From: Jann Horn Date: Sat, 29 Mar 2014 17:29:13 +0000 (+0100) Subject: commit all the code! X-Git-Url: http://git.thejh.net/?p=detour.git;a=commitdiff_plain;h=b6172a42cb1cc9e60c23696f20f5cdbd86e64cec;ds=sidebyside commit all the code! --- diff --git a/.gitignore b/.gitignore index 37c46df..df7f308 100644 --- a/.gitignore +++ b/.gitignore @@ -1 +1,6 @@ pulser +pulsehunter +pulserecord +random_bits + +out diff --git a/common.c b/common.c index d45457b..a2a4ca0 100644 --- a/common.c +++ b/common.c @@ -1,9 +1,116 @@ +// This file is basically utility functions from various places mashed together, including +// from my helper library libjh. + #include #include -#include #include #include #include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define JH_TCP_HINTS (&libjh_tcp_hints) +const struct addrinfo libjh_tcp_hints = { + .ai_flags = AI_ADDRCONFIG, + .ai_family = AF_UNSPEC, + .ai_socktype = SOCK_STREAM, + .ai_protocol = 0 +}; + +int netopen_server(const char *node /*NULL for ANY*/, const char *service, const struct addrinfo *hints) { + struct addrinfo hints_; + if (hints == &libjh_tcp_hints) { + hints_ = *hints; + hints_.ai_flags |= AI_PASSIVE; + hints_.ai_flags &= ~AI_ADDRCONFIG; + hints = &hints_; + } + + struct addrinfo *addrs; + int gai_res = getaddrinfo(node, service, hints, &addrs); + if (gai_res) return gai_res; + + int s = socket(addrs[0].ai_family, addrs[0].ai_socktype, addrs[0].ai_protocol); + if (s == -1) goto err_socket; + + if (bind(s, addrs[0].ai_addr, addrs[0].ai_addrlen)) goto err_bind_n_listen; + if (listen(s, 16)) goto err_bind_n_listen; + + freeaddrinfo(addrs); + return s; + +err_bind_n_listen:; + int errno_ = errno; + close(s); + errno = errno_; +err_socket: + freeaddrinfo(addrs); + return EAI_SYSTEM; +} + +ssize_t read_nointr(int fd, void *buf, size_t count, int *last_res) { + errno = 0; + size_t done = 0; + while (done < count) { + ssize_t part_res = read(fd, buf+done, count-done); + if (part_res == -1 && errno == EINTR) continue; + if (part_res <= 0) { + if (last_res) *last_res = part_res; + if (done) return done; + return part_res; + } + done += part_res; + } + if (last_res) *last_res = 1; + return done; +} + +void *slurp_fd(int fd, size_t *len_out) { + int errno_; + + size_t size_guess; + + struct stat st; + if (fstat(fd, &st) == 0) { + if (st.st_size > 0) { + size_guess = st.st_size; + } + } + + char *buf = NULL; + int done = 0; + + while (1) { + buf = realloc(buf, size_guess); + if (buf == NULL) return NULL; + int last_res; + ssize_t read_res = read_nointr(fd, buf+done, size_guess-done, &last_res); + if (last_res == -1) { errno_=errno; free(buf); errno=errno_; return NULL; } + done += read_res; + if (len_out) *len_out = done; + return buf; + } +} + +void *slurp_file(char *path, size_t *len_out) { + int fd = open(path, O_RDONLY|O_CLOEXEC); + if (fd == -1) return NULL; + char *res = slurp_fd(fd, len_out); + int errno_ = errno; + close(fd); + errno = errno_; + return res; +} time_t real_seconds(void) { struct timespec t; @@ -15,23 +122,23 @@ time_t real_seconds(void) { /* Subtract the `struct timeval' values X and Y, storing the result in RESULT. Return 1 if the difference is negative, otherwise 0. */ -int timeval_subtract (struct timeval *result, struct timeval *x, struct timeval *y) { +int timespec_subtract(struct timespec *result, struct timespec *x, struct timespec *y) { /* Perform the carry for the later subtraction by updating y. */ - if (x->tv_usec < y->tv_usec) { - int nsec = (y->tv_usec - x->tv_usec) / 1000000 + 1; - y->tv_usec -= 1000000 * nsec; + if (x->tv_nsec < y->tv_nsec) { + int nsec = (y->tv_nsec - x->tv_nsec) / 1000000000 + 1; + y->tv_nsec -= 1000000000 * nsec; y->tv_sec += nsec; } - if (x->tv_usec - y->tv_usec > 1000000) { - int nsec = (x->tv_usec - y->tv_usec) / 1000000; - y->tv_usec += 1000000 * nsec; + if (x->tv_nsec - y->tv_nsec > 1000000000) { + int nsec = (x->tv_nsec - y->tv_nsec) / 1000000000; + y->tv_nsec += 1000000000 * nsec; y->tv_sec -= nsec; } /* Compute the time remaining to wait. tv_usec is certainly positive. */ result->tv_sec = x->tv_sec - y->tv_sec; - result->tv_usec = x->tv_usec - y->tv_usec; + result->tv_nsec = x->tv_nsec - y->tv_nsec; /* Return 1 if result is negative. */ return x->tv_sec < y->tv_sec; diff --git a/compile.sh b/compile.sh index d4f14df..cc336cf 100755 --- a/compile.sh +++ b/compile.sh @@ -1,2 +1,6 @@ #!/bin/sh -gcc -Wall -o pulser pulser.c -std=gnu99 -lrt -ljh -lev +gcc -g -Wall -o pulser pulser.c -std=gnu99 -lrt +gcc -g -Wall -o pulserecord pulserecord.c -std=gnu99 -lpcap -lrt +gcc -g -Wall -o random_bits random_bits.c -std=gnu99 +gcc -g -Wall -o pulsehunter pulsehunter.c -std=gnu99 -lrt + diff --git a/devurandom.c b/devurandom.c new file mode 100644 index 0000000..8a7a7b6 --- /dev/null +++ b/devurandom.c @@ -0,0 +1,35 @@ +#include +#include +#include +#include + +/* it's really stupid that there isn't a syscall for this */ + +static int fd = -1; + +void randombytes(void *x_,unsigned long long xlen) +{ + unsigned char *x = x_; + int i; + + if (fd == -1) { + for (;;) { + fd = open("/dev/urandom",O_RDONLY); + if (fd != -1) break; + sleep(1); + } + } + + while (xlen > 0) { + if (xlen < 1048576) i = xlen; else i = 1048576; + + i = read(fd,x,i); + if (i < 1) { + sleep(1); + continue; + } + + x += i; + xlen -= i; + } +} diff --git a/pulsehunter.c b/pulsehunter.c new file mode 100644 index 0000000..f3a0de5 --- /dev/null +++ b/pulsehunter.c @@ -0,0 +1,27 @@ +#include "common.c" + +int main(int argc, char **argv) { + if (argc != 2) puts("invocation: ./pulsehunter "), exit(1); + char *bits = argv[1]; + int nbits = strlen(bits); + + if (chdir("out")) perror("unable to enter directory 