syslogd: close/re-resolve/reopen remote sockets on some sendto() errors
[oweals/busybox.git] / sysklogd / syslogd.c
index f4644f218d902c69a72cc8c096fa8178ddc441fc..37a119b3d7b2cd66e325cdb2195dcb9204373a3b 100644 (file)
  * Licensed under the GPL v2 or later, see the file LICENSE in this tarball.
  */
 
-#include "busybox.h"
-#include <paths.h>
-#include <sys/un.h>
-
-/* SYSLOG_NAMES defined to pull some extra junk from syslog.h */
+/*
+ * Done in syslogd_and_logger.c:
+#include "libbb.h"
 #define SYSLOG_NAMES
-#include <sys/syslog.h>
+#define SYSLOG_NAMES_CONST
+#include <syslog.h>
+*/
+
+#include <sys/un.h>
 #include <sys/uio.h>
 
-#define DEBUG 0
+#if ENABLE_FEATURE_REMOTE_LOG
+#include <netinet/in.h>
+#endif
 
-/* Path to the unix socket */
-static char *dev_log_name;
+#if ENABLE_FEATURE_IPC_SYSLOG
+#include <sys/ipc.h>
+#include <sys/sem.h>
+#include <sys/shm.h>
+#endif
 
-/* Path for the file where all log messages are written */
-static const char *logFilePath = "/var/log/messages";
 
-/* interval between marks in seconds */
-static int markInterval = 20 * 60;
+#define DEBUG 0
 
-/* level of messages to be locally logged */
-static int logLevel = 8;
+/* MARK code is not very useful, is bloat, and broken:
+ * can deadlock if alarmed to make MARK while writing to IPC buffer
+ * (semaphores are down but do_mark routine tries to down them again) */
+#undef SYSLOGD_MARK
 
-/* localhost's name */
-static char localHostName[64];
+/* Write locking does not seem to be useful either */
+#undef SYSLOGD_WRLOCK
 
-#if ENABLE_FEATURE_ROTATE_LOGFILE
-/* max size of message file before being rotated */
-static int logFileSize = 200 * 1024;
-/* number of rotated message files */
-static int logFileRotate = 1;
+enum {
+       MAX_READ = CONFIG_FEATURE_SYSLOGD_READ_BUFFER_SIZE,
+       DNS_WAIT_SEC = 2 * 60,
+};
+
+/* Semaphore operation structures */
+struct shbuf_ds {
+       int32_t size;   /* size of data - 1 */
+       int32_t tail;   /* end of message list */
+       char data[1];   /* data/messages */
+};
+
+#if ENABLE_FEATURE_REMOTE_LOG
+typedef struct {
+       int remoteFD;
+       unsigned last_dns_resolve;
+       len_and_sockaddr *remoteAddr;
+       const char *remoteHostname;
+} remoteHost_t;
 #endif
 
+/* Allows us to have smaller initializer. Ugly. */
+#define GLOBALS \
+       const char *logFilePath;                \
+       int logFD;                              \
+       /* interval between marks in seconds */ \
+       /*int markInterval;*/                   \
+       /* level of messages to be logged */    \
+       int logLevel;                           \
+IF_FEATURE_ROTATE_LOGFILE( \
+       /* max size of file before rotation */  \
+       unsigned logFileSize;                   \
+       /* number of rotated message files */   \
+       unsigned logFileRotate;                 \
+       unsigned curFileSize;                   \
+       smallint isRegular;                     \
+) \
+IF_FEATURE_IPC_SYSLOG( \
+       int shmid; /* ipc shared memory id */   \
+       int s_semid; /* ipc semaphore id */     \
+       int shm_size;                           \
+       struct sembuf SMwup[1];                 \
+       struct sembuf SMwdn[3];                 \
+)
+
+struct init_globals {
+       GLOBALS
+};
+
+struct globals {
+       GLOBALS
+
 #if ENABLE_FEATURE_REMOTE_LOG
-#include <netinet/in.h>
-/* udp socket for logging to remote host */
-static int remoteFD = -1;
-static struct sockaddr_in remoteAddr;
+       llist_t *remoteHosts;
 #endif
+#if ENABLE_FEATURE_IPC_SYSLOG
+       struct shbuf_ds *shbuf;
+#endif
+       time_t last_log_time;
+       /* localhost's name. We print only first 64 chars */
+       char *hostname;
+
+       /* We recv into recvbuf... */
+       char recvbuf[MAX_READ * (1 + ENABLE_FEATURE_SYSLOGD_DUP)];
+       /* ...then copy to parsebuf, escaping control chars */
+       /* (can grow x2 max) */
+       char parsebuf[MAX_READ*2];
+       /* ...then sprintf into printbuf, adding timestamp (15 chars),
+        * host (64), fac.prio (20) to the message */
+       /* (growth by: 15 + 64 + 20 + delims = ~110) */
+       char printbuf[MAX_READ*2 + 128];
+};
 
+static const struct init_globals init_data = {
+       .logFilePath = "/var/log/messages",
+       .logFD = -1,
+#ifdef SYSLOGD_MARK
+       .markInterval = 20 * 60,
+#endif
+       .logLevel = 8,
+#if ENABLE_FEATURE_ROTATE_LOGFILE
+       .logFileSize = 200 * 1024,
+       .logFileRotate = 1,
+#endif
+#if ENABLE_FEATURE_IPC_SYSLOG
+       .shmid = -1,
+       .s_semid = -1,
+       .shm_size = ((CONFIG_FEATURE_IPC_SYSLOG_BUFFER_SIZE)*1024), // default shm size
+       .SMwup = { {1, -1, IPC_NOWAIT} },
+       .SMwdn = { {0, 0}, {1, 0}, {1, +1} },
+#endif
+};
 
