (!found) {
if (check_signature_token(digest, DIGEST_LEN, tok, cert->identity_key, 0,
"key certificate")) {
goto err;
}
tok = find_by_keyword(tokens, K_DIR_KEY_CROSSCERT);
if (check_signature_token(cert->cache_info.identity_digest,
DIGEST_LEN,
tok,
cert->signing_key,
CST_NO_CHECK_OBJTYPE,
"key cross-certification")) {
goto err;
}
}
cert->cache_info.signed_descriptor_len = len;
cert->cache_info.signed_descriptor_body = tor_malloc(len+1);
memcpy(cert->cache_info.signed_descriptor_body, s, len);
cert->cache_info.signed_descriptor_body[len] = 0;
cert->cache_info.saved_location = SAVED_NOWHERE;
if (end_of_string) {
*end_of_string = eat_whitespace(eos);
}
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_free(tokens);
if (area) {
DUMP_AREA(area, "authority cert");
memarea_drop_all(area);
}
return cert;
err:
dump_desc(s_dup, "authority cert");
authority_cert_free(cert);
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_free(tokens);
if (area) {
DUMP_AREA(area, "authority cert");
memarea_drop_all(area);
}
return NULL;
}
/** Helper: given a string <b>s</b>, return the start of the next router-status
* object (starting with "r " at the start of a line). If none is found,
* return the start of the directory footer, or the next directory signature.
* If none is found, return the end of the string. */
static INLINE const char *
find_start_of_next_routerstatus(const char *s)
{
const char *eos, *footer, *sig;
if ((eos = strstr(s, "\nr ")))
++eos;
else
eos = s + strlen(s);
footer = tor_memstr(s, eos-s, "\ndirectory-footer");
sig = tor_memstr(s, eos-s, "\ndirectory-signature");
if (footer && sig)
return MIN(footer, sig) + 1;
else if (footer)
return footer+1;
else if (sig)
return sig+1;
else
return eos;
}
/** Given a string at *<b>s</b>, containing a routerstatus object, and an
* empty smartlist at <b>tokens</b>, parse and return the first router status
* object in the string, and advance *<b>s</b> to just after the end of the
* router status. Return NULL and advance *<b>s</b> on error.
*
* If <b>vote</b> and <b>vote_rs</b> are provided, don't allocate a fresh
* routerstatus but use <b>vote_rs</b> instead.
*
* If <b>consensus_method</b> is nonzero, this routerstatus is part of a
* consensus, and we should parse it according to the method used to
* make that consensus.
*
* Parse according to the syntax used by the consensus flavor <b>flav</b>.
**/
static routerstatus_t *
routerstatus_parse_entry_from_string(memarea_t *area,
const char **s, smartlist_t *tokens,
networkstatus_t *vote,
vote_routerstatus_t *vote_rs,
int consensus_method,
consensus_flavor_t flav)
{
const char *eos, *s_dup = *s;
routerstatus_t *rs = NULL;
directory_token_t *tok;
char timebuf[ISO_TIME_LEN+1];
struct in_addr in;
int offset = 0;
tor_assert(tokens);
tor_assert(bool_eq(vote, vote_rs));
if (!consensus_method)
flav = FLAV_NS;
tor_assert(flav == FLAV_NS || flav == FLAV_MICRODESC);
eos = find_start_of_next_routerstatus(*s);
if (tokenize_string(area,*s, eos, tokens, rtrstatus_token_table,0)) {
log_warn(LD_DIR, "Error tokenizing router status");
goto err;
}
if (smartlist_len(tokens) < 1) {
log_warn(LD_DIR, "Impossibly short router status");
goto err;
}
tok = find_by_keyword(tokens, K_R);
tor_assert(tok->n_args >= 7); /* guaranteed by GE(7) in K_R setup */
if (flav == FLAV_NS) {
if (tok->n_args < 8) {
log_warn(LD_DIR, "Too few arguments to r");
goto err;
}
} else if (flav == FLAV_MICRODESC) {
offset = -1; /* There is no identity digest */
}
if (vote_rs) {
rs = &vote_rs->status;
} else {
rs = tor_malloc_zero(sizeof(routerstatus_t));
}
if (!is_legal_nickname(tok->args[0])) {
log_warn(LD_DIR,
"Invalid nickname %s in router status; skipping.",
escaped(tok->args[0]));
goto err;
}
strlcpy(rs->nickname, tok->args[0], sizeof(rs->nickname));
if (digest_from_base64(rs->identity_digest, tok->args[1])) {
log_warn(LD_DIR, "Error decoding identity digest %s",
escaped(tok->args[1]));
goto err;
}
if (flav == FLAV_NS) {
if (digest_from_base64(rs->descriptor_digest, tok->args[2])) {
log_warn(LD_DIR, "Error decoding descriptor digest %s",
escaped(tok->args[2]));
goto err;
}
}
if (tor_snprintf(timebuf, sizeof(timebuf), "%s %s",
tok->args[3+offset], tok->args[4+offset]) < 0 ||
parse_iso_time(timebuf, &rs->published_on)<0) {
log_warn(LD_DIR, "Error parsing time '%s %s' [%d %d]",
tok->args[3+offset], tok->args[4+offset],
offset, (int)flav);
goto err;
}
if (tor_inet_aton(tok->args[5+offset], &in) == 0) {
log_warn(LD_DIR, "Error parsing router address in network-status %s",
escaped(tok->args[5+offset]));
goto err;
}
rs->addr = ntohl(in.s_addr);
rs->or_port = (uint16_t) tor_parse_long(tok->args[6+offset],
10,0,65535,NULL,NULL);
rs->dir_port = (uint16_t) tor_parse_long(tok->args[7+offset],
10,0,65535,NULL,NULL);
{
smartlist_t *a_lines = find_all_by_keyword(tokens, K_A);
if (a_lines) {
find_single_ipv6_orport(a_lines, &rs->ipv6_addr, &rs->ipv6_orport);
smartlist_free(a_lines);
}
}
tok = find_opt_by_keyword(tokens, K_S);
if (tok && vote) {
int i;
vote_rs->flags = 0;
for (i=0; i < tok->n_args; ++i) {
int p = smartlist_string_pos(vote->known_flags, tok->args[i]);
if (p >= 0) {
vote_rs->flags |= (U64_LITERAL(1)<<p);
} else {
log_warn(LD_DIR, "Flags line had a flag %s not listed in known_flags.",
escaped(tok->args[i]));
goto err;
}
}
} else if (tok) {
int i;
for (i=0; i < tok->n_args; ++i) {
if (!strcmp(tok->args[i], "Exit"))
rs->is_exit = 1;
else if (!strcmp(tok->args[i], "Stable"))
rs->is_stable = 1;
else if (!strcmp(tok->args[i], "Fast"))
rs->is_fast = 1;
else if (!strcmp(tok->args[i], "Running"))
rs->is_flagged_running = 1;
else if (!strcmp(tok->args[i], "Named"))
rs->is_named = 1;
else if (!strcmp(tok->args[i], "Valid"))
rs->is_valid = 1;
else if (!strcmp(tok->args[i], "Guard"))
rs->is_possible_guard = 1;
else if (!strcmp(tok->args[i], "BadExit"))
rs->is_bad_exit = 1;
else if (!strcmp(tok->args[i], "BadDirectory"))
rs->is_bad_directory = 1;
else if (!strcmp(tok->args[i], "Authority"))
rs->is_authority = 1;
else if (!strcmp(tok->args[i], "Unnamed") &&
consensus_method >= 2) {
/* Unnamed is computed right by consensus method 2 and later. */
rs->is_unnamed = 1;
} else if (!strcmp(tok->args[i], "HSDir")) {
rs->is_hs_dir = 1;
}
}
}
if ((tok = find_opt_by_keyword(tokens, K_V))) {
tor_assert(tok->n_args == 1);
rs->version_known = 1;
if (strcmpstart(tok->args[0], "Tor ")) {
rs->version_supports_microdesc_cache = 1;
rs->version_supports_optimistic_data = 1;
} else {
rs->version_supports_microdesc_cache =
tor_version_supports_microdescriptors(tok->args[0]);
rs->version_supports_optimistic_data =
tor_version_as_new_as(tok->args[0], "0.2.3.1-alpha");
rs->version_supports_extend2_cells =
tor_version_as_new_as(tok->args[0], "0.2.4.8-alpha");
}
if (vote_rs) {
vote_rs->version = tor_strdup(tok->args[0]);
}
}
/* handle weighting/bandwidth info */
if ((tok = find_opt_by_keyword(tokens, K_W))) {
int i;
for (i=0; i < tok->n_args; ++i) {
if (!strcmpstart(tok->args[i], "Bandwidth=")) {
int ok;
rs->bandwidth_kb =
(uint32_t)tor_parse_ulong(strchr(tok->args[i], '=')+1,
10, 0, UINT32_MAX,
&ok, NULL);
if (!ok) {
log_warn(LD_DIR, "Invalid Bandwidth %s", escaped(tok->args[i]));
goto err;
}
rs->has_bandwidth = 1;
} else if (!strcmpstart(tok->args[i], "Measured=") && vote_rs) {
int ok;
vote_rs->measured_bw_kb =
(uint32_t)tor_parse_ulong(strchr(tok->args[i], '=')+1,
10, 0, UINT32_MAX, &ok, NULL);
if (!ok) {
log_warn(LD_DIR, "Invalid Measured Bandwidth %s",
escaped(tok->args[i]));
goto err;
}
vote_rs->has_measured_bw = 1;
vote->has_measured_bws = 1;
} else if (!strcmpstart(tok->args[i], "Unmeasured=1")) {
rs->bw_is_unmeasured = 1;
}
}
}
/* parse exit policy summaries */
if ((tok = find_opt_by_keyword(tokens, K_P))) {
tor_assert(tok->n_args == 1);
if (strcmpstart(tok->args[0], "accept ") &&
strcmpstart(tok->args[0], "reject ")) {
log_warn(LD_DIR, "Unknown exit policy summary type %s.",
escaped(tok->args[0]));
goto err;
}
/* XXX weasel: parse this into ports and represent them somehow smart,
* maybe not here but somewhere on if we need it for the client.
* we should still parse it here to check it's valid tho.
*/
rs->exitsummary = tor_strdup(tok->args[0]);
rs->has_exitsummary = 1;
}
if (vote_rs) {
SMARTLIST_FOREACH_BEGIN(tokens, directory_token_t *, t) {
if (t->tp == K_M && t->n_args) {
vote_microdesc_hash_t *line =
tor_malloc(sizeof(vote_microdesc_hash_t));
line->next = vote_rs->microdesc;
line->microdesc_hash_line = tor_strdup(t->args[0]);
vote_rs->microdesc = line;
}
} SMARTLIST_FOREACH_END(t);
} else if (flav == FLAV_MICRODESC) {
tok = find_opt_by_keyword(tokens, K_M);
if (tok) {
tor_assert(tok->n_args);
if (digest256_from_base64(rs->descriptor_digest, tok->args[0])) {
log_warn(LD_DIR, "Error decoding microdescriptor digest %s",
escaped(tok->args[0]));
goto err;
}
} else {
log_info(LD_BUG, "Found an entry in networkstatus with no "
"microdescriptor digest. (Router %s ($%s) at %s:%d.)",
rs->nickname, hex_str(rs->identity_digest, DIGEST_LEN),
fmt_addr32(rs->addr), rs->or_port);
}
}
if (!strcasecmp(rs->nickname, UNNAMED_ROUTER_NICKNAME))
rs->is_named = 0;
goto done;
err:
dump_desc(s_dup, "routerstatus entry");
if (rs && !vote_rs)
routerstatus_free(rs);
rs = NULL;
done:
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_clear(tokens);
if (area) {
DUMP_AREA(area, "routerstatus entry");
memarea_clear(area);
}
*s = eos;
return rs;
}
int
compare_vote_routerstatus_entries(const void **_a, const void **_b)
{
const vote_routerstatus_t *a = *_a, *b = *_b;
return fast_memcmp(a->status.identity_digest, b->status.identity_digest,
DIGEST_LEN);
}
/** Verify the bandwidth weights of a network status document */
int
networkstatus_verify_bw_weights(networkstatus_t *ns, int consensus_method)
{
int64_t weight_scale;
int64_t G=0, M=0, E=0, D=0, T=0;
double Wgg, Wgm, Wgd, Wmg, Wmm, Wme, Wmd, Weg, Wem, Wee, Wed;
double Gtotal=0, Mtotal=0, Etotal=0;
const char *casename = NULL;
int valid = 1;
weight_scale = networkstatus_get_weight_scale_param(ns);
Wgg = networkstatus_get_bw_weight(ns, "Wgg", -1);
Wgm = networkstatus_get_bw_weight(ns, "Wgm", -1);
Wgd = networkstatus_get_bw_weight(ns, "Wgd", -1);
Wmg = networkstatus_get_bw_weight(ns, "Wmg", -1);
Wmm = networkstatus_get_bw_weight(ns, "Wmm", -1);
Wme = networkstatus_get_bw_weight(ns, "Wme", -1);
Wmd = networkstatus_get_bw_weight(ns, "Wmd", -1);
Weg = networkstatus_get_bw_weight(ns, "Weg", -1);
Wem = networkstatus_get_bw_weight(ns, "Wem", -1);
Wee = networkstatus_get_bw_weight(ns, "Wee", -1);
Wed = networkstatus_get_bw_weight(ns, "Wed", -1);
if (Wgg<0 || Wgm<0 || Wgd<0 || Wmg<0 || Wmm<0 || Wme<0 || Wmd<0 || Weg<0
|| Wem<0 || Wee<0 || Wed<0) {
log_warn(LD_BUG, "No bandwidth weights produced in consensus!");
return 0;
}
// First, sanity check basic summing properties that hold for all cases
// We use > 1 as the check for these because they are computed as integers.