'out'"), exit(1); + DIR *d = opendir("."); + if (!d) perror("unable to open directory 'out'"), exit(1); + for (struct dirent *e = (errno=0,readdir(d)); e; e = (errno=0,readdir(d))) { + if (e->d_name[0] == '.') continue; + size_t len; + char *data = slurp_file(e->d_name, &len); + if (!data) { perror("error while slurping dirent"); continue; } + int maxbits = 0; + for (int i=0; i<((int)len)-nbits; i++) { + int matching = 0; + for (int j=0; j maxbits) maxbits = matching; + } + printf("%d\t%s\n", maxbits, e->d_name); + free(data); + } + if (errno) perror("error while reading directory 'out'"), exit(1); + return 0; +} \ No newline at end of file diff --git a/pulserecord.c b/pulserecord.c new file mode 100644 index 0000000..f7bec56 --- /dev/null +++ b/pulserecord.c @@ -0,0 +1,224 @@ +// most of this copied from http://www.tcpdump.org/pcap.html and other places + +#include +#include "common.c" +#include "uthash.h" + + +/* Ethernet addresses are 6 bytes */ +#define ETHER_ADDR_LEN 6 + +/* ethernet headers are always exactly 14 bytes */ +#define SIZE_ETHERNET 14 + +/* Ethernet header */ +struct sniff_ethernet { + u_char ether_dhost[ETHER_ADDR_LEN]; /* Destination host address */ + u_char ether_shost[ETHER_ADDR_LEN]; /* Source host address */ + u_short ether_type; /* IP? ARP? RARP? etc */ +}; + +/* IP header */ +struct sniff_ip { + u_char ip_vhl; /* version << 4 | header length >> 2 */ + u_char ip_tos; /* type of service */ + u_short ip_len; /* total length */ + u_short ip_id; /* identification */ + u_short ip_off; /* fragment offset field */ +#define IP_RF 0x8000 /* reserved fragment flag */ +#define IP_DF 0x4000 /* dont fragment flag */ +#define IP_MF 0x2000 /* more fragments flag */ +#define IP_OFFMASK 0x1fff /* mask for fragmenting bits */ + u_char ip_ttl; /* time to live */ + u_char ip_p; /* protocol */ + u_short ip_sum; /* checksum */ + struct in_addr ip_src,ip_dst; /* source and dest address */ +}; +#define IP_HL(ip) (((ip)->ip_vhl) & 0x0f) +#define IP_V(ip) (((ip)->ip_vhl) >> 4) + +/* TCP header */ +typedef u_int tcp_seq; + +struct sniff_tcp { + u_short th_sport; /* source port */ + u_short th_dport; /* destination port */ + tcp_seq th_seq; /* sequence number */ + tcp_seq th_ack; /* acknowledgement number */ + u_char th_offx2; /* data offset, rsvd */ +#define TH_OFF(th) (((th)->th_offx2 & 0xf0) >> 4) + u_char th_flags; +#define TH_FIN 0x01 +#define TH_SYN 0x02 +#define TH_RST 0x04 +#define TH_PUSH 0x08 +#define TH_ACK 0x10 +#define TH_URG 0x20 +#define TH_ECE 0x40 +#define TH_CWR 0x80 +#define TH_FLAGS (TH_FIN|TH_SYN|TH_RST|TH_ACK|TH_URG|TH_ECE|TH_CWR) + u_short th_win; /* window */ + u_short th_sum; /* checksum */ + u_short th_urp; /* urgent pointer */ +}; + + + +struct con { + UT_hash_handle hh; + char name[47]; + unsigned int bytesA, bytesB; + unsigned int idle; +}; +struct con *cons = NULL; + +time_t t; +bool even; + + +void packet_handler(u_char *user, const struct pcap_pkthdr *h, const u_char *packet) { + unsigned int len = h->len; + + /* The ethernet header */ + //const struct sniff_ethernet *ethernet = (struct sniff_ethernet*)(packet); + /* The IP header */ + const struct sniff_ip *ip = (struct sniff_ip*)(packet + SIZE_ETHERNET); + u_int size_ip = IP_HL(ip)*4; + if (size_ip < 20) { + printf(" * Invalid IP header length: %u bytes\n", size_ip); + return; + } + /* The TCP header */ + const struct sniff_tcp *tcp = (struct sniff_tcp*)(packet + SIZE_ETHERNET + size_ip); + u_int size_tcp = TH_OFF(tcp)*4; + if (size_tcp < 20) { + printf(" * Invalid TCP header length: %u bytes\n", size_tcp); + return; + } + + // construct name string + // length of IP: 3+1+3+1+3+1+3=15 + // length of port: 5 + // total: 2*(15+1+5)+4+1=47 including '\0' + char name[47]; + uint32_t ipA = ntohl(ip->ip_src.s_addr); + uint32_t ipB = ntohl(ip->ip_dst.s_addr); + sprintf(name, "%u.%u.%u.%u:%u -> %u.%u.%u.%u:%u", + ipA>>24, (ipA>>16)&0xff, (ipA>>8)&0xff, ipA&0xff, ntohs(tcp->th_sport), + ipB>>24, (ipB>>16)&0xff, (ipB>>8)&0xff, ipB&0xff, ntohs(tcp->th_dport)); + + //printf("%s: %u\n", name, len); + + struct con *c; + HASH_FIND_STR(cons, name, c); + if (!c) { + c = calloc(1, sizeof(struct con)); + if (!c) printf("calloc fail"), exit(1); + strcpy(c->name, name); + HASH_ADD_STR(cons, name, c); + } + + if (even) + c->bytesB += len; + else + c->bytesA += len; + c->idle = 0; +} + +int main(int argc, char **argv) { + if (argc != 2) puts("invocation: ./pulserecord "), exit(1); + char *dev = argv[1]; + + setbuf(stdout, NULL); + char errbuf[PCAP_ERRBUF_SIZE]; + printf("Device: %s\n", dev); + + mkdir("out", 0700); + if (chdir("out")) perror("unable to enter directory 'out'"), exit(1); + + // We use a zero-timeout. The pcap manual says: + // "to_ms is the read time out in milliseconds (a value of 0 means + // no time out; on at least some platforms, this means that you may + // wait until a sufficient number of packets arrive before seeing + // any packets, so you should use a non-zero timeout)." + // That's simply not acceptable for us, so we can use a zero timeout + // just as well and tell everyone to use a sensible OS. :D + pcap_t *handle = pcap_open_live(dev, BUFSIZ, 1, 0, errbuf); + if (!handle) printf("can't open device %s: %s\n", dev, errbuf), exit(1); + if (pcap_datalink(handle) != DLT_EN10MB) + printf("Device %s doesn't provide Ethernet headers - not supported\n", dev), exit(1); + if (pcap_setnonblock(handle, 1, errbuf) == -1) printf("unable to go nonblocking\n"), exit(1); + int pcap_fd = pcap_get_selectable_fd(handle); + if (pcap_fd == -1) printf("unable to get a pcap fd\n"), exit(1); + + struct bpf_program fp; /* The compiled filter */ + if (pcap_compile(handle, &fp, "tcp", 