]> rtime.felk.cvut.cz Git - lisovros/iproute2_canprio.git/blobdiff - misc/ss.c
ss: use new INET_DIAG_SKMEMINFO option to get more memory information for tcp socket
[lisovros/iproute2_canprio.git] / misc / ss.c
index ec272f442798ae2a81d922c2596eb0ee3eef2653..cf529ef7bb49537890ed95ce8a1f9abcabe96a2e 100644 (file)
--- a/misc/ss.c
+++ b/misc/ss.c
 #include "libnetlink.h"
 #include "SNAPSHOT.h"
 
+#include <netinet/tcp.h>
+#include <linux/sock_diag.h>
 #include <linux/inet_diag.h>
-#include <linux/tcp.h>
-#include <net/tcp_states.h>
+#include <linux/unix_diag.h>
 
 int resolve_hosts = 0;
 int resolve_services = 1;
@@ -105,184 +106,244 @@ struct filter
 };
 
 struct filter default_filter = {
-       dbs: (1<<TCP_DB),
-       states: SS_ALL & ~((1<<SS_LISTEN)|(1<<SS_CLOSE)|(1<<SS_TIME_WAIT)|(1<<SS_SYN_RECV)),
-       families: (1<<AF_INET)|(1<<AF_INET6),
+       .dbs    =  (1<<TCP_DB),
+       .states = SS_ALL & ~((1<<SS_LISTEN)|(1<<SS_CLOSE)|(1<<SS_TIME_WAIT)|(1<<SS_SYN_RECV)),
+       .families= (1<<AF_INET)|(1<<AF_INET6),
 };
 
 struct filter current_filter;
 
-int generic_proc_open(char *env, char *name)
+static FILE *generic_proc_open(const char *env, const char *name)
 {
+       const char *p = getenv(env);
        char store[128];
-       char *p = getenv(env);
+
        if (!p) {
                p = getenv("PROC_ROOT") ? : "/proc";
                snprintf(store, sizeof(store)-1, "%s/%s", p, name);
                p = store;
        }
-       return open(store, O_RDONLY);
+
+       return fopen(p, "r");
 }
 
-int net_tcp_open(void)
+static FILE *net_tcp_open(void)
 {
        return generic_proc_open("PROC_NET_TCP", "net/tcp");
 }
 
-int net_tcp6_open(void)
+static FILE *net_tcp6_open(void)
 {
        return generic_proc_open("PROC_NET_TCP6", "net/tcp6");
 }
 
-int net_udp_open(void)
+static FILE *net_udp_open(void)
 {
        return generic_proc_open("PROC_NET_UDP", "net/udp");
 }
 
-int net_udp6_open(void)
+static FILE *net_udp6_open(void)
 {
        return generic_proc_open("PROC_NET_UDP6", "net/udp6");
 }
 
-int net_raw_open(void)
+static FILE *net_raw_open(void)
 {
        return generic_proc_open("PROC_NET_RAW", "net/raw");
 }
 
-int net_raw6_open(void)
+static FILE *net_raw6_open(void)
 {
        return generic_proc_open("PROC_NET_RAW6", "net/raw6");
 }
 
-int net_unix_open(void)
+static FILE *net_unix_open(void)
 {
        return generic_proc_open("PROC_NET_UNIX", "net/unix");
 }
 
-int net_packet_open(void)
+static FILE *net_packet_open(void)
 {
        return generic_proc_open("PROC_NET_PACKET", "net/packet");
 }
 
-int net_netlink_open(void)
+static FILE *net_netlink_open(void)
 {
        return generic_proc_open("PROC_NET_NETLINK", "net/netlink");
 }
 
-int slabinfo_open(void)
+static FILE *slabinfo_open(void)
 {
        return generic_proc_open("PROC_SLABINFO", "slabinfo");
 }
 
-int net_sockstat_open(void)
+static FILE *net_sockstat_open(void)
 {
        return generic_proc_open("PROC_NET_SOCKSTAT", "net/sockstat");
 }
 
-int net_sockstat6_open(void)
+static FILE *net_sockstat6_open(void)
 {
        return generic_proc_open("PROC_NET_SOCKSTAT6", "net/sockstat6");
 }
 
-int net_snmp_open(void)
+static FILE *net_snmp_open(void)
 {
        return generic_proc_open("PROC_NET_SNMP", "net/snmp");
 }
 
-int net_netstat_open(void)
+static FILE *ephemeral_ports_open(void)
 {
-       return generic_proc_open("PROC_NET_NETSTAT", "net/netstat");
+       return generic_proc_open("PROC_IP_LOCAL_PORT_RANGE", "sys/net/ipv4/ip_local_port_range");
 }
 
-int ephemeral_ports_open(void)
+struct user_ent {
+       struct user_ent *next;
+       unsigned int    ino;
+       int             pid;
+       int             fd;
+       char            process[0];
+};
+
+#define USER_ENT_HASH_SIZE     256
+struct user_ent *user_ent_hash[USER_ENT_HASH_SIZE];
+
+static int user_ent_hashfn(unsigned int ino)
 {
-       return generic_proc_open("PROC_IP_LOCAL_PORT_RANGE", "sys/net/ipv4/ip_local_port_range");
+       int val = (ino >> 24) ^ (ino >> 16) ^ (ino >> 8) ^ ino;
+
+       return val & (USER_ENT_HASH_SIZE - 1);
 }
 
