diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-17 06:40:13 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-17 06:40:13 +0000 |
commit | e9be59e1502a41bab9891d96d753102a7dafef0b (patch) | |
tree | c3b2da87c414881f4b53d0964f407c83492d813e /lib/clplumbing/ipcsocket.c | |
parent | Initial commit. (diff) | |
download | cluster-glue-e9be59e1502a41bab9891d96d753102a7dafef0b.tar.xz cluster-glue-e9be59e1502a41bab9891d96d753102a7dafef0b.zip |
Adding upstream version 1.0.12.upstream/1.0.12upstream
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to 'lib/clplumbing/ipcsocket.c')
-rw-r--r-- | lib/clplumbing/ipcsocket.c | 2767 |
1 files changed, 2767 insertions, 0 deletions
diff --git a/lib/clplumbing/ipcsocket.c b/lib/clplumbing/ipcsocket.c new file mode 100644 index 0000000..14c3504 --- /dev/null +++ b/lib/clplumbing/ipcsocket.c @@ -0,0 +1,2767 @@ +/* + * ipcsocket unix domain socket implementation of IPC abstraction. + * + * Copyright (c) 2002 Xiaoxiang Liu <xiliu@ncsa.uiuc.edu> + * + * Stream support (c) 2004,2006 David Lee <t.d.lee@durham.ac.uk> + * Note: many of the variable/function names "*socket*" should be + * interpreted as having a more generic "ipc-channel-type" meaning. + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, write to the Free Software + * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA + * + */ + +#include <lha_internal.h> + +#include <clplumbing/ipc.h> +#include <clplumbing/cl_log.h> +#include <clplumbing/realtime.h> +#include <clplumbing/cl_poll.h> + +#include <ha_msg.h> +/* avoid including cib.h - used in gshi's "late message" code to avoid + * printing insanely large messages + */ +#define F_CIB_CALLDATA "cib_calldata" + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <syslog.h> +#include <time.h> +#include <sched.h> +#include <sys/types.h> +#include <sys/stat.h> +#include <sys/param.h> +#include <sys/uio.h> +#ifdef HAVE_SYS_FILIO_H +# include <sys/filio.h> +#endif +#ifdef HAVE_SYS_SYSLIMITS_H +# include <sys/syslimits.h> +#endif +#ifdef HAVE_SYS_CRED_H +# include <sys/cred.h> +#endif +#ifdef HAVE_SYS_UCRED_H +# include <sys/ucred.h> +#endif + +/* For 'getpeerucred()' (Solaris 10 upwards) */ +#ifdef HAVE_UCRED_H +# include <ucred.h> +#endif + +#ifdef HAVE_SYS_SOCKET_H +# include <sys/socket.h> +#endif + +/* + * Normally use "socket" code. But on some OSes alternatives may be + * preferred (or necessary). + */ +#define HB_IPC_SOCKET 1 +#define HB_IPC_STREAM 2 +/* #define HB_IPC_ANOTHER 3 */ + +#ifndef HB_IPC_METHOD +# if defined(SO_PEERCRED) || defined(HAVE_GETPEEREID) \ + || defined(SCM_CREDS) || defined(HAVE_GETPEERUCRED) +# define HB_IPC_METHOD HB_IPC_SOCKET +# elif defined(HAVE_STROPTS_H) +# define HB_IPC_METHOD HB_IPC_STREAM +# else +# error. Surely we have sockets or streams... +# endif +#endif + +#if HB_IPC_METHOD == HB_IPC_SOCKET +# include <sys/poll.h> +# include <netinet/in.h> +# include <sys/un.h> +#elif HB_IPC_METHOD == HB_IPC_STREAM +# include <stropts.h> +#else +# error "IPC type invalid" +#endif + +#include <sys/ioctl.h> +#include <unistd.h> +#include <errno.h> +#include <fcntl.h> + +#ifndef UNIX_PATH_MAX +# define UNIX_PATH_MAX 108 +#endif + +#if HB_IPC_METHOD == HB_IPC_SOCKET + +# define MAX_LISTEN_NUM 128 + +# ifndef MSG_NOSIGNAL +# define MSG_NOSIGNAL 0 +# endif + +# ifndef AF_LOCAL +# define AF_LOCAL AF_UNIX +# endif + +#endif /* HB_IPC_METHOD */ + +/*********************************************************************** + * + * Determine the IPC authentication scheme... More machine dependent than + * we'd like, but don't know any better way... + * + ***********************************************************************/ +#ifdef SO_PEERCRED +# define USE_SO_PEERCRED +#elif HAVE_GETPEEREID +# define USE_GETPEEREID +#elif defined(SCM_CREDS) +# define USE_SCM_CREDS +#elif HAVE_GETPEERUCRED /* e.g. Solaris 10 upwards */ +# define USE_GETPEERUCRED +#elif HB_IPC_METHOD == HB_IPC_STREAM +# define USE_STREAM_CREDS +#else +# define USE_DUMMY_CREDS +/* This will make it compile, but attempts to authenticate + * will fail. This is a stopgap measure ;-) + */ +#endif + +#if HB_IPC_METHOD == HB_IPC_SOCKET + +# ifdef USE_BINDSTAT_CREDS +# ifndef SUN_LEN +# define SUN_LEN(ptr) ((size_t) (offsetof (sockaddr_un, sun_path) + strlen ((ptr)->sun_path)) +# endif +# endif + +#endif /* HB_IPC_METHOD */ + +/* wait connection private data. */ +struct SOCKET_WAIT_CONN_PRIVATE{ + /* the path name wich the connection will be built on. */ + char path_name[UNIX_PATH_MAX]; +#if HB_IPC_METHOD == HB_IPC_SOCKET + /* the domain socket. */ + int s; +#elif HB_IPC_METHOD == HB_IPC_STREAM + /* the streams pipe */ + int pipefds[2]; +#endif +}; + +/* channel private data. */ +struct SOCKET_CH_PRIVATE{ + /* the path name wich the connection will be built on. */ + char path_name[UNIX_PATH_MAX]; + /* the domain socket. */ + int s; + /* the size of expecting data for below buffered message buf_msg */ + int remaining_data; + +#if HB_IPC_METHOD == HB_IPC_SOCKET + /* The address of our peer - used by USE_BINDSTAT_CREDS version of + * socket_verify_auth() + */ + struct sockaddr_un *peer_addr; +#elif HB_IPC_METHOD == HB_IPC_STREAM + uid_t farside_uid; + gid_t farside_gid; +#endif + + /* the buf used to save unfinished message */ + struct IPC_MESSAGE *buf_msg; +}; + +struct IPC_Stats { + long nsent; + long noutqueued; + long send_count; + long nreceived; + long ninqueued; + long recv_count; + int last_recv_errno; + int last_recv_rc; + int last_send_errno; + int last_send_rc; +}; + +static struct IPC_Stats SocketIPCStats = {0, 0, 0, 0}; +extern int debug_level; + +/* unix domain socket implementations of IPC functions. */ + +static int socket_resume_io(struct IPC_CHANNEL *ch); + +static struct IPC_MESSAGE* socket_message_new(struct IPC_CHANNEL*ch +, int msg_len); + +struct IPC_WAIT_CONNECTION *socket_wait_conn_new(GHashTable* ch_attrs); + +/* *** FIXME: This is also declared in 'ocf_ipc.c'. */ +struct IPC_CHANNEL* socket_client_channel_new(GHashTable *attrs); + +static struct IPC_CHANNEL* socket_server_channel_new(int sockfd); + +static struct IPC_CHANNEL * channel_new(int sockfd, int conntype, const char *pathname); +static int client_channel_new_auth(int sockfd); +static int verify_creds(struct IPC_AUTH *auth_info, uid_t uid, gid_t gid); + +typedef void (*DelProc)(IPC_Message*); + +static struct IPC_MESSAGE * ipcmsg_new(struct IPC_CHANNEL* ch, + const void* data, int len, void* private, DelProc d); + +static pid_t socket_get_farside_pid(int sockfd); + +extern int (*ipc_pollfunc_ptr)(struct pollfd *, nfds_t, int); + +static int socket_resume_io_read(struct IPC_CHANNEL *ch, int*, gboolean read1anyway); + +static struct IPC_OPS socket_ops; +static gboolean ipc_time_debug_flag = TRUE; + +void +set_ipc_time_debug_flag(gboolean flag) +{ + ipc_time_debug_flag = flag; +} + +#ifdef IPC_TIME_DEBUG + +extern struct ha_msg* wirefmt2msg(const char* s, size_t length, int flag); +void cl_log_message (int log_level, const struct ha_msg *m); +int timediff(longclock_t t1, longclock_t t2); +void ha_msg_del(struct ha_msg* msg); +void ipc_time_debug(IPC_Channel* ch, IPC_Message* ipcmsg, int whichpos); + +#define SET_ENQUEUE_TIME(x,t) memcpy(&((struct SOCKET_MSG_HEAD*)x->msg_buf)->enqueue_time, &t, sizeof(longclock_t)) +#define SET_SEND_TIME(x,t) memcpy(&((struct SOCKET_MSG_HEAD*)x->msg_buf)->send_time, &t, sizeof(longclock_t)) +#define SET_RECV_TIME(x,t) memcpy(&((struct SOCKET_MSG_HEAD*)x->msg_buf)->recv_time, &t, sizeof(longclock_t)) +#define SET_DEQUEUE_TIME(x,t) memcpy(&((struct SOCKET_MSG_HEAD*)x->msg_buf)->dequeue_time, &t, sizeof(longclock_t)) + +static +longclock_t +get_enqueue_time(IPC_Message *ipcmsg) +{ + longclock_t t; + + memcpy(&t, + &(((struct SOCKET_MSG_HEAD *)ipcmsg->msg_buf)->enqueue_time), + sizeof(longclock_t)); + + return t; +} + +int +timediff(longclock_t t1, longclock_t t2) +{ + longclock_t remain; + + remain = sub_longclock(t1, t2); + + return longclockto_ms(remain); +} + +void +ipc_time_debug(IPC_Channel* ch, IPC_Message* ipcmsg, int whichpos) +{ + int msdiff = 0; + longclock_t lnow = time_longclock(); + char positions[4][16]={ + "enqueue", + "send", + "recv", + "dequeue"}; + + if (ipc_time_debug_flag == FALSE) { + return ; + } + + if (ipcmsg->msg_body == NULL + || ipcmsg->msg_buf == NULL) { + cl_log(LOG_ERR, "msg_body =%p, msg_bu=%p", + ipcmsg->msg_body, ipcmsg->msg_buf); + abort(); + return; + } + + switch(whichpos) { + case MSGPOS_ENQUEUE: + SET_ENQUEUE_TIME(ipcmsg, lnow); + break; + case MSGPOS_SEND: + SET_SEND_TIME(ipcmsg, lnow); + goto checktime; + case MSGPOS_RECV: + SET_RECV_TIME(ipcmsg, lnow); + goto