1, PCAP_NETMASK_UNKNOWN) == -1) { + printf("Couldn't parse filter: %s\n", pcap_geterr(handle)); + exit(1); + } + if (pcap_setfilter(handle, &fp) == -1) { + printf("Couldn't install filter: %s\n", pcap_geterr(handle)); + exit(1); + } + + t = round_up(real_seconds(), 2); + even = (t&3) == 0; /* are we processing part 2/2 of one encoded bit? */ + while (1) { + struct timespec dst, delta; + dst.tv_sec = t; + dst.tv_nsec = 0; + + // this inner loop runs until we hit a 2s-boundary + while (1) { + fd_set rfds; + FD_ZERO(&rfds); + FD_SET(pcap_fd, &rfds); + struct timespec cur; + int r = clock_gettime(CLOCK_REALTIME, &cur); + assert(r==0); + if (timespec_subtract(&delta, &dst, &cur)) break; + r = pselect(pcap_fd+1, &rfds, NULL, NULL, &delta, NULL); + if (r == -1) perror("select failed"), exit(1); + if (r == 0) { + // timeout + break; + } else { + r = pcap_dispatch(handle, -1, packet_handler, NULL); + if (r < 0) printf("pcap_dispatch failed\n"), exit(1); + } + } + + if (even) { // cycle complete + struct con *c, *tmp; + HASH_ITER(hh, cons, c, tmp) { + if (c->idle >= 10) { + HASH_DEL(cons, c); + free(c); + c = NULL; // just to be sure + continue; + } + char bit = '_'; + if (!c->idle) + bit = (c->bytesA < c->bytesB) ? '1' : '0'; + int fd = open(c->name, O_WRONLY|O_APPEND|O_CREAT, 0666); + if (fd == -1) { + perror("unable to open confile"); + } else { +w:; ssize_t r = write(fd, &bit, 1); + if (r == -1 && errno == EINTR) goto w; + if (r == -1) perror("confile write failed"); + if (r == 0) puts("confile write failed"); + close(fd); + } + c->bytesA = 0; + c->bytesB = 0; + c->idle++; + } + } + + t += 2; + even = !even; + } + + return 0; +} diff --git a/random_bits.c b/random_bits.c new file mode 100644 index 0000000..4377244 --- /dev/null +++ b/random_bits.c @@ -0,0 +1,13 @@ +#include "devurandom.c" +#include +#include + +int main(void) { + // yup, this is cheap and slow :D + char bits[65]; + randombytes(bits, sizeof(bits)); + for (int i=0; i /* memcmp,strlen */ +#include /* ptrdiff_t */ +#include /* exit() */ + +/* These macros use decltype or the earlier __typeof GNU extension. + As decltype is only available in newer compilers (VS2010 or gcc 4.3+ + when compiling c++ source) this code uses whatever method is needed + or, for VS2008 where neither is available, uses casting workarounds. */ +#ifdef _MSC_VER /* MS compiler */ +#if _MSC_VER >= 1600 && defined(__cplusplus) /* VS2010 or newer in C++ mode */ +#define DECLTYPE(x) (decltype(x)) +#else /* VS2008 or older (or VS2010 in C mode) */ +#define NO_DECLTYPE +#define DECLTYPE(x) +#endif +#else /* GNU, Sun and other compilers */ +#define DECLTYPE(x) (__typeof(x)) +#endif + +#ifdef NO_DECLTYPE +#define DECLTYPE_ASSIGN(dst,src) \ +do { \ + char **_da_dst = (char**)(&(dst)); \ + *_da_dst = (char*)(src); \ +} while(0) +#else +#define DECLTYPE_ASSIGN(dst,src) \ +do { \ + (dst) = DECLTYPE(dst)(src); \ +} while(0) +#endif + +/* a number of the hash function use uint32_t which isn't defined on win32 */ +#ifdef _MSC_VER +typedef unsigned int uint32_t; +typedef unsigned char uint8_t; +#else +#include /* uint32_t */ +#endif + +#define UTHASH_VERSION 1.9.9 + +#ifndef uthash_fatal +#define uthash_fatal(msg) exit(-1) /* fatal error (out of memory,etc) */ +#endif +#ifndef uthash_malloc +#define uthash_malloc(sz) malloc(sz) /* malloc fcn */ +#endif +#ifndef uthash_free +#define uthash_free(ptr,sz) free(ptr) /* free fcn */ +#endif + +#ifndef uthash_noexpand_fyi +#define uthash_noexpand_fyi(tbl) /* can be defined to log noexpand */ +#endif +#ifndef uthash_expand_fyi +#define uthash_expand_fyi(tbl) /* can be defined to log expands */ +#endif + +/* initial number of buckets */ +#define HASH_INITIAL_NUM_BUCKETS 32 /* initial number of buckets */ +#define HASH_INITIAL_NUM_BUCKETS_LOG2 5 /* lg2 of initial number of buckets */ +#define HASH_BKT_CAPACITY_THRESH 10 /* expand when bucket count reaches */ + +/* calculate the element whose hash handle address is hhe */ +#define ELMT_FROM_HH(tbl,hhp) ((void*)(((char*)(hhp)) - ((tbl)->hho))) + +#define HASH_FIND(hh,head,keyptr,keylen,out) \ +do { \ + unsigned _hf_bkt,_hf_hashv; \ + out=NULL; \ + if (head) { \ + HASH_FCN(keyptr,keylen, (head)->hh.tbl->num_buckets, _hf_hashv, _hf_bkt); \ + if (HASH_BLOOM_TEST((head)->hh.tbl, _hf_hashv)) { \ + HASH_FIND_IN_BKT((head)->hh.tbl, hh, (head)->hh.tbl->buckets[ _hf_bkt ], \ + keyptr,keylen,out); \ + } \ + } \ +} while (0) + +#ifdef HASH_BLOOM +#define HASH_BLOOM_BITLEN (1ULL << HASH_BLOOM) +#define HASH_BLOOM_BYTELEN (HASH_BLOOM_BITLEN/8) + ((HASH_BLOOM_BITLEN%8) ? 1:0) +#define HASH_BLOOM_MAKE(tbl) \ +do { \ + (tbl)->bloom_nbits = HASH_BLOOM; \ + (tbl)->bloom_bv = (uint8_t*)uthash_malloc(HASH_BLOOM_BYTELEN); \ + if (!((tbl)->bloom_bv)) { uthash_fatal( "out of memory"); } \ + memset((tbl)->bloom_bv, 0, HASH_BLOOM_BYTELEN); \ + (tbl)->bloom_sig = HASH_BLOOM_SIGNATURE; \ +} while (0) + +#define HASH_BLOOM_FREE(tbl) \ +do { \ + uthash_free((tbl)->bloom_bv, HASH_BLOOM_BYTELEN); \ +} while (0) + +#define HASH_BLOOM_BITSET(bv,idx) (bv[(idx)/8] |= (1U << ((idx)%8))) +#define HASH_BLOOM_BITTEST(bv,idx) (bv[(idx)/8] & (1U << ((idx)%8))) + +#define HASH_BLOOM_ADD(tbl,hashv) \ + HASH_BLOOM_BITSET((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1))) + +#define HASH_BLOOM_TEST(tbl,hashv) \ + HASH_BLOOM_BITTEST((tbl)->bloom_bv, (hashv & (uint32_t)((1ULL << (tbl)->bloom_nbits) - 1))) + +#else +#define HASH_BLOOM_MAKE(tbl) +#define HASH_BLOOM_FREE(tbl) +#define HASH_BLOOM_ADD(tbl,hashv) +#define HASH_BLOOM_TEST(tbl,hashv) (1) +#define HASH_BLOOM_BYTELEN 0 +#endif + +#define HASH_MAKE_TABLE(hh,head) \ +do { \ + (head)->hh.tbl = (UT_hash_table*)uthash_malloc( \ + sizeof(UT_hash_table)); \ + if (!