-int find_users(int ino, char *buf, int buflen)
+static void user_ent_add(unsigned int ino, const char *process, int pid, int fd)
 {
-       char pattern[64];
-       int  pattern_len;
-       char *ptr = buf;
-       char name[1024];
-       DIR *dir;
-       struct dirent *d;
-       int cnt = 0;
-       int nameoff;
+       struct user_ent *p, **pp;
+       int str_len;
 
-       if (!ino)
-               return 0;
+       str_len = strlen(process) + 1;
+       p = malloc(sizeof(struct user_ent) + str_len);
+       if (!p)
+               abort();
+       p->next = NULL;
+       p->ino = ino;
+       p->pid = pid;
+       p->fd = fd;
+       strcpy(p->process, process);
+
+       pp = &user_ent_hash[user_ent_hashfn(ino)];
+       p->next = *pp;
+       *pp = p;
+}
 
-       sprintf(pattern, "socket:[%d]", ino);
-       pattern_len = strlen(pattern);
+static void user_ent_hash_build(void)
+{
+       const char *root = getenv("PROC_ROOT") ? : "/proc/";
+       struct dirent *d;
+       char name[1024];
+       int nameoff;
+       DIR *dir;
 
-       strncpy(name, getenv("PROC_ROOT") ? : "/proc/", sizeof(name)/2);
-       name[sizeof(name)/2] = 0;
-       if (strlen(name) == 0 ||
-           name[strlen(name)-1] != '/')
+       strcpy(name, root);
+       if (strlen(name) == 0 || name[strlen(name)-1] != '/')
                strcat(name, "/");
+
        nameoff = strlen(name);
-       if ((dir = opendir(name)) == NULL)
-               return 0;
+
+       dir = opendir(name);
+       if (!dir)
+               return;
 
        while ((d = readdir(dir)) != NULL) {
-               DIR *dir1;
                struct dirent *d1;
-               int pid;
-               int pos;
-               char crap;
                char process[16];
+               int pid, pos;
+               DIR *dir1;
+               char crap;
 
                if (sscanf(d->d_name, "%d%c", &pid, &crap) != 1)
                        continue;
 
-               sprintf(name+nameoff, "%d/fd/", pid);
+               sprintf(name + nameoff, "%d/fd/", pid);
                pos = strlen(name);
                if ((dir1 = opendir(name)) == NULL)
                        continue;
 
-               process[0] = 0;
+               process[0] = '\0';
 
                while ((d1 = readdir(dir1)) != NULL) {
-                       int fd, n;
+                       const char *pattern = "socket:[";
+                       unsigned int ino;
                        char lnk[64];
+                       int fd;
+                       ssize_t link_len;
 
                        if (sscanf(d1->d_name, "%d%c", &fd, &crap) != 1)
                                continue;
 
                        sprintf(name+pos, "%d", fd);
-                       n = readlink(name, lnk, sizeof(lnk)-1);
-                       if (n != pattern_len ||
-                           memcmp(lnk, pattern, n))
+
+                       link_len = readlink(name, lnk, sizeof(lnk)-1);
+                       if (link_len == -1)
                                continue;
+                       lnk[link_len] = '\0';
 
-                       if (ptr-buf >= buflen-1)
-                               break;
+                       if (strncmp(lnk, pattern, strlen(pattern)))
+                               continue;
 
-                       if (process[0] == 0) {
+                       sscanf(lnk, "socket:[%u]", &ino);
+
+                       if (process[0] == '\0') {
                                char tmp[1024];
                                FILE *fp;
-                               snprintf(tmp, sizeof(tmp), "%s/%d/stat",
-                                        getenv("PROC_ROOT") ? : "/proc", pid);
+
+                               snprintf(tmp, sizeof(tmp), "%s/%d/stat", root, pid);
                                if ((fp = fopen(tmp, "r")) != NULL) {
                                        fscanf(fp, "%*d (%[^)])", process);
                                        fclose(fp);
                                }
                        }
 
-                       snprintf(ptr, buflen-(ptr-buf), "(\"%s\",%d,%d),", process, pid, fd);
-                       ptr += strlen(ptr);
-                       cnt++;
+                       user_ent_add(ino, process, pid, fd);
                }
                closedir(dir1);
        }
        closedir(dir);
+}
+
+int find_users(unsigned ino, char *buf, int buflen)
+{
+       struct user_ent *p;
+       int cnt = 0;
+       char *ptr;
+
+       if (!ino)
+               return 0;
+
+       p = user_ent_hash[user_ent_hashfn(ino)];
+       ptr = buf;
+       while (p) {
+               if (p->ino != ino)
+                       goto next;
+
+               if (ptr - buf >= buflen - 1)
+                       break;
+
+               snprintf(ptr, buflen - (ptr - buf),
+                        "(\"%s\",%d,%d),",
+                        p->process, p->pid, p->fd);
+               ptr += strlen(ptr);
+               cnt++;
+
+       next:
+               p = p->next;
+       }
+
        if (ptr != buf)
-               ptr[-1] = 0;
+               ptr[-1] = '\0';
+
        return cnt;
 }
 
-
 /* Get stats from slab */
 
 struct slabstat
@@ -313,7 +374,8 @@ int get_slabstat(struct slabstat *s)
 
        memset(s, 0, sizeof(*s));
 