checktime; + case MSGPOS_DEQUEUE: + SET_DEQUEUE_TIME(ipcmsg, lnow); + + checktime: + msdiff = timediff(lnow, get_enqueue_time(ipcmsg)); + if (msdiff > MAXIPCTIME) { + struct ha_msg* hamsg = NULL; + cl_log(LOG_WARNING, + " message delayed from enqueue to %s %d ms " + "(enqueue-time=%lu, peer pid=%d) ", + positions[whichpos], + msdiff, + longclockto_ms(get_enqueue_time(ipcmsg)), + ch->farside_pid); + + (void)hamsg; +#if 0 + hamsg = wirefmt2msg(ipcmsg->msg_body, ipcmsg->msg_len, 0); + if (hamsg != NULL) { + struct ha_msg *crm_data = NULL; + crm_data = cl_get_struct( + hamsg, F_CRM_DATA); + + if(crm_data == NULL) { + crm_data = cl_get_struct( + hamsg, F_CIB_CALLDATA); + } + if(crm_data != NULL) { + cl_msg_remove_value( + hamsg, crm_data); + } + + cl_log_message(LOG_DEBUG, hamsg); + ha_msg_del(hamsg); + } else { + if (!ipcmsg) { + cl_log(LOG_ERR, + "IPC msg 0x%lx is unallocated" + , (gulong)ipcmsg); + return; + } + if (!ipcmsg->msg_body) { + cl_log(LOG_ERR, + "IPC msg body 0x%lx is unallocated" + , (gulong)ipcmsg->msg_body); + return; + } + } +#endif + + } + break; + default: + cl_log(LOG_ERR, "wrong position value in IPC:%d", whichpos); + return; + } +} +#endif + +void dump_ipc_info(const IPC_Channel* chan); + +#undef AUDIT_CHANNELS + +#ifndef AUDIT_CHANNELS +# define CHANAUDIT(ch) /*NOTHING */ +#else +# define CHANAUDIT(ch) socket_chan_audit(ch) +# define MAXPID 65535 + +static void +socket_chan_audit(const struct IPC_CHANNEL* ch) +{ + int badch = FALSE; + + struct SOCKET_CH_PRIVATE *chp; + struct stat b; + + if ((chp = ch->ch_private) == NULL) { + cl_log(LOG_CRIT, "Bad ch_private"); + badch = TRUE; + } + if (ch->ops != &socket_ops) { + cl_log(LOG_CRIT, "Bad socket_ops"); + badch = TRUE; + } + if (ch->ch_status == IPC_DISCONNECT) { + return; + } + if (!IPC_ISRCONN(ch)) { + cl_log(LOG_CRIT, "Bad ch_status [%d]", ch->ch_status); + badch = TRUE; + } + if (ch->farside_pid < 0 || ch->farside_pid > MAXPID) { + cl_log(LOG_CRIT, "Bad farside_pid"); + badch = TRUE; + } + if (fstat(chp->s, &b) < 0) { + badch = TRUE; + } else if ((b.st_mode & S_IFMT) != S_IFSOCK) { + cl_log(LOG_CRIT, "channel @ 0x%lx: not a socket" + , (unsigned long)ch); + badch = TRUE; + } + if (chp->remaining_data < 0) { + cl_log(LOG_CRIT, "Negative remaining_data"); + badch = TRUE; + } + if (chp->remaining_data < 0 || chp->remaining_data > MAXMSG) { + cl_log(LOG_CRIT, "Excessive/bad remaining_data"); + badch = TRUE; + } + if (chp->remaining_data && chp->buf_msg == NULL) { + cl_log(LOG_CRIT + , "inconsistent remaining_data [%ld]/buf_msg[0x%lx]" + , (long)chp->remaining_data, (unsigned long)chp->buf_msg); + badch = TRUE; + } + if (chp->remaining_data == 0 && chp->buf_msg != NULL) { + cl_log(LOG_CRIT + , "inconsistent remaining_data [%ld]/buf_msg[0x%lx] (2)" + , (long)chp->remaining_data, (unsigned long)chp->buf_msg); + badch = TRUE; + } + if (ch->send_queue == NULL || ch->recv_queue == NULL) { + cl_log(LOG_CRIT, "bad send/recv queue"); + badch = TRUE; + } + if (ch->recv_queue->current_qlen < 0 + || ch->recv_queue->current_qlen > ch->recv_queue->max_qlen) { + cl_log(LOG_CRIT, "bad recv queue"); + badch = TRUE; + } + if (ch->send_queue->current_qlen < 0 + || ch->send_queue->current_qlen > ch->send_queue->max_qlen) { + cl_log(LOG_CRIT, "bad send_queue"); + badch = TRUE; + } + if (badch) { + cl_log(LOG_CRIT, "Bad channel @ 0x%lx", (unsigned long)ch); + dump_ipc_info(ch); + abort(); + } +} +#endif + +#ifdef CHEAT_CHECKS +long SeqNums[32]; + +static long +cheat_get_sequence(IPC_Message* msg) +{ + const char header [] = "String-"; + size_t header_len = sizeof(header)-1; + char * body; + + if (msg == NULL || msg->msg_len < sizeof(header) + || msg->msg_len > sizeof(header) + 10 + || strncmp(msg->msg_body, header, header_len) != 0) { + return -1L; + } + body = msg->msg_body; + return atol(body+header_len); +} +static char SavedReadBody[32]; +static char SavedReceivedBody[32]; +static char SavedQueuedBody[32]; +static char SavedSentBody[32]; +#ifndef MIN +# define MIN(a,b) (a < b ? a : b) +#endif + +static void +save_body(struct IPC_MESSAGE *msg, char * savearea, size_t length) +{ + int mlen = strnlen(msg->msg_body, MIN(length, msg->msg_len)); + memcpy(savearea, msg->msg_body, mlen); + savearea[mlen] = EOS; +} + +static void +audit_readmsgq_msg(gpointer msg, gpointer user_data) +{ + long cheatseq = cheat_get_sequence(msg); + + if (cheatseq < SeqNums[1] || cheatseq > SeqNums[2]) { + cl_log(LOG_ERR + , "Read Q Message %ld not in range [%ld:%ld]" + , cheatseq, SeqNums[1], SeqNums[2]); + } +} + +static void +saveandcheck(struct IPC_CHANNEL * ch, struct IPC_MESSAGE* msg, char * savearea +, size_t savesize, long* lastseq, const char * text) +{ + long cheatseq = cheat_get_sequence(msg); + + save_body(msg, savearea, savesize); + if (*lastseq != 0 ) { + if (cheatseq != *lastseq +1) { + int j; + cl_log(LOG_ERR + , "%s packets out of sequence! %ld versus %ld [pid %d]" + , text, cheatseq, *lastseq, (int)getpid()); + dump_ipc_info(ch); + for (j=0; j < 4; ++j) { + cl_log(LOG_DEBUG + , "SeqNums[%d] = %ld" + , j, SeqNums[j]); + } + cl_log(LOG_ERR + , "SocketIPCStats.nsent = %ld" + , SocketIPCStats.nsent); + cl_log(LOG_ERR + , "SocketIPCStats.noutqueued = %ld" + , SocketIPCStats.noutqueued); + cl_log(LOG_ERR + , "SocketIPCStats.nreceived = %ld" + , SocketIPCStats.nreceived); + cl_log(LOG_ERR + , "SocketIPCStats.ninqueued = %ld" + , SocketIPCStats.ninqueued); + } + + } + g_list_foreach(ch->recv_queue->queue, audit_readmsgq_msg, NULL); + if (cheatseq > 0) { + *lastseq = cheatseq; + } +} + +# define CHECKFOO(which, ch, msg, area, text) { \ + saveandcheck(ch,msg,area,sizeof(area),SeqNums+which,text); \ + } +#else +# define CHECKFOO(which, ch, msg, area, text) /* Nothing */ +#endif + +static void +dump_msg(struct IPC_MESSAGE *msg, const char * label) +{ +#ifdef CHEAT_CHECKS + cl_log(LOG_DEBUG, "%s packet (length %d) [%s] %ld pid %d" + , label, (int)msg->msg_len, (char*)msg->msg_body + , cheat_get_sequence(msg), (int)getpid()); +#else + cl_log(LOG_DEBUG, "%s length %d [%s] pid %d" + , label, (int)msg->msg_len, (char*)msg->msg_body + , (int)getpid()); +#endif +} + +static void +dump_msgq_msg(gpointer data, gpointer user_data) +{ + dump_msg(data, user_data); +} + +void +dump_ipc_info(const IPC_Channel* chan) +{ + char squeue[] = "Send queue"; + char rqueue[] = "Receive queue"; +#ifdef CHEAT_CHECKS + cl_log(LOG_DEBUG, "Saved Last Body read[%s]", SavedReadBody); + cl_log(LOG_DEBUG, "Saved Last Body received[%s]", SavedReceivedBody); + cl_log(LOG_DEBUG, "Saved Last Body Queued[%s]", SavedQueuedBody); + cl_log(LOG_DEBUG, "Saved Last Body Sent[%s]", SavedSentBody); +#endif + g_list_foreach(chan->send_queue->queue, dump_msgq_msg, squeue); + g_list_foreach(chan->recv_queue->queue, dump_msgq_msg, rqueue); + CHANAUDIT(chan); +} + +/* destroy socket wait channel */ +static void +socket_destroy_wait_conn(struct IPC_WAIT_CONNECTION * wait_conn) +{ + struct SOCKET_WAIT_CONN_PRIVATE * wc = wait_conn->ch_private; + + if (wc != NULL) { +#if HB_IPC_METHOD == HB_IPC_SOCKET + if (wc->s >= 0) { + if (debug_level > 1) { + cl_log(LOG_DEBUG + , "%s: closing socket %d" + , __FUNCTION__, wc->s); + } + close(wc->s); + cl_poll_ignore(wc->s); + unlink(wc->path_name); + wc->s = -1; + } +#elif HB_IPC_METHOD == HB_IPC_STREAM + cl_poll_ignore(wc->pipefds[0]); + if (wc->pipefds[0] >= 0) { + if (debug_level > 1) { + cl_log(LOG_DEBUG + , "%s: closing pipe[0] %d" + , __FUNCTION__, wc->pipefds[0]); + } + wc->pipefds[0] = -1; + } + if (wc->pipefds[1] >= 0) { + if (debug_level > 1) { + cl_log(LOG_DEBUG + , "%s: closing pipe[1] %d" + , __FUNCTION__, wc->pipefds[1]); + } + wc->pipefds[0] = -1; + } + unlink(wc->path_name); +#endif + g_free(wc); + } + g_free((void*) wait_conn); +} + +/* return a fd which can be listened on for new connections. */ +static int +socket_wait_selectfd(struct IPC_WAIT_CONNECTION *wait_conn) +{ + struct SOCKET_WAIT_CONN_PRIVATE * wc = wait_conn->ch_private; + +#if HB_IPC_METHOD == HB_IPC_SOCKET + return (wc == NULL ? -1 : wc->s); +#elif HB_IPC_METHOD == HB_IPC_STREAM + return (wc == NULL ? -1 : wc->pipefds[0]); +#endif +} + +/* socket accept connection. */ +static struct IPC_CHANNEL* +socket_accept_connection(struct IPC_WAIT_CONNECTION * wait_conn +, struct IPC_AUTH *auth_info) +{ + struct IPC_CHANNEL * ch = NULL; + int s; + int new_sock; + struct SOCKET_WAIT_CONN_PRIVATE* conn_private; + struct SOCKET_CH_PRIVATE * ch_private ; + int auth_result = IPC_FAIL; + int saveerrno=errno; + gboolean was_error = FALSE; +#if HB_IPC_METHOD == HB_IPC_SOCKET + /* make peer_addr a pointer so it can be used