((head)->hh.tbl)) { uthash_fatal( "out of memory"); } \ + memset((head)->hh.tbl, 0, sizeof(UT_hash_table)); \ + (head)->hh.tbl->tail = &((head)->hh); \ + (head)->hh.tbl->num_buckets = HASH_INITIAL_NUM_BUCKETS; \ + (head)->hh.tbl->log2_num_buckets = HASH_INITIAL_NUM_BUCKETS_LOG2; \ + (head)->hh.tbl->hho = (char*)(&(head)->hh) - (char*)(head); \ + (head)->hh.tbl->buckets = (UT_hash_bucket*)uthash_malloc( \ + HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \ + if (! (head)->hh.tbl->buckets) { uthash_fatal( "out of memory"); } \ + memset((head)->hh.tbl->buckets, 0, \ + HASH_INITIAL_NUM_BUCKETS*sizeof(struct UT_hash_bucket)); \ + HASH_BLOOM_MAKE((head)->hh.tbl); \ + (head)->hh.tbl->signature = HASH_SIGNATURE; \ +} while(0) + +#define HASH_ADD(hh,head,fieldname,keylen_in,add) \ + HASH_ADD_KEYPTR(hh,head,&((add)->fieldname),keylen_in,add) + +#define HASH_REPLACE(hh,head,fieldname,keylen_in,add,replaced) \ +do { \ + replaced=NULL; \ + HASH_FIND(hh,head,&((add)->fieldname),keylen_in,replaced); \ + if (replaced!=NULL) { \ + HASH_DELETE(hh,head,replaced); \ + }; \ + HASH_ADD(hh,head,fieldname,keylen_in,add); \ +} while(0) + +#define HASH_ADD_KEYPTR(hh,head,keyptr,keylen_in,add) \ +do { \ + unsigned _ha_bkt; \ + (add)->hh.next = NULL; \ + (add)->hh.key = (char*)(keyptr); \ + (add)->hh.keylen = (unsigned)(keylen_in); \ + if (!(head)) { \ + head = (add); \ + (head)->hh.prev = NULL; \ + HASH_MAKE_TABLE(hh,head); \ + } else { \ + (head)->hh.tbl->tail->next = (add); \ + (add)->hh.prev = ELMT_FROM_HH((head)->hh.tbl, (head)->hh.tbl->tail); \ + (head)->hh.tbl->tail = &((add)->hh); \ + } \ + (head)->hh.tbl->num_items++; \ + (add)->hh.tbl = (head)->hh.tbl; \ + HASH_FCN(keyptr,keylen_in, (head)->hh.tbl->num_buckets, \ + (add)->hh.hashv, _ha_bkt); \ + HASH_ADD_TO_BKT((head)->hh.tbl->buckets[_ha_bkt],&(add)->hh); \ + HASH_BLOOM_ADD((head)->hh.tbl,(add)->hh.hashv); \ + HASH_EMIT_KEY(hh,head,keyptr,keylen_in); \ + HASH_FSCK(hh,head); \ +} while(0) + +#define HASH_TO_BKT( hashv, num_bkts, bkt ) \ +do { \ + bkt = ((hashv) & ((num_bkts) - 1)); \ +} while(0) + +/* delete "delptr" from the hash table. + * "the usual" patch-up process for the app-order doubly-linked-list. + * The use of _hd_hh_del below deserves special explanation. + * These used to be expressed using (delptr) but that led to a bug + * if someone used the same symbol for the head and deletee, like + * HASH_DELETE(hh,users,users); + * We want that to work, but by changing the head (users) below + * we were forfeiting our ability to further refer to the deletee (users) + * in the patch-up process. Solution: use scratch space to + * copy the deletee pointer, then the latter references are via that + * scratch pointer rather than through the repointed (users) symbol. + */ +#define HASH_DELETE(hh,head,delptr) \ +do { \ + unsigned _hd_bkt; \ + struct UT_hash_handle *_hd_hh_del; \ + if ( ((delptr)->hh.prev == NULL) && ((delptr)->hh.next == NULL) ) { \ + uthash_free((head)->hh.tbl->buckets, \ + (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ + HASH_BLOOM_FREE((head)->hh.tbl); \ + uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ + head = NULL; \ + } else { \ + _hd_hh_del = &((delptr)->hh); \ + if ((delptr) == ELMT_FROM_HH((head)->hh.tbl,(head)->hh.tbl->tail)) { \ + (head)->hh.tbl->tail = \ + (UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ + (head)->hh.tbl->hho); \ + } \ + if ((delptr)->hh.prev) { \ + ((UT_hash_handle*)((ptrdiff_t)((delptr)->hh.prev) + \ + (head)->hh.tbl->hho))->next = (delptr)->hh.next; \ + } else { \ + DECLTYPE_ASSIGN(head,(delptr)->hh.next); \ + } \ + if (_hd_hh_del->next) { \ + ((UT_hash_handle*)((ptrdiff_t)_hd_hh_del->next + \ + (head)->hh.tbl->hho))->prev = \ + _hd_hh_del->prev; \ + } \ + HASH_TO_BKT( _hd_hh_del->hashv, (head)->hh.tbl->num_buckets, _hd_bkt); \ + HASH_DEL_IN_BKT(hh,(head)->hh.tbl->buckets[_hd_bkt], _hd_hh_del); \ + (head)->hh.tbl->num_items--; \ + } \ + HASH_FSCK(hh,head); \ +} while (0) + + +/* convenience forms of HASH_FIND/HASH_ADD/HASH_DEL */ +#define HASH_FIND_STR(head,findstr,out) \ + HASH_FIND(hh,head,findstr,strlen(findstr),out) +#define HASH_ADD_STR(head,strfield,add) \ + HASH_ADD(hh,head,strfield[0],strlen(add->strfield),add) +#define HASH_REPLACE_STR(head,strfield,add,replaced) \ + HASH_REPLACE(hh,head,strfield,strlen(add->strfield),add,replaced) +#define HASH_FIND_INT(head,findint,out) \ + HASH_FIND(hh,head,findint,sizeof(int),out) +#define HASH_ADD_INT(head,intfield,add) \ + HASH_ADD(hh,head,intfield,sizeof(int),add) +#define HASH_REPLACE_INT(head,intfield,add,replaced) \ + HASH_REPLACE(hh,head,intfield,sizeof(int),add,replaced) +#define HASH_FIND_PTR(head,findptr,out) \ + HASH_FIND(hh,head,findptr,sizeof(void *),out) +#define HASH_ADD_PTR(head,ptrfield,add) \ + HASH_ADD(hh,head,ptrfield,sizeof(void *),add) +#define HASH_REPLACE_PTR(head,ptrfield,add,replaced) \ + HASH_REPLACE(hh,head,ptrfield,sizeof(void *),add,replaced) +#define HASH_DEL(head,delptr) \ + HASH_DELETE(hh,head,delptr) + +/* HASH_FSCK checks hash integrity on every add/delete when HASH_DEBUG is defined. + * This is for uthash developer only; it compiles away if HASH_DEBUG isn't defined. + */ +#ifdef HASH_DEBUG +#define HASH_OOPS(...) do { fprintf(stderr,__VA_ARGS__); exit(-1); } while (0) +#define HASH_FSCK(hh,head) \ +do { \ + unsigned _bkt_i; \ + unsigned _count, _bkt_count; \ + char *_prev; \ + struct UT_hash_handle *_thh; \ + if (head) { \ + _count = 0; \ + for( _bkt_i = 0; _bkt_i < (head)->hh.tbl->num_buckets; _bkt_i++) { \ + _bkt_count = 0; \ + _thh = (head)->hh.tbl->buckets[_bkt_i].hh_head; \ + _prev = NULL; \ + while (_thh) { \ + if (_prev != (char*)(_thh->hh_prev)) { \ + HASH_OOPS("invalid hh_prev %p, actual %p\n", \ + _thh->hh_prev, _prev ); \ + } \ + _bkt_count++; \ + _prev = (char*)(_thh); \ + _thh = _thh->hh_next; \ + } \ + _count += _bkt_count; \ + if ((head)->hh.tbl->buckets[_bkt_i].count != _bkt_count) { \ + HASH_OOPS("invalid bucket count %d, actual %d\n", \ + (head)->hh.tbl->buckets[_bkt_i].count, _bkt_count); \ + } \ + } \ + if (_count != (head)->hh.tbl->num_items) { \ + HASH_OOPS("invalid hh item count %d, actual %d\n", \ + (head)->hh.tbl->num_items, _count ); \ + } \ + /* traverse hh in app order; check next/prev integrity, count */ \ + _count = 0; \ + _prev = NULL; \ + _thh = &(head)->hh; \ + while (_thh) { \ + _count++; \ + if (_prev !