-       if ((fp = fdopen(slabinfo_open(), "r")) == NULL)
+       fp = slabinfo_open();
+       if (!fp)
                return -1;
 
        cnt = sizeof(*s)/sizeof(int);
@@ -377,9 +439,9 @@ struct tcpstat
        int             timer;
        int             timeout;
        int             retrs;
-       int             ino;
+       unsigned        ino;
        int             probes;
-       int             uid;
+       unsigned        uid;
        int             refcnt;
        unsigned long long sk;
        int             rto, ato, qack, cwnd, ssthresh;
@@ -419,13 +481,13 @@ const char *print_ms_timer(int timeout)
        if (msecs)
                sprintf(buf+strlen(buf), "%03dms", msecs);
        return buf;
-};
+}
 
 const char *print_hz_timer(int timeout)
 {
-       int hz = get_hz();
+       int hz = get_user_hz();
        return print_ms_timer(((timeout*1000) + hz-1)/hz);
-};
+}
 
 struct scache
 {
@@ -464,6 +526,7 @@ void init_service_resolver(void)
                                }
                        }
                }
+               pclose(fp);
        }
 }
 
@@ -478,7 +541,7 @@ static int ip_local_port_min, ip_local_port_max;
 static int is_ephemeral(int port)
 {
        if (!ip_local_port_min) {
-               FILE *f = fdopen(ephemeral_ports_open(), "r");
+               FILE *f = ephemeral_ports_open();
                if (f) {
                        fscanf(f, "%d %d",
                               &ip_local_port_min, &ip_local_port_max);
@@ -655,7 +718,7 @@ int run_ssfilter(struct ssfilter *f, struct tcpstat *s)
                        return s->lport < 0;
 
                 if (!low) {
-                       FILE *fp = fdopen(ephemeral_ports_open(), "r");
+                       FILE *fp = ephemeral_ports_open();
                        if (fp) {
                                fscanf(fp, "%d%d", &low, &high);
                                fclose(fp);
@@ -749,7 +812,7 @@ static int ssfilter_bytecompile(struct ssfilter *f, char **bytecode)
        {
                if (!(*bytecode=malloc(4))) abort();
                ((struct inet_diag_bc_op*)*bytecode)[0] = (struct inet_diag_bc_op){ INET_DIAG_BC_AUTO, 4, 8 };
-               return 8;
+               return 4;
        }
                case SSF_DCOND:
                case SSF_SCOND:
@@ -1103,7 +1166,7 @@ void *parse_hostcond(char *addr)
        return res;
 }
 
-static int tcp_show_line(char *line, struct filter *f, int family)
+static int tcp_show_line(char *line, const struct filter *f, int family)
 {
        struct tcpstat s;
        char *loc, *rem, *data;
@@ -1157,7 +1220,7 @@ static int tcp_show_line(char *line, struct filter *f, int family)
                return 0;
 
        opt[0] = 0;
-       n = sscanf(data, "%x %x:%x %x:%x %x %d %d %d %d %llx %d %d %d %d %d %[^\n]\n",
+       n = sscanf(data, "%x %x:%x %x:%x %x %d %d %u %d %llx %d %d %d %d %d %[^\n]\n",
                   &s.state, &s.wq, &s.rq,
                   &s.timer, &s.timeout, &s.retrs, &s.uid, &s.probes, &s.ino,
                   &s.refcnt, &s.sk, &s.rto, &s.ato, &s.qack,
@@ -1194,10 +1257,11 @@ static int tcp_show_line(char *line, struct filter *f, int family)
                }
        }
        if (show_tcpinfo) {
-               if (s.rto && s.rto != 3*get_hz())
-                       printf(" rto:%g", (double)s.rto/get_hz());
+               int hz = get_user_hz();
+               if (s.rto && s.rto != 3*hz)
+                       printf(" rto:%g", (double)s.rto/hz);
                if (s.ato)
-                       printf(" ato:%g", (double)s.ato/get_hz());
+                       printf(" ato:%g", (double)s.ato/hz);
                if (s.cwnd != 2)
                        printf(" cwnd:%d", s.cwnd);
                if (s.ssthresh != -1)
@@ -1215,7 +1279,7 @@ static int tcp_show_line(char *line, struct filter *f, int family)
        if (show_details) {
                if (s.uid)
                        printf(" uid:%u", (unsigned)s.uid);
-               printf(" ino:%u", (unsigned)s.ino);
+               printf(" ino:%u", s.ino);
                printf(" sk:%llx", s.sk);
                if (opt[0])
                        printf(" opt:\"%s\"", opt);
@@ -1225,68 +1289,30 @@ static int tcp_show_line(char *line, struct filter *f, int family)
        return 0;
 }
 
-static int generic_record_read(int fd, char *buf, int bufsize,
-                       int (*worker)(char*, struct filter *, int),
-                       struct filter *f, int fam)
+static int generic_record_read(FILE *fp,
+                              int (*worker)(char*, const struct filter *, int),
+                              const struct filter *f, int fam)
 {
-       int n;
-       int recsize;
-       int eof = 0;
-       char *p;
+       char line[256];
 