by the + * USE_BINDSTAT_CREDS implementation of socket_verify_auth() + */ + struct sockaddr_un * peer_addr = NULL; + socklen_t sin_size; +#elif HB_IPC_METHOD == HB_IPC_STREAM + struct strrecvfd strrecvfd; +#endif + + /* get select fd */ + + s = wait_conn->ops->get_select_fd(wait_conn); + if (s < 0) { + cl_log(LOG_ERR, "get_select_fd: invalid fd"); + return NULL; + } + + /* Get client connection. */ +#if HB_IPC_METHOD == HB_IPC_SOCKET + peer_addr = g_new(struct sockaddr_un, 1); + *peer_addr->sun_path = '\0'; + sin_size = sizeof(struct sockaddr_un); + new_sock = accept(s, (struct sockaddr *)peer_addr, &sin_size); +#elif HB_IPC_METHOD == HB_IPC_STREAM + if (ioctl(s, I_RECVFD, &strrecvfd) == -1) { + new_sock = -1; + } + else { + new_sock = strrecvfd.fd; + } +#endif + saveerrno=errno; + if (new_sock == -1) { + if (errno != EAGAIN && errno != EWOULDBLOCK) { + cl_perror("socket_accept_connection: accept(sock=%d)" + , s); + } + was_error = TRUE; + + } else { + if ((ch = socket_server_channel_new(new_sock)) == NULL) { + cl_log(LOG_ERR + , "socket_accept_connection:" + " Can't create new channel"); + was_error = TRUE; + } else { + conn_private=(struct SOCKET_WAIT_CONN_PRIVATE*) + ( wait_conn->ch_private); + ch_private = (struct SOCKET_CH_PRIVATE *)(ch->ch_private); + strncpy(ch_private->path_name,conn_private->path_name + , sizeof(conn_private->path_name)); + +#if HB_IPC_METHOD == HB_IPC_SOCKET + ch_private->peer_addr = peer_addr; +#elif HB_IPC_METHOD == HB_IPC_STREAM + ch_private->farside_uid = strrecvfd.uid; + ch_private->farside_gid = strrecvfd.gid; +#endif + } + } + + /* Verify the client authorization information. */ + if(was_error == FALSE) { + auth_result = ch->ops->verify_auth(ch, auth_info); + if (auth_result == IPC_OK) { + ch->ch_status = IPC_CONNECT; + ch->farside_pid = socket_get_farside_pid(new_sock); + return ch; + } + saveerrno=errno; + } + +#if HB_IPC_METHOD == HB_IPC_SOCKET + g_free(peer_addr); + peer_addr = NULL; +#endif + errno=saveerrno; + return NULL; +} + +/* + * Called by socket_destroy(). Disconnect the connection + * and set ch_status to IPC_DISCONNECT. + * + * parameters : + * ch (IN) the pointer to the channel. + * + * return values : + * IPC_OK the connection is disconnected successfully. + * IPC_FAIL operation fails. +*/ + +static int +socket_disconnect(struct IPC_CHANNEL* ch) +{ + struct SOCKET_CH_PRIVATE* conn_info; + + conn_info = (struct SOCKET_CH_PRIVATE*) ch->ch_private; + if (debug_level > 1) { + cl_log(LOG_DEBUG + , "%s(sock=%d, ch=0x%lx){" + , __FUNCTION__ + , conn_info->s, (unsigned long)ch); + } +#if 0 + if (ch->ch_status != IPC_DISCONNECT) { + cl_log(LOG_INFO, "forced disconnect for fd %d", conn_info->s); + } +#endif + if (ch->ch_status == IPC_CONNECT) { + socket_resume_io(ch); + } + + if (conn_info->s >= 0) { + if (debug_level > 1) { + cl_log(LOG_DEBUG + , "%s: closing socket %d" + , __FUNCTION__, conn_info->s); + } + close(conn_info->s); + cl_poll_ignore(conn_info->s); + conn_info->s = -1; + } + ch->ch_status = IPC_DISCONNECT; + if (debug_level > 1) { + cl_log(LOG_DEBUG, "}/*%s(sock=%d, ch=0x%lx)*/" + , __FUNCTION__, conn_info->s, (unsigned long)ch); + } + return IPC_OK; +} + +/* + * destroy a ipc queue and clean all memory space assigned to this queue. + * parameters: + * q (IN) the pointer to the queue which should be destroied. + * + * FIXME: This function does not free up messages that might + * be in the queue. + */ + +static void +socket_destroy_queue(struct IPC_QUEUE * q) +{ + g_list_free(q->queue); + + g_free((void *) q); +} + +static void +socket_destroy_channel(struct IPC_CHANNEL * ch) +{ + --ch->refcount; + if (ch->refcount > 0) { + return; + } + if (ch->ch_status == IPC_CONNECT) { + socket_resume_io(ch); + } + if (debug_level > 1) { + cl_log(LOG_DEBUG, "socket_destroy(ch=0x%lx){" + , (unsigned long)ch); + } + socket_disconnect(ch); + socket_destroy_queue(ch->send_queue); + socket_destroy_queue(ch->recv_queue); + + if (ch->pool) { + ipc_bufpool_unref(ch->pool); + } + + if (ch->ch_private != NULL) { +#if HB_IPC_METHOD == HB_IPC_SOCKET + struct SOCKET_CH_PRIVATE *priv = (struct SOCKET_CH_PRIVATE *) + ch->ch_private; + if(priv->peer_addr != NULL) { + if (*priv->peer_addr->sun_path) { + unlink(priv->peer_addr->sun_path); + } + g_free((void*)(priv->peer_addr)); + } +#endif + g_free((void*)(ch->ch_private)); + } + memset(ch, 0xff, sizeof(*ch)); + g_free((void*)ch); + if (debug_level > 1) { + cl_log(LOG_DEBUG, "}/*socket_destroy(ch=0x%lx)*/" + , (unsigned long)ch); + } +} + +static int +socket_check_disc_pending(struct IPC_CHANNEL* ch) +{ + int rc; + struct pollfd sockpoll; + + if (ch->ch_status == IPC_DISCONNECT) { + cl_log(LOG_ERR, "check_disc_pending() already disconnected"); + return IPC_BROKEN; + } + if (ch->recv_queue->current_qlen > 0) { + return IPC_OK; + } + sockpoll.fd = ch->ops->get_recv_select_fd(ch); + sockpoll.events = POLLIN; + + rc = ipc_pollfunc_ptr(&sockpoll, 1, 0); + + if (rc < 0) { + cl_log(LOG_INFO + , "socket_check_disc_pending() bad poll call"); + ch->ch_status = IPC_DISCONNECT; + return IPC_BROKEN; + } + + if (sockpoll.revents & POLLHUP) { + if (sockpoll.revents & POLLIN) { + ch->ch_status = IPC_DISC_PENDING; + } else { +#if 1 + cl_log(LOG_INFO, "HUP without input"); +#endif + ch->ch_status = IPC_DISCONNECT; + return IPC_BROKEN; + } + } + if (sockpoll.revents & POLLIN) { + int dummy; + socket_resume_io_read(ch, &dummy, FALSE); + } + return IPC_OK; +} + +static int +socket_initiate_connection(struct IPC_CHANNEL * ch) +{ + struct SOCKET_CH_PRIVATE* conn_info; +#if HB_IPC_METHOD == HB_IPC_SOCKET + struct sockaddr_un peer_addr; /* connector's address information */ +#elif HB_IPC_METHOD == HB_IPC_STREAM +#endif + + conn_info = (struct SOCKET_CH_PRIVATE*) ch->ch_private; + +#if HB_IPC_METHOD == HB_IPC_SOCKET + /* Prepare the socket */ + memset(&peer_addr, 0, sizeof(peer_addr)); + peer_addr.sun_family = AF_LOCAL; /* host byte order */ + + if (strlen(conn_info->path_name) >= sizeof(peer_addr.sun_path)) { + return IPC_FAIL; + } + strncpy(peer_addr.sun_path, conn_info->path_name, sizeof(peer_addr.sun_path)); + + /* Send connection request */ + if (connect(conn_info->s, (struct sockaddr *)&peer_addr + , sizeof(struct sockaddr_un)) == -1) { + return IPC_FAIL; + } +#elif HB_IPC_METHOD == HB_IPC_STREAM + +#endif + + ch->ch_status = IPC_CONNECT; + ch->farside_pid = socket_get_farside_pid(conn_info->s); + return IPC_OK; +} + +static void +socket_set_high_flow_callback(IPC_Channel* ch, + flow_callback_t callback, + void* userdata) { + ch->high_flow_callback = callback; + ch->high_flow_userdata = userdata; +} + +static void +socket_set_low_flow_callback(IPC_Channel* ch, + flow_callback_t callback, + void* userdata) { + ch->low_flow_callback = callback; + ch->low_flow_userdata = userdata; +} + +static void +socket_check_flow_control(struct IPC_CHANNEL* ch, + int orig_qlen, + int curr_qlen) +{ + if (!IPC_ISRCONN(ch)) { + return; + } + + if (curr_qlen >= ch->high_flow_mark + && ch->high_flow_callback) { + ch->high_flow_callback(ch, ch->high_flow_userdata); + } + + if (curr_qlen <= ch->low_flow_mark + && orig_qlen > ch->low_flow_mark + && ch->low_flow_callback) { + ch->low_flow_callback(ch, ch->low_flow_userdata); + } +} + +static int +socket_send(struct IPC_CHANNEL * ch, struct IPC_MESSAGE* msg) +{ + int orig_qlen; + int diff; + struct IPC_MESSAGE* newmsg; + + if (msg->msg_len > MAXMSG) { + cl_log(LOG_ERR, "%s: sorry, cannot send messages " + "bigger than %d (requested %lu)", + __FUNCTION__, MAXMSG, (unsigned long)msg->msg_len); + return IPC_FAIL; + } + if (msg->msg_len < 0) { + cl_log(LOG_ERR, "socket_send: " + "invalid message"); + return IPC_FAIL; + } + + if (ch->ch_status != IPC_CONNECT) { + return IPC_FAIL; + } + + ch->ops->resume_io(ch); + + if (ch->send_queue->maxqlen_cnt && + time(NULL) - ch->send_queue->last_maxqlen_warn >= 60) { + cl_log(LOG_ERR, "%u messages dropped on a non-blocking channel (send queue maximum length %d)", + ch->send_queue->maxqlen_cnt, (int)ch->send_queue->max_qlen); + ch->send_queue->maxqlen_cnt = 0; + } + if ( !ch->should_send_block && + ch->send_queue->current_qlen >= ch->send_queue->max_qlen) { + if (!ch->send_queue->maxqlen_cnt) { + ch->send_queue->last_maxqlen_warn = time(NULL); + } + ch->send_queue->maxqlen_cnt++; + + if (ch->should_block_fail) { + return IPC_FAIL; + } else { + return IPC_OK; + } + } + + while (ch->send_queue->current_qlen >= ch->send_queue->max_qlen) { + if (ch->ch_status != IPC_CONNECT) { + cl_log(LOG_WARNING, "socket_send:" + " message queue exceeded and IPC not connected"); + return IPC_FAIL; + } + cl_shortsleep(); + ch->ops->resume_io(ch); + } + + /* add