// Sometimes there are rounding errors.
if (fabs(Wmm - weight_scale) > 1) {
log_warn(LD_BUG, "Wmm=%f != "I64_FORMAT,
Wmm, I64_PRINTF_ARG(weight_scale));
valid = 0;
}
if (fabs(Wem - Wee) > 1) {
log_warn(LD_BUG, "Wem=%f != Wee=%f", Wem, Wee);
valid = 0;
}
if (fabs(Wgm - Wgg) > 1) {
log_warn(LD_BUG, "Wgm=%f != Wgg=%f", Wgm, Wgg);
valid = 0;
}
if (fabs(Weg - Wed) > 1) {
log_warn(LD_BUG, "Wed=%f != Weg=%f", Wed, Weg);
valid = 0;
}
if (fabs(Wgg + Wmg - weight_scale) > 0.001*weight_scale) {
log_warn(LD_BUG, "Wgg=%f != "I64_FORMAT" - Wmg=%f", Wgg,
I64_PRINTF_ARG(weight_scale), Wmg);
valid = 0;
}
if (fabs(Wee + Wme - weight_scale) > 0.001*weight_scale) {
log_warn(LD_BUG, "Wee=%f != "I64_FORMAT" - Wme=%f", Wee,
I64_PRINTF_ARG(weight_scale), Wme);
valid = 0;
}
if (fabs(Wgd + Wmd + Wed - weight_scale) > 0.001*weight_scale) {
log_warn(LD_BUG, "Wgd=%f + Wmd=%f + Wed=%f != "I64_FORMAT,
Wgd, Wmd, Wed, I64_PRINTF_ARG(weight_scale));
valid = 0;
}
Wgg /= weight_scale;
Wgm /= weight_scale;
Wgd /= weight_scale;
Wmg /= weight_scale;
Wmm /= weight_scale;
Wme /= weight_scale;
Wmd /= weight_scale;
Weg /= weight_scale;
Wem /= weight_scale;
Wee /= weight_scale;
Wed /= weight_scale;
// Then, gather G, M, E, D, T to determine case
SMARTLIST_FOREACH_BEGIN(ns->routerstatus_list, routerstatus_t *, rs) {
int is_exit = 0;
if (consensus_method >= MIN_METHOD_TO_CUT_BADEXIT_WEIGHT) {
/* Bug #2203: Don't count bad exits as exits for balancing */
is_exit = rs->is_exit && !rs->is_bad_exit;
} else {
is_exit = rs->is_exit;
}
if (rs->has_bandwidth) {
T += rs->bandwidth_kb;
if (is_exit && rs->is_possible_guard) {
D += rs->bandwidth_kb;
Gtotal += Wgd*rs->bandwidth_kb;
Mtotal += Wmd*rs->bandwidth_kb;
Etotal += Wed*rs->bandwidth_kb;
} else if (is_exit) {
E += rs->bandwidth_kb;
Mtotal += Wme*rs->bandwidth_kb;
Etotal += Wee*rs->bandwidth_kb;
} else if (rs->is_possible_guard) {
G += rs->bandwidth_kb;
Gtotal += Wgg*rs->bandwidth_kb;
Mtotal += Wmg*rs->bandwidth_kb;
} else {
M += rs->bandwidth_kb;
Mtotal += Wmm*rs->bandwidth_kb;
}
} else {
log_warn(LD_BUG, "Missing consensus bandwidth for router %s",
routerstatus_describe(rs));
}
} SMARTLIST_FOREACH_END(rs);
// Finally, check equality conditions depending upon case 1, 2 or 3
// Full equality cases: 1, 3b
// Partial equality cases: 2b (E=G), 3a (M=E)
// Fully unknown: 2a
if (3*E >= T && 3*G >= T) {
// Case 1: Neither are scarce
casename = "Case 1";
if (fabs(Etotal-Mtotal) > 0.01*MAX(Etotal,Mtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Etotal %f != Mtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Etotal, Mtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
if (fabs(Etotal-Gtotal) > 0.01*MAX(Etotal,Gtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Etotal %f != Gtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Etotal, Gtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
if (fabs(Gtotal-Mtotal) > 0.01*MAX(Gtotal,Mtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Mtotal %f != Gtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Mtotal, Gtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
} else if (3*E < T && 3*G < T) {
int64_t R = MIN(E, G);
int64_t S = MAX(E, G);
/*
* Case 2: Both Guards and Exits are scarce
* Balance D between E and G, depending upon
* D capacity and scarcity. Devote no extra
* bandwidth to middle nodes.
*/
if (R+D < S) { // Subcase a
double Rtotal, Stotal;
if (E < G) {
Rtotal = Etotal;
Stotal = Gtotal;
} else {
Rtotal = Gtotal;
Stotal = Etotal;
}
casename = "Case 2a";
// Rtotal < Stotal
if (Rtotal > Stotal) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Rtotal %f > Stotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Rtotal, Stotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
// Rtotal < T/3
if (3*Rtotal > T) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: 3*Rtotal %f > T "
I64_FORMAT". G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT
" D="I64_FORMAT" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Rtotal*3, I64_PRINTF_ARG(T),
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
// Stotal < T/3
if (3*Stotal > T) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: 3*Stotal %f > T "
I64_FORMAT". G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT
" D="I64_FORMAT" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Stotal*3, I64_PRINTF_ARG(T),
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
// Mtotal > T/3
if (3*Mtotal < T) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: 3*Mtotal %f < T "
I64_FORMAT". "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Mtotal*3, I64_PRINTF_ARG(T),
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
} else { // Subcase b: R+D > S
casename = "Case 2b";
/* Check the rare-M redirect case. */
if (D != 0 && 3*M < T) {
casename = "Case 2b (balanced)";
if (fabs(Etotal-Mtotal) > 0.01*MAX(Etotal,Mtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Etotal %f != Mtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Etotal, Mtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
if (fabs(Etotal-Gtotal) > 0.01*MAX(Etotal,Gtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Etotal %f != Gtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Etotal, Gtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
if (fabs(Gtotal-Mtotal) > 0.01*MAX(Gtotal,Mtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Mtotal %f != Gtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Mtotal, Gtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
} else {
if (fabs(Etotal-Gtotal) > 0.01*MAX(Etotal,Gtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Etotal %f != Gtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Etotal, Gtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
}
}
} else { // if (E < T/3 || G < T/3) {
int64_t S = MIN(E, G);
int64_t NS = MAX(E, G);
if (3*(S+D) < T) { // Subcase a:
double Stotal;
double NStotal;
if (G < E) {
casename = "Case 3a (G scarce)";
Stotal = Gtotal;
NStotal = Etotal;
} else { // if (G >= E) {
casename = "Case 3a (E scarce)";
NStotal = Gtotal;
Stotal = Etotal;
}
// Stotal < T/3
if (3*Stotal > T) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: 3*Stotal %f > T "
I64_FORMAT". G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT
" D="I64_FORMAT" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Stotal*3, I64_PRINTF_ARG(T),
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
if (NS >= M) {
if (fabs(NStotal-Mtotal) > 0.01*MAX(NStotal,Mtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: NStotal %f != Mtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, NStotal, Mtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
} else {
// if NS < M, NStotal > T/3 because only one of G or E is scarce
if (3*NStotal < T) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: 3*NStotal %f < T "
I64_FORMAT". G="I64_FORMAT" M="I64_FORMAT
" E="I64_FORMAT" D="I64_FORMAT" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, NStotal*3, I64_PRINTF_ARG(T),
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
}
} else { // Subcase b: S+D >= T/3
casename = "Case 3b";
if (fabs(Etotal-Mtotal) > 0.01*MAX(Etotal,Mtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Etotal %f != Mtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Etotal, Mtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
if (fabs(Etotal-Gtotal) > 0.01*MAX(Etotal,Gtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Etotal %f != Gtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Etotal, Gtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
if (fabs(Gtotal-Mtotal) > 0.01*MAX(Gtotal,Mtotal)) {
log_warn(LD_DIR,
"Bw Weight Failure for %s: Mtotal %f != Gtotal %f. "
"G="I64_FORMAT" M="I64_FORMAT" E="I64_FORMAT" D="I64_FORMAT
" T="I64_FORMAT". "
"Wgg=%f Wgd=%f Wmg=%f Wme=%f Wmd=%f Wee=%f Wed=%f",
casename, Mtotal, Gtotal,
I64_PRINTF_ARG(G), I64_PRINTF_ARG(M), I64_PRINTF_ARG(E),
I64_PRINTF_ARG(D), I64_PRINTF_ARG(T),
Wgg, Wgd, Wmg, Wme, Wmd, Wee, Wed);
valid = 0;
}
}
}
if (valid)
log_notice(LD_DIR, "Bandwidth-weight %s is verified and valid.",
casename);
return valid;
}
/** Parse a v3 networkstatus vote, opinion, or consensus (depending on
* ns_type), from <b>s</b>, and return the result. Return NULL on failure. */
networkstatus_t *
networkstatus_parse_vote_from_string(const char *s, const char **eos_out,
networkstatus_type_t ns_type)
{
smartlist_t *tokens = smartlist_new();
smartlist_t *rs_tokens = NULL, *footer_tokens = NULL;
networkstatus_voter_info_t *voter = NULL;
networkstatus_t *ns = NULL;
digests_t ns_digests;
const char *cert, *end_of_header, *end_of_footer, *s_dup = s;
directory_token_t *tok;
int ok;
struct in_addr in;
int i, inorder, n_signatures = 0;
memarea_t *area = NULL, *rs_area = NULL;
consensus_flavor_t flav = FLAV_NS;
char *last_kwd=NULL;
tor_assert(s);
if (eos_out)
*eos_out = NULL;
if (router_get_networkstatus_v3_hashes(s, &ns_digests)) {
log_warn(LD_DIR, "Unable to compute digest of network-status");
goto err;
}
area = memarea_new();
end_of_header = find_start_of_next_routerstatus(s);
if (tokenize_string(area, s, end_of_header, tokens,
(ns_type == NS_TYPE_CONSENSUS) ?