-       /* Load the first chunk and calculate record length from it. */
-       n = read(fd, buf, bufsize);
-       if (n < 0)
+       /* skip header */
+       if (fgets(line, sizeof(line), fp) == NULL)
                goto outerr;
-       /* I _know_ that this is wrong, do not remind. :-)
-        * But this works nowadays. */
-       if (n < bufsize)
-               eof = 1;
-       p = memchr(buf, '\n', n);
-       if (p == NULL || (p-buf) >= n)
-               goto outwrongformat;
-       recsize = (p-buf)+1;
-       p = buf+recsize;
-
-       for (;;) {
-               while ((p+recsize) - buf <= n) {
-                       if (p[recsize-1] != '\n')
-                               goto outwrongformat;
-                       p[recsize-1] = 0;
-                       if (worker(p, f, fam) < 0)
-                               goto done;
-                       p += recsize;
-               }
-               if (!eof) {
-                       int remains = (buf+bufsize) - p;
-                       memcpy(buf, p, remains);
-                       p = buf+remains;
-                       n = read(fd, p, (buf+bufsize) - p);
-                       if (n < 0)
-                               goto outerr;
-                       if (n < (buf+bufsize) - p) {
-                               eof = 1;
-                               if (n == 0) {
-                                       if (remains)
-                                               goto outwrongformat;
-                                       goto done;
-                               }
-                       }
-                       n += remains;
-                       p = buf;
-               } else {
-                       if (p != buf+n)
-                               goto outwrongformat;
-                       goto done;
+
+       while (fgets(line, sizeof(line), fp) != NULL) {
+               int n = strlen(line);
+               if (n == 0 || line[n-1] != '\n') {
+                       errno = -EINVAL;
+                       return -1;
                }
-       }
-done:
-       return 0;
+               line[n-1] = 0;
 
-outwrongformat:
-       errno = EINVAL;
+               if (worker(line, f, fam) < 0)
+                       return 0;
+       }
 outerr:
-       return -1;
+
+       return ferror(fp) ? -1 : 0;
 }
 
 static char *sprint_bw(char *buf, double bw)
@@ -1310,7 +1336,17 @@ static void tcp_show_info(const struct nlmsghdr *nlh, struct inet_diag_msg *r)
        parse_rtattr(tb, INET_DIAG_MAX, (struct rtattr*)(r+1),
                     nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*r)));
 
-       if (tb[INET_DIAG_MEMINFO]) {
+       if (tb[INET_DIAG_SKMEMINFO]) {
+               const __u32 *skmeminfo =  RTA_DATA(tb[INET_DIAG_SKMEMINFO]);
+               printf(" skmem:(r%u,rb%u,t%u,tb%u,f%u,w%u,o%u)",
+                       skmeminfo[SK_MEMINFO_RMEM_ALLOC],
+                       skmeminfo[SK_MEMINFO_RCVBUF],
+                       skmeminfo[SK_MEMINFO_WMEM_ALLOC],
+                       skmeminfo[SK_MEMINFO_SNDBUF],
+                       skmeminfo[SK_MEMINFO_FWD_ALLOC],
+                       skmeminfo[SK_MEMINFO_WMEM_QUEUED],
+                       skmeminfo[SK_MEMINFO_OPTMEM]);
+       }else if (tb[INET_DIAG_MEMINFO]) {
                const struct inet_diag_meminfo *minfo
                        = RTA_DATA(tb[INET_DIAG_MEMINFO]);
                printf(" mem:(r%u,w%u,f%u,t%u)",
@@ -1339,10 +1375,12 @@ static void tcp_show_info(const struct nlmsghdr *nlh, struct inet_diag_msg *r)
                                printf(" sack");
                        if (info->tcpi_options & TCPI_OPT_ECN)
                                printf(" ecn");
+                       if (info->tcpi_options & TCPI_OPT_ECN_SEEN)
+                               printf(" ecnseen");
                }
 
                if (tb[INET_DIAG_CONG])
-                       printf("%s", (char *) RTA_DATA(tb[INET_DIAG_CONG]));
+                       printf(" %s", rta_getattr_str(tb[INET_DIAG_CONG]));
 
                if (info->tcpi_options & TCPI_OPT_WSCALE)
                        printf(" wscale:%d,%d", info->tcpi_snd_wscale,
@@ -1384,7 +1422,7 @@ static void tcp_show_info(const struct nlmsghdr *nlh, struct inet_diag_msg *r)
        }
 }
 
-int tcp_show_sock(struct nlmsghdr *nlh, struct filter *f)
+static int tcp_show_sock(struct nlmsghdr *nlh, struct filter *f)
 {
        struct inet_diag_msg *r = NLMSG_DATA(nlh);
        struct tcpstat s;
@@ -1432,10 +1470,11 @@ int tcp_show_sock(struct nlmsghdr *nlh, struct filter *f)
        if (show_details) {
                if (r->idiag_uid)
                        printf(" uid:%u", (unsigned)r->idiag_uid);
-               printf(" ino:%u", (unsigned)r->idiag_inode);
-               printf(" sk:%08x", r->id.idiag_cookie[0]);
+               printf(" ino:%u", r->idiag_inode);
+               printf(" sk:");
                if (r->id.idiag_cookie[1] != 0)
                        printf("%08x", r->id.idiag_cookie[1]);
+               printf("%08x", r->id.idiag_cookie[0]);
        }
        if (show_mem || show_tcpinfo) {
                printf("\n\t");
@@ -1447,7 +1486,7 @@ int tcp_show_sock(struct nlmsghdr *nlh, struct filter *f)
        return 0;
 }
 