=(char*)(_thh->prev)) { \ + HASH_OOPS("invalid prev %p, actual %p\n", \ + _thh->prev, _prev ); \ + } \ + _prev = (char*)ELMT_FROM_HH((head)->hh.tbl, _thh); \ + _thh = ( _thh->next ? (UT_hash_handle*)((char*)(_thh->next) + \ + (head)->hh.tbl->hho) : NULL ); \ + } \ + if (_count != (head)->hh.tbl->num_items) { \ + HASH_OOPS("invalid app item count %d, actual %d\n", \ + (head)->hh.tbl->num_items, _count ); \ + } \ + } \ +} while (0) +#else +#define HASH_FSCK(hh,head) +#endif + +/* When compiled with -DHASH_EMIT_KEYS, length-prefixed keys are emitted to + * the descriptor to which this macro is defined for tuning the hash function. + * The app can #include to get the prototype for write(2). */ +#ifdef HASH_EMIT_KEYS +#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) \ +do { \ + unsigned _klen = fieldlen; \ + write(HASH_EMIT_KEYS, &_klen, sizeof(_klen)); \ + write(HASH_EMIT_KEYS, keyptr, fieldlen); \ +} while (0) +#else +#define HASH_EMIT_KEY(hh,head,keyptr,fieldlen) +#endif + +/* default to Jenkin's hash unless overridden e.g. DHASH_FUNCTION=HASH_SAX */ +#ifdef HASH_FUNCTION +#define HASH_FCN HASH_FUNCTION +#else +#define HASH_FCN HASH_JEN +#endif + +/* The Bernstein hash function, used in Perl prior to v5.6 */ +#define HASH_BER(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _hb_keylen=keylen; \ + char *_hb_key=(char*)(key); \ + (hashv) = 0; \ + while (_hb_keylen--) { (hashv) = ((hashv) * 33) + *_hb_key++; } \ + bkt = (hashv) & (num_bkts-1); \ +} while (0) + + +/* SAX/FNV/OAT/JEN hash functions are macro variants of those listed at + * http://eternallyconfuzzled.com/tuts/algorithms/jsw_tut_hashing.aspx */ +#define HASH_SAX(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _sx_i; \ + char *_hs_key=(char*)(key); \ + hashv = 0; \ + for(_sx_i=0; _sx_i < keylen; _sx_i++) \ + hashv ^= (hashv << 5) + (hashv >> 2) + _hs_key[_sx_i]; \ + bkt = hashv & (num_bkts-1); \ +} while (0) + +#define HASH_FNV(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _fn_i; \ + char *_hf_key=(char*)(key); \ + hashv = 2166136261UL; \ + for(_fn_i=0; _fn_i < keylen; _fn_i++) \ + hashv = (hashv * 16777619) ^ _hf_key[_fn_i]; \ + bkt = hashv & (num_bkts-1); \ +} while(0) + +#define HASH_OAT(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _ho_i; \ + char *_ho_key=(char*)(key); \ + hashv = 0; \ + for(_ho_i=0; _ho_i < keylen; _ho_i++) { \ + hashv += _ho_key[_ho_i]; \ + hashv += (hashv << 10); \ + hashv ^= (hashv >> 6); \ + } \ + hashv += (hashv << 3); \ + hashv ^= (hashv >> 11); \ + hashv += (hashv << 15); \ + bkt = hashv & (num_bkts-1); \ +} while(0) + +#define HASH_JEN_MIX(a,b,c) \ +do { \ + a -= b; a -= c; a ^= ( c >> 13 ); \ + b -= c; b -= a; b ^= ( a << 8 ); \ + c -= a; c -= b; c ^= ( b >> 13 ); \ + a -= b; a -= c; a ^= ( c >> 12 ); \ + b -= c; b -= a; b ^= ( a << 16 ); \ + c -= a; c -= b; c ^= ( b >> 5 ); \ + a -= b; a -= c; a ^= ( c >> 3 ); \ + b -= c; b -= a; b ^= ( a << 10 ); \ + c -= a; c -= b; c ^= ( b >> 15 ); \ +} while (0) + +#define HASH_JEN(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned _hj_i,_hj_j,_hj_k; \ + unsigned char *_hj_key=(unsigned char*)(key); \ + hashv = 0xfeedbeef; \ + _hj_i = _hj_j = 0x9e3779b9; \ + _hj_k = (unsigned)(keylen); \ + while (_hj_k >= 12) { \ + _hj_i += (_hj_key[0] + ( (unsigned)_hj_key[1] << 8 ) \ + + ( (unsigned)_hj_key[2] << 16 ) \ + + ( (unsigned)_hj_key[3] << 24 ) ); \ + _hj_j += (_hj_key[4] + ( (unsigned)_hj_key[5] << 8 ) \ + + ( (unsigned)_hj_key[6] << 16 ) \ + + ( (unsigned)_hj_key[7] << 24 ) ); \ + hashv += (_hj_key[8] + ( (unsigned)_hj_key[9] << 8 ) \ + + ( (unsigned)_hj_key[10] << 16 ) \ + + ( (unsigned)_hj_key[11] << 24 ) ); \ + \ + HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ + \ + _hj_key += 12; \ + _hj_k -= 12; \ + } \ + hashv += keylen; \ + switch ( _hj_k ) { \ + case 11: hashv += ( (unsigned)_hj_key[10] << 24 ); \ + case 10: hashv += ( (unsigned)_hj_key[9] << 16 ); \ + case 9: hashv += ( (unsigned)_hj_key[8] << 8 ); \ + case 8: _hj_j += ( (unsigned)_hj_key[7] << 24 ); \ + case 7: _hj_j += ( (unsigned)_hj_key[6] << 16 ); \ + case 6: _hj_j += ( (unsigned)_hj_key[5] << 8 ); \ + case 5: _hj_j += _hj_key[4]; \ + case 4: _hj_i += ( (unsigned)_hj_key[3] << 24 ); \ + case 3: _hj_i += ( (unsigned)_hj_key[2] << 16 ); \ + case 2: _hj_i += ( (unsigned)_hj_key[1] << 8 ); \ + case 1: _hj_i += _hj_key[0]; \ + } \ + HASH_JEN_MIX(_hj_i, _hj_j, hashv); \ + bkt = hashv & (num_bkts-1); \ +} while(0) + +/* The Paul Hsieh hash function */ +#undef get16bits +#if (defined(__GNUC__) && defined(__i386__)) || defined(__WATCOMC__) \ + || defined(_MSC_VER) || defined (__BORLANDC__) || defined (__TURBOC__) +#define get16bits(d) (*((const uint16_t *) (d))) +#endif + +#if !defined (get16bits) +#define get16bits(d) ((((uint32_t)(((const uint8_t *)(d))[1])) << 8) \ + +(uint32_t)(((const uint8_t *)(d))[0]) ) +#endif +#define HASH_SFH(key,keylen,num_bkts,hashv,bkt) \ +do { \ + unsigned char *_sfh_key=(unsigned char*)(key); \ + uint32_t _sfh_tmp, _sfh_len = keylen; \ + \ + int _sfh_rem = _sfh_len & 3; \ + _sfh_len >>= 2; \ + hashv = 0xcafebabe; \ + \ + /* Main loop */ \ + for (;_sfh_len > 0; _sfh_len--) { \ + hashv += get16bits (_sfh_key); \ + _sfh_tmp = (uint32_t)(get16bits (_sfh_key+2)) << 11 ^ hashv; \ + hashv = (hashv << 16) ^ _sfh_tmp; \ + _sfh_key += 2*sizeof (uint16_t); \ + hashv += hashv >> 11; \ + } \ + \ + /* Handle end cases */ \ + switch (_sfh_rem) { \ + case 3: hashv += get16bits (_sfh_key); \ + hashv ^= hashv << 16; \ + hashv ^= (uint32_t)(_sfh_key[sizeof (uint16_t)] << 18); \ + hashv += hashv >> 11; \ + break; \ + case 2: hashv += get16bits (_sfh_key); \ + hashv ^= hashv << 11; \ + hashv += hashv >> 17; \ + break; \ + case 1: hashv += *_sfh_key; \ + hashv ^= hashv << 10; \ + hashv += hashv >> 1; \ + } \ + \ + /* Force "avalanching" of final 127 bits */ \ + hashv ^= hashv << 3; \ + hashv += hashv >> 5; \ + hashv ^= hashv << 4; \ + hashv += hashv >> 17; \ + hashv ^= hashv << 25; \ + hashv += hashv >> 6; \ + bkt = hashv & (num_bkts-1); \ +} while(0) + +#ifdef HASH_USING_NO_STRICT_ALIASING +/* The MurmurHash exploits some CPU's (x86,x86_64) tolerance for unaligned reads. + * For other types of CPU's (e.g. Sparc) an unaligned read causes a bus error. + * MurmurHash uses the faster approach only on CPU's where we know it's safe. + * + * Note the preprocessor built-in defines can be emitted using: + * + * gcc -m64 -dM -E - < /dev/null (on gcc) + * cc -## a.c (where a.c is a simple test file) (Sun Studio) + */ +#if (defined(__i386__) || defined(__x86_64__) || defined(_M_IX86)) +#define MUR_GETBLOCK(p,i) p[i] +#else /* non intel */ +#define MUR_PLUS0_ALIGNED(p) (((unsigned long)p & 0x3) == 0) +#define MUR_PLUS1_ALIGNED(p) (((unsigned long)p & 0x3) == 1) +#define MUR_PLUS2_ALIGNED(p) (((unsigned long)p & 0x3) == 2) +#define MUR_PLUS3_ALIGNED(p) (((unsigned long)p & 0x3) == 3) +#define WP(p) ((uint32_t*)((unsigned long)(p) & ~3UL)) +#if (defined(__BIG_ENDIAN__) || defined(SPARC) || defined(__ppc__) || defined(__ppc64__)) +#define MUR_THREE_ONE(p) ((((*WP(p))&0x00ffffff) << 8) | (((*(WP(p)+1))&0xff000000) >> 24)) +#define MUR_TWO_TWO(p) ((((*WP(p))&0x0000ffff) <<16) | (((*(WP(p)+1))&0xffff0000) >> 16)) +#define MUR_ONE_THREE(p) ((((*WP(p))&0x000000ff) <<24) | (((*(WP(p)+1))&0xffffff00) >> 8)) +#else /* assume little endian non-intel */ +#define MUR_THREE_ONE(p) ((((*WP(p))&0xffffff00) >> 8) | (((*(WP(p)+1))&0x000000ff) << 24)) +#define MUR_TWO_TWO(p) ((((*WP(p))&0xffff0000) >>16) | (((*(WP(p)+1))&0x0000ffff) << 16)) +#define MUR_ONE_THREE(p) ((((*WP(p))&0xff000000) >>24) | (((*(WP(p)+1))&0x00ffffff) << 8)) +#endif +#define MUR_GETBLOCK(p,i) (MUR_PLUS0_ALIGNED(p) ? ((p)[i]) : \ + (MUR_PLUS1_ALIGNED(p) ? MUR_THREE_ONE(p) : \ + (MUR_PLUS2_ALIGNED(p) ? MUR_TWO_TWO(p) : \ + MUR_ONE_THREE(p)))) +#endif +#define MUR_ROTL32(x,r) (((x) << (r)) | ((x) >> (32 - (r)))) +#define MUR_FMIX(_h) \ +do { \ + _h ^= _h >> 16; \ + _h *= 0x85ebca6b; \ + _h ^= _h >> 13; \ + _h *= 0xc2b2ae35l; \ + _h ^= _h >> 16; \ +} while(0) + +#define HASH_MUR(key,keylen,num_bkts,hashv,bkt) \ +do { \ + const uint8_t *_mur_data = (const uint8_t*)(key); \ + const int _mur_nblocks = (keylen) / 4; \ + uint32_t _mur_h1 = 0xf88D5353; \ + uint32_t _mur_c1 = 0xcc9e2d51; \ + uint32_t _mur_c2 = 0x1b873593; \ + uint32_t _mur_k1 = 0; \ + const uint8_t *_mur_tail; \ + const uint32_t *_mur_blocks = (const uint32_t*)(_mur_data+_mur_nblocks*4); \ + int _mur_i; \ + for(_mur_i = -_mur_nblocks; _mur_i; _mur_i++) { \ + _mur_k1 = MUR_GETBLOCK(_mur_blocks,_mur_i); \ + _mur_k1 *= _mur_c1; \ + _mur_k1 = MUR_ROTL32(_mur_k1,15); \ + _mur_k1 *= _mur_c2; \ + \ + _mur_h1 ^= _mur_k1; \ + _mur_h1 = MUR_ROTL32(_mur_h1,13); \ + _mur_h1 = _mur_h1*5+0xe6546b64; \ + } \ + _mur_tail = (const uint8_t*)(_mur_data + _mur_nblocks*4); \ + _mur_k1=0; \ + switch((keylen) & 3) { \ + case 3: _mur_k1 ^= _mur_tail[2] << 16; \ + case 2: _mur_k1 ^= _mur_tail[1] << 8; \ + case 1: _mur_k1 ^= _mur_tail[0]; \ + _mur_k1 *= _mur_c1; \ + _mur_k1 = MUR_ROTL32(_mur_k1,15); \ + _mur_k1 *= _mur_c2; \ + _mur_h1 ^= _mur_k1; \ + } \ + _mur_h1 ^= (keylen); \ + MUR_FMIX(_mur_h1); \ + hashv = _mur_h1; \ + bkt = hashv & (num_bkts-1); \ +} while(0) +#endif /* HASH_USING_NO_STRICT_ALIASING */ + +/* key comparison function; return 0 if keys equal */ +#define HASH_KEYCMP(a,b,len) memcmp(a,b,len) + +/* iterate over items in a known bucket to find desired item */ +#define HASH_FIND_IN_BKT(tbl,hh,head,keyptr,keylen_in,out) \ +do { \ + if (head.hh_head) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,head.hh_head)); \ + else out=NULL; \ + while (out) { \ + if ((out)->hh.keylen == keylen_in) { \ + if ((HASH_KEYCMP((out)->hh.key,keyptr,keylen_in)) == 0) break; \ + } \ + if ((out)->hh.hh_next) DECLTYPE_ASSIGN(out,ELMT_FROM_HH(tbl,(out)->hh.hh_next)); \ + else out = NULL; \ + } \ +} while(0) + +/* add an item to a bucket */ +#define HASH_ADD_TO_BKT(head,addhh) \ +do { \ + head.count++; \ + (addhh)->hh_next = head.hh_head; \ + (addhh)->hh_prev = NULL; \ + if (head.hh_head) { (head).hh_head->hh_prev = (addhh); } \ + (head).hh_head=addhh; \ + if (head.count >= ((head.expand_mult+1) * HASH_BKT_CAPACITY_THRESH) \ + && (addhh)->tbl->noexpand != 1) { \ + HASH_EXPAND_BUCKETS((addhh)->tbl); \ + } \ +} while(0) + +/* remove an item from a given bucket */ +#define HASH_DEL_IN_BKT(hh,head,hh_del) \ + (head).count--; \ + if ((head).hh_head == hh_del) { \ + (head).hh_head = hh_del->hh_next; \ + } \ + if (hh_del->hh_prev) { \ + hh_del->hh_prev->hh_next = hh_del->hh_next; \ + } \ + if (hh_del->hh_next) { \ + hh_del->hh_next->hh_prev = hh_del->hh_prev; \ + } + +/* Bucket expansion has the effect of doubling the number of