-/* NB: we may need 2x this amount on stack... */
-enum { MAX_READ = 1024 };
+#define G (*ptr_to_globals)
+#define INIT_G() do { \
+       SET_PTR_TO_GLOBALS(memcpy(xzalloc(sizeof(G)), &init_data, sizeof(init_data))); \
+} while (0)
 
 
-/* options */
-/* Correct regardless of combination of CONFIG_xxx */
+/* Options */
 enum {
        OPTBIT_mark = 0, // -m
        OPTBIT_nofork, // -n
        OPTBIT_outfile, // -O
        OPTBIT_loglevel, // -l
        OPTBIT_small, // -S
-       USE_FEATURE_ROTATE_LOGFILE(OPTBIT_filesize   ,) // -s
-       USE_FEATURE_ROTATE_LOGFILE(OPTBIT_rotatecnt  ,) // -b
-       USE_FEATURE_REMOTE_LOG(    OPTBIT_remote     ,) // -R
-       USE_FEATURE_REMOTE_LOG(    OPTBIT_localtoo   ,) // -L
-       USE_FEATURE_IPC_SYSLOG(    OPTBIT_circularlog,) // -C
+       IF_FEATURE_ROTATE_LOGFILE(OPTBIT_filesize   ,)  // -s
+       IF_FEATURE_ROTATE_LOGFILE(OPTBIT_rotatecnt  ,)  // -b
+       IF_FEATURE_REMOTE_LOG(    OPTBIT_remotelog  ,)  // -R
+       IF_FEATURE_REMOTE_LOG(    OPTBIT_locallog   ,)  // -L
+       IF_FEATURE_IPC_SYSLOG(    OPTBIT_circularlog,)  // -C
+       IF_FEATURE_SYSLOGD_DUP(   OPTBIT_dup        ,)  // -D
 
        OPT_mark        = 1 << OPTBIT_mark    ,
        OPT_nofork      = 1 << OPTBIT_nofork  ,
        OPT_outfile     = 1 << OPTBIT_outfile ,
        OPT_loglevel    = 1 << OPTBIT_loglevel,
        OPT_small       = 1 << OPTBIT_small   ,
-       OPT_filesize    = USE_FEATURE_ROTATE_LOGFILE((1 << OPTBIT_filesize   )) + 0,
-       OPT_rotatecnt   = USE_FEATURE_ROTATE_LOGFILE((1 << OPTBIT_rotatecnt  )) + 0,
-       OPT_remotelog   = USE_FEATURE_REMOTE_LOG(    (1 << OPTBIT_remote     )) + 0,
-       OPT_locallog    = USE_FEATURE_REMOTE_LOG(    (1 << OPTBIT_localtoo   )) + 0,
-       OPT_circularlog = USE_FEATURE_IPC_SYSLOG(    (1 << OPTBIT_circularlog)) + 0,
+       OPT_filesize    = IF_FEATURE_ROTATE_LOGFILE((1 << OPTBIT_filesize   )) + 0,
+       OPT_rotatecnt   = IF_FEATURE_ROTATE_LOGFILE((1 << OPTBIT_rotatecnt  )) + 0,
+       OPT_remotelog   = IF_FEATURE_REMOTE_LOG(    (1 << OPTBIT_remotelog  )) + 0,
+       OPT_locallog    = IF_FEATURE_REMOTE_LOG(    (1 << OPTBIT_locallog   )) + 0,
+       OPT_circularlog = IF_FEATURE_IPC_SYSLOG(    (1 << OPTBIT_circularlog)) + 0,
+       OPT_dup         = IF_FEATURE_SYSLOGD_DUP(   (1 << OPTBIT_dup        )) + 0,
 };
 #define OPTION_STR "m:nO:l:S" \
-       USE_FEATURE_ROTATE_LOGFILE("s:" ) \
-       USE_FEATURE_ROTATE_LOGFILE("b:" ) \
-       USE_FEATURE_REMOTE_LOG(    "R:" ) \
-       USE_FEATURE_REMOTE_LOG(    "L"  ) \
-       USE_FEATURE_IPC_SYSLOG(    "C::")
+       IF_FEATURE_ROTATE_LOGFILE("s:" ) \
+       IF_FEATURE_ROTATE_LOGFILE("b:" ) \
+       IF_FEATURE_REMOTE_LOG(    "R:" ) \
+       IF_FEATURE_REMOTE_LOG(    "L"  ) \
+       IF_FEATURE_IPC_SYSLOG(    "C::") \
+       IF_FEATURE_SYSLOGD_DUP(   "D"  )
 #define OPTION_DECL *opt_m, *opt_l \
-       USE_FEATURE_ROTATE_LOGFILE(,*opt_s) \
-       USE_FEATURE_ROTATE_LOGFILE(,*opt_b) \
-       USE_FEATURE_REMOTE_LOG(    ,*opt_R) \
-       USE_FEATURE_IPC_SYSLOG(    ,*opt_C = NULL)
-#define OPTION_PARAM &opt_m, &logFilePath, &opt_l \
-       USE_FEATURE_ROTATE_LOGFILE(,&opt_s) \
-       USE_FEATURE_ROTATE_LOGFILE(,&opt_b) \
-       USE_FEATURE_REMOTE_LOG(    ,&opt_R) \
-       USE_FEATURE_IPC_SYSLOG(    ,&opt_C)
+       IF_FEATURE_ROTATE_LOGFILE(,*opt_s) \
+       IF_FEATURE_ROTATE_LOGFILE(,*opt_b) \
+       IF_FEATURE_IPC_SYSLOG(    ,*opt_C = NULL)
+#define OPTION_PARAM &opt_m, &G.logFilePath, &opt_l \
+       IF_FEATURE_ROTATE_LOGFILE(,&opt_s) \
+       IF_FEATURE_ROTATE_LOGFILE(,&opt_b) \
+       IF_FEATURE_REMOTE_LOG(    ,&remoteAddrList) \
+       IF_FEATURE_IPC_SYSLOG(    ,&opt_C)
 