networkstatus_consensus_token_table :
networkstatus_token_table, 0)) {
log_warn(LD_DIR, "Error tokenizing network-status vote header");
goto err;
}
ns = tor_malloc_zero(sizeof(networkstatus_t));
memcpy(&ns->digests, &ns_digests, sizeof(ns_digests));
tok = find_by_keyword(tokens, K_NETWORK_STATUS_VERSION);
tor_assert(tok);
if (tok->n_args > 1) {
int flavor = networkstatus_parse_flavor_name(tok->args[1]);
if (flavor < 0) {
log_warn(LD_DIR, "Can't parse document with unknown flavor %s",
escaped(tok->args[1]));
goto err;
}
ns->flavor = flav = flavor;
}
if (flav != FLAV_NS && ns_type != NS_TYPE_CONSENSUS) {
log_warn(LD_DIR, "Flavor found on non-consensus networkstatus.");
goto err;
}
if (ns_type != NS_TYPE_CONSENSUS) {
const char *end_of_cert = NULL;
if (!(cert = strstr(s, "\ndir-key-certificate-version")))
goto err;
++cert;
ns->cert = authority_cert_parse_from_string(cert, &end_of_cert);
if (!ns->cert || !end_of_cert || end_of_cert > end_of_header)
goto err;
}
tok = find_by_keyword(tokens, K_VOTE_STATUS);
tor_assert(tok->n_args);
if (!strcmp(tok->args[0], "vote")) {
ns->type = NS_TYPE_VOTE;
} else if (!strcmp(tok->args[0], "consensus")) {
ns->type = NS_TYPE_CONSENSUS;
} else if (!strcmp(tok->args[0], "opinion")) {
ns->type = NS_TYPE_OPINION;
} else {
log_warn(LD_DIR, "Unrecognized vote status %s in network-status",
escaped(tok->args[0]));
goto err;
}
if (ns_type != ns->type) {
log_warn(LD_DIR, "Got the wrong kind of v3 networkstatus.");
goto err;
}
if (ns->type == NS_TYPE_VOTE || ns->type == NS_TYPE_OPINION) {
tok = find_by_keyword(tokens, K_PUBLISHED);
if (parse_iso_time(tok->args[0], &ns->published))
goto err;
ns->supported_methods = smartlist_new();
tok = find_opt_by_keyword(tokens, K_CONSENSUS_METHODS);
if (tok) {
for (i=0; i < tok->n_args; ++i)
smartlist_add(ns->supported_methods, tor_strdup(tok->args[i]));
} else {
smartlist_add(ns->supported_methods, tor_strdup("1"));
}
} else {
tok = find_opt_by_keyword(tokens, K_CONSENSUS_METHOD);
if (tok) {
ns->consensus_method = (int)tor_parse_long(tok->args[0], 10, 1, INT_MAX,
&ok, NULL);
if (!ok)
goto err;
} else {
ns->consensus_method = 1;
}
}
tok = find_by_keyword(tokens, K_VALID_AFTER);
if (parse_iso_time(tok->args[0], &ns->valid_after))
goto err;
tok = find_by_keyword(tokens, K_FRESH_UNTIL);
if (parse_iso_time(tok->args[0], &ns->fresh_until))
goto err;
tok = find_by_keyword(tokens, K_VALID_UNTIL);
if (parse_iso_time(tok->args[0], &ns->valid_until))
goto err;
tok = find_by_keyword(tokens, K_VOTING_DELAY);
tor_assert(tok->n_args >= 2);
ns->vote_seconds =
(int) tor_parse_long(tok->args[0], 10, 0, INT_MAX, &ok, NULL);
if (!ok)
goto err;
ns->dist_seconds =
(int) tor_parse_long(tok->args[1], 10, 0, INT_MAX, &ok, NULL);
if (!ok)
goto err;
if (ns->valid_after + MIN_VOTE_INTERVAL > ns->fresh_until) {
log_warn(LD_DIR, "Vote/consensus freshness interval is too short");
goto err;
}
if (ns->valid_after + MIN_VOTE_INTERVAL*2 > ns->valid_until) {
log_warn(LD_DIR, "Vote/consensus liveness interval is too short");
goto err;
}
if (ns->vote_seconds < MIN_VOTE_SECONDS) {
log_warn(LD_DIR, "Vote seconds is too short");
goto err;
}
if (ns->dist_seconds < MIN_DIST_SECONDS) {
log_warn(LD_DIR, "Dist seconds is too short");
goto err;
}
if ((tok = find_opt_by_keyword(tokens, K_CLIENT_VERSIONS))) {
ns->client_versions = tor_strdup(tok->args[0]);
}
if ((tok = find_opt_by_keyword(tokens, K_SERVER_VERSIONS))) {
ns->server_versions = tor_strdup(tok->args[0]);
}
tok = find_by_keyword(tokens, K_KNOWN_FLAGS);
ns->known_flags = smartlist_new();
inorder = 1;
for (i = 0; i < tok->n_args; ++i) {
smartlist_add(ns->known_flags, tor_strdup(tok->args[i]));
if (i>0 && strcmp(tok->args[i-1], tok->args[i])>= 0) {
log_warn(LD_DIR, "%s >= %s", tok->args[i-1], tok->args[i]);
inorder = 0;
}
}
if (!inorder) {
log_warn(LD_DIR, "known-flags not in order");
goto err;
}
if (ns->type != NS_TYPE_CONSENSUS &&
smartlist_len(ns->known_flags) > MAX_KNOWN_FLAGS_IN_VOTE) {
/* If we allowed more than 64 flags in votes, then parsing them would make
* us invoke undefined behavior whenever we used 1<<flagnum to do a
* bit-shift. This is only for votes and opinions: consensus users don't
* care about flags they don't recognize, and so don't build a bitfield
* for them. */
log_warn(LD_DIR, "Too many known-flags in consensus vote or opinion");
goto err;
}
tok = find_opt_by_keyword(tokens, K_PARAMS);
if (tok) {
int any_dups = 0;
inorder = 1;
ns->net_params = smartlist_new();
for (i = 0; i < tok->n_args; ++i) {
int ok=0;
char *eq = strchr(tok->args[i], '=');
size_t eq_pos;
if (!eq) {
log_warn(LD_DIR, "Bad element '%s' in params", escaped(tok->args[i]));
goto err;
}
eq_pos = eq-tok->args[i];
tor_parse_long(eq+1, 10, INT32_MIN, INT32_MAX, &ok, NULL);
if (!ok) {
log_warn(LD_DIR, "Bad element '%s' in params", escaped(tok->args[i]));
goto err;
}
if (i > 0 && strcmp(tok->args[i-1], tok->args[i]) >= 0) {
log_warn(LD_DIR, "%s >= %s", tok->args[i-1], tok->args[i]);
inorder = 0;
}
if (last_kwd && eq_pos == strlen(last_kwd) &&
fast_memeq(last_kwd, tok->args[i], eq_pos)) {
log_warn(LD_DIR, "Duplicate value for %s parameter",
escaped(tok->args[i]));
any_dups = 1;
}
tor_free(last_kwd);
last_kwd = tor_strndup(tok->args[i], eq_pos);
smartlist_add(ns->net_params, tor_strdup(tok->args[i]));
}
if (!inorder) {
log_warn(LD_DIR, "params not in order");
goto err;
}
if (any_dups) {
log_warn(LD_DIR, "Duplicate in parameters");
goto err;
}
}
ns->voters = smartlist_new();
SMARTLIST_FOREACH_BEGIN(tokens, directory_token_t *, _tok) {
tok = _tok;
if (tok->tp == K_DIR_SOURCE) {
tor_assert(tok->n_args >= 6);
if (voter)
smartlist_add(ns->voters, voter);
voter = tor_malloc_zero(sizeof(networkstatus_voter_info_t));
voter->sigs = smartlist_new();
if (ns->type != NS_TYPE_CONSENSUS)
memcpy(voter->vote_digest, ns_digests.d[DIGEST_SHA1], DIGEST_LEN);
voter->nickname = tor_strdup(tok->args[0]);
if (strlen(tok->args[1]) != HEX_DIGEST_LEN ||
base16_decode(voter->identity_digest, sizeof(voter->identity_digest),
tok->args[1], HEX_DIGEST_LEN) < 0) {
log_warn(LD_DIR, "Error decoding identity digest %s in "
"network-status vote.", escaped(tok->args[1]));
goto err;
}
if (ns->type != NS_TYPE_CONSENSUS &&
tor_memneq(ns->cert->cache_info.identity_digest,
voter->identity_digest, DIGEST_LEN)) {
log_warn(LD_DIR,"Mismatch between identities in certificate and vote");
goto err;
}
if (ns->type != NS_TYPE_CONSENSUS) {
if (authority_cert_is_blacklisted(ns->cert)) {
log_warn(LD_DIR, "Rejecting vote signature made with blacklisted "
"signing key %s",
hex_str(ns->cert->signing_key_digest, DIGEST_LEN));
goto err;
}
}
voter->address = tor_strdup(tok->args[2]);
if (!tor_inet_aton(tok->args[3], &in)) {
log_warn(LD_DIR, "Error decoding IP address %s in network-status.",
escaped(tok->args[3]));
goto err;
}
voter->addr = ntohl(in.s_addr);
voter->dir_port = (uint16_t)
tor_parse_long(tok->args[4], 10, 0, 65535, &ok, NULL);
if (!ok)
goto err;
voter->or_port = (uint16_t)
tor_parse_long(tok->args[5], 10, 0, 65535, &ok, NULL);
if (!ok)
goto err;
} else if (tok->tp == K_CONTACT) {
if (!voter || voter->contact) {
log_warn(LD_DIR, "contact element is out of place.");
goto err;
}
voter->contact = tor_strdup(tok->args[0]);
} else if (tok->tp == K_VOTE_DIGEST) {
tor_assert(ns->type == NS_TYPE_CONSENSUS);
tor_assert(tok->n_args >= 1);
if (!voter || ! tor_digest_is_zero(voter->vote_digest)) {
log_warn(LD_DIR, "vote-digest element is out of place.");
goto err;
}
if (strlen(tok->args[0]) != HEX_DIGEST_LEN ||
base16_decode(voter->vote_digest, sizeof(voter->vote_digest),
tok->args[0], HEX_DIGEST_LEN) < 0) {
log_warn(LD_DIR, "Error decoding vote digest %s in "
"network-status consensus.", escaped(tok->args[0]));
goto err;
}
}
} SMARTLIST_FOREACH_END(_tok);
if (voter) {
smartlist_add(ns->voters, voter);
voter = NULL;
}
if (smartlist_len(ns->voters) == 0) {
log_warn(LD_DIR, "Missing dir-source elements in a vote networkstatus.");
goto err;
} else if (ns->type != NS_TYPE_CONSENSUS && smartlist_len(ns->voters) != 1) {
log_warn(LD_DIR, "Too many dir-source elements in a vote networkstatus.");
goto err;
}
if (ns->type != NS_TYPE_CONSENSUS &&
(tok = find_opt_by_keyword(tokens, K_LEGACY_DIR_KEY))) {
int bad = 1;
if (strlen(tok->args[0]) == HEX_DIGEST_LEN) {
networkstatus_voter_info_t *voter = smartlist_get(ns->voters, 0);
if (base16_decode(voter->legacy_id_digest, DIGEST_LEN,
tok->args[0], HEX_DIGEST_LEN)<0)
bad = 1;
else
bad = 0;
}
if (bad) {
log_warn(LD_DIR, "Invalid legacy key digest %s on vote.",
escaped(tok->args[0]));
}
}
/* Parse routerstatus lines. */
rs_tokens = smartlist_new();
rs_area = memarea_new();
s = end_of_header;
ns->routerstatus_list = smartlist_new();
while (!strcmpstart(s, "r ")) {
if (ns->type != NS_TYPE_CONSENSUS) {
vote_routerstatus_t *rs = tor_malloc_zero(sizeof(vote_routerstatus_t));
if (routerstatus_parse_entry_from_string(rs_area, &s, rs_tokens, ns,
rs, 0, 0))
smartlist_add(ns->routerstatus_list, rs);
else {
tor_free(rs->version);
tor_free(rs);
}
} else {
routerstatus_t *rs;
if ((rs = routerstatus_parse_entry_from_string(rs_area, &s, rs_tokens,
NULL, NULL,
ns->consensus_method,
flav)))
smartlist_add(ns->routerstatus_list, rs);
}
}
for (i = 1; i < smartlist_len(ns->routerstatus_list); ++i) {
routerstatus_t *rs1, *rs2;
if (ns->type != NS_TYPE_CONSENSUS) {
vote_routerstatus_t *a = smartlist_get(ns->routerstatus_list, i-1);
vote_routerstatus_t *b = smartlist_get(ns->routerstatus_list, i);
rs1 = &a->status; rs2 = &b->status;
} else {
rs1 = smartlist_get(ns->routerstatus_list, i-1);
rs2 = smartlist_get(ns->routerstatus_list, i);
}
if (fast_memcmp(rs1->identity_digest, rs2->identity_digest, DIGEST_LEN)
>= 0) {
log_warn(LD_DIR, "Vote networkstatus entries not sorted by identity "
"digest");
goto err;
}
}
/* Parse footer; check signature. */
footer_tokens = smartlist_new();
if ((end_of_footer = strstr(s, "\nnetwork-status-version ")))
++end_of_footer;
else
end_of_footer = s + strlen(s);
if (tokenize_string(area,s, end_of_footer, footer_tokens,
networkstatus_vote_footer_token_table, 0)) {
log_warn(LD_DIR, "Error tokenizing network-status vote footer.");
goto err;
}
{
int found_sig = 0;
SMARTLIST_FOREACH_BEGIN(footer_tokens, directory_token_t *, _tok) {
tok = _tok;
if (tok->tp == K_DIRECTORY_SIGNATURE)
found_sig = 1;
else if (found_sig) {
log_warn(LD_DIR, "Extraneous token after first directory-signature");
goto err;
}
} SMARTLIST_FOREACH_END(_tok);
}
if ((tok = find_opt_by_keyword(footer_tokens, K_DIRECTORY_FOOTER))) {
if (tok != smartlist_get(footer_tokens, 0)) {
log_warn(LD_DIR, "Misplaced directory-footer token");
goto err;
}
}
tok = find_opt_by_keyword(footer_tokens, K_BW_WEIGHTS);
if (tok) {
ns->weight_params = smartlist_new();
for (i = 0; i < tok->n_args; ++i) {
int ok=0;
char *eq = strchr(tok->args[i], '=');
if (!eq) {
log_warn(LD_DIR, "Bad element '%s' in weight params",
escaped(tok->args[i]));
goto err;
}
tor_parse_long(eq+1, 10, INT32_MIN, INT32_MAX, &ok, NULL);
if (!ok) {
log_warn(LD_DIR, "Bad element '%s' in params", escaped(tok->args[i]));
goto err;
}
smartlist_add(ns->weight_params, tor_strdup(tok->args[i]));
}
}
SMARTLIST_FOREACH_BEGIN(footer_tokens, directory_token_t *, _tok) {
char declared_identity[DIGEST_LEN];
networkstatus_voter_info_t *v;
document_signature_t *sig;
const char *id_hexdigest = NULL;
const char *sk_hexdigest = NULL;
digest_algorithm_t alg = DIGEST_SHA1;
tok = _tok;
if (tok->tp != K_DIRECTORY_SIGNATURE)
continue;
tor_assert(tok->n_args >= 2);
if (tok->n_args == 2) {
id_hexdigest = tok->args[0];
sk_hexdigest = tok->args[1];
} else {
const char *algname = tok->args[0];
int a;
id_hexdigest = tok->args[1];
sk_hexdigest = tok->args[2];
a = crypto_digest_algorithm_parse_name(algname);
if (a<0) {
log_warn(LD_DIR, "Unknown digest algorithm %s; skipping",
escaped(algname));
continue;
}
alg = a;
}
if (!tok->object_type ||
strcmp(tok->object_type, "SIGNATURE") ||
tok->object_size < 128 || tok->object_size > 512) {
log_warn(LD_DIR, "Bad object type or length on directory-signature");
goto err;
}
if (strlen(id_hexdigest) != HEX_DIGEST_LEN ||
base16_decode(declared_identity, sizeof(declared_identity),
id_hexdigest, HEX_DIGEST_LEN) < 0) {
log_warn(LD_DIR, "Error decoding declared identity %s in "
"network-status vote.", escaped(id_hexdigest));
goto err;
}
if (!(v = networkstatus_get_voter_by_id(ns, declared_identity))) {
log_warn(LD_DIR, "ID on signature on network-status vote does not match "
"any declared directory source.");