-int tcp_show_netlink(struct filter *f, FILE *dump_fp, int socktype)
+static int tcp_show_netlink(struct filter *f, FILE *dump_fp, int socktype)
 {
        int fd;
        struct sockaddr_nl nladdr;
@@ -1476,8 +1515,10 @@ int tcp_show_netlink(struct filter *f, FILE *dump_fp, int socktype)
        memset(&req.r, 0, sizeof(req.r));
        req.r.idiag_family = AF_INET;
        req.r.idiag_states = f->states;
-       if (show_mem)
+       if (show_mem) {
                req.r.idiag_ext |= (1<<(INET_DIAG_MEMINFO-1));
+               req.r.idiag_ext |= (1<<(INET_DIAG_SKMEMINFO-1));
+       }
 
        if (show_tcpinfo) {
                req.r.idiag_ext |= (1<<(INET_DIAG_INFO-1));
@@ -1505,8 +1546,10 @@ int tcp_show_netlink(struct filter *f, FILE *dump_fp, int socktype)
                .msg_iovlen = f->f ? 3 : 1,
        };
 
-       if (sendmsg(fd, &msg, 0) < 0)
+       if (sendmsg(fd, &msg, 0) < 0) {
+               close(fd);
                return -1;
+       }
 
        iov[0] = (struct iovec){
                .iov_base = buf,
@@ -1534,6 +1577,7 @@ int tcp_show_netlink(struct filter *f, FILE *dump_fp, int socktype)
                }
                if (status == 0) {
                        fprintf(stderr, "EOF on netlink\n");
+                       close(fd);
                        return 0;
                }
 
@@ -1543,27 +1587,41 @@ int tcp_show_netlink(struct filter *f, FILE *dump_fp, int socktype)
                h = (struct nlmsghdr*)buf;
                while (NLMSG_OK(h, status)) {
                        int err;
+                       struct inet_diag_msg *r = NLMSG_DATA(h);
 
                        if (/*h->nlmsg_pid != rth->local.nl_pid ||*/
                            h->nlmsg_seq != 123456)
                                goto skip_it;
 
-                       if (h->nlmsg_type == NLMSG_DONE)
+                       if (h->nlmsg_type == NLMSG_DONE) {
+                               close(fd);
                                return 0;
+                       }
                        if (h->nlmsg_type == NLMSG_ERROR) {
                                struct nlmsgerr *err = (struct nlmsgerr*)NLMSG_DATA(h);
                                if (h->nlmsg_len < NLMSG_LENGTH(sizeof(struct nlmsgerr))) {
                                        fprintf(stderr, "ERROR truncated\n");
                                } else {
                                        errno = -err->error;
+                                       if (errno == EOPNOTSUPP) {
+                                               close(fd);
+                                               return -1;
+                                       }
                                        perror("TCPDIAG answers");
                                }
+                               close(fd);
                                return 0;
                        }
                        if (!dump_fp) {
+                               if (!(f->families & (1<<r->idiag_family))) {
+                                       h = NLMSG_NEXT(h, status);
+                                       continue;
+                               }
                                err = tcp_show_sock(h, NULL);
-                               if (err < 0)
+                               if (err < 0) {
+                                       close(fd);
                                        return err;
+                               }
                        }
 
 skip_it:
@@ -1578,10 +1636,11 @@ skip_it:
                        exit(1);
                }
        }
+       close(fd);
        return 0;
 }
 
-int tcp_show_netlink_file(struct filter *f)
+static int tcp_show_netlink_file(struct filter *f)
 {
        FILE    *fp;
        char    buf[8192];
@@ -1637,9 +1696,9 @@ int tcp_show_netlink_file(struct filter *f)
        }
 }
 
-int tcp_show(struct filter *f, int socktype)
+static int tcp_show(struct filter *f, int socktype)
 {
-       int fd = -1;
+       FILE *fp = NULL;
        char *buf = NULL;
        int bufsize = 64*1024;
 
@@ -1654,6 +1713,7 @@ int tcp_show(struct filter *f, int socktype)
 
        /* Sigh... We have to parse /proc/net/tcp... */
 
+
        /* Estimate amount of sockets and try to allocate
         * huge buffer to read all the table at one read.
         * Limit it by 16MB though. The assumption is: as soon as
@@ -1681,18 +1741,21 @@ int tcp_show(struct filter *f, int socktype)
        }
 
        if (f->families & (1<<AF_INET)) {
-               if ((fd = net_tcp_open()) < 0)
+               if ((fp = net_tcp_open()) == NULL)
                        goto outerr;
-               if (generic_record_read(fd, buf, bufsize, tcp_show_line, f, AF_INET))
+
+               setbuffer(fp, buf, bufsize);
+               if (generic_record_read(fp, tcp_show_line, f, AF_INET))
                        goto outerr;
-               close(fd);
+               fclose(fp);
        }
 
        if ((f->families & (1<<AF_INET6)) &&
-           (fd = net_tcp6_open()) >= 0) {
-               if (generic_record_read(fd, buf, bufsize, tcp_show_line, f, AF_INET6))
+           (fp = net_tcp6_open()) != NULL) {
+               setbuffer(fp, buf, bufsize);
+               if (generic_record_read(fp, tcp_show_line, f, AF_INET6))
                        goto outerr;
-               close(fd);
+               fclose(fp);
        }
 
        free(buf);
@@ -1703,15 +1766,15 @@ outerr:
                int saved_errno = errno;
                if (buf)
                        free(buf);
-               if (fd >= 0)
-                       close(fd);
+               if (fp)
+                       fclose(fp);
                errno = saved_errno;
                return -1;
        } while (0);
 }
 