 
 /* circular buffer variables/structures */
 #if ENABLE_FEATURE_IPC_SYSLOG
 
-
 #if CONFIG_FEATURE_IPC_SYSLOG_BUFFER_SIZE < 4
 #error Sorry, you must set the syslogd buffer size to at least 4KB.
 #error Please check CONFIG_FEATURE_IPC_SYSLOG_BUFFER_SIZE
 #endif
 
-#include <sys/ipc.h>
-#include <sys/sem.h>
-#include <sys/shm.h>
-
-/* our shared key */
-#define KEY_ID ((long)0x414e4547) /* "GENA" */
-
-// Semaphore operation structures
-static struct shbuf_ds {
-       int32_t size;   // size of data written
-       int32_t head;   // start of message list
-       int32_t tail;   // end of message list
-       char data[1];   // data/messages
-} *shbuf;               // shared memory pointer
-
-static int shmid = -1;   // ipc shared memory id
-static int s_semid = -1; // ipc semaphore id
-static int shm_size = ((CONFIG_FEATURE_IPC_SYSLOG_BUFFER_SIZE)*1024);  // default shm size
+/* our shared key (syslogd.c and logread.c must be in sync) */
+enum { KEY_ID = 0x414e4547 }; /* "GENA" */
 
 static void ipcsyslog_cleanup(void)
 {
-       if (shmid != -1) {
-               shmdt(shbuf);
+       if (G.shmid != -1) {
+               shmdt(G.shbuf);
        }
-       if (shmid != -1) {
-               shmctl(shmid, IPC_RMID, NULL);
+       if (G.shmid != -1) {
+               shmctl(G.shmid, IPC_RMID, NULL);
        }
-       if (s_semid != -1) {
-               semctl(s_semid, 0, IPC_RMID, 0);
+       if (G.s_semid != -1) {
+               semctl(G.s_semid, 0, IPC_RMID, 0);
        }
 }
 
 static void ipcsyslog_init(void)
 {
        if (DEBUG)
-               printf("shmget(%lx, %d,...)\n", KEY_ID, shm_size);
+               printf("shmget(%x, %d,...)\n", (int)KEY_ID, G.shm_size);
 
-       shmid = shmget(KEY_ID, shm_size, IPC_CREAT | 1023);
-       if (shmid == -1) {
+       G.shmid = shmget(KEY_ID, G.shm_size, IPC_CREAT | 0644);
+       if (G.shmid == -1) {
                bb_perror_msg_and_die("shmget");
        }
 
-       shbuf = shmat(shmid, NULL, 0);
-       if (!shbuf) {
+       G.shbuf = shmat(G.shmid, NULL, 0);
+       if (G.shbuf == (void*) -1L) { /* shmat has bizarre error return */
                bb_perror_msg_and_die("shmat");
        }
 
-       shbuf->size = shm_size - offsetof(struct shbuf_ds, data);
-       shbuf->head = shbuf->tail = 0;
+       memset(G.shbuf, 0, G.shm_size);
+       G.shbuf->size = G.shm_size - offsetof(struct shbuf_ds, data) - 1;
+       /*G.shbuf->tail = 0;*/
 
        // we'll trust the OS to set initial semval to 0 (let's hope)
-       s_semid = semget(KEY_ID, 2, IPC_CREAT | IPC_EXCL | 1023);
-       if (s_semid == -1) {
+       G.s_semid = semget(KEY_ID, 2, IPC_CREAT | IPC_EXCL | 1023);
+       if (G.s_semid == -1) {
                if (errno == EEXIST) {
-                       s_semid = semget(KEY_ID, 2, 0);
-                       if (s_semid != -1)
+                       G.s_semid = semget(KEY_ID, 2, 0);
+                       if (G.s_semid != -1)
                                return;
                }
                bb_perror_msg_and_die("semget");
        }
 }
 
-/* write message to buffer */
+/* Write message to shared mem buffer */
 static void log_to_shmem(const char *msg, int len)
 {
-       static /*const*/ struct sembuf SMwup[1] = { {1, -1, IPC_NOWAIT} };
-       static /*const*/ struct sembuf SMwdn[3] = { {0, 0}, {1, 0}, {1, +1} };
-
        int old_tail, new_tail;
-       char *c;
 
-       if (semop(s_semid, SMwdn, 3) == -1) {
+       if (semop(G.s_semid, G.SMwdn, 3) == -1) {
                bb_perror_msg_and_die("SMwdn");
        }
 