goto err;
}
sig = tor_malloc_zero(sizeof(document_signature_t));
memcpy(sig->identity_digest, v->identity_digest, DIGEST_LEN);
sig->alg = alg;
if (strlen(sk_hexdigest) != HEX_DIGEST_LEN ||
base16_decode(sig->signing_key_digest, sizeof(sig->signing_key_digest),
sk_hexdigest, HEX_DIGEST_LEN) < 0) {
log_warn(LD_DIR, "Error decoding declared signing key digest %s in "
"network-status vote.", escaped(sk_hexdigest));
tor_free(sig);
goto err;
}
if (ns->type != NS_TYPE_CONSENSUS) {
if (tor_memneq(declared_identity, ns->cert->cache_info.identity_digest,
DIGEST_LEN)) {
log_warn(LD_DIR, "Digest mismatch between declared and actual on "
"network-status vote.");
tor_free(sig);
goto err;
}
}
if (voter_get_sig_by_algorithm(v, sig->alg)) {
/* We already parsed a vote with this algorithm from this voter. Use the
first one. */
log_fn(LOG_PROTOCOL_WARN, LD_DIR, "We received a networkstatus "
"that contains two votes from the same voter with the same "
"algorithm. Ignoring the second vote.");
tor_free(sig);
continue;
}
if (ns->type != NS_TYPE_CONSENSUS) {
if (check_signature_token(ns_digests.d[DIGEST_SHA1], DIGEST_LEN,
tok, ns->cert->signing_key, 0,
"network-status vote")) {
tor_free(sig);
goto err;
}
sig->good_signature = 1;
} else {
if (tok->object_size >= INT_MAX || tok->object_size >= SIZE_T_CEILING) {
tor_free(sig);
goto err;
}
sig->signature = tor_memdup(tok->object_body, tok->object_size);
sig->signature_len = (int) tok->object_size;
}
smartlist_add(v->sigs, sig);
++n_signatures;
} SMARTLIST_FOREACH_END(_tok);
if (! n_signatures) {
log_warn(LD_DIR, "No signatures on networkstatus vote.");
goto err;
} else if (ns->type == NS_TYPE_VOTE && n_signatures != 1) {
log_warn(LD_DIR, "Received more than one signature on a "
"network-status vote.");
goto err;
}
if (eos_out)
*eos_out = end_of_footer;
goto done;
err:
dump_desc(s_dup, "v3 networkstatus");
networkstatus_vote_free(ns);
ns = NULL;
done:
if (tokens) {
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_free(tokens);
}
if (voter) {
if (voter->sigs) {
SMARTLIST_FOREACH(voter->sigs, document_signature_t *, sig,
document_signature_free(sig));
smartlist_free(voter->sigs);
}
tor_free(voter->nickname);
tor_free(voter->address);
tor_free(voter->contact);
tor_free(voter);
}
if (rs_tokens) {
SMARTLIST_FOREACH(rs_tokens, directory_token_t *, t, token_clear(t));
smartlist_free(rs_tokens);
}
if (footer_tokens) {
SMARTLIST_FOREACH(footer_tokens, directory_token_t *, t, token_clear(t));
smartlist_free(footer_tokens);
}
if (area) {
DUMP_AREA(area, "v3 networkstatus");
memarea_drop_all(area);
}
if (rs_area)
memarea_drop_all(rs_area);
tor_free(last_kwd);
return ns;
}
/** Return the digests_t that holds the digests of the
* <b>flavor_name</b>-flavored networkstatus according to the detached
* signatures document <b>sigs</b>, allocating a new digests_t as neeeded. */
static digests_t *
detached_get_digests(ns_detached_signatures_t *sigs, const char *flavor_name)
{
digests_t *d = strmap_get(sigs->digests, flavor_name);
if (!d) {
d = tor_malloc_zero(sizeof(digests_t));
strmap_set(sigs->digests, flavor_name, d);
}
return d;
}
/** Return the list of signatures of the <b>flavor_name</b>-flavored
* networkstatus according to the detached signatures document <b>sigs</b>,
* allocating a new digests_t as neeeded. */
static smartlist_t *
detached_get_signatures(ns_detached_signatures_t *sigs,
const char *flavor_name)
{
smartlist_t *sl = strmap_get(sigs->signatures, flavor_name);
if (!sl) {
sl = smartlist_new();
strmap_set(sigs->signatures, flavor_name, sl);
}
return sl;
}
/** Parse a detached v3 networkstatus signature document between <b>s</b> and
* <b>eos</b> and return the result. Return -1 on failure. */
ns_detached_signatures_t *
networkstatus_parse_detached_signatures(const char *s, const char *eos)
{
/* XXXX there is too much duplicate shared between this function and
* networkstatus_parse_vote_from_string(). */
directory_token_t *tok;
memarea_t *area = NULL;
digests_t *digests;
smartlist_t *tokens = smartlist_new();
ns_detached_signatures_t *sigs =
tor_malloc_zero(sizeof(ns_detached_signatures_t));
sigs->digests = strmap_new();
sigs->signatures = strmap_new();
if (!eos)
eos = s + strlen(s);
area = memarea_new();
if (tokenize_string(area,s, eos, tokens,
networkstatus_detached_signature_token_table, 0)) {
log_warn(LD_DIR, "Error tokenizing detached networkstatus signatures");
goto err;
}
/* Grab all the digest-like tokens. */
SMARTLIST_FOREACH_BEGIN(tokens, directory_token_t *, _tok) {
const char *algname;
digest_algorithm_t alg;
const char *flavor;
const char *hexdigest;
size_t expected_length;
tok = _tok;
if (tok->tp == K_CONSENSUS_DIGEST) {
algname = "sha1";
alg = DIGEST_SHA1;
flavor = "ns";
hexdigest = tok->args[0];
} else if (tok->tp == K_ADDITIONAL_DIGEST) {
int a = crypto_digest_algorithm_parse_name(tok->args[1]);
if (a<0) {
log_warn(LD_DIR, "Unrecognized algorithm name %s", tok->args[0]);
continue;
}
alg = (digest_algorithm_t) a;
flavor = tok->args[0];
algname = tok->args[1];
hexdigest = tok->args[2];
} else {
continue;
}
expected_length =
(alg == DIGEST_SHA1) ? HEX_DIGEST_LEN : HEX_DIGEST256_LEN;
if (strlen(hexdigest) != expected_length) {
log_warn(LD_DIR, "Wrong length on consensus-digest in detached "
"networkstatus signatures");
goto err;
}
digests = detached_get_digests(sigs, flavor);
tor_assert(digests);
if (!tor_mem_is_zero(digests->d[alg], DIGEST256_LEN)) {
log_warn(LD_DIR, "Multiple digests for %s with %s on detached "
"signatures document", flavor, algname);
continue;
}
if (base16_decode(digests->d[alg], DIGEST256_LEN,
hexdigest, strlen(hexdigest)) < 0) {
log_warn(LD_DIR, "Bad encoding on consensus-digest in detached "
"networkstatus signatures");
goto err;
}
} SMARTLIST_FOREACH_END(_tok);
tok = find_by_keyword(tokens, K_VALID_AFTER);
if (parse_iso_time(tok->args[0], &sigs->valid_after)) {
log_warn(LD_DIR, "Bad valid-after in detached networkstatus signatures");
goto err;
}
tok = find_by_keyword(tokens, K_FRESH_UNTIL);
if (parse_iso_time(tok->args[0], &sigs->fresh_until)) {
log_warn(LD_DIR, "Bad fresh-until in detached networkstatus signatures");
goto err;
}
tok = find_by_keyword(tokens, K_VALID_UNTIL);
if (parse_iso_time(tok->args[0], &sigs->valid_until)) {
log_warn(LD_DIR, "Bad valid-until in detached networkstatus signatures");
goto err;
}
SMARTLIST_FOREACH_BEGIN(tokens, directory_token_t *, _tok) {
const char *id_hexdigest;
const char *sk_hexdigest;
const char *algname;
const char *flavor;
digest_algorithm_t alg;
char id_digest[DIGEST_LEN];
char sk_digest[DIGEST_LEN];
smartlist_t *siglist;
document_signature_t *sig;
int is_duplicate;
tok = _tok;
if (tok->tp == K_DIRECTORY_SIGNATURE) {
tor_assert(tok->n_args >= 2);
flavor = "ns";
algname = "sha1";
id_hexdigest = tok->args[0];
sk_hexdigest = tok->args[1];
} else if (tok->tp == K_ADDITIONAL_SIGNATURE) {
tor_assert(tok->n_args >= 4);
flavor = tok->args[0];
algname = tok->args[1];
id_hexdigest = tok->args[2];
sk_hexdigest = tok->args[3];
} else {
continue;
}
{
int a = crypto_digest_algorithm_parse_name(algname);
if (a<0) {
log_warn(LD_DIR, "Unrecognized algorithm name %s", algname);
continue;
}
alg = (digest_algorithm_t) a;
}
if (!tok->object_type ||
strcmp(tok->object_type, "SIGNATURE") ||
tok->object_size < 128 || tok->object_size > 512) {
log_warn(LD_DIR, "Bad object type or length on directory-signature");
goto err;
}
if (strlen(id_hexdigest) != HEX_DIGEST_LEN ||
base16_decode(id_digest, sizeof(id_digest),
id_hexdigest, HEX_DIGEST_LEN) < 0) {
log_warn(LD_DIR, "Error decoding declared identity %s in "
"network-status vote.", escaped(id_hexdigest));
goto err;
}
if (strlen(sk_hexdigest) != HEX_DIGEST_LEN ||
base16_decode(sk_digest, sizeof(sk_digest),
sk_hexdigest, HEX_DIGEST_LEN) < 0) {
log_warn(LD_DIR, "Error decoding declared signing key digest %s in "
"network-status vote.", escaped(sk_hexdigest));
goto err;
}
siglist = detached_get_signatures(sigs, flavor);
is_duplicate = 0;
SMARTLIST_FOREACH(siglist, document_signature_t *, dsig, {
if (dsig->alg == alg &&
tor_memeq(id_digest, dsig->identity_digest, DIGEST_LEN) &&
tor_memeq(sk_digest, dsig->signing_key_digest, DIGEST_LEN)) {
is_duplicate = 1;
}
});
if (is_duplicate) {
log_warn(LD_DIR, "Two signatures with identical keys and algorithm "
"found.");
continue;
}
sig = tor_malloc_zero(sizeof(document_signature_t));
sig->alg = alg;
memcpy(sig->identity_digest, id_digest, DIGEST_LEN);
memcpy(sig->signing_key_digest, sk_digest, DIGEST_LEN);
if (tok->object_size >= INT_MAX || tok->object_size >= SIZE_T_CEILING) {
tor_free(sig);
goto err;
}
sig->signature = tor_memdup(tok->object_body, tok->object_size);
sig->signature_len = (int) tok->object_size;
smartlist_add(siglist, sig);
} SMARTLIST_FOREACH_END(_tok);
goto done;
err:
ns_detached_signatures_free(sigs);
sigs = NULL;
done:
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_free(tokens);
if (area) {
DUMP_AREA(area, "detached signatures");
memarea_drop_all(area);
}
return sigs;
}
/** Parse the addr policy in the string <b>s</b> and return it. If
* assume_action is nonnegative, then insert its action (ADDR_POLICY_ACCEPT or
* ADDR_POLICY_REJECT) for items that specify no action.
*
* The addr_policy_t returned by this function can have its address set to
* AF_UNSPEC for '*'. Use policy_expand_unspec() to turn this into a pair
* of AF_INET and AF_INET6 items.
*/
addr_policy_t *
router_parse_addr_policy_item_from_string(const char *s, int assume_action)
{
directory_token_t *tok = NULL;
const char *cp, *eos;
/* Longest possible policy is "accept ffff:ffff:..255/ffff:...255:0-65535".
* But note that there can be an arbitrary amount of space between the
* accept and the address:mask/port element. */
char line[TOR_ADDR_BUF_LEN*2 + 32];
addr_policy_t *r;
memarea_t *area = NULL;
s = eat_whitespace(s);
if ((*s == '*' || TOR_ISDIGIT(*s)) && assume_action >= 0) {
if (tor_snprintf(line, sizeof(line), "%s %s",
assume_action == ADDR_POLICY_ACCEPT?"accept":"reject", s)<0) {
log_warn(LD_DIR, "Policy %s is too long.", escaped(s));
return NULL;
}
cp = line;
tor_strlower(line);
} else { /* assume an already well-formed address policy line */
cp = s;
}
eos = cp + strlen(cp);
area = memarea_new();
tok = get_next_token(area, &cp, eos, routerdesc_token_table);
if (tok->tp == ERR_) {
log_warn(LD_DIR, "Error reading address policy: %s", tok->error);
goto err;
}
if (tok->tp != K_ACCEPT && tok->tp != K_ACCEPT6 &&
tok->tp != K_REJECT && tok->tp != K_REJECT6) {
log_warn(LD_DIR, "Expected 'accept' or 'reject'.");
goto err;
}
r = router_parse_addr_policy(tok, TAPMP_EXTENDED_STAR);
goto done;
err:
r = NULL;
done:
token_clear(tok);
if (area) {
DUMP_AREA(area, "policy item");
memarea_drop_all(area);
}
return r;
}
/** Add an exit policy stored in the token <b>tok</b> to the router info in
* <b>router</b>. Return 0 on success, -1 on failure. */
static int
router_add_exit_policy(routerinfo_t *router, directory_token_t *tok)
{
addr_policy_t *newe;
newe = router_parse_addr_policy(tok, 0);
if (!newe)
return -1;
if (! router->exit_policy)
router->exit_policy = smartlist_new();
if (((tok->tp == K_ACCEPT6 || tok->tp == K_REJECT6) &&
tor_addr_family(&newe->addr) == AF_INET)
||
((tok->tp == K_ACCEPT || tok->tp == K_REJECT) &&
tor_addr_family(&newe->addr) == AF_INET6)) {
log_warn(LD_DIR, "Mismatch between field type and address type in exit "
"policy");
addr_policy_free(newe);
return -1;
}
smartlist_add(router->exit_policy, newe);
return 0;
}
/** Given a K_ACCEPT or K_REJECT token and a router, create and return
* a new exit_policy_t corresponding to the token. */
static addr_policy_t *
router_parse_addr_policy(directory_token_t *tok, unsigned fmt_flags)
{
addr_policy_t newe;
char *arg;
tor_assert(tok->tp == K_REJECT || tok->tp == K_REJECT6 ||
tok->tp == K_ACCEPT || tok->tp == K_ACCEPT6);
if (tok->n_args != 1)
return NULL;
arg = tok->args[0];
if (!strcmpstart(arg,"private"))
return router_parse_addr_policy_private(tok);
memset(&newe, 0, sizeof(newe));
if (tok->tp == K_REJECT || tok->tp == K_REJECT6)
newe.policy_type = ADDR_POLICY_REJECT;
else
newe.policy_type = ADDR_POLICY_ACCEPT;
if (tor_addr_parse_mask_ports(arg, fmt_flags, &newe.addr, &newe.maskbits,
&newe.prt_min, &newe.prt_max) < 0) {
log_warn(LD_DIR,"Couldn't parse line %s. Dropping", escaped(arg));
return NULL;
}
return addr_policy_get_canonical_entry(&newe);
}
/** Parse an exit policy line of the format "accept/reject private:...".