 
-int dgram_show_line(char *line, struct filter *f, int family)
+int dgram_show_line(char *line, const struct filter *f, int family)
 {
        struct tcpstat s;
        char *loc, *rem, *data;
@@ -1765,7 +1828,7 @@ int dgram_show_line(char *line, struct filter *f, int family)
                return 0;
 
        opt[0] = 0;
-       n = sscanf(data, "%x %x:%x %*x:%*x %*x %d %*d %d %d %llx %[^\n]\n",
+       n = sscanf(data, "%x %x:%x %*x:%*x %*x %d %*d %u %d %llx %[^\n]\n",
               &s.state, &s.wq, &s.rq,
               &s.uid, &s.ino,
               &s.refcnt, &s.sk, opt);
@@ -1792,7 +1855,7 @@ int dgram_show_line(char *line, struct filter *f, int family)
        if (show_details) {
                if (s.uid)
                        printf(" uid=%u", (unsigned)s.uid);
-               printf(" ino=%u", (unsigned)s.ino);
+               printf(" ino=%u", s.ino);
                printf(" sk=%llx", s.sk);
                if (opt[0])
                        printf(" opt:\"%s\"", opt);
@@ -1805,33 +1868,31 @@ int dgram_show_line(char *line, struct filter *f, int family)
 
 int udp_show(struct filter *f)
 {
-       int fd = -1;
-       char buf[8192];
-       int  bufsize = sizeof(buf);
+       FILE *fp = NULL;
 
        dg_proto = UDP_PROTO;
 
        if (f->families&(1<<AF_INET)) {
-               if ((fd = net_udp_open()) < 0)
+               if ((fp = net_udp_open()) == NULL)
                        goto outerr;
-               if (generic_record_read(fd, buf, bufsize, dgram_show_line, f, AF_INET))
+               if (generic_record_read(fp, dgram_show_line, f, AF_INET))
                        goto outerr;
-               close(fd);
+               fclose(fp);
        }
 
        if ((f->families&(1<<AF_INET6)) &&
-           (fd = net_udp6_open()) >= 0) {
-               if (generic_record_read(fd, buf, bufsize, dgram_show_line, f, AF_INET6))
+           (fp = net_udp6_open()) != NULL) {
+               if (generic_record_read(fp, dgram_show_line, f, AF_INET6))
                        goto outerr;
-               close(fd);
+               fclose(fp);
        }
        return 0;
 
 outerr:
        do {
                int saved_errno = errno;
-               if (fd >= 0)
-                       close(fd);
+               if (fp)
+                       fclose(fp);
                errno = saved_errno;
                return -1;
        } while (0);
@@ -1839,33 +1900,31 @@ outerr:
 
 int raw_show(struct filter *f)
 {
-       int fd = -1;
-       char buf[8192];
-       int  bufsize = sizeof(buf);
+       FILE *fp = NULL;
 
        dg_proto = RAW_PROTO;
 
        if (f->families&(1<<AF_INET)) {
-               if ((fd = net_raw_open()) < 0)
+               if ((fp = net_raw_open()) == NULL)
                        goto outerr;
-               if (generic_record_read(fd, buf, bufsize, dgram_show_line, f, AF_INET))
+               if (generic_record_read(fp, dgram_show_line, f, AF_INET))
                        goto outerr;
-               close(fd);
+               fclose(fp);
        }
 
        if ((f->families&(1<<AF_INET6)) &&
-           (fd = net_raw6_open()) >= 0) {
-               if (generic_record_read(fd, buf, bufsize, dgram_show_line, f, AF_INET6))
+           (fp = net_raw6_open()) != NULL) {
+               if (generic_record_read(fp, dgram_show_line, f, AF_INET6))
                        goto outerr;
-               close(fd);
+               fclose(fp);
        }
        return 0;
 
 outerr:
        do {
                int saved_errno = errno;
-               if (fd >= 0)
-                       close(fd);
+               if (fp)
+                       fclose(fp);
                errno = saved_errno;
                return -1;
        } while (0);
@@ -1961,6 +2020,157 @@ void unix_list_print(struct unixstat *list, struct filter *f)
        }
 }
 