        /* Circular Buffer Algorithm:
         * --------------------------
-         * tail == position where to store next syslog message.
-        * head == position of next message to retrieve ("print").
-        * if head == tail, there is no "unprinted" messages left.
-        * head is typically advanced by separate "reader" program,
-        * but if there isn't one, we have to do it ourself.
-        * messages are NUL-separated.
+        * tail == position where to store next syslog message.
+        * tail's max value is (shbuf->size - 1)
+        * Last byte of buffer is never used and remains NUL.
         */
        len++; /* length with NUL included */
  again:
-       old_tail = shbuf->tail;
+       old_tail = G.shbuf->tail;
        new_tail = old_tail + len;
-       if (new_tail < shbuf->size) {
-               /* No need to move head if shbuf->head <= old_tail,
-                * else... */
-               if (old_tail < shbuf->head && shbuf->head <= new_tail) {
-                       /* ...need to move head forward */
-                       c = memchr(shbuf->data + new_tail, '\0',
-                                          shbuf->size - new_tail);
-                       if (!c) /* no NUL ahead of us, wrap around */
-                               c = memchr(shbuf->data, '\0', old_tail);
-                       if (!c) { /* still nothing? point to this msg... */
-                               shbuf->head = old_tail;
-                       } else {
-                               /* convert pointer to offset + skip NUL */
-                               shbuf->head = c - shbuf->data + 1;
-                       }
-               }
+       if (new_tail < G.shbuf->size) {
                /* store message, set new tail */
-               memcpy(shbuf->data + old_tail, msg, len);
-               shbuf->tail = new_tail;
+               memcpy(G.shbuf->data + old_tail, msg, len);
+               G.shbuf->tail = new_tail;
        } else {
-               /* we need to break up the message and wrap it around */
                /* k == available buffer space ahead of old tail */
-               int k = shbuf->size - old_tail - 1;
-               if (shbuf->head > old_tail) {
-                       /* we are going to overwrite head, need to
-                        * move it out of the way */
-                       c = memchr(shbuf->data, '\0', old_tail);
-                       if (!c) { /* nothing? point to this msg... */
-                               shbuf->head = old_tail;
-                       } else { /* convert pointer to offset + skip NUL */
-                               shbuf->head = c - shbuf->data + 1;
-                       }
-               }
+               int k = G.shbuf->size - old_tail;
                /* copy what fits to the end of buffer, and repeat */
-               memcpy(shbuf->data + old_tail, msg, k);
+               memcpy(G.shbuf->data + old_tail, msg, k);
                msg += k;
                len -= k;
-               shbuf->tail = 0;
+               G.shbuf->tail = 0;
                goto again;
        }
-       if (semop(s_semid, SMwup, 1) == -1) {
+       if (semop(G.s_semid, G.SMwup, 1) == -1) {
                bb_perror_msg_and_die("SMwup");
        }
        if (DEBUG)
-               printf("head:%d tail:%d\n", shbuf->head, shbuf->tail);
+               printf("tail:%d\n", G.shbuf->tail);
 }
 #else
 void ipcsyslog_cleanup(void);
 void ipcsyslog_init(void);
 void log_to_shmem(const char *msg);
-
-
 #endif /* FEATURE_IPC_SYSLOG */
 
 
 /* Print a message to the log file. */
-static void log_locally(char *msg)
+static void log_locally(time_t now, char *msg)
 {
-       int fd, len = strlen(msg);
+#ifdef SYSLOGD_WRLOCK
+       struct flock fl;
+#endif
+       int len = strlen(msg);
 
 #if ENABLE_FEATURE_IPC_SYSLOG
-       if ((option_mask32 & OPT_circularlog) && shbuf) {
+       if ((option_mask32 & OPT_circularlog) && G.shbuf) {
                log_to_shmem(msg, len);
                return;
        }
 #endif
+       if (G.logFD >= 0) {
+               /* Reopen log file every second. This allows admin
+                * to delete the file and not worry about restarting us.
+                * This costs almost nothing since it happens
+                * _at most_ once a second.
+                */
+               if (!now)
+                       now = time(NULL);
+               if (G.last_log_time != now) {
+                       G.last_log_time = now;
+                       close(G.logFD);
+                       goto reopen;
+               }
+       } else {
+ reopen:
+               G.logFD = open(G.logFilePath, O_WRONLY | O_CREAT
+                                       | O_NOCTTY | O_APPEND | O_NONBLOCK,
+                                       0666);
+               if (G.logFD < 0) {
+                       /* cannot open logfile? - print to /dev/console then */
+                       int fd = device_open(DEV_CONSOLE, O_WRONLY | O_NOCTTY | O_NONBLOCK);
+                       if (fd < 0)
+                               fd = 2; /* then stderr, dammit */
+                       full_write(fd, msg, len);
+                       if (fd != 2)
+                               close(fd);
+                       return;
+               }
+#if ENABLE_FEATURE_ROTATE_LOGFILE
+               {
+                       struct stat statf;
+                       G.isRegular = (fstat(G.logFD, &statf) == 0 && S_ISREG(statf.st_mode));
+                       /* bug (mostly harmless): can wrap around if file > 4gb */
+                       G.curFileSize = statf.st_size;
+               }
+#endif
+       }
 
- again:
-       fd = device_open(logFilePath, O_WRONLY | O_CREAT
-                                       | O_NOCTTY | O_APPEND | O_NONBLOCK);
-       if (fd >= 0) {
-               struct flock fl;
-
-               fl.l_whence = SEEK_SET;
-               fl.l_start = 0;
-               fl.l_len = 1;
-               fl.l_type = F_WRLCK;
-               fcntl(fd, F_SETLKW, &fl);
+#ifdef SYSLOGD_WRLOCK
+       fl.l_whence = SEEK_SET;
+       fl.l_start = 0;
+       fl.l_len = 1;
+       fl.l_type = F_WRLCK;
+       fcntl(G.logFD, F_SETLKW, &fl);
+#endif
 