* This didn't exist until Tor 0.1.1.15, so nobody should generate it in
* router descriptors until earlier versions are obsolete.
*/
static addr_policy_t *
router_parse_addr_policy_private(directory_token_t *tok)
{
const char *arg;
uint16_t port_min, port_max;
addr_policy_t result;
arg = tok->args[0];
if (strcmpstart(arg, "private"))
return NULL;
arg += strlen("private");
arg = (char*) eat_whitespace(arg);
if (!arg || *arg != ':')
return NULL;
if (parse_port_range(arg+1, &port_min, &port_max)<0)
return NULL;
memset(&result, 0, sizeof(result));
if (tok->tp == K_REJECT || tok->tp == K_REJECT6)
result.policy_type = ADDR_POLICY_REJECT;
else
result.policy_type = ADDR_POLICY_ACCEPT;
result.is_private = 1;
result.prt_min = port_min;
result.prt_max = port_max;
return addr_policy_get_canonical_entry(&result);
}
/** Log and exit if <b>t</b> is malformed */
void
assert_addr_policy_ok(smartlist_t *lst)
{
if (!lst) return;
SMARTLIST_FOREACH(lst, addr_policy_t *, t, {
tor_assert(t->policy_type == ADDR_POLICY_REJECT ||
t->policy_type == ADDR_POLICY_ACCEPT);
tor_assert(t->prt_min <= t->prt_max);
});
}
/*
* Low-level tokenizer for router descriptors and directories.
*/
/** Free all resources allocated for <b>tok</b> */
static void
token_clear(directory_token_t *tok)
{
if (tok->key)
crypto_pk_free(tok->key);
}
#define ALLOC_ZERO(sz) memarea_alloc_zero(area,sz)
#define ALLOC(sz) memarea_alloc(area,sz)
#define STRDUP(str) memarea_strdup(area,str)
#define STRNDUP(str,n) memarea_strndup(area,(str),(n))
#define RET_ERR(msg) \
STMT_BEGIN \
if (tok) token_clear(tok); \
tok = ALLOC_ZERO(sizeof(directory_token_t)); \
tok->tp = ERR_; \
tok->error = STRDUP(msg); \
goto done_tokenizing; \
STMT_END
/** Helper: make sure that the token <b>tok</b> with keyword <b>kwd</b> obeys
* the object syntax of <b>o_syn</b>. Allocate all storage in <b>area</b>.
* Return <b>tok</b> on success, or a new ERR_ token if the token didn't
* conform to the syntax we wanted.
**/
static INLINE directory_token_t *
token_check_object(memarea_t *area, const char *kwd,
directory_token_t *tok, obj_syntax o_syn)
{
char ebuf[128];
switch (o_syn) {
case NO_OBJ:
/* No object is allowed for this token. */
if (tok->object_body) {
tor_snprintf(ebuf, sizeof(ebuf), "Unexpected object for %s", kwd);
RET_ERR(ebuf);
}
if (tok->key) {
tor_snprintf(ebuf, sizeof(ebuf), "Unexpected public key for %s", kwd);
RET_ERR(ebuf);
}
break;
case NEED_OBJ:
/* There must be a (non-key) object. */
if (!tok->object_body) {
tor_snprintf(ebuf, sizeof(ebuf), "Missing object for %s", kwd);
RET_ERR(ebuf);
}
break;
case NEED_KEY_1024: /* There must be a 1024-bit public key. */
case NEED_SKEY_1024: /* There must be a 1024-bit private key. */
if (tok->key && crypto_pk_num_bits(tok->key) != PK_BYTES*8) {
tor_snprintf(ebuf, sizeof(ebuf), "Wrong size on key for %s: %d bits",
kwd, crypto_pk_num_bits(tok->key));
RET_ERR(ebuf);
}
/* fall through */
case NEED_KEY: /* There must be some kind of key. */
if (!tok->key) {
tor_snprintf(ebuf, sizeof(ebuf), "Missing public key for %s", kwd);
RET_ERR(ebuf);
}
if (o_syn != NEED_SKEY_1024) {
if (crypto_pk_key_is_private(tok->key)) {
tor_snprintf(ebuf, sizeof(ebuf),
"Private key given for %s, which wants a public key", kwd);
RET_ERR(ebuf);
}
} else { /* o_syn == NEED_SKEY_1024 */
if (!crypto_pk_key_is_private(tok->key)) {
tor_snprintf(ebuf, sizeof(ebuf),
"Public key given for %s, which wants a private key", kwd);
RET_ERR(ebuf);
}
}
break;
case OBJ_OK:
/* Anything goes with this token. */
break;
}
done_tokenizing:
return tok;
}
/** Helper: parse space-separated arguments from the string <b>s</b> ending at
* <b>eol</b>, and store them in the args field of <b>tok</b>. Store the
* number of parsed elements into the n_args field of <b>tok</b>. Allocate
* all storage in <b>area</b>. Return the number of arguments parsed, or
* return -1 if there was an insanely high number of arguments. */
static INLINE int
get_token_arguments(memarea_t *area, directory_token_t *tok,
const char *s, const char *eol)
{
/** Largest number of arguments we'll accept to any token, ever. */
#define MAX_ARGS 512
char *mem = memarea_strndup(area, s, eol-s);
char *cp = mem;
int j = 0;
char *args[MAX_ARGS];
while (*cp) {
if (j == MAX_ARGS)
return -1;
args[j++] = cp;
cp = (char*)find_whitespace(cp);
if (!cp || !*cp)
break; /* End of the line. */
*cp++ = '\0';
cp = (char*)eat_whitespace(cp);
}
tok->n_args = j;
tok->args = memarea_memdup(area, args, j*sizeof(char*));
return j;
#undef MAX_ARGS
}
/** Helper function: read the next token from *s, advance *s to the end of the
* token, and return the parsed token. Parse *<b>s</b> according to the list
* of tokens in <b>table</b>.
*/
static directory_token_t *
get_next_token(memarea_t *area,
const char **s, const char *eos, token_rule_t *table)
{
/** Reject any object at least this big; it is probably an overflow, an
* attack, a bug, or some other nonsense. */
#define MAX_UNPARSED_OBJECT_SIZE (128*1024)
/** Reject any line at least this big; it is probably an overflow, an
* attack, a bug, or some other nonsense. */
#define MAX_LINE_LENGTH (128*1024)
const char *next, *eol, *obstart;
size_t obname_len;
int i;
directory_token_t *tok;
obj_syntax o_syn = NO_OBJ;
char ebuf[128];
const char *kwd = "";
tor_assert(area);
tok = ALLOC_ZERO(sizeof(directory_token_t));
tok->tp = ERR_;
/* Set *s to first token, eol to end-of-line, next to after first token */
*s = eat_whitespace_eos(*s, eos); /* eat multi-line whitespace */
tor_assert(eos >= *s);
eol = memchr(*s, '\n', eos-*s);
if (!eol)
eol = eos;
if (eol - *s > MAX_LINE_LENGTH) {
RET_ERR("Line far too long");
}
next = find_whitespace_eos(*s, eol);
if (!strcmp_len(*s, "opt", next-*s)) {
/* Skip past an "opt" at the start of the line. */
*s = eat_whitespace_eos_no_nl(next, eol);
next = find_whitespace_eos(*s, eol);
} else if (*s == eos) { /* If no "opt", and end-of-line, line is invalid */
RET_ERR("Unexpected EOF");
}
/* Search the table for the appropriate entry. (I tried a binary search
* instead, but it wasn't any faster.) */
for (i = 0; table[i].t ; ++i) {
if (!strcmp_len(*s, table[i].t, next-*s)) {
/* We've found the keyword. */
kwd = table[i].t;
tok->tp = table[i].v;
o_syn = table[i].os;
*s = eat_whitespace_eos_no_nl(next, eol);
/* We go ahead whether there are arguments or not, so that tok->args is
* always set if we want arguments. */
if (table[i].concat_args) {
/* The keyword takes the line as a single argument */
tok->args = ALLOC(sizeof(char*));
tok->args[0] = STRNDUP(*s,eol-*s); /* Grab everything on line */
tok->n_args = 1;
} else {
/* This keyword takes multiple arguments. */
if (get_token_arguments(area, tok, *s, eol)<0) {
tor_snprintf(ebuf, sizeof(ebuf),"Far too many arguments to %s", kwd);
RET_ERR(ebuf);
}
*s = eol;
}
if (tok->n_args < table[i].min_args) {
tor_snprintf(ebuf, sizeof(ebuf), "Too few arguments to %s", kwd);
RET_ERR(ebuf);
} else if (tok->n_args > table[i].max_args) {
tor_snprintf(ebuf, sizeof(ebuf), "Too many arguments to %s", kwd);
RET_ERR(ebuf);
}
break;
}
}
if (tok->tp == ERR_) {
/* No keyword matched; call it an "K_opt" or "A_unrecognized" */
if (*s < eol && **s == '@')
tok->tp = A_UNKNOWN_;
else
tok->tp = K_OPT;
tok->args = ALLOC(sizeof(char*));
tok->args[0] = STRNDUP(*s, eol-*s);
tok->n_args = 1;
o_syn = OBJ_OK;
}
/* Check whether there's an object present */
*s = eat_whitespace_eos(eol, eos); /* Scan from end of first line */
tor_assert(eos >= *s);
eol = memchr(*s, '\n', eos-*s);
if (!eol || eol-*s<11 || strcmpstart(*s, "-----BEGIN ")) /* No object. */
goto check_object;
obstart = *s; /* Set obstart to start of object spec */
if (*s+16 >= eol || memchr(*s+11,'\0',eol-*s-16) || /* no short lines, */
strcmp_len(eol-5, "-----", 5) || /* nuls or invalid endings */
(eol-*s) > MAX_UNPARSED_OBJECT_SIZE) { /* name too long */
RET_ERR("Malformed object: bad begin line");
}
tok->object_type = STRNDUP(*s+11, eol-*s-16);
obname_len = eol-*s-16; /* store objname length here to avoid a strlen() */
*s = eol+1; /* Set *s to possible start of object data (could be eos) */
/* Go to the end of the object */
next = tor_memstr(*s, eos-*s, "-----END ");
if (!next) {
RET_ERR("Malformed object: missing object end line");
}
tor_assert(eos >= next);
eol = memchr(next, '\n', eos-next);
if (!eol) /* end-of-line marker, or eos if there's no '\n' */
eol = eos;
/* Validate the ending tag, which should be 9 + NAME + 5 + eol */
if ((size_t)(eol-next) != 9+obname_len+5 ||
strcmp_len(next+9, tok->object_type, obname_len) ||
strcmp_len(eol-5, "-----", 5)) {
tor_snprintf(ebuf, sizeof(ebuf), "Malformed object: mismatched end tag %s",
tok->object_type);
ebuf[sizeof(ebuf)-1] = '\0';
RET_ERR(ebuf);
}
if (next - *s > MAX_UNPARSED_OBJECT_SIZE)
RET_ERR("Couldn't parse object: missing footer or object much too big.");
if (!strcmp(tok->object_type, "RSA PUBLIC KEY")) { /* If it's a public key */
tok->key = crypto_pk_new();
if (crypto_pk_read_public_key_from_string(tok->key, obstart, eol-obstart))
RET_ERR("Couldn't parse public key.");
} else if (!strcmp(tok->object_type, "RSA PRIVATE KEY")) { /* private key */
tok->key = crypto_pk_new();
if (crypto_pk_read_private_key_from_string(tok->key, obstart, eol-obstart))
RET_ERR("Couldn't parse private key.");
} else { /* If it's something else, try to base64-decode it */
int r;
tok->object_body = ALLOC(next-*s); /* really, this is too much RAM. */
r = base64_decode(tok->object_body, next-*s, *s, next-*s);
if (r<0)
RET_ERR("Malformed object: bad base64-encoded data");
tok->object_size = r;
}
*s = eol;
check_object:
tok = token_check_object(area, kwd, tok, o_syn);
done_tokenizing:
return tok;
#undef RET_ERR
#undef ALLOC
#undef ALLOC_ZERO
#undef STRDUP
#undef STRNDUP
}
/** Read all tokens from a string between <b>start</b> and <b>end</b>, and add
* them to <b>out</b>. Parse according to the token rules in <b>table</b>.
* Caller must free tokens in <b>out</b>. If <b>end</b> is NULL, use the
* entire string.