+static int unix_show_sock(struct nlmsghdr *nlh, struct filter *f)
+{
+       struct unix_diag_msg *r = NLMSG_DATA(nlh);
+       struct rtattr *tb[UNIX_DIAG_MAX+1];
+       char name[128];
+       int peer_ino;
+       int rqlen;
+
+       parse_rtattr(tb, UNIX_DIAG_MAX, (struct rtattr*)(r+1),
+                    nlh->nlmsg_len - NLMSG_LENGTH(sizeof(*r)));
+
+       if (netid_width)
+               printf("%-*s ", netid_width,
+                               r->udiag_type == SOCK_STREAM ? "u_str" : "u_dgr");
+       if (state_width)
+               printf("%-*s ", state_width, sstate_name[r->udiag_state]);
+
+       if (tb[UNIX_DIAG_RQLEN])
+               rqlen = *(int *)RTA_DATA(tb[UNIX_DIAG_RQLEN]);
+       else
+               rqlen = 0;
+
+       printf("%-6d %-6d ", rqlen, 0);
+
+       if (tb[UNIX_DIAG_NAME]) {
+               int len = RTA_PAYLOAD(tb[UNIX_DIAG_NAME]);
+
+               memcpy(name, RTA_DATA(tb[UNIX_DIAG_NAME]), len);
+               name[len] = '\0';
+               if (name[0] == '\0')
+                       name[0] = '@';
+       } else
+               sprintf(name, "*");
+
+       if (tb[UNIX_DIAG_PEER])
+               peer_ino = *(int *)RTA_DATA(tb[UNIX_DIAG_PEER]);
+       else
+               peer_ino = 0;
+
+       printf("%*s %-*d %*s %-*d",
+                       addr_width, name,
+                       serv_width, r->udiag_ino,
+                       addr_width, "*", /* FIXME */
+                       serv_width, peer_ino);
+
+       if (show_users) {
+               char ubuf[4096];
+               if (find_users(r->udiag_ino, ubuf, sizeof(ubuf)) > 0)
+                       printf(" users:(%s)", ubuf);
+       }
+
+       printf("\n");
+
+       return 0;
+}
+
+static int unix_show_netlink(struct filter *f, FILE *dump_fp)
+{
+       int fd;
+       struct {
+               struct nlmsghdr nlh;
+               struct unix_diag_req r;
+       } req;
+       char    buf[8192];
+
+       if ((fd = socket(AF_NETLINK, SOCK_RAW, NETLINK_INET_DIAG)) < 0)
+               return -1;
+
+       memset(&req, 0, sizeof(req));
+       req.nlh.nlmsg_len = sizeof(req);
+       req.nlh.nlmsg_type = SOCK_DIAG_BY_FAMILY;
+       req.nlh.nlmsg_flags = NLM_F_ROOT|NLM_F_MATCH|NLM_F_REQUEST;
+       req.nlh.nlmsg_seq = 123456;
+
+       req.r.sdiag_family = AF_UNIX;
+       req.r.udiag_states = f->states;
+       req.r.udiag_show = UDIAG_SHOW_NAME | UDIAG_SHOW_PEER | UDIAG_SHOW_RQLEN;
+
+       if (send(fd, &req, sizeof(req), 0) < 0) {
+               close(fd);
+               return -1;
+       }
+
+       while (1) {
+               ssize_t status;
+               struct nlmsghdr *h;
+               struct sockaddr_nl nladdr;
+               socklen_t slen = sizeof(nladdr);
+
+               status = recvfrom(fd, buf, sizeof(buf), 0,
+                                 (struct sockaddr *) &nladdr, &slen);
+               if (status < 0) {
+                       if (errno == EINTR)
+                               continue;
+                       perror("OVERRUN");
+                       continue;
+               }
+               if (status == 0) {
+                       fprintf(stderr, "EOF on netlink\n");
+                       goto close_it;
+               }
+
+               if (dump_fp)
+                       fwrite(buf, 1, NLMSG_ALIGN(status), dump_fp);
+
+               h = (struct nlmsghdr*)buf;
+               while (NLMSG_OK(h, status)) {
+                       int err;
+
+                       if (/*h->nlmsg_pid != rth->local.nl_pid ||*/
+                           h->nlmsg_seq != 123456)
+                               goto skip_it;
+
+                       if (h->nlmsg_type == NLMSG_DONE)
+                               goto close_it;
+
+                       if (h->nlmsg_type == NLMSG_ERROR) {
+                               struct nlmsgerr *err = (struct nlmsgerr*)NLMSG_DATA(h);
+                               if (h->nlmsg_len < NLMSG_LENGTH(sizeof(struct nlmsgerr))) {
+                                       fprintf(stderr, "ERROR truncated\n");
+                               } else {
+                                       errno = -err->error;
+                                       if (errno != ENOENT)
+                                               fprintf(stderr, "UDIAG answers %d\n", errno);
+                               }
+                               close(fd);
+                               return -1;
+                       }
+                       if (!dump_fp) {
+                               err = unix_show_sock(h, f);
+                               if (err < 0) {
+                                       close(fd);
+                                       return err;
+                               }
+                       }
+
+skip_it:
+                       h = NLMSG_NEXT(h, status);
+               }
+
+               if (status) {
+                       fprintf(stderr, "!!!Remnant of size %zd\n", status);
+                       exit(1);
+               }
+       }
+
+close_it:
+       close(fd);
+       return 0;
+}
+
 int unix_show(struct filter *f)
 {
        FILE *fp;
@@ -1970,7 +2180,11 @@ int unix_show(struct filter *f)
        int  cnt;
        struct unixstat *list = NULL;
 
-       if ((fp = fdopen(net_unix_open(), "r")) == NULL)
+       if (!getenv("PROC_NET_UNIX") && !getenv("PROC_ROOT")
+           && unix_show_netlink(f, NULL) == 0)
+               return 0;
+
+       if ((fp = net_unix_open()) == NULL)
                return -1;
        fgets(buf, sizeof(buf)-1, fp);
 
@@ -2030,7 +2244,7 @@ int unix_show(struct filter *f)
                        cnt = 0;
                }
        }
-
+       fclose(fp);
        if (list) {
                unix_list_print(list, f);
                unix_list_free(list);
@@ -2058,7 +2272,7 @@ int packet_show(struct filter *f)
        if (!(f->states & (1<<SS_CLOSE)))
                return 0;
 
-       if ((fp = fdopen(net_packet_open(), "r")) == NULL)
+       if ((fp = net_packet_open()) == NULL)
                return -1;
        fgets(buf, sizeof(buf)-1, fp);
 
@@ -2131,7 +2345,7 @@ int netlink_show(struct filter *f)
        if (!(f->states & (1<<SS_CLOSE)))
                return 0;
 