 #if ENABLE_FEATURE_ROTATE_LOGFILE
-               if (logFileSize) {
-                       struct stat statf;
-                       int r = fstat(fd, &statf);
-                       if (!r && (statf.st_mode & S_IFREG)
-                        && (lseek(fd, 0, SEEK_END) > logFileSize)
-                       ) {
-                               if (logFileRotate) { /* always 0..99 */
-                                       int i = strlen(logFilePath) + 3 + 1;
-                                       char oldFile[i];
-                                       char newFile[i];
-                                       i = logFileRotate - 1;
-                                       /* rename: f.8 -> f.9; f.7 -> f.8; ... */
-                                       while (1) {
-                                               sprintf(newFile, "%s.%d", logFilePath, i);
-                                               if (i == 0) break;
-                                               sprintf(oldFile, "%s.%d", logFilePath, --i);
-                                               rename(oldFile, newFile);
-                                       }
-                                       /* newFile == "f.0" now */
-                                       rename(logFilePath, newFile);
-                                       fl.l_type = F_UNLCK;
-                                       fcntl(fd, F_SETLKW, &fl);
-                                       close(fd);
-                                       goto again;
-                               }
-                               ftruncate(fd, 0);
+       if (G.logFileSize && G.isRegular && G.curFileSize > G.logFileSize) {
+               if (G.logFileRotate) { /* always 0..99 */
+                       int i = strlen(G.logFilePath) + 3 + 1;
+                       char oldFile[i];
+                       char newFile[i];
+                       i = G.logFileRotate - 1;
+                       /* rename: f.8 -> f.9; f.7 -> f.8; ... */
+                       while (1) {
+                               sprintf(newFile, "%s.%d", G.logFilePath, i);
+                               if (i == 0) break;
+                               sprintf(oldFile, "%s.%d", G.logFilePath, --i);
+                               /* ignore errors - file might be missing */
+                               rename(oldFile, newFile);
                        }
-               }
+                       /* newFile == "f.0" now */
+                       rename(G.logFilePath, newFile);
+#ifdef SYSLOGD_WRLOCK
+                       fl.l_type = F_UNLCK;
+                       fcntl(G.logFD, F_SETLKW, &fl);
 #endif
-               full_write(fd, msg, len);
-               fl.l_type = F_UNLCK;
-               fcntl(fd, F_SETLKW, &fl);
-               close(fd);
-       } else {
-               /* cannot open logfile? - print to /dev/console then */
-               fd = device_open(_PATH_CONSOLE, O_WRONLY | O_NOCTTY | O_NONBLOCK);
-               if (fd < 0)
-                       fd = 2; /* then stderr, dammit */
-               full_write(fd, msg, len);
-               if (fd != 2)
-                       close(fd);
+                       close(G.logFD);
+                       goto reopen;
+               }
+               ftruncate(G.logFD, 0);
        }
+       G.curFileSize +=
+#endif
+                       full_write(G.logFD, msg, len);
+#ifdef SYSLOGD_WRLOCK
+       fl.l_type = F_UNLCK;
+       fcntl(G.logFD, F_SETLKW, &fl);
+#endif
 }
 
 static void parse_fac_prio_20(int pri, char *res20)
 {
-       CODE *c_pri, *c_fac;
+       const CODE *c_pri, *c_fac;
 
        if (pri != 0) {
                c_fac = facilitynames;
                while (c_fac->c_name) {
                        if (c_fac->c_val != (LOG_FAC(pri) << 3)) {
-                               c_fac++; continue;
+                               c_fac++;
+                               continue;
                        }
                        /* facility is found, look for prio */
                        c_pri = prioritynames;
                        while (c_pri->c_name) {
                                if (c_pri->c_val != LOG_PRI(pri)) {
-                                       c_pri++; continue;
+                                       c_pri++;
+                                       continue;
                                }
                                snprintf(res20, 20, "%s.%s",
                                                c_fac->c_name, c_pri->c_name);
@@ -349,241 +413,332 @@ static void parse_fac_prio_20(int pri, char *res20)
        }
 }
 
-/* len parameter is used only for "is there a timestamp?" check
- * NB: some callers cheat and supply 0 when they know
- * that there is no timestamp, short-cutting the test */
+/* len parameter is used only for "is there a timestamp?" check.
+ * NB: some callers cheat and supply len==0 when they know
+ * that there is no timestamp, short-circuiting the test. */
 static void timestamp_and_log(int pri, char *msg, int len)
 {
-       time_t now;
        char *timestamp;
+       time_t now;
 
+       /* Jan 18 00:11:22 msg... */
+       /* 01234567890123456 */
        if (len < 16 || msg[3] != ' ' || msg[6] != ' '
         || msg[9] != ':' || msg[12] != ':' || msg[15] != ' '
        ) {
                time(&now);
-               timestamp = ctime(&now) + 4;
+               timestamp = ctime(&now) + 4; /* skip day of week */
        } else {
+               now = 0;
                timestamp = msg;
                msg += 16;
        }
        timestamp[15] = '\0';
 
-       /* Log message locally (to file or shared mem) */
-       if (!ENABLE_FEATURE_REMOTE_LOG || (option_mask32 & OPT_locallog)) {
-               if (LOG_PRI(pri) < logLevel) {
-                       if (option_mask32 & OPT_small)
-                               msg = xasprintf("%s %s\n", timestamp, msg);
-                       else {
-                               char res[20];
-                               parse_fac_prio_20(pri, res);
-                               msg = xasprintf("%s %s %s %s\n", timestamp, localHostName, res, msg);
-                       }
-                       log_locally(msg);
-                       free(msg);
-               }
+       if (option_mask32 & OPT_small)
+               sprintf(G.printbuf, "%s %s\n", timestamp, msg);
+       else {
+               char res[20];
+               parse_fac_prio_20(pri, res);
+               sprintf(G.printbuf, "%s %.64s %s %s\n", timestamp, G.hostname, res, msg);
        }
+
+       /* Log message locally (to file or shared mem) */
+       log_locally(now, G.printbuf);
+}
+
+static void timestamp_and_log_internal(const char *msg)
+{
+       /* -L, or no -R */
+       if (ENABLE_FEATURE_REMOTE_LOG && !(option_mask32 & OPT_locallog))
+               return;
+       timestamp_and_log(LOG_SYSLOG | LOG_INFO, (char*)msg, 0);
 }
 