*/
static int
tokenize_string(memarea_t *area,
const char *start, const char *end, smartlist_t *out,
token_rule_t *table, int flags)
{
const char **s;
directory_token_t *tok = NULL;
int counts[NIL_];
int i;
int first_nonannotation;
int prev_len = smartlist_len(out);
tor_assert(area);
s = &start;
if (!end) {
end = start+strlen(start);
} else {
/* it's only meaningful to check for nuls if we got an end-of-string ptr */
if (memchr(start, '\0', end-start)) {
log_warn(LD_DIR, "parse error: internal NUL character.");
return -1;
}
}
for (i = 0; i < NIL_; ++i)
counts[i] = 0;
SMARTLIST_FOREACH(out, const directory_token_t *, t, ++counts[t->tp]);
while (*s < end && (!tok || tok->tp != EOF_)) {
tok = get_next_token(area, s, end, table);
if (tok->tp == ERR_) {
log_warn(LD_DIR, "parse error: %s", tok->error);
token_clear(tok);
return -1;
}
++counts[tok->tp];
smartlist_add(out, tok);
*s = eat_whitespace_eos(*s, end);
}
if (flags & TS_NOCHECK)
return 0;
if ((flags & TS_ANNOTATIONS_OK)) {
first_nonannotation = -1;
for (i = 0; i < smartlist_len(out); ++i) {
tok = smartlist_get(out, i);
if (tok->tp < MIN_ANNOTATION || tok->tp > MAX_ANNOTATION) {
first_nonannotation = i;
break;
}
}
if (first_nonannotation < 0) {
log_warn(LD_DIR, "parse error: item contains only annotations");
return -1;
}
for (i=first_nonannotation; i < smartlist_len(out); ++i) {
tok = smartlist_get(out, i);
if (tok->tp >= MIN_ANNOTATION && tok->tp <= MAX_ANNOTATION) {
log_warn(LD_DIR, "parse error: Annotations mixed with keywords");
return -1;
}
}
if ((flags & TS_NO_NEW_ANNOTATIONS)) {
if (first_nonannotation != prev_len) {
log_warn(LD_DIR, "parse error: Unexpected annotations.");
return -1;
}
}
} else {
for (i=0; i < smartlist_len(out); ++i) {
tok = smartlist_get(out, i);
if (tok->tp >= MIN_ANNOTATION && tok->tp <= MAX_ANNOTATION) {
log_warn(LD_DIR, "parse error: no annotations allowed.");
return -1;
}
}
first_nonannotation = 0;
}
for (i = 0; table[i].t; ++i) {
if (counts[table[i].v] < table[i].min_cnt) {
log_warn(LD_DIR, "Parse error: missing %s element.", table[i].t);
return -1;
}
if (counts[table[i].v] > table[i].max_cnt) {
log_warn(LD_DIR, "Parse error: too many %s elements.", table[i].t);
return -1;
}
if (table[i].pos & AT_START) {
if (smartlist_len(out) < 1 ||
(tok = smartlist_get(out, first_nonannotation))->tp != table[i].v) {
log_warn(LD_DIR, "Parse error: first item is not %s.", table[i].t);
return -1;
}
}
if (table[i].pos & AT_END) {
if (smartlist_len(out) < 1 ||
(tok = smartlist_get(out, smartlist_len(out)-1))->tp != table[i].v) {
log_warn(LD_DIR, "Parse error: last item is not %s.", table[i].t);
return -1;
}
}
}
return 0;
}
/** Find the first token in <b>s</b> whose keyword is <b>keyword</b>; return
* NULL if no such keyword is found.
*/
static directory_token_t *
find_opt_by_keyword(smartlist_t *s, directory_keyword keyword)
{
SMARTLIST_FOREACH(s, directory_token_t *, t, if (t->tp == keyword) return t);
return NULL;
}
/** Find the first token in <b>s</b> whose keyword is <b>keyword</b>; fail
* with an assert if no such keyword is found.
*/
static directory_token_t *
find_by_keyword_(smartlist_t *s, directory_keyword keyword,
const char *keyword_as_string)
{
directory_token_t *tok = find_opt_by_keyword(s, keyword);
if (PREDICT_UNLIKELY(!tok)) {
log_err(LD_BUG, "Missing %s [%d] in directory object that should have "
"been validated. Internal error.", keyword_as_string, (int)keyword);
tor_assert(tok);
}
return tok;
}
/** If there are any directory_token_t entries in <b>s</b> whose keyword is
* <b>k</b>, return a newly allocated smartlist_t containing all such entries,
* in the same order in which they occur in <b>s</b>. Otherwise return
* NULL. */
static smartlist_t *
find_all_by_keyword(smartlist_t *s, directory_keyword k)
{
smartlist_t *out = NULL;
SMARTLIST_FOREACH(s, directory_token_t *, t,
if (t->tp == k) {
if (!out)
out = smartlist_new();
smartlist_add(out, t);
});
return out;
}
/** Return a newly allocated smartlist of all accept or reject tokens in
* <b>s</b>.
*/
static smartlist_t *
find_all_exitpolicy(smartlist_t *s)
{
smartlist_t *out = smartlist_new();
SMARTLIST_FOREACH(s, directory_token_t *, t,
if (t->tp == K_ACCEPT || t->tp == K_ACCEPT6 ||
t->tp == K_REJECT || t->tp == K_REJECT6)
smartlist_add(out,t));
return out;
}
/** Helper function for <b>router_get_hash_impl</b>: given <b>s</b>,
* <b>s_len</b>, <b>start_str</b>, <b>end_str</b>, and <b>end_c</b> with the
* same semantics as in that function, set *<b>start_out</b> (inclusive) and
* *<b>end_out</b> (exclusive) to the boundaries of the string to be hashed.
*
* Return 0 on success and -1 on failure.
*/
static int
router_get_hash_impl_helper(const char *s, size_t s_len,
const char *start_str,
const char *end_str, char end_c,
const char **start_out, const char **end_out)
{
const char *start, *end;
start = tor_memstr(s, s_len, start_str);
if (!start) {
log_warn(LD_DIR,"couldn't find start of hashed material \"%s\"",start_str);
return -1;
}
if (start != s && *(start-1) != '\n') {
log_warn(LD_DIR,
"first occurrence of \"%s\" is not at the start of a line",
start_str);
return -1;
}
end = tor_memstr(start+strlen(start_str),
s_len - (start-s) - strlen(start_str), end_str);
if (!end) {
log_warn(LD_DIR,"couldn't find end of hashed material \"%s\"",end_str);
return -1;
}
end = memchr(end+strlen(end_str), end_c, s_len - (end-s) - strlen(end_str));
if (!end) {
log_warn(LD_DIR,"couldn't find EOL");
return -1;
}
++end;
*start_out = start;
*end_out = end;
return 0;
}
/** Compute the digest of the substring of <b>s</b> taken from the first
* occurrence of <b>start_str</b> through the first instance of c after the
* first subsequent occurrence of <b>end_str</b>; store the 20-byte or 32-byte
* result in <b>digest</b>; return 0 on success.
*
* If no such substring exists, return -1.
*/
static int
router_get_hash_impl(const char *s, size_t s_len, char *digest,
const char *start_str,
const char *end_str, char end_c,
digest_algorithm_t alg)
{
const char *start=NULL, *end=NULL;
if (router_get_hash_impl_helper(s,s_len,start_str,end_str,end_c,
&start,&end)<0)
return -1;
if (alg == DIGEST_SHA1) {
if (crypto_digest(digest, start, end-start)) {
log_warn(LD_BUG,"couldn't compute digest");
return -1;
}
} else {
if (crypto_digest256(digest, start, end-start, alg)) {
log_warn(LD_BUG,"couldn't compute digest");
return -1;
}
}
return 0;
}
/** As router_get_hash_impl, but compute all hashes. */
static int
router_get_hashes_impl(const char *s, size_t s_len, digests_t *digests,
const char *start_str,
const char *end_str, char end_c)
{
const char *start=NULL, *end=NULL;
if (router_get_hash_impl_helper(s,s_len,start_str,end_str,end_c,
&start,&end)<0)
return -1;
if (crypto_digest_all(digests, start, end-start)) {
log_warn(LD_BUG,"couldn't compute digests");
return -1;
}
return 0;
}
/** Assuming that s starts with a microdesc, return the start of the
* *NEXT* one. Return NULL on "not found." */
static const char *
find_start_of_next_microdesc(const char *s, const char *eos)
{
int started_with_annotations;
s = eat_whitespace_eos(s, eos);
if (!s)
return NULL;
#define CHECK_LENGTH() STMT_BEGIN \
if (s+32 > eos) \
return NULL; \
STMT_END
#define NEXT_LINE() STMT_BEGIN \
s = memchr(s, '\n', eos-s); \
if (!s || s+1 >= eos) \
return NULL; \
s++; \
STMT_END
CHECK_LENGTH();
started_with_annotations = (*s == '@');
if (started_with_annotations) {
/* Start by advancing to the first non-annotation line. */
while (*s == '@')
NEXT_LINE();
}
CHECK_LENGTH();
/* Now we should be pointed at an onion-key line. If we are, then skip
* it. */
if (!strcmpstart(s, "onion-key"))
NEXT_LINE();
/* Okay, now we're pointed at the first line of the microdescriptor which is
not an annotation or onion-key. The next line that _is_ an annotation or
onion-key is the start of the next microdescriptor. */
while (s+32 < eos) {
if (*s == '@' || !strcmpstart(s, "onion-key"))
return s;
NEXT_LINE();
}
return NULL;
#undef CHECK_LENGTH
#undef NEXT_LINE
}
/** Parse as many microdescriptors as are found from the string starting at
* <b>s</b> and ending at <b>eos</b>. If allow_annotations is set, read any
* annotations we recognize and ignore ones we don't.
*
* If <b>saved_location</b> isn't SAVED_IN_CACHE, make a local copy of each
* descriptor in the body field of each microdesc_t.
*
* Return all newly
* parsed microdescriptors in a newly allocated smartlist_t. */
smartlist_t *
microdescs_parse_from_string(const char *s, const char *eos,
int allow_annotations,
saved_location_t where)
{
smartlist_t *tokens;
smartlist_t *result;
microdesc_t *md = NULL;
memarea_t *area;
const char *start = s;
const char *start_of_next_microdesc;
int flags = allow_annotations ? TS_ANNOTATIONS_OK : 0;
const int copy_body = (where != SAVED_IN_CACHE);
directory_token_t *tok;
if (!eos)
eos = s + strlen(s);
s = eat_whitespace_eos(s, eos);
area = memarea_new();
result = smartlist_new();
tokens = smartlist_new();
while (s < eos) {
start_of_next_microdesc = find_start_of_next_microdesc(s, eos);
if (!start_of_next_microdesc)
start_of_next_microdesc = eos;
if (tokenize_string(area, s, start_of_next_microdesc, tokens,
microdesc_token_table, flags)) {
log_warn(LD_DIR, "Unparseable microdescriptor");
goto next;
}
md = tor_malloc_zero(sizeof(microdesc_t));
{
const char *cp = tor_memstr(s, start_of_next_microdesc-s,
"onion-key");
tor_assert(cp);
md->bodylen = start_of_next_microdesc - cp;
md->saved_location = where;
if (copy_body)
md->body = tor_memdup_nulterm(cp, md->bodylen);
else
md->body = (char*)cp;
md->off = cp - start;
}
if ((tok = find_opt_by_keyword(tokens, A_LAST_LISTED))) {
if (parse_iso_time(tok->args[0], &md->last_listed)) {
log_warn(LD_DIR, "Bad last-listed time in microdescriptor");
goto next;
}
}
tok = find_by_keyword(tokens, K_ONION_KEY);
if (!crypto_pk_public_exponent_ok(tok->key)) {
log_warn(LD_DIR,
"Relay's onion key had invalid exponent.");
goto next;
}
md->onion_pkey = tok->key;
tok->key = NULL;
if ((tok = find_opt_by_keyword(tokens, K_ONION_KEY_NTOR))) {
curve25519_public_key_t k;
tor_assert(tok->n_args >= 1);
if (curve25519_public_from_base64(&k, tok->args[0]) < 0) {
log_warn(LD_DIR, "Bogus ntor-onion-key in microdesc");
goto next;
}
md->onion_curve25519_pkey =
tor_memdup(&k, sizeof(curve25519_public_key_t));
}
{
smartlist_t *a_lines = find_all_by_keyword(tokens, K_A);
if (a_lines) {
find_single_ipv6_orport(a_lines, &md->ipv6_addr, &md->ipv6_orport);
smartlist_free(a_lines);
}
}
if ((tok = find_opt_by_keyword(tokens, K_FAMILY))) {
int i;
md->family = smartlist_new();
for (i=0;i<tok->n_args;++i) {
if (!is_legal_nickname_or_hexdigest(tok->args[i])) {
log_warn(LD_DIR, "Illegal nickname %s in family line",
escaped(tok->args[i]));
goto next;
}
smartlist_add(md->family, tor_strdup(tok->args[i]));
}
}
if ((tok = find_opt_by_keyword(tokens, K_P))) {
md->exit_policy = parse_short_policy(tok->args[0]);
}
if ((tok = find_opt_by_keyword(tokens, K_P6))) {
md->ipv6_exit_policy = parse_short_policy(tok->args[0]);
}
crypto_digest256(md->digest, md->body, md->bodylen, DIGEST_SHA256);
smartlist_add(result, md);
md = NULL;
next:
microdesc_free(md);
md = NULL;
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
memarea_clear(area);
smartlist_clear(tokens);
s = start_of_next_microdesc;
}
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
memarea_drop_all(area);
smartlist_free(tokens);
return result;
}
/** Return true iff this Tor version can answer directory questions
* about microdescriptors. */
int
tor_version_supports_microdescriptors(const char *platform)
{
return tor_version_as_new_as(platform, "0.2.3.1-alpha");
}
/** Parse the Tor version of the platform string <b>platform</b>,
* and compare it to the version in <b>cutoff</b>. Return 1 if
* the router is at least as new as the cutoff, else return 0.
*/
int
tor_version_as_new_as(const char *platform, const char *cutoff)
{
tor_version_t cutoff_version, router_version;
char *s, *s2, *start;
char tmp[128];
tor_assert(platform);
if (tor_version_parse(cutoff, &cutoff_version)<0) {
log_warn(LD_BUG,"cutoff version '%s' unparseable.",cutoff);
return 0;
}
if (strcmpstart(platform,"Tor ")) /* nonstandard Tor; be safe and say yes */
return 1;
start = (char *)eat_whitespace(platform+3);
if (!*start) return 0;
s = (char *)find_whitespace(start); /* also finds '\0', which is fine */
s2 = (char*)eat_whitespace(s);
if (!strcmpstart(s2, "(r") || !strcmpstart(s2, "(git-"))
s = (char*)find_whitespace(s2);
if ((size_t)(s-start+1) >= sizeof(tmp)) /* too big, no */
return 0;
strlcpy(tmp, start, s-start+1);
if (tor_version_parse(tmp, &router_version)<0) {
log_info(LD_DIR,"Router version '%s' unparseable.",tmp);
return 1; /* be safe and say yes */
}
/* Here's why we don't need to do any special handling for svn revisions:
* - If neither has an svn revision, we're fine.