the message into the send queue */ + CHECKFOO(0,ch, msg, SavedQueuedBody, "queued message"); + SocketIPCStats.noutqueued++; + + diff = 0; + if (msg->msg_buf ) { + diff = (char*)msg->msg_body - (char*)msg->msg_buf; + } + if ( diff < (int)sizeof(struct SOCKET_MSG_HEAD) ) { + /* either we don't have msg->msg_buf set + * or we don't have enough bytes for socket head + * we delete this message and creates + * a new one and delete the old one + */ + + newmsg= socket_message_new(ch, msg->msg_len); + if (newmsg == NULL) { + cl_log(LOG_ERR, "socket_resume_io_write: " + "allocating memory for new ipc msg failed"); + return IPC_FAIL; + } + + memcpy(newmsg->msg_body, msg->msg_body, msg->msg_len); + + if(msg->msg_done) { + msg->msg_done(msg); + }; + msg = newmsg; + } +#ifdef IPC_TIME_DEBUG + ipc_time_debug(ch,msg, MSGPOS_ENQUEUE); +#endif + ch->send_queue->queue = g_list_append(ch->send_queue->queue, + msg); + orig_qlen = ch->send_queue->current_qlen++; + + socket_check_flow_control(ch, orig_qlen, orig_qlen +1 ); + + /* resume io */ + ch->ops->resume_io(ch); + return IPC_OK; +} + +static int +socket_recv(struct IPC_CHANNEL * ch, struct IPC_MESSAGE** message) +{ + GList *element; + + int nbytes; + int result; + + socket_resume_io(ch); + result = socket_resume_io_read(ch, &nbytes, TRUE); + + *message = NULL; + + if (ch->recv_queue->current_qlen == 0) { + return result != IPC_OK ? result : IPC_FAIL; + /*return IPC_OK;*/ + } + element = g_list_first(ch->recv_queue->queue); + + if (element == NULL) { + /* Internal accounting error, but correctable */ + cl_log(LOG_ERR + , "recv failure: qlen (%ld) > 0, but no message found." + , (long)ch->recv_queue->current_qlen); + ch->recv_queue->current_qlen = 0; + return IPC_FAIL; + } + *message = (struct IPC_MESSAGE *) (element->data); +#ifdef IPC_TIME_DEBUG + ipc_time_debug(ch, *message, MSGPOS_DEQUEUE); +#endif + + CHECKFOO(1,ch, *message, SavedReadBody, "read message"); + SocketIPCStats.nreceived++; + ch->recv_queue->queue = g_list_remove(ch->recv_queue->queue + , element->data); + ch->recv_queue->current_qlen--; + return IPC_OK; +} + +static int +socket_check_poll(struct IPC_CHANNEL * ch +, struct pollfd * sockpoll) +{ + if (ch->ch_status == IPC_DISCONNECT) { + return IPC_OK; + } + if (sockpoll->revents & POLLHUP) { + /* If input present, or this is an output-only poll... */ + if (sockpoll->revents & POLLIN + || (sockpoll-> events & POLLIN) == 0 ) { + ch->ch_status = IPC_DISC_PENDING; + return IPC_OK; + } +#if 1 + cl_log(LOG_INFO, "socket_check_poll(): HUP without input"); +#endif + ch->ch_status = IPC_DISCONNECT; + return IPC_BROKEN; + + } else if (sockpoll->revents & (POLLNVAL|POLLERR)) { + /* Have we already closed the socket? */ + if (fcntl(sockpoll->fd, F_GETFL) < 0) { + cl_perror("socket_check_poll(pid %d): bad fd [%d]" + , (int) getpid(), sockpoll->fd); + ch->ch_status = IPC_DISCONNECT; + return IPC_OK; + } + cl_log(LOG_ERR + , "revents failure: fd %d, flags 0x%x" + , sockpoll->fd, sockpoll->revents); + errno = EINVAL; + return IPC_FAIL; + } + return IPC_OK; +} + +static int +socket_waitfor(struct IPC_CHANNEL * ch +, gboolean (*finished)(struct IPC_CHANNEL * ch)) +{ + struct pollfd sockpoll; + + CHANAUDIT(ch); + if (finished(ch)) { + return IPC_OK; + } + + if (ch->ch_status == IPC_DISCONNECT) { + return IPC_BROKEN; + } + sockpoll.fd = ch->ops->get_recv_select_fd(ch); + + while (!finished(ch) && IPC_ISRCONN(ch)) { + int rc; + + sockpoll.events = POLLIN; + + /* Cannot call resume_io after the call to finished() + * and before the call to poll because we might + * change the state of the thing finished() is + * waiting for. + * This means that the poll call below would be + * not only pointless, but might + * make us hang forever waiting for this + * event which has already happened + */ + if (ch->send_queue->current_qlen > 0) { + sockpoll.events |= POLLOUT; + } + + rc = ipc_pollfunc_ptr(&sockpoll, 1, -1); + + if (rc < 0) { + return (errno == EINTR ? IPC_INTR : IPC_FAIL); + } + + rc = socket_check_poll(ch, &sockpoll); + if (sockpoll.revents & POLLIN) { + socket_resume_io(ch); + } + if (rc != IPC_OK) { + CHANAUDIT(ch); + return rc; + } + } + + CHANAUDIT(ch); + return IPC_OK; +} + +static int +socket_waitin(struct IPC_CHANNEL * ch) +{ + return socket_waitfor(ch, ch->ops->is_message_pending); +} +static gboolean +socket_is_output_flushed(struct IPC_CHANNEL * ch) +{ + return ! ch->ops->is_sending_blocked(ch); +} + +static int +socket_waitout(struct IPC_CHANNEL * ch) +{ + int rc; + CHANAUDIT(ch); + rc = socket_waitfor(ch, socket_is_output_flushed); + + if (rc != IPC_OK) { + cl_log(LOG_ERR + , "socket_waitout failure: rc = %d", rc); + } else if (ch->ops->is_sending_blocked(ch)) { + cl_log(LOG_ERR, "socket_waitout output still blocked"); + } + CHANAUDIT(ch); + return rc; +} + +static gboolean +socket_is_message_pending(struct IPC_CHANNEL * ch) +{ + int nbytes; + + socket_resume_io_read(ch, &nbytes, TRUE); + ch->ops->resume_io(ch); + if (ch->recv_queue->current_qlen > 0) { + return TRUE; + } + + return !IPC_ISRCONN(ch); +} + +static gboolean +socket_is_output_pending(struct IPC_CHANNEL * ch) +{ + socket_resume_io(ch); + return ch->ch_status == IPC_CONNECT + && ch->send_queue->current_qlen > 0; +} + +static gboolean +socket_is_sendq_full(struct IPC_CHANNEL * ch) +{ + ch->ops->resume_io(ch); + return(ch->send_queue->current_qlen == ch->send_queue->max_qlen); +} + +static gboolean +socket_is_recvq_full(struct IPC_CHANNEL * ch) +{ + ch->ops->resume_io(ch); + return(ch->recv_queue->current_qlen == ch->recv_queue->max_qlen); +} + +static int +socket_get_conntype(struct IPC_CHANNEL* ch) +{ + return ch->conntype; +} + +static int +socket_assert_auth(struct IPC_CHANNEL *ch, GHashTable *auth) +{ + cl_log(LOG_ERR + , "the assert_auth function for domain socket is not implemented"); + return IPC_FAIL; +} + +static int +socket_resume_io_read(struct IPC_CHANNEL *ch, int* nbytes, gboolean read1anyway) +{ + struct SOCKET_CH_PRIVATE* conn_info; + int retcode = IPC_OK; + struct pollfd sockpoll; + int debug_loopcount = 0; + int debug_bytecount = 0; + size_t maxqlen = ch->recv_queue->max_qlen; + struct ipc_bufpool* pool = ch->pool; + int nmsgs = 0; + int spaceneeded; + *nbytes = 0; + + CHANAUDIT(ch); + conn_info = (struct SOCKET_CH_PRIVATE *) ch->ch_private; + + if (ch->ch_status == IPC_DISCONNECT) { + return IPC_BROKEN; + } + + if (pool == NULL) { + ch->pool = pool = ipc_bufpool_new(0); + if (pool == NULL) { + cl_log(LOG_ERR, "socket_resume_io_read: " + "memory allocation for ipc pool failed"); + return IPC_FAIL; + } + } + + if (ipc_bufpool_full(pool, ch, &spaceneeded)) { + struct ipc_bufpool* newpool; + + newpool = ipc_bufpool_new(spaceneeded); + if (newpool == NULL) { + cl_log(LOG_ERR, "socket_resume_io_read: " + "memory allocation for a new ipc pool failed"); + return IPC_FAIL; + } + + ipc_bufpool_partial_copy(newpool, pool); + ipc_bufpool_unref(pool); + ch->pool = pool = newpool; + } + if (maxqlen <= 0 && read1anyway) { + maxqlen = 1; + } + if (ch->recv_queue->current_qlen < maxqlen && retcode == IPC_OK) { + void * msg_begin; + int msg_len; + int len; +#if HB_IPC_METHOD == HB_IPC_STREAM + struct strbuf d; + int flags, rc; +#endif + + CHANAUDIT(ch); + ++debug_loopcount; + + len = ipc_bufpool_spaceleft(pool); + msg_begin = pool->currpos; + + CHANAUDIT(ch); + + /* Now try to receive some data */ + +#if HB_IPC_METHOD == HB_IPC_SOCKET + msg_len = recv(conn_info->s, msg_begin, len, MSG_DONTWAIT); +#elif HB_IPC_METHOD == HB_IPC_STREAM + d.maxlen = len; + d.len = 0; + d.buf = msg_begin; + flags = 0; + rc = getmsg(conn_info->s, NULL, &d, &flags); + msg_len = (rc < 0) ? rc : d.len; +#endif + SocketIPCStats.last_recv_rc = msg_len; + SocketIPCStats.last_recv_errno = errno; + ++SocketIPCStats.recv_count; + + /* Did we get an error? */ + if (msg_len < 0) { + switch (errno) { + case EAGAIN: + if (ch->ch_status==IPC_DISC_PENDING) { + ch->ch_status =IPC_DISCONNECT; + retcode = IPC_BROKEN; + } + break; + + case ECONNREFUSED: + case ECONNRESET: + ch->ch_status = IPC_DISC_PENDING; + retcode= socket_check_disc_pending(ch); + break; + + default: + cl_perror("socket_resume_io_read" + ": unknown recv error, peerpid=%d", + ch->farside_pid); + ch->ch_status = IPC_DISCONNECT; + retcode = IPC_FAIL; + break; + } + + } else if (msg_len == 0) { + ch->ch_status = IPC_DISC_PENDING; + if(ch->recv_queue->current_qlen <= 0) { + ch->ch_status = IPC_DISCONNECT; + retcode = IPC_FAIL; + } + } else { + /* We read something! */ + /* Note that all previous cases break out of the loop */ + debug_bytecount += msg_len; + *nbytes = msg_len; + nmsgs = ipc_bufpool_update(pool, ch, msg_len, ch->recv_queue) ; + + if (nmsgs < 0) { + /* we didn't like the other side */ + cl_log(LOG_ERR, "socket_resume_io_read: " + "disconnecting the other side"); + ch->ch_status = IPC_DISCONNECT; + retcode = IPC_FAIL; + } else { + SocketIPCStats.ninqueued += nmsgs; + } + } + } + + /* Check for errors uncaught by recv() */ + /* NOTE: It doesn't seem right we have to do this every time */ + /* FIXME?? */ + + memset(&sockpoll,0, sizeof(struct pollfd)); + if ((retcode == IPC_OK) + && (sockpoll.fd = conn_info->s) >= 0) { + /* Just check for errors, not for data */ + sockpoll.events = 0; + ipc_pollfunc_ptr(&sockpoll, 1, 0); + retcode = socket_check_poll(ch, &sockpoll); + } + + CHANAUDIT(ch); + if (retcode != IPC_OK) { + return retcode; + } + + return IPC_ISRCONN(ch) ? IPC_OK : IPC_BROKEN; +} + +static int +socket_resume_io_write(struct IPC_CHANNEL *ch, int* nmsg) +{ + int retcode = IPC_OK; + struct SOCKET_CH_PRIVATE* conn_info; + + CHANAUDIT(ch); + *nmsg = 0; + conn_info = (struct SOCKET_CH_PRIVATE *) ch->ch_private; + + while (ch->ch_status == IPC_CONNECT + && retcode == IPC_OK + && ch->send_queue->current_qlen > 0) { + + GList * element; + struct IPC_MESSAGE * msg; + struct SOCKET_MSG_HEAD head; + struct IPC_MESSAGE* oldmsg = NULL; + int sendrc = 0; + struct IPC_MESSAGE* newmsg; + char* p; + unsigned int bytes_remaining; + int diff; + + CHANAUDIT(ch); + element = g_list_first(ch->send_queue->queue); + if (element == NULL) { + /* OOPS! - correct consistency problem */ + ch->send_queue->current_qlen = 0; + break; + } + msg = (struct IPC_MESSAGE *) (element->data); + + diff = 0; + if (msg->msg_buf ) { + diff = (char*)msg->msg_body - (char*)msg->msg_buf; + } + if ( diff < (int)sizeof(struct SOCKET_MSG_HEAD) ) { + /* either we don't have msg->msg_buf set + * or we don't have enough bytes for socket head + * we delete this message and creates + * a new one and delete the old one + */ + + newmsg= socket_message_new(ch, msg->msg_len); + if (newmsg == NULL) { + cl_log(LOG_ERR, "socket_resume_io_write: " + "allocating memory for new ipc msg failed"); + return IPC_FAIL; + } + + memcpy(newmsg->msg_body, msg->msg_body, msg->msg_len); + oldmsg = msg; + msg = newmsg; + } + + head.msg_len = msg->msg_len; + head.magic = HEADMAGIC; + memcpy(msg->msg_buf, &head, sizeof(struct SOCKET_MSG_HEAD)); + + if (ch->bytes_remaining == 0) { + /*we start to send a new message*/ +#ifdef IPC_TIME_DEBUG + ipc_time_debug(ch, msg, MSGPOS_SEND); +#endif + bytes_remaining = msg->msg_len + ch->msgpad; + p = msg->msg_buf; + } else { + bytes_remaining = ch->bytes_remaining; + p = ((char*)msg->msg_buf) + msg->msg_len + ch->msgpad + - bytes_remaining; + + } + + sendrc = 0; + + do { +#if HB_IPC_METHOD == HB_IPC_STREAM + struct strbuf d; + int msglen, putmsgrc; +#endif + + CHANAUDIT(ch); + +#if HB_IPC_METHOD == HB_IPC_SOCKET + sendrc = send(conn_info->s, p + , bytes_remaining, (MSG_DONTWAIT|MSG_NOSIGNAL)); +#elif HB_IPC_METHOD == HB_IPC_STREAM + d.maxlen = 0; + d.len = msglen = bytes_remaining; + d.buf = p; + putmsgrc = putmsg(conn_info->s, NULL, &d, 0); + sendrc = putmsgrc == 0 ? msglen : -1; +#endif + SocketIPCStats.last_send_rc = sendrc; + SocketIPCStats.last_send_errno = errno; + ++SocketIPCStats.send_count; + + if (sendrc <= 0) { + break; + } else { + p = p + sendrc; + bytes_remaining -= sendrc; + } + + } while(bytes_remaining > 0 ); + + ch->bytes_remaining = bytes_remaining; + + if (sendrc < 0) { + switch (errno) { + case EAGAIN: + retcode = IPC_OK; + break; + case EPIPE: + ch->ch_status = IPC_DISC_PENDING; + socket_check_disc_pending(ch); + retcode = IPC_BROKEN; + break; + default: + cl_perror("socket_resume_io_write" + ": send2 bad errno"); + ch->ch_status = IPC_DISCONNECT; + retcode = IPC_FAIL; + break; + } + break; + } else { + int orig_qlen; + + CHECKFOO(3,ch, msg, SavedSentBody, "sent message") + + if (oldmsg) { + if (msg->msg_done != NULL) { + msg->msg_done(msg); + } + msg=oldmsg; + } + + if(ch->bytes_remaining ==0) { + ch->send_queue->queue = g_list_remove(ch->send_queue->queue, msg); + if (msg->msg_done != NULL) { + msg->msg_done(msg); + } + + SocketIPCStats.nsent++; + orig_qlen = ch->send_queue->current_qlen--; + socket_check_flow_control(ch, orig_qlen, orig_qlen -1 ); + (*nmsg)++; + } + } + } + CHANAUDIT(ch); + if (retcode != IPC_OK) { + return retcode; + } + return IPC_ISRCONN(ch) ? IPC_OK : IPC_BROKEN; +} + +static int +socket_resume_io(struct IPC_CHANNEL *ch) +{ + int rc1 = IPC_OK; + int rc2 = IPC_OK; + int nwmsg = 1; + int nbytes_r = 1; + gboolean OKonce = FALSE; + + CHANAUDIT(ch); + if (!IPC_ISRCONN(ch)) { + return IPC_BROKEN; + } + + do { + if (nbytes_r > 0) { + rc1 = socket_resume_io_read(ch, &nbytes_r, FALSE); + } + if (nwmsg > 0) { + nwmsg = 0; + rc2 = socket_resume_io_write(ch, &nwmsg); + } + if (rc1 == IPC_OK || rc2 == IPC_OK) { + OKonce = TRUE; + } + } while ((nbytes_r > 0 || nwmsg > 0) && IPC_ISRCONN(ch)); + + if (IPC_ISRCONN(ch)) { + if (rc1 != IPC_OK) { + cl_log(LOG_ERR + , "socket_resume_io_read() failure"); + } + if (rc2 != IPC_OK && IPC_CONNECT == ch->ch_status) { + cl_log(LOG_ERR + , "socket_resume_io_write() failure"); + } + } else { + return (OKonce ? IPC_OK : IPC_BROKEN); + } + + return (rc1 != IPC_OK ? rc1 : rc2); +} + +static int +socket_get_recv_fd(struct IPC_CHANNEL *ch) +{ + struct SOCKET_CH_PRIVATE* chp = ch ->ch_private; + + return (chp == NULL ? -1 : chp->s); +} + +static int +socket_get_send_fd(struct IPC_CHANNEL *ch) +{ + return socket_get_recv_fd(ch); +} + +static void +socket_adjust_buf(struct IPC_CHANNEL *ch, int optname, unsigned q_len) +{ + const char *direction = optname == SO_SNDBUF ? "snd" : "rcv"; + int fd = socket_get_send_fd(ch); + unsigned byte; + + /* Arbitrary scaling. + * DEFAULT_MAX_QLEN is 64, default socket buf is often 64k to 128k, + * at least on those linux I checked. + * Keep that ratio, and allow for some overhead. */ + if (q_len == 0) + /* client does not want anything, + * reduce system buffers as well */ + byte = 4096; + else if (q_len < 512) + byte = (32 + q_len) * 1024; + else + byte = q_len * 1024; + + if (0 == setsockopt(fd, SOL_SOCKET, optname, &byte, sizeof(byte))) { + if (debug_level > 1) { + cl_log(LOG_DEBUG, "adjusted %sbuf size to %u", + direction, byte); + } + } else { + /* If this fails, you may need to adjust net.core.rmem_max, + * ...wmem_max, or equivalent */ + cl_log(LOG_WARNING, "adjust %sbuf size to %u failed: %s", + direction, byte, strerror(errno)); + } +} + +static int +socket_set_send_qlen (struct IPC_CHANNEL* ch, int q_len) +{ + /* This seems more like an assertion failure than a normal error */ + if (ch->send_queue == NULL) { + return IPC_FAIL; + } + socket_adjust_buf(ch, SO_SNDBUF, q_len); + ch->send_queue->max_qlen = q_len; + return IPC_OK; +} + +static int +socket_set_recv_qlen (struct IPC_CHANNEL* ch, int q_len) +{ + /* This seems more like an assertion failure than a normal error */ + if (ch->recv_queue == NULL) { + return IPC_FAIL; + } + socket_adjust_buf(ch, SO_RCVBUF, q_len); + ch->recv_queue->max_qlen = q_len; + return IPC_OK; +} + +static int ipcmsg_count_allocated = 0; +static int ipcmsg_count_freed = 0; +void socket_ipcmsg_dump_stats(void); +void +socket_ipcmsg_dump_stats(void) { + cl_log(LOG_INFO, "ipcsocket ipcmsg allocated=%d, freed=%d, diff=%d", + ipcmsg_count_allocated, + ipcmsg_count_freed, + ipcmsg_count_allocated - ipcmsg_count_freed); +} + +static void +socket_del_ipcmsg(IPC_Message* m) +{ + if (m == NULL) { + cl_log(LOG_ERR, "socket_del_ipcmsg:" + "msg is NULL"); + return; + } + + if (m->msg_body) { + memset(m->msg_body, 0, m->msg_len); + } + if (m->msg_buf) { + g_free(m->msg_buf); + } + + memset(m, 0, sizeof(*m)); + g_free(m); + + ipcmsg_count_freed ++; +} + +static IPC_Message* +socket_new_ipcmsg(IPC_Channel* ch, const void* data, int len, void* private) +{ + IPC_Message* hdr; + + if (ch == NULL || len < 0) { + cl_log(LOG_ERR, "socket_new_ipcmsg:" + " invalid parameter"); + return NULL; + } + + if (ch->msgpad > MAX_MSGPAD) { + cl_log(LOG_ERR, "socket_new_ipcmsg: too many pads " + "something is wrong"); + return NULL; + } + + hdr = ipcmsg_new(ch, data, len, private, socket_del_ipcmsg); + + if (hdr) ipcmsg_count_allocated ++; + + return hdr; +} + +static +struct IPC_MESSAGE * +ipcmsg_new(struct IPC_CHANNEL * ch, const void* data, int len, void* private, + DelProc delproc) +{ + struct IPC_MESSAGE * hdr; + char* copy = NULL; + char* buf; + char* body; + + if ((hdr = g_new(struct IPC_MESSAGE, 1)) == NULL) { + return NULL; + } + memset(hdr, 0, sizeof(*hdr)); + + if (len > 0) { + if ((copy = (char*)g_malloc(ch->msgpad + len)) == NULL) { + g_free(hdr); + return NULL; + } + if (data) { + memcpy(copy + ch->msgpad, data, len); + } + buf = copy; + body = copy + ch->msgpad;; + } else { + len = 0; + buf = body = NULL; + } + hdr->msg_len = len; + hdr->msg_buf = buf; + hdr->msg_body = body; + hdr->msg_ch = ch; + hdr->msg_done = delproc; + hdr->msg_private = private; + + return hdr; +} + +static int +socket_get_chan_status(IPC_Channel* ch) +{ + socket_resume_io(ch); + return ch->ch_status; +} + +/* socket object of the function table */ +static struct IPC_WAIT_OPS socket_wait_ops = { + socket_destroy_wait_conn, + socket_wait_selectfd, + socket_accept_connection, +}; + +/* + * create a new ipc queue whose length = 0 and inner queue = NULL. + * return the pointer to a new ipc queue or NULL is the queue can't be created. + */ + +static struct IPC_QUEUE* +socket_queue_new(void) +{ + struct IPC_QUEUE *temp_queue; + + /* temp queue with length = 0 and inner queue = NULL. */ + temp_queue = g_new(struct IPC_QUEUE, 1); + temp_queue->current_qlen = 0; + temp_queue->max_qlen = DEFAULT_MAX_QLEN; + temp_queue->queue = NULL; + temp_queue->last_maxqlen_warn = 0; + temp_queue->maxqlen_cnt = 0; + return temp_queue; +} + +/* + * socket_wait_conn_new: + * Called by ipc_wait_conn_constructor to get a new socket + * waiting connection. + * (better explanation of this role might be nice) + * + * Parameters : + * ch_attrs (IN) the attributes used to create this connection. + * + * Return : + * the pointer to the new waiting connection or NULL if the connection + * can't be created. + * + * NOTE : + * for domain socket implementation, the only attribute needed is path name. + * so the user should + * create the hash table like this: + * GHashTable * attrs; + * attrs = g_hash_table_new(g_str_hash, g_str_equal); + * g_hash_table_insert(attrs, PATH_ATTR, path_name); + * here PATH_ATTR is defined as "path". + * + * NOTE : + * The streams implementation uses "Streams Programming Guide", Solaris 8, + * as its guide (sample code near end of "Configuration" chapter 11). + */ +struct IPC_WAIT_CONNECTION * +socket_wait_conn_new(GHashTable *ch_attrs) +{ + struct IPC_WAIT_CONNECTION * temp_ch; + char *path_name; + char *mode_attr; + int s, flags; + struct SOCKET_WAIT_CONN_PRIVATE *wait_private; + mode_t s_mode; +#if HB_IPC_METHOD == HB_IPC_SOCKET + struct sockaddr_un my_addr; +#elif HB_IPC_METHOD == HB_IPC_STREAM + int pipefds[2]; +#endif + + path_name = (char *) g_hash_table_lookup(ch_attrs, IPC_PATH_ATTR); + mode_attr = (char *) g_hash_table_lookup(ch_attrs, IPC_MODE_ATTR); + + if (mode_attr != NULL) { + s_mode = (mode_t)strtoul((const char *)mode_attr, NULL, 8); + } else { + s_mode = 0777; + } + if (path_name == NULL) { + return NULL; + } + +#if HB_IPC_METHOD == HB_IPC_SOCKET + /* prepare the unix domain socket */ + if ((s = socket(AF_LOCAL, SOCK_STREAM, 0)) == -1) { + cl_perror("socket_wait_conn_new: socket() failure"); + return NULL; + } + + if (unlink(path_name) < 0 && errno != ENOENT) { + cl_perror("socket_wait_conn_new: unlink failure(%s)", + path_name); + } + memset(&my_addr, 0, sizeof(my_addr)); + my_addr.sun_family = AF_LOCAL; /* host byte order */ + + if (strlen(path_name) >= sizeof(my_addr.sun_path)) { + close(s); + return NULL; + } + + strncpy(my_addr.sun_path, path_name, sizeof(my_addr.sun_path)); + + if (bind(s, (struct sockaddr *)&my_addr, sizeof(my_addr)) == -1) { + cl_perror("socket_wait_conn_new: trying to create in %s bind:" + , path_name); + close(s); + return NULL; + } +#elif HB_IPC_METHOD == HB_IPC_STREAM + /* Set up the communication channel the clients will use to us (server) */ + if (pipe(pipefds) == -1) { + cl_perror("pipe() failure"); + return NULL; + } + + /* Let clients have unique connections to us */ + if (ioctl(pipefds[1], I_PUSH, "connld") == -1) { + cl_perror("ioctl(%d, I_PUSH, \"connld\") failure", pipefds[1]); + return NULL; + } + + if (unlink(path_name) < 0 && errno != ENOENT) { + cl_perror("socket_wait_conn_new: unlink failure(%s)", + path_name); + } + + if (mkfifo(path_name, s_mode) == -1) { + cl_perror("socket_wait_conn_new: mkfifo(%s, ...) failure", path_name); + return NULL; + } + + if (fattach(pipefds[1], path_name) == -1) { + cl_perror("socket_wait_conn_new: fattach(..., %s) failure", path_name); + return NULL; + } + + /* the pseudo-socket is the other part of the pipe */ + s = pipefds[0]; +#endif + + /* Change the permission of the socket */ + if (chmod(path_name,s_mode) < 0) { + cl_perror("socket_wait_conn_new: failure trying to chmod %s" + , path_name); + close(s); + return NULL; + } + +#if HB_IPC_METHOD == HB_IPC_SOCKET + /* listen to the socket */ + if (listen(s, MAX_LISTEN_NUM) == -1) { + cl_perror("socket_wait_conn_new: listen(MAX_LISTEN_NUM)"); + close(s); + return NULL; + } +#elif HB_IPC_METHOD == HB_IPC_STREAM + +#endif + + flags = fcntl(s, F_GETFL); + if (flags == -1) { + cl_perror("socket_wait_conn_new: cannot read file descriptor flags"); + close(s); + return NULL; + } + flags |= O_NONBLOCK; + if (fcntl(s, F_SETFL, flags) < 0) { + cl_perror("socket_wait_conn_new: cannot set O_NONBLOCK"); + close(s); + return NULL; + } + + wait_private = g_new(struct SOCKET_WAIT_CONN_PRIVATE, 1); +#if HB_IPC_METHOD == HB_IPC_SOCKET + wait_private->s = s; +#elif HB_IPC_METHOD == HB_IPC_STREAM + wait_private->pipefds[0] = pipefds[0]; + wait_private->pipefds[1] = pipefds[1]; +#endif + strncpy(wait_private->path_name, path_name, sizeof(wait_private->path_name)); + temp_ch = g_new(struct IPC_WAIT_CONNECTION, 1); + temp_ch->ch_private = (void *) wait_private; + temp_ch->ch_status = IPC_WAIT; + temp_ch->ops = (struct IPC_WAIT_OPS *)&socket_wait_ops; + + return temp_ch; +} + +/* + * will be called by ipc_channel_constructor to create a new socket channel. + * parameters : + * attrs (IN) the hash table of the attributes used to create this channel. + * + * return: + * the pointer to the new waiting channel or NULL if the channel can't be created. +*/ + +struct IPC_CHANNEL * +socket_client_channel_new(GHashTable *ch_attrs) { + char *path_name; + int sockfd; + + /* + * I don't really understand why the client and the server use different + * parameter names... + * + * It's a really bad idea to store both integers and strings + * in the same table. + * + * Maybe we need an internal function with a different set of parameters? + */ + + /* + * if we want to seperate them. I suggest + * <client side> + * user call ipc_channel_constructor(ch_type,attrs) to create a new channel. + * ipc_channel_constructor() call socket_channel_new(GHashTable*)to + * create a new socket channel. + * <server side> + * wait_conn->accept_connection() will call another function to create a + * new channel. This function will take socketfd as the parameter to + * create a socket channel. + */ + + path_name = (char *) g_hash_table_lookup(ch_attrs, IPC_PATH_ATTR); + if (path_name == NULL) { + return NULL; + } + +#if HB_IPC_METHOD == HB_IPC_SOCKET + /* prepare the socket */ + if ((sockfd = socket(AF_LOCAL, SOCK_STREAM, 0)) == -1) { + cl_perror("socket_client_channel_new: socket"); + return NULL; + } +#elif HB_IPC_METHOD == HB_IPC_STREAM + sockfd = open(path_name, O_RDWR|O_NONBLOCK); + if (sockfd == -1) { + cl_perror("socket_client_channel_new: open(%s, ...) failure", path_name); + return NULL; + } +#endif + + if (client_channel_new_auth(sockfd) < 0) { + close(sockfd); + return NULL; + } + return channel_new(sockfd, IPC_CLIENT, path_name); +} + +static +int client_channel_new_auth(int sockfd) { +#ifdef USE_BINDSTAT_CREDS + char rand_id[16]; + char uuid_str_tmp[40]; + struct sockaddr_un sock_addr; + + /* Prepare the socket */ + memset(&sock_addr, 0, sizeof(sock_addr)); + sock_addr.sun_family = AF_UNIX; + + /* make sure socket paths never clash */ + uuid_generate(rand_id); + uuid_unparse(rand_id, uuid_str_tmp); + + snprintf(sock_addr.sun_path, sizeof(sock_addr.sun_path), + "%s/%s", HA_VARLIBHBDIR, uuid_str_tmp); + + unlink(sock_addr.sun_path); + if(bind(sockfd, (struct sockaddr*)&sock_addr, SUN_LEN(&sock_addr)) < 0) { + perror("Client bind() failure"); + return 0; + } +#endif + + return 0; +} + +static +struct IPC_CHANNEL * +socket_server_channel_new(int sockfd) { + return channel_new(sockfd, IPC_SERVER, "?"); +} + +static +struct IPC_CHANNEL * +channel_new(int sockfd, int conntype, const char *path_name) { + struct IPC_CHANNEL * temp_ch; + struct SOCKET_CH_PRIVATE* conn_info; + int flags; + + if (path_name == NULL || strlen(path_name) >= sizeof(conn_info->path_name)) { + return NULL; + } + + temp_ch = g_new(struct IPC_CHANNEL, 1); + if (temp_ch == NULL) { + cl_log(LOG_ERR, "channel_new: allocating memory for channel failed"); + return NULL; + } + memset(temp_ch, 0, sizeof(struct IPC_CHANNEL)); + + conn_info = g_new(struct SOCKET_CH_PRIVATE, 1); + + flags = fcntl(sockfd, F_GETFL); + if (flags == -1) { + cl_perror("channel_new: cannot read file descriptor flags"); + g_free(conn_info); conn_info = NULL; + g_free(temp_ch); + if (conntype == IPC_CLIENT) close(sockfd); + return NULL; + } + flags |= O_NONBLOCK; + if (fcntl(sockfd, F_SETFL, flags) < 0) { + cl_perror("channel_new: cannot set O_NONBLOCK"); + g_free(conn_info); conn_info = NULL; + g_free(temp_ch); + if (conntype == IPC_CLIENT) close(sockfd); + return NULL; + } + + conn_info->s = sockfd; + conn_info->remaining_data = 0; + conn_info->buf_msg = NULL; +#if HB_IPC_METHOD == HB_IPC_SOCKET + conn_info->peer_addr = NULL; +#endif + strncpy(conn_info->path_name, path_name, sizeof(conn_info->path_name)); + +#ifdef DEBUG + cl_log(LOG_INFO, "Initializing socket %d to DISCONNECT", sockfd); +#endif + temp_ch->ch_status = IPC_DISCONNECT; + temp_ch->ch_private = (void*) conn_info; + temp_ch->ops = (struct IPC_OPS *)&socket_ops; + temp_ch->msgpad = sizeof(struct SOCKET_MSG_HEAD); + temp_ch->bytes_remaining = 0; + temp_ch->should_send_block = FALSE; + temp_ch->should_block_fail = TRUE; + temp_ch->send_queue = socket_queue_new(); + temp_ch->recv_queue = socket_queue_new(); + temp_ch->pool = NULL; + temp_ch->high_flow_mark = temp_ch->send_queue->max_qlen; + temp_ch->low_flow_mark = -1; + temp_ch->conntype = conntype; + temp_ch->refcount = 0; + temp_ch->farside_uid = -1; + temp_ch->farside_gid = -1; + + return temp_ch; +} + +/* + * Create a new pair of pre-connected IPC channels similar to + * the result of pipe(2), or socketpair(2). + */ + +int +ipc_channel_pair(IPC_Channel* channels[2]) +{ + int sockets[2]; + int rc; + int j; + const char *pname; + +#if HB_IPC_METHOD == HB_IPC_SOCKET + pname = "[socketpair]"; + + if ((rc = socketpair(AF_LOCAL, SOCK_STREAM, 0, sockets)) < 0) { + return IPC_FAIL; + } +#elif HB_IPC_METHOD == HB_IPC_STREAM + pname = "[pipe]"; + + if ((rc = pipe(sockets)) < 0) { + return IPC_FAIL; + } + rc = 0; + for (j=0; j < 2; ++j) { + if (fcntl(sockets[j], F_SETFL, O_NONBLOCK) < 0) { + cl_perror("ipc_channel_pair: cannot set O_NONBLOCK"); + rc = -1; + } + } + if (rc < 0) { + close(sockets[0]); + close(sockets[1]); + return IPC_FAIL; + } +#endif + + if ((channels[0] = socket_server_channel_new(sockets[0])) == NULL) { + close(sockets[0]); + close(sockets[1]); + return IPC_FAIL; + } + if ((channels[1] = socket_server_channel_new(sockets[1])) == NULL) { + close(sockets[0]); + close(sockets[1]); + channels[0]->ops->destroy(channels[0]); + return IPC_FAIL; + } + for (j=0; j < 2; ++j) { + struct SOCKET_CH_PRIVATE* p = channels[j]->ch_private; + channels[j]->ch_status = IPC_CONNECT; + channels[j]->conntype = IPC_PEER; + /* Valid, but not terribly meaningful */ + channels[j]->farside_pid = getpid(); + strncpy(p->path_name, pname, sizeof(p->path_name)); + } + + return IPC_OK; +} + +/* brief free the memory space allocated to msg and destroy msg. */ + +static void +socket_free_message(struct IPC_MESSAGE * msg) { +#if 0 + memset(msg->msg_body, 0xff, msg->msg_len); +#endif + if (msg->msg_buf) { + g_free(msg->msg_buf); + } else { + g_free(msg->msg_body); + } +#if 0 + memset(msg, 0xff, sizeof(*msg)); +#endif + g_free((void *)msg); +} + +/* + * create a new ipc message whose msg_body's length is msg_len. + * + * parameters : + * msg_len (IN) the length of this message body in this message. + * + * return : + * the pointer to the new message or NULL if the message can't be created. + */ + +static struct IPC_MESSAGE* +socket_message_new(struct IPC_CHANNEL *ch, int msg_len) +{ + return ipcmsg_new(ch, NULL, msg_len, NULL, socket_free_message); +} + +/*********************************************************************** + * + * IPC authentication schemes... More machine dependent than + * we'd like, but don't know any better way... + * + ***********************************************************************/ + +static int +verify_creds(struct IPC_AUTH *auth_info, uid_t uid, gid_t gid) +{ + int ret = IPC_FAIL; + + if (!auth_info || (!auth_info->uid && !auth_info->gid)) { + return IPC_OK; + } + if ( auth_info->uid + && (g_hash_table_lookup(auth_info->uid + , GUINT_TO_POINTER((guint)uid)) != NULL)) { + ret = IPC_OK; + } else if (auth_info->gid + && (g_hash_table_lookup(auth_info->gid + , GUINT_TO_POINTER((guint)gid)) != NULL)) { + ret = IPC_OK; + } + return ret; +} + +/*********************************************************************** + * SO_PEERCRED VERSION... (Linux) + ***********************************************************************/ + +#ifdef USE_SO_PEERCRED +/* verify the authentication information. */ +static int +socket_verify_auth(struct IPC_CHANNEL* ch, struct IPC_AUTH * auth_info) +{ + struct SOCKET_CH_PRIVATE * conn_info; + int ret = IPC_FAIL; + struct ucred cred; + socklen_t n = sizeof(cred); + + if (ch == NULL || ch->ch_private == NULL) { + return IPC_FAIL; + } + if (auth_info == NULL + || (auth_info->uid == NULL && auth_info->gid == NULL)) { + ret = IPC_OK; /* no restriction for authentication */ + } + + /* Get the credential information for our peer */ + conn_info = (struct SOCKET_CH_PRIVATE *) ch->ch_private; + if (getsockopt(conn_info->s, SOL_SOCKET, SO_PEERCRED, &cred, &n) != 0 + || (size_t)n != sizeof(cred)) { + return ret; + } + + ch->farside_uid = cred.uid; + ch->farside_gid = cred.gid; + if (ret == IPC_OK) { + return ret; + } +#if 0 + cl_log(LOG_DEBUG, "SO_PEERCRED returned [%d, (%ld:%ld)]" + , cred.pid, (long)cred.uid, (long)cred.uid); + cl_log(LOG_DEBUG, "Verifying authentication: cred.uid=%d cred.gid=%d" + , cred.uid, cred.gid); + cl_log(LOG_DEBUG, "Verifying authentication: uidptr=0x%lx gidptr=0x%lx" + , (unsigned long)auth_info->uid + , (unsigned long)auth_info->gid); +#endif + /* verify the credential information. */ + return verify_creds(auth_info, cred.uid, cred.gid); +} + +/* get farside pid for our peer process */ + +static +pid_t +socket_get_farside_pid(int sockfd) +{ + socklen_t n; + struct ucred *cred; + pid_t f_pid; + + /* Get the credential information from peer */ + n = sizeof(struct ucred); + cred = g_new(struct ucred, 1); + if (getsockopt(sockfd, SOL_SOCKET, SO_PEERCRED, cred, &n) != 0) { + g_free(cred); + return -1; + } + + f_pid = cred->pid; + g_free(cred); + return f_pid; +} +#endif /* SO_PEERCRED version */ + +#ifdef USE_GETPEEREID +/* + * This is implemented in OpenBSD and FreeBSD. + * + * It's not a half-bad interface... + * + * This should probably be our standard way of doing it, and put it + * as a replacement library. That would simplify things... + */ + +static int +socket_verify_auth(struct IPC_CHANNEL* ch, struct IPC_AUTH * auth_info) +{ + struct SOCKET_CH_PRIVATE *conn_info; + uid_t euid; + gid_t egid; + int ret = IPC_FAIL; + + if (auth_info == NULL + || (auth_info->uid == NULL && auth_info->gid == NULL)) { + ret = IPC_OK; /* no restriction for authentication */ + } + conn_info = (struct SOCKET_CH_PRIVATE *) ch->ch_private; + + if (getpeereid(conn_info->s, &euid, &egid) < 0) { + cl_perror("getpeereid() failure"); + return ret; + } + + ch->farside_uid = euid; + ch->farside_gid = egid; + + /* verify the credential information. */ + return verify_creds(auth_info, euid, egid); +} + +static +pid_t +socket_get_farside_pid(int sock) +{ + return -1; +} +#endif /* USE_GETPEEREID */ + +/*********************************************************************** + * SCM_CREDS VERSION... (*BSD systems) + ***********************************************************************/ +#ifdef USE_SCM_CREDS +/* FIXME! Need to implement SCM_CREDS mechanism for BSD-based systems + * This isn't an emergency, but should be done in the future... + * Hint: * Postgresql does both types of authentication... + * see src/backend/libpq/auth.c + * Not clear its SO_PEERCRED implementation works though ;-) + */ + +/* Done.... Haven't tested yet. */ +static int +socket_verify_auth(struct IPC_CHANNEL* ch, struct IPC_AUTH * auth_info) +{ + struct msghdr msg; + /* Credentials structure */ + +#define EXTRASPACE 0 + +#ifdef HAVE_STRUCT_CMSGCRED + /* FreeBSD */ + typedef struct cmsgcred Cred; +# define crRuid cmcred_uid +# define crEuid cmcred_euid +# define crRgid cmcred_gid +# define crEgid cmcred_groups[0] /* Best guess */ +# define crpid cmcred_pid +# define crngrp cmcred_ngroups +# define crgrps cmcred_groups + +#elif HAVE_STRUCT_FCRED + /* Stevens' book */ + typedef struct fcred Cred; +# define crRuid