+/* tmpbuf[len] is a NUL byte (set by caller), but there can be other,
+ * embedded NULs. Split messages on each of these NULs, parse prio,
+ * escape control chars and log each locally. */
 static void split_escape_and_log(char *tmpbuf, int len)
 {
-       char line[len * 2 + 1]; /* gcc' cheap alloca */
        char *p = tmpbuf;
 
        tmpbuf += len;
        while (p < tmpbuf) {
                char c;
-               char *q = line;
+               char *q = G.parsebuf;
                int pri = (LOG_USER | LOG_NOTICE);
 
                if (*p == '<') {
-                       /* Parse the magic priority number. */
+                       /* Parse the magic priority number */
                        pri = bb_strtou(p + 1, &p, 10);
-                       if (*p == '>') p++;
-                       if (pri & ~(LOG_FACMASK | LOG_PRIMASK)) {
+                       if (*p == '>')
+                               p++;
+                       if (pri & ~(LOG_FACMASK | LOG_PRIMASK))
                                pri = (LOG_USER | LOG_NOTICE);
-                       }
                }
 
                while ((c = *p++)) {
                        if (c == '\n')
                                c = ' ';
-                       if (!(c & ~0x1f)) {
+                       if (!(c & ~0x1f) && c != '\t') {
                                *q++ = '^';
                                c += '@'; /* ^@, ^A, ^B... */
                        }
                        *q++ = c;
                }
                *q = '\0';
-               /* now log it */
-               timestamp_and_log(pri, line, q - line);
-       }
-}
 
-static void quit_signal(int sig)
-{
-       timestamp_and_log(LOG_SYSLOG | LOG_INFO, "System log daemon exiting", 0);
-       puts("System log daemon exiting");
-       unlink(dev_log_name);
-       if (ENABLE_FEATURE_IPC_SYSLOG)
-               ipcsyslog_cleanup();
-       exit(1);
+               /* Now log it */
+               if (LOG_PRI(pri) < G.logLevel)
+                       timestamp_and_log(pri, G.parsebuf, q - G.parsebuf);
+       }
 }
 
+#ifdef SYSLOGD_MARK
 static void do_mark(int sig)
 {
-       if (markInterval) {
-               timestamp_and_log(LOG_SYSLOG | LOG_INFO, "-- MARK --", 0);
-               alarm(markInterval);
+       if (G.markInterval) {
+               timestamp_and_log_internal("-- MARK --");
+               alarm(G.markInterval);
        }
 }
+#endif
 
-static void do_syslogd(void) ATTRIBUTE_NORETURN;
-static void do_syslogd(void)
+/* Don't inline: prevent struct sockaddr_un to take up space on stack
+ * permanently */
+static NOINLINE int create_socket(void)
 {
        struct sockaddr_un sunx;
-       socklen_t addrLength;
        int sock_fd;
-       fd_set fds;
-
-       /* Set up signal handlers. */
-       signal(SIGINT, quit_signal);
-       signal(SIGTERM, quit_signal);
-       signal(SIGQUIT, quit_signal);
-       signal(SIGHUP, SIG_IGN);
-       signal(SIGCHLD, SIG_IGN);
-#ifdef SIGCLD
-       signal(SIGCLD, SIG_IGN);
-#endif
-       signal(SIGALRM, do_mark);
-       alarm(markInterval);
-
-       dev_log_name = xmalloc_realpath(_PATH_LOG);
-       if (!dev_log_name)
-               dev_log_name = _PATH_LOG;
-
-       /* Unlink old /dev/log (or object it points to) */
-       unlink(dev_log_name);
+       char *dev_log_name;
 
        memset(&sunx, 0, sizeof(sunx));
        sunx.sun_family = AF_UNIX;
-       strncpy(sunx.sun_path, dev_log_name, sizeof(sunx.sun_path));
-       sock_fd = xsocket(AF_UNIX, SOCK_DGRAM, 0);
-       addrLength = sizeof(sunx.sun_family) + strlen(sunx.sun_path);
-       if (bind(sock_fd, (struct sockaddr *) &sunx, addrLength) < 0) {
-               bb_perror_msg_and_die("cannot connect to socket %s", dev_log_name);
+
+       /* Unlink old /dev/log or object it points to. */
+       /* (if it exists, bind will fail) */
+       strcpy(sunx.sun_path, "/dev/log");
+       dev_log_name = xmalloc_follow_symlinks("/dev/log");
+       if (dev_log_name) {
+               safe_strncpy(sunx.sun_path, dev_log_name, sizeof(sunx.sun_path));
+               free(dev_log_name);
        }
+       unlink(sunx.sun_path);
 
-       if (chmod(dev_log_name, 0666) < 0) {
-               bb_perror_msg_and_die("cannot set permission on %s", dev_log_name);
+       sock_fd = xsocket(AF_UNIX, SOCK_DGRAM, 0);
+       xbind(sock_fd, (struct sockaddr *) &sunx, sizeof(sunx));
+       chmod("/dev/log", 0666);
+
+       return sock_fd;
+}
+
+#if ENABLE_FEATURE_REMOTE_LOG
+static int try_to_resolve_remote(remoteHost_t *rh)
+{
+       if (!rh->remoteAddr) {
+               unsigned now = monotonic_sec();
+
+               /* Don't resolve name too often - DNS timeouts can be big */
+               if ((now - rh->last_dns_resolve) < DNS_WAIT_SEC)
+                       return -1;
+               rh->last_dns_resolve = now;
+               rh->remoteAddr = host2sockaddr(rh->remoteHostname, 514);
+               if (!rh->remoteAddr)
+                       return -1;
        }
+       return xsocket(rh->remoteAddr->u.sa.sa_family, SOCK_DGRAM, 0);
+}
+#endif
+
+static void do_syslogd(void) NORETURN;
+static void do_syslogd(void)
+{
+       int sock_fd;
+#if ENABLE_FEATURE_REMOTE_LOG
+       llist_t *item;
+#endif
+#if ENABLE_FEATURE_SYSLOGD_DUP
+       int last_sz = -1;
+       char *last_buf;
+       char *recvbuf = G.recvbuf;
+#else
+#define recvbuf (G.recvbuf)
+#endif
+
+       /* Set up signal handlers (so that they interrupt read()) */
+       signal_no_SA_RESTART_empty_mask(SIGTERM, record_signo);
+       signal_no_SA_RESTART_empty_mask(SIGINT, record_signo);
+       //signal_no_SA_RESTART_empty_mask(SIGQUIT, record_signo);
+       signal(SIGHUP, SIG_IGN);
+#ifdef SYSLOGD_MARK
+       signal(SIGALRM, do_mark);
+       alarm(G.markInterval);
+#endif
+       sock_fd = create_socket();
+
        if (ENABLE_FEATURE_IPC_SYSLOG && (option_mask32 & OPT_circularlog)) {
                ipcsyslog_init();
        }
 