* - If the router doesn't have an svn revision, we can't assume that it
* is "at least" any svn revision, so we need to return 0.
* - If the target version doesn't have an svn revision, any svn revision
* (or none at all) is good enough, so return 1.
* - If both target and router have an svn revision, we compare them.
*/
return tor_version_compare(&router_version, &cutoff_version) >= 0;
}
/** Parse a tor version from <b>s</b>, and store the result in <b>out</b>.
* Return 0 on success, -1 on failure. */
int
tor_version_parse(const char *s, tor_version_t *out)
{
char *eos=NULL;
const char *cp=NULL;
/* Format is:
* "Tor " ? NUM dot NUM dot NUM [ ( pre | rc | dot ) NUM [ - tag ] ]
*/
tor_assert(s);
tor_assert(out);
memset(out, 0, sizeof(tor_version_t));
if (!strcasecmpstart(s, "Tor "))
s += 4;
/* Get major. */
out->major = (int)strtol(s,&eos,10);
if (!eos || eos==s || *eos != '.') return -1;
cp = eos+1;
/* Get minor */
out->minor = (int) strtol(cp,&eos,10);
if (!eos || eos==cp || *eos != '.') return -1;
cp = eos+1;
/* Get micro */
out->micro = (int) strtol(cp,&eos,10);
if (!eos || eos==cp) return -1;
if (!*eos) {
out->status = VER_RELEASE;
out->patchlevel = 0;
return 0;
}
cp = eos;
/* Get status */
if (*cp == '.') {
out->status = VER_RELEASE;
++cp;
} else if (0==strncmp(cp, "pre", 3)) {
out->status = VER_PRE;
cp += 3;
} else if (0==strncmp(cp, "rc", 2)) {
out->status = VER_RC;
cp += 2;
} else {
return -1;
}
/* Get patchlevel */
out->patchlevel = (int) strtol(cp,&eos,10);
if (!eos || eos==cp) return -1;
cp = eos;
/* Get status tag. */
if (*cp == '-' || *cp == '.')
++cp;
eos = (char*) find_whitespace(cp);
if (eos-cp >= (int)sizeof(out->status_tag))
strlcpy(out->status_tag, cp, sizeof(out->status_tag));
else {
memcpy(out->status_tag, cp, eos-cp);
out->status_tag[eos-cp] = 0;
}
cp = eat_whitespace(eos);
if (!strcmpstart(cp, "(r")) {
cp += 2;
out->svn_revision = (int) strtol(cp,&eos,10);
} else if (!strcmpstart(cp, "(git-")) {
char *close_paren = strchr(cp, ')');
int hexlen;
char digest[DIGEST_LEN];
if (! close_paren)
return -1;
cp += 5;
if (close_paren-cp > HEX_DIGEST_LEN)
return -1;
hexlen = (int)(close_paren-cp);
memwipe(digest, 0, sizeof(digest));
if ( hexlen == 0 || (hexlen % 2) == 1)
return -1;
if (base16_decode(digest, hexlen/2, cp, hexlen))
return -1;
memcpy(out->git_tag, digest, hexlen/2);
out->git_tag_len = hexlen/2;
}
return 0;
}
/** Compare two tor versions; Return <0 if a < b; 0 if a ==b, >0 if a >
* b. */
int
tor_version_compare(tor_version_t *a, tor_version_t *b)
{
int i;
tor_assert(a);
tor_assert(b);
/* We take this approach to comparison to ensure the same (bogus!) behavior
* on all inputs as we would have seen before bug #21278 was fixed. The
* only important difference here is that this method doesn't cause
* a signed integer underflow.
*/
#define CMP(field) do { \
unsigned aval = (unsigned) a->field; \
unsigned bval = (unsigned) b->field; \
int result = (int) (aval - bval); \
if (result < 0) \
return -1; \
else if (result > 0) \
return 1; \
} while (0)
CMP(major);
CMP(minor);
CMP(micro);
CMP(status);
CMP(patchlevel);
if ((i = strcmp(a->status_tag, b->status_tag)))
return i;
CMP(svn_revision);
CMP(git_tag_len);
if (a->git_tag_len)
return fast_memcmp(a->git_tag, b->git_tag, a->git_tag_len);
else
return 0;
#undef CMP
}
/** Return true iff versions <b>a</b> and <b>b</b> belong to the same series.
*/
int
tor_version_same_series(tor_version_t *a, tor_version_t *b)
{
tor_assert(a);
tor_assert(b);
return ((a->major == b->major) &&
(a->minor == b->minor) &&
(a->micro == b->micro));
}
/** Helper: Given pointers to two strings describing tor versions, return -1
* if _a precedes _b, 1 if _b precedes _a, and 0 if they are equivalent.
* Used to sort a list of versions. */
static int
compare_tor_version_str_ptr_(const void **_a, const void **_b)
{
const char *a = *_a, *b = *_b;
int ca, cb;
tor_version_t va, vb;
ca = tor_version_parse(a, &va);
cb = tor_version_parse(b, &vb);
/* If they both parse, compare them. */
if (!ca && !cb)
return tor_version_compare(&va,&vb);
/* If one parses, it comes first. */
if (!ca && cb)
return -1;
if (ca && !cb)
return 1;
/* If neither parses, compare strings. Also, the directory server admin
** needs to be smacked upside the head. But Tor is tolerant and gentle. */
return strcmp(a,b);
}
/** Sort a list of string-representations of versions in ascending order. */
void
sort_version_list(smartlist_t *versions, int remove_duplicates)
{
smartlist_sort(versions, compare_tor_version_str_ptr_);
if (remove_duplicates)
smartlist_uniq(versions, compare_tor_version_str_ptr_, tor_free_);
}
/** Parse and validate the ASCII-encoded v2 descriptor in <b>desc</b>,
* write the parsed descriptor to the newly allocated *<b>parsed_out</b>, the
* binary descriptor ID of length DIGEST_LEN to <b>desc_id_out</b>, the
* encrypted introduction points to the newly allocated
* *<b>intro_points_encrypted_out</b>, their encrypted size to
* *<b>intro_points_encrypted_size_out</b>, the size of the encoded descriptor
* to *<b>encoded_size_out</b>, and a pointer to the possibly next
* descriptor to *<b>next_out</b>; return 0 for success (including validation)
* and -1 for failure.
*/
int
rend_parse_v2_service_descriptor(rend_service_descriptor_t **parsed_out,
char *desc_id_out,
char **intro_points_encrypted_out,
size_t *intro_points_encrypted_size_out,
size_t *encoded_size_out,
const char **next_out, const char *desc)
{
rend_service_descriptor_t *result =
tor_malloc_zero(sizeof(rend_service_descriptor_t));
char desc_hash[DIGEST_LEN];
const char *eos;
smartlist_t *tokens = smartlist_new();
directory_token_t *tok;
char secret_id_part[DIGEST_LEN];
int i, version, num_ok=1;
smartlist_t *versions;
char public_key_hash[DIGEST_LEN];
char test_desc_id[DIGEST_LEN];
memarea_t *area = NULL;
tor_assert(desc);
/* Check if desc starts correctly. */
if (strncmp(desc, "rendezvous-service-descriptor ",
strlen("rendezvous-service-descriptor "))) {
log_info(LD_REND, "Descriptor does not start correctly.");
goto err;
}
/* Compute descriptor hash for later validation. */
if (router_get_hash_impl(desc, strlen(desc), desc_hash,
"rendezvous-service-descriptor ",
"\nsignature", '\n', DIGEST_SHA1) < 0) {
log_warn(LD_REND, "Couldn't compute descriptor hash.");
goto err;
}
/* Determine end of string. */
eos = strstr(desc, "\nrendezvous-service-descriptor ");
if (!eos)
eos = desc + strlen(desc);
else
eos = eos + 1;
/* Check length. */
if (eos-desc > REND_DESC_MAX_SIZE) {
/* XXX023 If we are parsing this descriptor as a server, this
* should be a protocol warning. */
log_warn(LD_REND, "Descriptor length is %d which exceeds "
"maximum rendezvous descriptor size of %d bytes.",
(int)(eos-desc), REND_DESC_MAX_SIZE);
goto err;
}
/* Tokenize descriptor. */
area = memarea_new();
if (tokenize_string(area, desc, eos, tokens, desc_token_table, 0)) {
log_warn(LD_REND, "Error tokenizing descriptor.");
goto err;
}
/* Set next to next descriptor, if available. */
*next_out = eos;
/* Set length of encoded descriptor. */
*encoded_size_out = eos - desc;
/* Check min allowed length of token list. */
if (smartlist_len(tokens) < 7) {
log_warn(LD_REND, "Impossibly short descriptor.");
goto err;
}
/* Parse base32-encoded descriptor ID. */
tok = find_by_keyword(tokens, R_RENDEZVOUS_SERVICE_DESCRIPTOR);
tor_assert(tok == smartlist_get(tokens, 0));
tor_assert(tok->n_args == 1);
if (strlen(tok->args[0]) != REND_DESC_ID_V2_LEN_BASE32 ||
strspn(tok->args[0], BASE32_CHARS) != REND_DESC_ID_V2_LEN_BASE32) {
log_warn(LD_REND, "Invalid descriptor ID: '%s'", tok->args[0]);
goto err;
}
if (base32_decode(desc_id_out, DIGEST_LEN,
tok->args[0], REND_DESC_ID_V2_LEN_BASE32) < 0) {
log_warn(LD_REND, "Descriptor ID contains illegal characters: %s",
tok->args[0]);
goto err;
}
/* Parse descriptor version. */
tok = find_by_keyword(tokens, R_VERSION);
tor_assert(tok->n_args == 1);
result->version =
(int) tor_parse_long(tok->args[0], 10, 0, INT_MAX, &num_ok, NULL);
if (result->version != 2 || !num_ok) {
/* If it's <2, it shouldn't be under this format. If the number
* is greater than 2, we bumped it because we broke backward
* compatibility. See how version numbers in our other formats
* work. */
log_warn(LD_REND, "Unrecognized descriptor version: %s",
escaped(tok->args[0]));
goto err;
}
/* Parse public key. */
tok = find_by_keyword(tokens, R_PERMANENT_KEY);
result->pk = tok->key;
tok->key = NULL; /* Prevent free */
/* Parse secret ID part. */
tok = find_by_keyword(tokens, R_SECRET_ID_PART);
tor_assert(tok->n_args == 1);
if (strlen(tok->args[0]) != REND_SECRET_ID_PART_LEN_BASE32 ||
strspn(tok->args[0], BASE32_CHARS) != REND_SECRET_ID_PART_LEN_BASE32) {
log_warn(LD_REND, "Invalid secret ID part: '%s'", tok->args[0]);
goto err;
}
if (base32_decode(secret_id_part, DIGEST_LEN, tok->args[0], 32) < 0) {
log_warn(LD_REND, "Secret ID part contains illegal characters: %s",
tok->args[0]);
goto err;
}
/* Parse publication time -- up-to-date check is done when storing the
* descriptor. */
tok = find_by_keyword(tokens, R_PUBLICATION_TIME);
tor_assert(tok->n_args == 1);
if (parse_iso_time(tok->args[0], &result->timestamp) < 0) {
log_warn(LD_REND, "Invalid publication time: '%s'", tok->args[0]);
goto err;
}
/* Parse protocol versions. */
tok = find_by_keyword(tokens, R_PROTOCOL_VERSIONS);
tor_assert(tok->n_args == 1);
versions = smartlist_new();
smartlist_split_string(versions, tok->args[0], ",",
SPLIT_SKIP_SPACE|SPLIT_IGNORE_BLANK, 0);
for (i = 0; i < smartlist_len(versions); i++) {
version = (int) tor_parse_long(smartlist_get(versions, i),
10, 0, INT_MAX, &num_ok, NULL);
if (!num_ok) /* It's a string; let's ignore it. */
continue;
if (version >= REND_PROTOCOL_VERSION_BITMASK_WIDTH)
/* Avoid undefined left-shift behaviour. */
continue;
result->protocols |= 1 << version;
}
SMARTLIST_FOREACH(versions, char *, cp, tor_free(cp));
smartlist_free(versions);
/* Parse encrypted introduction points. Don't verify. */
tok = find_opt_by_keyword(tokens, R_INTRODUCTION_POINTS);
if (tok) {
if (strcmp(tok->object_type, "MESSAGE")) {
log_warn(LD_DIR, "Bad object type: introduction points should be of "
"type MESSAGE");
goto err;
}
*intro_points_encrypted_out = tor_memdup(tok->object_body,
tok->object_size);
*intro_points_encrypted_size_out = tok->object_size;
} else {
*intro_points_encrypted_out = NULL;
*intro_points_encrypted_size_out = 0;
}
/* Parse and verify signature. */
tok = find_by_keyword(tokens, R_SIGNATURE);
note_crypto_pk_op(VERIFY_RTR);
if (check_signature_token(desc_hash, DIGEST_LEN, tok, result->pk, 0,
"v2 rendezvous service descriptor") < 0)
goto err;
/* Verify that descriptor ID belongs to public key and secret ID part. */
crypto_pk_get_digest(result->pk, public_key_hash);
rend_get_descriptor_id_bytes(test_desc_id, public_key_hash,
secret_id_part);
if (tor_memneq(desc_id_out, test_desc_id, DIGEST_LEN)) {
log_warn(LD_REND, "Parsed descriptor ID does not match "
"computed descriptor ID.");
goto err;
}
goto done;
err:
rend_service_descriptor_free(result);
result = NULL;
done:
if (tokens) {
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_free(tokens);
}
if (area)
memarea_drop_all(area);
*parsed_out = result;
if (result)
return 0;
return -1;
}
/** Decrypt the encrypted introduction points in <b>ipos_encrypted</b> of
* length <b>ipos_encrypted_size</b> using <b>descriptor_cookie</b> and
* write the result to a newly allocated string that is pointed to by
* <b>ipos_decrypted</b> and its length to <b>ipos_decrypted_size</b>.