-       if ((fp = fdopen(net_netlink_open(), "r")) == NULL)
+       if ((fp = net_netlink_open()) == NULL)
                return -1;
        fgets(buf, sizeof(buf)-1, fp);
 
@@ -2217,7 +2431,7 @@ int get_snmp_int(char *proto, char *key, int *result)
 
        *result = 0;
 
-       if ((fp = fdopen(net_snmp_open(), "r")) == NULL)
+       if ((fp = net_snmp_open()) == NULL)
                return -1;
 
        while (fgets(buf, sizeof(buf), fp) != NULL) {
@@ -2310,13 +2524,13 @@ int get_sockstat(struct sockstat *s)
 
        memset(s, 0, sizeof(*s));
 
-       if ((fp = fdopen(net_sockstat_open(), "r")) == NULL)
+       if ((fp = net_sockstat_open()) == NULL)
                return -1;
        while(fgets(buf, sizeof(buf), fp) != NULL)
                get_sockstat_line(buf, s);
        fclose(fp);
 
-       if ((fp = fdopen(net_sockstat6_open(), "r")) == NULL)
+       if ((fp = net_sockstat6_open()) == NULL)
                return 0;
        while(fgets(buf, sizeof(buf), fp) != NULL)
                get_sockstat_line(buf, s);
@@ -2365,12 +2579,9 @@ int print_summary(void)
        return 0;
 }
 
-
-static void usage(void) __attribute__((noreturn));
-
-static void usage(void)
+static void _usage(FILE *dest)
 {
-       fprintf(stderr,
+       fprintf(dest,
 "Usage: ss [ OPTIONS ]\n"
 "       ss [ OPTIONS ] [ FILTER ]\n"
 "   -h, --help         this message\n"
@@ -2396,12 +2607,26 @@ static void usage(void)
 "   -x, --unix         display only Unix domain sockets\n"
 "   -f, --family=FAMILY display sockets of type FAMILY\n"
 "\n"
-"   -A, --query=QUERY\n"
+"   -A, --query=QUERY, --socket=QUERY\n"
 "       QUERY := {all|inet|tcp|udp|raw|unix|packet|netlink}[,QUERY]\n"
 "\n"
+"   -D, --diag=FILE     Dump raw information about TCP sockets to FILE\n"
 "   -F, --filter=FILE   read filter information from FILE\n"
 "       FILTER := [ state TCP-STATE ] [ EXPRESSION ]\n"
                );
+}
+
+static void help(void) __attribute__((noreturn));
+static void help(void)
+{
+       _usage(stdout);
+       exit(0);
+}
+
+static void usage(void) __attribute__((noreturn));
+static void usage(void)
+{
+       _usage(stderr);
        exit(-1);
 }
 
@@ -2453,8 +2678,9 @@ static const struct option long_opts[] = {
        { "packet", 0, 0, '0' },
        { "family", 1, 0, 'f' },
        { "socket", 1, 0, 'A' },
+       { "query", 1, 0, 'A' },
        { "summary", 0, 0, 's' },
-       { "diag", 0, 0, 'D' },
+       { "diag", 1, 0, 'D' },
        { "filter", 1, 0, 'F' },
        { "version", 0, 0, 'V' },
        { "help", 0, 0, 'h' },
@@ -2500,6 +2726,7 @@ int main(int argc, char *argv[])
                        break;
                case 'p':
                        show_users++;
+                       user_ent_hash_build();
                        break;
                case 'd':
                        current_filter.dbs |= (1<<DCCP_DB);
@@ -2525,7 +2752,7 @@ int main(int argc, char *argv[])
                        current_filter.states = SS_ALL;
                        break;
                case 'l':
-                       current_filter.states = (1<<SS_LISTEN);
+                       current_filter.states = (1<<SS_LISTEN) | (1<<SS_CLOSE);
                        break;
                case '4':
                        preferred_family = AF_INET;
@@ -2548,7 +2775,7 @@ int main(int argc, char *argv[])
                        else if (strcmp(optarg, "netlink") == 0)
                                preferred_family = AF_NETLINK;
                        else if (strcmp(optarg, "help") == 0)
-                               usage();
+                               help();
                        else {
                                fprintf(stderr, "ss: \"%s\" is invalid family\n", optarg);
                                usage();
@@ -2630,6 +2857,7 @@ int main(int argc, char *argv[])
                        exit(0);
                case 'h':
                case '?':
+                       help();
                default:
                        usage();
                }
@@ -2662,9 +2890,7 @@ int main(int argc, char *argv[])
                int mask2;
                if (preferred_family == AF_INET ||
                    preferred_family == AF_INET6) {
-                       mask2= (1<<TCP_DB);
-                       if (!do_default)
-                               mask2 = (1<<UDP_DB)|(1<<RAW_DB);
+                       mask2= current_filter.dbs;
                } else if (preferred_family == AF_PACKET) {
                        mask2 = PACKET_DBM;
                } else if (preferred_family == AF_UNIX) {
@@ -2800,7 +3026,6 @@ int main(int argc, char *argv[])
               addr_width, "Local Address", serv_width, "Port",
               addr_width, "Peer Address", serv_width, "Port");
 
-//printf("%08x %08x %08x\n", current_filter.dbs, current_filter.states, current_filter.families);
        fflush(stdout);
 
        if (current_filter.dbs & (1<<NETLINK_DB))