-       timestamp_and_log(LOG_SYSLOG | LOG_INFO, "syslogd started: BusyBox v" BB_VER, 0);
+       timestamp_and_log_internal("syslogd started: BusyBox v" BB_VER);
 
-       for (;;) {
-               FD_ZERO(&fds);
-               FD_SET(sock_fd, &fds);
+       while (!bb_got_signal) {
+               ssize_t sz;
 
-               if (select(sock_fd + 1, &fds, NULL, NULL, NULL) < 0) {
-                       if (errno == EINTR) {
-                               /* alarm may have happened. */
-                               continue;
-                       }
-                       bb_perror_msg_and_die("select");
+#if ENABLE_FEATURE_SYSLOGD_DUP
+               last_buf = recvbuf;
+               if (recvbuf == G.recvbuf)
+                       recvbuf = G.recvbuf + MAX_READ;
+               else
+                       recvbuf = G.recvbuf;
+#endif
+ read_again:
+               sz = read(sock_fd, recvbuf, MAX_READ - 1);
+               if (sz < 0) {
+                       if (!bb_got_signal)
+                               bb_perror_msg("read from /dev/log");
+                       break;
                }
 
-               if (FD_ISSET(sock_fd, &fds)) {
-                       int i;
-#define tmpbuf bb_common_bufsiz1
-                       i = recv(sock_fd, tmpbuf, MAX_READ, 0);
-                       if (i <= 0)
-                               bb_perror_msg_and_die("UNIX socket error");
-                       /* TODO: maybe supress duplicates? */
+               /* Drop trailing '\n' and NULs (typically there is one NUL) */
+               while (1) {
+                       if (sz == 0)
+                               goto read_again;
+                       /* man 3 syslog says: "A trailing newline is added when needed".
+                        * However, neither glibc nor uclibc do this:
+                        * syslog(prio, "test")   sends "test\0" to /dev/log,
+                        * syslog(prio, "test\n") sends "test\n\0".
+                        * IOW: newline is passed verbatim!
+                        * I take it to mean that it's syslogd's job
+                        * to make those look identical in the log files. */
+                       if (recvbuf[sz-1] != '\0' && recvbuf[sz-1] != '\n')
+                               break;
+                       sz--;
+               }
+#if ENABLE_FEATURE_SYSLOGD_DUP
+               if ((option_mask32 & OPT_dup) && (sz == last_sz))
+                       if (memcmp(last_buf, recvbuf, sz) == 0)
+                               continue;
+               last_sz = sz;
+#endif
 #if ENABLE_FEATURE_REMOTE_LOG
-                       /* We are not modifying log messages in any way before send */
-                       /* Remote site cannot trust _us_ anyway and need to do validation again */
-                       if (option_mask32 & OPT_remotelog) {
-                               if (-1 == remoteFD) {
-                                       remoteFD = socket(AF_INET, SOCK_DGRAM, 0);
-                               }
-                               if (-1 != remoteFD) {
-                                       /* send message to remote logger, ignore possible error */
-                                       sendto(remoteFD, tmpbuf, i, MSG_DONTWAIT,
-                                               (struct sockaddr *) &remoteAddr,
-                                               sizeof(remoteAddr));
+               /* Stock syslogd sends it '\n'-terminated
+                * over network, mimic that */
+               recvbuf[sz] = '\n';
+
+               /* We are not modifying log messages in any way before send */
+               /* Remote site cannot trust _us_ anyway and need to do validation again */
+               for (item = G.remoteHosts; item != NULL; item = item->link) {
+                       remoteHost_t *rh = (remoteHost_t *)item->data;
+
+                       if (rh->remoteFD == -1) {
+                               rh->remoteFD = try_to_resolve_remote(rh);
+                               if (rh->remoteFD == -1)
+                                       continue;
+                       }
+
+                       /* Send message to remote logger.
+                        * On some errors, close and set remoteFD to -1
+                        * so that DNS resolution is retried.
+                        */
+                       if (sendto(rh->remoteFD, recvbuf, sz+1,
+                                       MSG_DONTWAIT | MSG_NOSIGNAL,
+                                       &(rh->remoteAddr->u.sa), rh->remoteAddr->len) == -1
+                       ) {
+                               switch (errno) {
+                               case ECONNRESET:
+                               case ENOTCONN: /* paranoia */
+                               case EPIPE:
+                                       close(rh->remoteFD);
+                                       rh->remoteFD = -1;
+                                       free(rh->remoteAddr);
+                                       rh->remoteAddr = NULL;
                                }
                        }
+               }
 #endif
-                       tmpbuf[i] = '\0';
-                       split_escape_and_log(tmpbuf, i);
-#undef tmpbuf
-               } /* FD_ISSET() */
-       } /* for */
+               if (!ENABLE_FEATURE_REMOTE_LOG || (option_mask32 & OPT_locallog)) {
+                       recvbuf[sz] = '\0'; /* ensure it *is* NUL terminated */
+                       split_escape_and_log(recvbuf, sz);
+               }
+       } /* while (!bb_got_signal) */
+
+       timestamp_and_log_internal("syslogd exiting");
+       puts("syslogd exiting");
+       if (ENABLE_FEATURE_IPC_SYSLOG)
+               ipcsyslog_cleanup();
+       kill_myself_with_sig(bb_got_signal);
+#undef recvbuf
 }
 