* Return 0 if decryption was successful and -1 otherwise. */
int
rend_decrypt_introduction_points(char **ipos_decrypted,
size_t *ipos_decrypted_size,
const char *descriptor_cookie,
const char *ipos_encrypted,
size_t ipos_encrypted_size)
{
tor_assert(ipos_encrypted);
tor_assert(descriptor_cookie);
if (ipos_encrypted_size < 2) {
log_warn(LD_REND, "Size of encrypted introduction points is too "
"small.");
return -1;
}
if (ipos_encrypted[0] == (int)REND_BASIC_AUTH) {
char iv[CIPHER_IV_LEN], client_id[REND_BASIC_AUTH_CLIENT_ID_LEN],
session_key[CIPHER_KEY_LEN], *dec;
int declen, client_blocks;
size_t pos = 0, len, client_entries_len;
crypto_digest_t *digest;
crypto_cipher_t *cipher;
client_blocks = (int) ipos_encrypted[1];
client_entries_len = client_blocks * REND_BASIC_AUTH_CLIENT_MULTIPLE *
REND_BASIC_AUTH_CLIENT_ENTRY_LEN;
if (ipos_encrypted_size < 2 + client_entries_len + CIPHER_IV_LEN + 1) {
log_warn(LD_REND, "Size of encrypted introduction points is too "
"small.");
return -1;
}
memcpy(iv, ipos_encrypted + 2 + client_entries_len, CIPHER_IV_LEN);
digest = crypto_digest_new();
crypto_digest_add_bytes(digest, descriptor_cookie, REND_DESC_COOKIE_LEN);
crypto_digest_add_bytes(digest, iv, CIPHER_IV_LEN);
crypto_digest_get_digest(digest, client_id,
REND_BASIC_AUTH_CLIENT_ID_LEN);
crypto_digest_free(digest);
for (pos = 2; pos < 2 + client_entries_len;
pos += REND_BASIC_AUTH_CLIENT_ENTRY_LEN) {
if (tor_memeq(ipos_encrypted + pos, client_id,
REND_BASIC_AUTH_CLIENT_ID_LEN)) {
/* Attempt to decrypt introduction points. */
cipher = crypto_cipher_new(descriptor_cookie);
if (crypto_cipher_decrypt(cipher, session_key, ipos_encrypted
+ pos + REND_BASIC_AUTH_CLIENT_ID_LEN,
CIPHER_KEY_LEN) < 0) {
log_warn(LD_REND, "Could not decrypt session key for client.");
crypto_cipher_free(cipher);
return -1;
}
crypto_cipher_free(cipher);
len = ipos_encrypted_size - 2 - client_entries_len - CIPHER_IV_LEN;
dec = tor_malloc_zero(len + 1);
declen = crypto_cipher_decrypt_with_iv(session_key, dec, len,
ipos_encrypted + 2 + client_entries_len,
ipos_encrypted_size - 2 - client_entries_len);
if (declen < 0) {
log_warn(LD_REND, "Could not decrypt introduction point string.");
tor_free(dec);
return -1;
}
if (fast_memcmpstart(dec, declen, "introduction-point ")) {
log_warn(LD_REND, "Decrypted introduction points don't "
"look like we could parse them.");
tor_free(dec);
continue;
}
*ipos_decrypted = dec;
*ipos_decrypted_size = declen;
return 0;
}
}
log_warn(LD_REND, "Could not decrypt introduction points. Please "
"check your authorization for this service!");
return -1;
} else if (ipos_encrypted[0] == (int)REND_STEALTH_AUTH) {
char *dec;
int declen;
if (ipos_encrypted_size < CIPHER_IV_LEN + 2) {
log_warn(LD_REND, "Size of encrypted introduction points is too "
"small.");
return -1;
}
dec = tor_malloc_zero(ipos_encrypted_size - CIPHER_IV_LEN - 1 + 1);
declen = crypto_cipher_decrypt_with_iv(descriptor_cookie, dec,
ipos_encrypted_size -
CIPHER_IV_LEN - 1,
ipos_encrypted + 1,
ipos_encrypted_size - 1);
if (declen < 0) {
log_warn(LD_REND, "Decrypting introduction points failed!");
tor_free(dec);
return -1;
}
*ipos_decrypted = dec;
*ipos_decrypted_size = declen;
return 0;
} else {
log_warn(LD_REND, "Unknown authorization type number: %d",
ipos_encrypted[0]);
return -1;
}
}
/** Parse the encoded introduction points in <b>intro_points_encoded</b> of
* length <b>intro_points_encoded_size</b> and write the result to the
* descriptor in <b>parsed</b>; return the number of successfully parsed
* introduction points or -1 in case of a failure. */
int
rend_parse_introduction_points(rend_service_descriptor_t *parsed,
const char *intro_points_encoded,
size_t intro_points_encoded_size)
{
const char *current_ipo, *end_of_intro_points;
smartlist_t *tokens = NULL;
directory_token_t *tok;
rend_intro_point_t *intro;
extend_info_t *info;
int result, num_ok=1;
memarea_t *area = NULL;
tor_assert(parsed);
/** Function may only be invoked once. */
tor_assert(!parsed->intro_nodes);
if (!intro_points_encoded || intro_points_encoded_size == 0) {
log_warn(LD_REND, "Empty or zero size introduction point list");
goto err;
}
/* Consider one intro point after the other. */
current_ipo = intro_points_encoded;
end_of_intro_points = intro_points_encoded + intro_points_encoded_size;
tokens = smartlist_new();
parsed->intro_nodes = smartlist_new();
area = memarea_new();
while (!fast_memcmpstart(current_ipo, end_of_intro_points-current_ipo,
"introduction-point ")) {
/* Determine end of string. */
const char *eos = tor_memstr(current_ipo, end_of_intro_points-current_ipo,
"\nintroduction-point ");
if (!eos)
eos = end_of_intro_points;
else
eos = eos+1;
tor_assert(eos <= intro_points_encoded+intro_points_encoded_size);
/* Free tokens and clear token list. */
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_clear(tokens);
memarea_clear(area);
/* Tokenize string. */
if (tokenize_string(area, current_ipo, eos, tokens, ipo_token_table, 0)) {
log_warn(LD_REND, "Error tokenizing introduction point");
goto err;
}
/* Advance to next introduction point, if available. */
current_ipo = eos;
/* Check minimum allowed length of introduction point. */
if (smartlist_len(tokens) < 5) {
log_warn(LD_REND, "Impossibly short introduction point.");
goto err;
}
/* Allocate new intro point and extend info. */
intro = tor_malloc_zero(sizeof(rend_intro_point_t));
info = intro->extend_info = tor_malloc_zero(sizeof(extend_info_t));
/* Parse identifier. */
tok = find_by_keyword(tokens, R_IPO_IDENTIFIER);
if (base32_decode(info->identity_digest, DIGEST_LEN,
tok->args[0], REND_INTRO_POINT_ID_LEN_BASE32) < 0) {
log_warn(LD_REND, "Identity digest contains illegal characters: %s",
tok->args[0]);
rend_intro_point_free(intro);
goto err;
}
/* Write identifier to nickname. */
info->nickname[0] = '$';
base16_encode(info->nickname + 1, sizeof(info->nickname) - 1,
info->identity_digest, DIGEST_LEN);
/* Parse IP address. */
tok = find_by_keyword(tokens, R_IPO_IP_ADDRESS);
if (tor_addr_parse(&info->addr, tok->args[0])<0) {
log_warn(LD_REND, "Could not parse introduction point address.");
rend_intro_point_free(intro);
goto err;
}
if (tor_addr_family(&info->addr) != AF_INET) {
log_warn(LD_REND, "Introduction point address was not ipv4.");
rend_intro_point_free(intro);
goto err;
}
/* Parse onion port. */
tok = find_by_keyword(tokens, R_IPO_ONION_PORT);
info->port = (uint16_t) tor_parse_long(tok->args[0],10,1,65535,
&num_ok,NULL);
if (!info->port || !num_ok) {
log_warn(LD_REND, "Introduction point onion port %s is invalid",
escaped(tok->args[0]));
rend_intro_point_free(intro);
goto err;
}
/* Parse onion key. */
tok = find_by_keyword(tokens, R_IPO_ONION_KEY);
if (!crypto_pk_public_exponent_ok(tok->key)) {
log_warn(LD_REND,
"Introduction point's onion key had invalid exponent.");
rend_intro_point_free(intro);
goto err;
}
info->onion_key = tok->key;
tok->key = NULL; /* Prevent free */
/* Parse service key. */
tok = find_by_keyword(tokens, R_IPO_SERVICE_KEY);
if (!crypto_pk_public_exponent_ok(tok->key)) {
log_warn(LD_REND,
"Introduction point key had invalid exponent.");
rend_intro_point_free(intro);
goto err;
}
intro->intro_key = tok->key;
tok->key = NULL; /* Prevent free */
/* Add extend info to list of introduction points. */
smartlist_add(parsed->intro_nodes, intro);
}
result = smartlist_len(parsed->intro_nodes);
goto done;
err:
result = -1;
done:
/* Free tokens and clear token list. */
if (tokens) {
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_free(tokens);
}
if (area)
memarea_drop_all(area);
return result;
}
/** Parse the content of a client_key file in <b>ckstr</b> and add
* rend_authorized_client_t's for each parsed client to
* <b>parsed_clients</b>. Return the number of parsed clients as result
* or -1 for failure. */
int
rend_parse_client_keys(strmap_t *parsed_clients, const char *ckstr)
{
int result = -1;
smartlist_t *tokens;
directory_token_t *tok;
const char *current_entry = NULL;
memarea_t *area = NULL;
if (!ckstr || strlen(ckstr) == 0)
return -1;
tokens = smartlist_new();
/* Begin parsing with first entry, skipping comments or whitespace at the
* beginning. */
area = memarea_new();
current_entry = eat_whitespace(ckstr);
while (!strcmpstart(current_entry, "client-name ")) {
rend_authorized_client_t *parsed_entry;
size_t len;
char descriptor_cookie_tmp[REND_DESC_COOKIE_LEN+2];
/* Determine end of string. */
const char *eos = strstr(current_entry, "\nclient-name ");
if (!eos)
eos = current_entry + strlen(current_entry);
else
eos = eos + 1;
/* Free tokens and clear token list. */
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_clear(tokens);
memarea_clear(area);
/* Tokenize string. */
if (tokenize_string(area, current_entry, eos, tokens,
client_keys_token_table, 0)) {
log_warn(LD_REND, "Error tokenizing client keys file.");
goto err;
}
/* Advance to next entry, if available. */
current_entry = eos;
/* Check minimum allowed length of token list. */
if (smartlist_len(tokens) < 2) {
log_warn(LD_REND, "Impossibly short client key entry.");
goto err;
}
/* Parse client name. */
tok = find_by_keyword(tokens, C_CLIENT_NAME);
tor_assert(tok == smartlist_get(tokens, 0));
tor_assert(tok->n_args == 1);
len = strlen(tok->args[0]);
if (len < 1 || len > 19 ||
strspn(tok->args[0], REND_LEGAL_CLIENTNAME_CHARACTERS) != len) {
log_warn(LD_CONFIG, "Illegal client name: %s. (Length must be "
"between 1 and 19, and valid characters are "
"[A-Za-z0-9+-_].)", tok->args[0]);
goto err;
}
/* Check if client name is duplicate. */
if (strmap_get(parsed_clients, tok->args[0])) {
log_warn(LD_CONFIG, "HiddenServiceAuthorizeClient contains a "
"duplicate client name: '%s'. Ignoring.", tok->args[0]);
goto err;
}
parsed_entry = tor_malloc_zero(sizeof(rend_authorized_client_t));
parsed_entry->client_name = tor_strdup(tok->args[0]);
strmap_set(parsed_clients, parsed_entry->client_name, parsed_entry);
/* Parse client key. */
tok = find_opt_by_keyword(tokens, C_CLIENT_KEY);
if (tok) {
parsed_entry->client_key = tok->key;
tok->key = NULL; /* Prevent free */
}
/* Parse descriptor cookie. */
tok = find_by_keyword(tokens, C_DESCRIPTOR_COOKIE);
tor_assert(tok->n_args == 1);
if (strlen(tok->args[0]) != REND_DESC_COOKIE_LEN_BASE64 + 2) {
log_warn(LD_REND, "Descriptor cookie has illegal length: %s",
escaped(tok->args[0]));
goto err;
}
/* The size of descriptor_cookie_tmp needs to be REND_DESC_COOKIE_LEN+2,
* because a base64 encoding of length 24 does not fit into 16 bytes in all
* cases. */
if (base64_decode(descriptor_cookie_tmp, sizeof(descriptor_cookie_tmp),
tok->args[0], strlen(tok->args[0]))
!= REND_DESC_COOKIE_LEN) {
log_warn(LD_REND, "Descriptor cookie contains illegal characters: "
"%s", escaped(tok->args[0]));
goto err;
}
memcpy(parsed_entry->descriptor_cookie, descriptor_cookie_tmp,
REND_DESC_COOKIE_LEN);
}
result = strmap_size(parsed_clients);
goto done;
err:
result = -1;
done:
/* Free tokens and clear token list. */
SMARTLIST_FOREACH(tokens, directory_token_t *, t, token_clear(t));
smartlist_free(tokens);
if (area)
memarea_drop_all(area);
return result;
}