fc_uid +# define crRgid fc_rgid +# define crEgid fc_gid +# define crngrp fc_ngroups +# define crgrps fc_groups + +#elif HAVE_STRUCT_SOCKCRED + /* NetBSD */ + typedef struct sockcred Cred; +# define crRuid sc_uid +# define crEuid sc_euid +# define crRgid sc_gid +# define crEgid sc_egid +# define crngrp sc_ngroups +# define crgrps sc_groups +# undef EXTRASPACE +# define EXTRASPACE SOCKCREDSIZE(ngroups) + +#elif HAVE_STRUCT_CRED + typedef struct cred Cred; +#define cruid c_uid + +#elif HAVE_STRUCT_UCRED + typedef struct ucred Cred; + + /* reuse this define for the moment */ +# if HAVE_STRUCT_UCRED_DARWIN +# define crEuid cr_uid +# define crEgid cr_groups[0] /* Best guess */ +# define crgrps cr_groups +# define crngrp cr_ngroups +# else +# define crEuid c_uid +# define crEgid c_gid +# endif +#else +# error "No credential type found!" +#endif + + struct SOCKET_CH_PRIVATE *conn_info; + int ret = IPC_FAIL; + char buf; + + /* Compute size without padding */ + #define CMSGSIZE (sizeof(struct cmsghdr)+(sizeof(Cred))+EXTRASPACE) + + union { + char mem[CMSGSIZE]; + struct cmsghdr hdr; + Cred credu; + }cmsgmem; + Cred cred; + + /* Point to start of first structure */ + struct cmsghdr *cmsg = &cmsgmem.hdr; + + if (auth_info == NULL + || (auth_info->uid == NULL && auth_info->gid == NULL)) { + ret = IPC_OK; /* no restriction for authentication */ + } + conn_info = (struct SOCKET_CH_PRIVATE *) ch->ch_private; + + memset(&msg, 0, sizeof(msg)); + msg.msg_iov = g_new(struct iovec, 1); + msg.msg_iovlen = 1; + msg.msg_control = (char *) cmsg; + msg.msg_controllen = CMSGSIZE; + memset(cmsg, 0, sizeof(cmsgmem)); + + /* + * The one character which is received here is not meaningful; its + * purpose is only to make sure that recvmsg() blocks long enough for + * the other side to send its credentials. + */ + msg.msg_iov->iov_base = &buf; + msg.msg_iov->iov_len = 1; + + if (recvmsg(conn_info->s, &msg, 0) < 0 + || cmsg->cmsg_len < CMSGSIZE + || cmsg->cmsg_type != SCM_CREDS) { + cl_perror("can't get credential information from peer"); + return ret; + } + + /* Avoid alignment issues - just copy it! */ + memcpy(&cred, CMSG_DATA(cmsg), sizeof(cred)); + + ch->farside_uid = cred.crEuid; + ch->farside_gid = cred.crEgid; + if (ret == IPC_OK) { + return ret; + } + + /* verify the credential information. */ + return verify_creds(auth_info, cred.crEuid, cred.crEgid); +} + +/* + * FIXME! Need to implement SCM_CREDS mechanism for BSD-based systems + * this is similar to the SCM_CREDS mechanism for verify_auth() function. + * here we just want to get the pid of the other side from the credential + * information. + */ + +static +pid_t +socket_get_farside_pid(int sock) +{ + /* FIXME! */ + return -1; +} +#endif /* SCM_CREDS version */ + +/*********************************************************************** + * Bind/Stat VERSION... (Supported on OSX/Darwin and 4.3+BSD at least...) + * + * This is for use on systems such as OSX-Darwin where + * none of the other options is available. + * + * This implementation has been adapted from "Advanced Programming + * in the Unix Environment", Section 15.5.2, by W. Richard Stevens. + * + */ +#ifdef USE_BINDSTAT_CREDS + +static int +socket_verify_auth(struct IPC_CHANNEL* ch, struct IPC_AUTH * auth_info) +{ + int len = 0; + int ret = IPC_FAIL; + struct stat stat_buf; + struct sockaddr_un *peer_addr = NULL; + struct SOCKET_CH_PRIVATE *ch_private = NULL; + + if(ch != NULL) { + ch_private = (struct SOCKET_CH_PRIVATE *)(ch->ch_private); + if(ch_private != NULL) { + peer_addr = ch_private->peer_addr; + } + } + + if(ch == NULL) { + cl_log(LOG_ERR, "No channel to authenticate"); + return IPC_FAIL; + + } else if (auth_info == NULL + || (auth_info->uid == NULL && auth_info->gid == NULL)) { + ret = IPC_OK; /* no restriction for authentication */ + + } + + if(ch_private == NULL) { + cl_log(LOG_ERR, "No channel private data available"); + return ret; + + } else if(peer_addr == NULL) { + cl_log(LOG_ERR, "No peer information available"); + return ret; + } + + len = SUN_LEN(peer_addr); + + if(len < 1) { + cl_log(LOG_ERR, "No peer information available"); + return ret; + } + peer_addr->sun_path[len] = 0; + stat(peer_addr->sun_path, &stat_buf); + + ch->farside_uid = stat_buf.st_uid; + ch->farside_gid = stat_buf.st_gid; + if (ret == IPC_OK) { + return ret; + } + + if ((auth_info->uid == NULL || g_hash_table_size(auth_info->uid) == 0) + && auth_info->gid != NULL + && g_hash_table_size(auth_info->gid) != 0) { + cl_log(LOG_WARNING, + "GID-Only IPC security is not supported" + " on this platform."); + return IPC_BROKEN; + } + + /* verify the credential information. */ + return verify_creds(auth_info, stat_buf.st_uid, stat_buf.st_gid); +} + +static pid_t +socket_get_farside_pid(int sock) +{ + return -1; +} +#endif /* Bind/stat version */ + +/*********************************************************************** + * USE_STREAM_CREDS VERSION... (e.g. Solaris pre-10) + ***********************************************************************/ +#ifdef USE_STREAM_CREDS +static int +socket_verify_auth(struct IPC_CHANNEL* ch, struct IPC_AUTH * auth_info) +{ + struct SOCKET_CH_PRIVATE *conn_info; + + if (ch == NULL || ch->ch_private == NULL) { + return IPC_FAIL; + } + + conn_info = (struct SOCKET_CH_PRIVATE *) ch->ch_private; + + ch->farside_uid = conn_info->farside_uid; + ch->farside_gid = conn_info->farside_gid; + + /* verify the credential information. */ + return verify_creds(auth_info, + conn_info->farside_uid, conn_info->farside_gid); +} + +static +pid_t +socket_get_farside_pid(int sock) +{ + return -1; +} +#endif + +/*********************************************************************** + * GETPEERUCRED VERSION... (e.g. Solaris 10 upwards) + ***********************************************************************/ + +#ifdef USE_GETPEERUCRED +/* verify the authentication information. */ +static int +socket_verify_auth(struct IPC_CHANNEL* ch, struct IPC_AUTH * auth_info) +{ + struct SOCKET_CH_PRIVATE *conn_info; + ucred_t *ucred = NULL; + int rc = IPC_FAIL; + + if (ch == NULL || ch->ch_private == NULL) { + return IPC_FAIL; + } + + conn_info = (struct SOCKET_CH_PRIVATE *) ch->ch_private; + + if (auth_info == NULL + || (auth_info->uid == NULL && auth_info->gid == NULL)) { + rc = IPC_OK; /* no restriction for authentication */ + } + + if (getpeerucred(conn_info->s, &ucred) < 0) { + cl_perror("getpeereid() failure"); + return rc; + } + + ch->farside_uid = ucred_geteuid(ucred); + ch->farside_gid = ucred_getegid(ucred); + if (rc == IPC_OK) { + return rc; + } + + /* verify the credential information. */ + rc = verify_creds(auth_info, + ucred_geteuid(ucred), ucred_getegid(ucred)); + ucred_free(ucred); + return rc; +} + +static +pid_t +socket_get_farside_pid(int sockfd) +{ + ucred_t *ucred = NULL; + pid_t pid; + + if (getpeerucred(sockfd, &ucred) < 0) { + cl_perror("getpeereid() failure"); + return IPC_FAIL; + } + + pid = ucred_getpid(ucred); + + ucred_free(ucred); + + return pid; +} +#endif + +/*********************************************************************** + * DUMMY VERSION... (other systems...) + * + * Other options that seem to be out there include + * SCM_CREDENTIALS and LOCAL_CREDS + * There are some kludgy things you can do with SCM_RIGHTS + * to pass an fd which could only be opened by the user id to + * validate the user id, but I don't know of a similar kludge which + * would work for group ids. And, even the uid one will fail + * if normal users are allowed to give away (chown) files. + * + * Unfortunately, this set of authentication routines have become + * very important to this API and its users (like heartbeat). + * + ***********************************************************************/ + +#ifdef USE_DUMMY_CREDS +static int +socket_verify_auth(struct IPC_CHANNEL* ch, struct IPC_AUTH * auth_info) +{ + return IPC_FAIL; +} + +static +pid_t +socket_get_farside_pid(int sock) +{ + return -1; +} +#endif /* Dummy version */ + +/* socket object of the function table */ +static struct IPC_OPS socket_ops = { + socket_destroy_channel, + socket_initiate_connection, + socket_verify_auth, + socket_assert_auth, + socket_send, + socket_recv, + socket_waitin, + socket_waitout, + socket_is_message_pending, + socket_is_output_pending, + socket_resume_io, + socket_get_send_fd, + socket_get_recv_fd, + socket_set_send_qlen, + socket_set_recv_qlen, + socket_set_high_flow_callback, + socket_set_low_flow_callback, + socket_new_ipcmsg, + socket_get_chan_status, + socket_is_sendq_full, + socket_is_recvq_full, + socket_get_conntype, + socket_disconnect, +}; |