buckets + * and redistributing the items into the new buckets. Ideally the + * items will distribute more or less evenly into the new buckets + * (the extent to which this is true is a measure of the quality of + * the hash function as it applies to the key domain). + * + * With the items distributed into more buckets, the chain length + * (item count) in each bucket is reduced. Thus by expanding buckets + * the hash keeps a bound on the chain length. This bounded chain + * length is the essence of how a hash provides constant time lookup. + * + * The calculation of tbl->ideal_chain_maxlen below deserves some + * explanation. First, keep in mind that we're calculating the ideal + * maximum chain length based on the *new* (doubled) bucket count. + * In fractions this is just n/b (n=number of items,b=new num buckets). + * Since the ideal chain length is an integer, we want to calculate + * ceil(n/b). We don't depend on floating point arithmetic in this + * hash, so to calculate ceil(n/b) with integers we could write + * + * ceil(n/b) = (n/b) + ((n%b)?1:0) + * + * and in fact a previous version of this hash did just that. + * But now we have improved things a bit by recognizing that b is + * always a power of two. We keep its base 2 log handy (call it lb), + * so now we can write this with a bit shift and logical AND: + * + * ceil(n/b) = (n>>lb) + ( (n & (b-1)) ? 1:0) + * + */ +#define HASH_EXPAND_BUCKETS(tbl) \ +do { \ + unsigned _he_bkt; \ + unsigned _he_bkt_i; \ + struct UT_hash_handle *_he_thh, *_he_hh_nxt; \ + UT_hash_bucket *_he_new_buckets, *_he_newbkt; \ + _he_new_buckets = (UT_hash_bucket*)uthash_malloc( \ + 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ + if (!_he_new_buckets) { uthash_fatal( "out of memory"); } \ + memset(_he_new_buckets, 0, \ + 2 * tbl->num_buckets * sizeof(struct UT_hash_bucket)); \ + tbl->ideal_chain_maxlen = \ + (tbl->num_items >> (tbl->log2_num_buckets+1)) + \ + ((tbl->num_items & ((tbl->num_buckets*2)-1)) ? 1 : 0); \ + tbl->nonideal_items = 0; \ + for(_he_bkt_i = 0; _he_bkt_i < tbl->num_buckets; _he_bkt_i++) \ + { \ + _he_thh = tbl->buckets[ _he_bkt_i ].hh_head; \ + while (_he_thh) { \ + _he_hh_nxt = _he_thh->hh_next; \ + HASH_TO_BKT( _he_thh->hashv, tbl->num_buckets*2, _he_bkt); \ + _he_newbkt = &(_he_new_buckets[ _he_bkt ]); \ + if (++(_he_newbkt->count) > tbl->ideal_chain_maxlen) { \ + tbl->nonideal_items++; \ + _he_newbkt->expand_mult = _he_newbkt->count / \ + tbl->ideal_chain_maxlen; \ + } \ + _he_thh->hh_prev = NULL; \ + _he_thh->hh_next = _he_newbkt->hh_head; \ + if (_he_newbkt->hh_head) _he_newbkt->hh_head->hh_prev = \ + _he_thh; \ + _he_newbkt->hh_head = _he_thh; \ + _he_thh = _he_hh_nxt; \ + } \ + } \ + uthash_free( tbl->buckets, tbl->num_buckets*sizeof(struct UT_hash_bucket) ); \ + tbl->num_buckets *= 2; \ + tbl->log2_num_buckets++; \ + tbl->buckets = _he_new_buckets; \ + tbl->ineff_expands = (tbl->nonideal_items > (tbl->num_items >> 1)) ? \ + (tbl->ineff_expands+1) : 0; \ + if (tbl->ineff_expands > 1) { \ + tbl->noexpand=1; \ + uthash_noexpand_fyi(tbl); \ + } \ + uthash_expand_fyi(tbl); \ +} while(0) + + +/* This is an adaptation of Simon Tatham's O(n log(n)) mergesort */ +/* Note that HASH_SORT assumes the hash handle name to be hh. + * HASH_SRT was added to allow the hash handle name to be passed in. */ +#define HASH_SORT(head,cmpfcn) HASH_SRT(hh,head,cmpfcn) +#define HASH_SRT(hh,head,cmpfcn) \ +do { \ + unsigned _hs_i; \ + unsigned _hs_looping,_hs_nmerges,_hs_insize,_hs_psize,_hs_qsize; \ + struct UT_hash_handle *_hs_p, *_hs_q, *_hs_e, *_hs_list, *_hs_tail; \ + if (head) { \ + _hs_insize = 1; \ + _hs_looping = 1; \ + _hs_list = &((head)->hh); \ + while (_hs_looping) { \ + _hs_p = _hs_list; \ + _hs_list = NULL; \ + _hs_tail = NULL; \ + _hs_nmerges = 0; \ + while (_hs_p) { \ + _hs_nmerges++; \ + _hs_q = _hs_p; \ + _hs_psize = 0; \ + for ( _hs_i = 0; _hs_i < _hs_insize; _hs_i++ ) { \ + _hs_psize++; \ + _hs_q = (UT_hash_handle*)((_hs_q->next) ? \ + ((void*)((char*)(_hs_q->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + if (! (_hs_q) ) break; \ + } \ + _hs_qsize = _hs_insize; \ + while ((_hs_psize > 0) || ((_hs_qsize > 0) && _hs_q )) { \ + if (_hs_psize == 0) { \ + _hs_e = _hs_q; \ + _hs_q = (UT_hash_handle*)((_hs_q->next) ? \ + ((void*)((char*)(_hs_q->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + _hs_qsize--; \ + } else if ( (_hs_qsize == 0) || !(_hs_q) ) { \ + _hs_e = _hs_p; \ + if (_hs_p){ \ + _hs_p = (UT_hash_handle*)((_hs_p->next) ? \ + ((void*)((char*)(_hs_p->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + } \ + _hs_psize--; \ + } else if (( \ + cmpfcn(DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_p)), \ + DECLTYPE(head)(ELMT_FROM_HH((head)->hh.tbl,_hs_q))) \ + ) <= 0) { \ + _hs_e = _hs_p; \ + if (_hs_p){ \ + _hs_p = (UT_hash_handle*)((_hs_p->next) ? \ + ((void*)((char*)(_hs_p->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + } \ + _hs_psize--; \ + } else { \ + _hs_e = _hs_q; \ + _hs_q = (UT_hash_handle*)((_hs_q->next) ? \ + ((void*)((char*)(_hs_q->next) + \ + (head)->hh.tbl->hho)) : NULL); \ + _hs_qsize--; \ + } \ + if ( _hs_tail ) { \ + _hs_tail->next = ((_hs_e) ? \ + ELMT_FROM_HH((head)->hh.tbl,_hs_e) : NULL); \ + } else { \ + _hs_list = _hs_e; \ + } \ + if (_hs_e) { \ + _hs_e->prev = ((_hs_tail) ? \ + ELMT_FROM_HH((head)->hh.tbl,_hs_tail) : NULL); \ + } \ + _hs_tail = _hs_e; \ + } \ + _hs_p = _hs_q; \ + } \ + if (_hs_tail){ \ + _hs_tail->next = NULL; \ + } \ + if ( _hs_nmerges <= 1 ) { \ + _hs_looping=0; \ + (head)->hh.tbl->tail = _hs_tail; \ + DECLTYPE_ASSIGN(head,ELMT_FROM_HH((head)->hh.tbl, _hs_list)); \ + } \ + _hs_insize *= 2; \ + } \ + HASH_FSCK(hh,head); \ + } \ +} while (0) + +/* This function selects items from one hash into another hash. + * The end result is that the selected items have dual presence + * in both hashes. There is no copy of the items made; rather + * they are added into the new hash through a secondary hash + * hash handle that must be present in the structure. */ +#define HASH_SELECT(hh_dst, dst, hh_src, src, cond) \ +do { \ + unsigned _src_bkt, _dst_bkt; \ + void *_last_elt=NULL, *_elt; \ + UT_hash_handle *_src_hh, *_dst_hh, *_last_elt_hh=NULL; \ + ptrdiff_t _dst_hho = ((char*)(&(dst)->hh_dst) - (char*)(dst)); \ + if (src) { \ + for(_src_bkt=0; _src_bkt < (src)->hh_src.tbl->num_buckets; _src_bkt++) { \ + for(_src_hh = (src)->hh_src.tbl->buckets[_src_bkt].hh_head; \ + _src_hh; \ + _src_hh = _src_hh->hh_next) { \ + _elt = ELMT_FROM_HH((src)->hh_src.tbl, _src_hh); \ + if (cond(_elt)) { \ + _dst_hh = (UT_hash_handle*)(((char*)_elt) + _dst_hho); \ + _dst_hh->key = _src_hh->key; \ + _dst_hh->keylen = _src_hh->keylen; \ + _dst_hh->hashv = _src_hh->hashv; \ + _dst_hh->prev = _last_elt; \ + _dst_hh->next = NULL; \ + if (_last_elt_hh) { _last_elt_hh->next = _elt; } \ + if (!dst) { \ + DECLTYPE_ASSIGN(dst,_elt); \ + HASH_MAKE_TABLE(hh_dst,dst); \ + } else { \ + _dst_hh->tbl = (dst)->hh_dst.tbl; \ + } \ + HASH_TO_BKT(_dst_hh->hashv, _dst_hh->tbl->num_buckets, _dst_bkt); \ + HASH_ADD_TO_BKT(_dst_hh->tbl->buckets[_dst_bkt],_dst_hh); \ + (dst)->hh_dst.tbl->num_items++; \ + _last_elt = _elt; \ + _last_elt_hh = _dst_hh; \ + } \ + } \ + } \ + } \ + HASH_FSCK(hh_dst,dst); \ +} while (0) + +#define HASH_CLEAR(hh,head) \ +do { \ + if (head) { \ + uthash_free((head)->hh.tbl->buckets, \ + (head)->hh.tbl->num_buckets*sizeof(struct UT_hash_bucket)); \ + HASH_BLOOM_FREE((head)->hh.tbl); \ + uthash_free((head)->hh.tbl, sizeof(UT_hash_table)); \ + (head)=NULL; \ + } \ +} while(0) + +#define HASH_OVERHEAD(hh,head) \ + (size_t)((((head)->hh.tbl->num_items * sizeof(UT_hash_handle)) + \ + ((head)->hh.tbl->num_buckets * sizeof(UT_hash_bucket)) + \ + (sizeof(UT_hash_table)) + \ + (HASH_BLOOM_BYTELEN))) + +#ifdef NO_DECLTYPE +#define HASH_ITER(hh,head,el,tmp) \ +for((el)=(head), (*(char**)(&(tmp)))=(char*)((head)?(head)->hh.next:NULL); \ + el; (el)=(tmp),(*(char**)(&(tmp)))=(char*)((tmp)?(tmp)->hh.next:NULL)) +#else +#define HASH_ITER(hh,head,el,tmp) \ +for((el)=(head),(tmp)=DECLTYPE(el)((head)?(head)->hh.next:NULL); \ + el; (el)=(tmp),(tmp)=DECLTYPE(el)((tmp)?(tmp)->hh.next:NULL)) +#endif + +/* obtain a count of items in the hash */ +#define HASH_COUNT(head) HASH_CNT(hh,head) +#define HASH_CNT(hh,head) ((head)?((head)->hh.tbl->num_items):0) + +typedef struct UT_hash_bucket { + struct UT_hash_handle *hh_head; + unsigned count; + + /* expand_mult is normally set to 0. In this situation, the max chain length + * threshold is enforced at its default value, HASH_BKT_CAPACITY_THRESH. (If + * the bucket's chain exceeds this length, bucket expansion is triggered). + * However, setting expand_mult to a non-zero value delays bucket expansion + * (that would be triggered by additions to this particular bucket) + * until its chain length reaches a *multiple* of HASH_BKT_CAPACITY_THRESH. + * (The multiplier is simply expand_mult+1). The whole idea of this + * multiplier is to reduce bucket expansions, since they are expensive, in + * situations where we know that a particular bucket tends to be overused. + * It is better to let its chain length grow to a longer yet-still-bounded + * value, than to do an O(n) bucket expansion too often. + */ + unsigned expand_mult; + +} UT_hash_bucket; + +/* random signature used only to find hash tables in external analysis */ +#define HASH_SIGNATURE 0xa0111fe1 +#define HASH_BLOOM_SIGNATURE 0xb12220f2 + +typedef struct UT_hash_table { + UT_hash_bucket *buckets; + unsigned num_buckets, log2_num_buckets; + unsigned num_items; + struct UT_hash_handle *tail; /* tail hh in app order, for fast append */ + ptrdiff_t hho; /* hash handle offset (byte pos of hash handle in element */ + + /* in an ideal situation (all buckets used equally), no bucket would have + * more than ceil(#items/#buckets) items. that's the ideal chain length. */ + unsigned ideal_chain_maxlen; + + /* nonideal_items is the number of items in the hash whose chain position + * exceeds the ideal chain maxlen. these items pay the penalty for an uneven + * hash distribution; reaching them in a chain traversal takes >ideal steps */ + unsigned nonideal_items; + + /* ineffective expands occur when a bucket doubling was performed, but + * afterward, more than half the items in the hash had nonideal chain + * positions. If this happens on two consecutive expansions we inhibit any + * further expansion, as it's not helping; this happens when the hash + * function isn't a good fit for the key domain. When expansion is inhibited + * the hash will still work, albeit no longer in constant time. */ + unsigned ineff_expands, noexpand; + + uint32_t signature; /* used only to find hash tables in external analysis */ +#ifdef HASH_BLOOM + uint32_t bloom_sig; /* used only to test bloom exists in external analysis */ + uint8_t *bloom_bv; + char bloom_nbits; +#endif + +} UT_hash_table; + +typedef struct UT_hash_handle { + struct UT_hash_table *tbl; + void *prev; /* prev element in app order */ + void *next; /* next element in app order */ + struct UT_hash_handle *hh_prev; /* previous hh in bucket order */ + struct UT_hash_handle *hh_next; /* next hh in bucket order */ + void *key; /* ptr to enclosing struct's key */ + unsigned keylen; /* enclosing struct's key len */ + unsigned hashv; /* result of hash-fcn(key) */ +} UT_hash_handle; + +#endif /* UTHASH_H */