-int syslogd_main(int argc, char **argv)
+int syslogd_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
+int syslogd_main(int argc UNUSED_PARAM, char **argv)
 {
+       int opts;
        char OPTION_DECL;
-       char *p;
-
-       /* do normal option parsing */
-       opt_complementary = "=0"; /* no non-option params */
-       getopt32(argc, argv, OPTION_STR, OPTION_PARAM);
-       if (option_mask32 & OPT_mark) // -m
-               markInterval = xatou_range(opt_m, 0, INT_MAX/60) * 60;
-       //if (option_mask32 & OPT_nofork) // -n
-       //if (option_mask32 & OPT_outfile) // -O
-       if (option_mask32 & OPT_loglevel) // -l
-               logLevel = xatou_range(opt_l, 1, 8);
-       //if (option_mask32 & OPT_small) // -S
-#if ENABLE_FEATURE_ROTATE_LOGFILE
-       if (option_mask32 & OPT_filesize) // -s
-               logFileSize = xatou_range(opt_s, 0, INT_MAX/1024) * 1024;
-       if (option_mask32 & OPT_rotatecnt) // -b
-               logFileRotate = xatou_range(opt_b, 0, 99);
+#if ENABLE_FEATURE_REMOTE_LOG
+       llist_t *remoteAddrList = NULL;
 #endif
+
+       INIT_G();
+
+       /* No non-option params, -R can occur multiple times */
+       opt_complementary = "=0" IF_FEATURE_REMOTE_LOG(":R::");
+       opts = getopt32(argv, OPTION_STR, OPTION_PARAM);
 #if ENABLE_FEATURE_REMOTE_LOG
-       if (option_mask32 & OPT_remotelog) { // -R
-               int port = 514;
-               p = strchr(opt_R, ':');
-               if (p) {
-                       *p++ = '\0';
-                       port = xatou16(p);
-               }
-               remoteAddr.sin_family = AF_INET;
-               /* FIXME: looks ip4-specific. need to do better */
-               remoteAddr.sin_addr = *(struct in_addr *) *(xgethostbyname(opt_R)->h_addr_list);
-               remoteAddr.sin_port = htons(port);
+       while (remoteAddrList) {
+               remoteHost_t *rh = xzalloc(sizeof(*rh));
+               rh->remoteHostname = llist_pop(&remoteAddrList);
+               rh->remoteFD = -1;
+               rh->last_dns_resolve = monotonic_sec() - DNS_WAIT_SEC - 1;
+               llist_add_to(&G.remoteHosts, rh);
        }
-       //if (option_mask32 & OPT_locallog) // -L
+#endif
+
+#ifdef SYSLOGD_MARK
+       if (opts & OPT_mark) // -m
+               G.markInterval = xatou_range(opt_m, 0, INT_MAX/60) * 60;
+#endif
+       //if (opts & OPT_nofork) // -n
+       //if (opts & OPT_outfile) // -O
+       if (opts & OPT_loglevel) // -l
+               G.logLevel = xatou_range(opt_l, 1, 8);
+       //if (opts & OPT_small) // -S
+#if ENABLE_FEATURE_ROTATE_LOGFILE
+       if (opts & OPT_filesize) // -s
+               G.logFileSize = xatou_range(opt_s, 0, INT_MAX/1024) * 1024;
+       if (opts & OPT_rotatecnt) // -b
+               G.logFileRotate = xatou_range(opt_b, 0, 99);
 #endif
 #if ENABLE_FEATURE_IPC_SYSLOG
-       if ((option_mask32 & OPT_circularlog) && opt_C) // -C
-               shm_size = xatoul_range(opt_C, 4, INT_MAX/1024) * 1024;
+       if (opt_C) // -Cn
+               G.shm_size = xatoul_range(opt_C, 4, INT_MAX/1024) * 1024;
 #endif
 
        /* If they have not specified remote logging, then log locally */
-       if (ENABLE_FEATURE_REMOTE_LOG && !(option_mask32 & OPT_remotelog))
+       if (ENABLE_FEATURE_REMOTE_LOG && !(opts & OPT_remotelog)) // -R
                option_mask32 |= OPT_locallog;
 
        /* Store away localhost's name before the fork */
-       gethostname(localHostName, sizeof(localHostName));
-       p = strchr(localHostName, '.');
-       if (p) {
-               *p = '\0';
-       }
+       G.hostname = safe_gethostname();
+       *strchrnul(G.hostname, '.') = '\0';
 
-       if (!(option_mask32 & OPT_nofork)) {
-#ifdef BB_NOMMU
-               vfork_daemon_rexec(0, 1, argc, argv, "-n");
-#else
-               xdaemon(0, 1);
-#endif
+       if (!(opts & OPT_nofork)) {
+               bb_daemonize_or_rexec(DAEMON_CHDIR_ROOT, argv);
        }
-       umask(0);
+       //umask(0); - why??
+       write_pidfile("/var/run/syslogd.pid");
        do_syslogd();
-
-       return EXIT_SUCCESS;
+       /* return EXIT_SUCCESS; */
 }
+
+/* Clean up. Needed because we are included from syslogd_and_logger.c */
+#undef DEBUG
+#undef SYSLOGD_MARK
+#undef SYSLOGD_WRLOCK
+#undef G
+#undef GLOBALS
+#undef INIT_G
+#undef OPTION_STR
+#undef OPTION_DECL
+#undef OPTION_PARAM