erts: Fix child_setup partial read

When calling read it may return a partial result. So we make
sure that all the data is read before returning.

The problem has been observed when reading the ACK message,
but we use the new routine for all reads in the child so that
we know that it will work everywhere.

1 files changed, 55 insertions, 45 deletions

diff --git a/erts/emulator/sys/unix/erl_child_setup.c b/erts/emulator/sys/unix/erl_child_setup.c
index a746306601..3dc7fcf313 100644
--- a/erts/emulator/sys/unix/erl_child_setup.c
+++ b/erts/emulator/sys/unix/erl_child_setup.c
@@ -136,6 +136,43 @@ void sys_sigrelease(int sig)
     sigprocmask(SIG_UNBLOCK, &mask, (sigset_t *)NULL);
 }
 
+
+/* This version of read/write makes sure to read/write the entire size before
+   returning. Normal read/write can handle partial results which we do not want. */
+static ssize_t read_all(int fd, char *buff, size_t size) {
+    ssize_t res, pos = 0;
+    do {
+        if ((res = read(fd, buff + pos, size - pos)) < 0) {
+            if (errno == ERRNO_BLOCK || errno == EINTR)
+                continue;
+            return res;
+        }
+        if (res == 0) {
+            errno = EPIPE;
+            return -1;
+        }
+        pos += res;
+    } while(size - pos != 0);
+    return pos;
+}
+
+static ssize_t write_all(int fd, const char *buff, size_t size) {
+    ssize_t res, pos = 0;
+    do {
+        if ((res = write(fd, buff + pos, size - pos)) < 0) {
+            if (errno == ERRNO_BLOCK || errno == EINTR)
+                continue;
+            return res;
+        }
+        if (res == 0) {
+            errno = EPIPE;
+            return -1;
+        }
+        pos += res;
+    } while (size - pos != 0);
+    return pos;
+}
+
 static void add_os_pid_to_port_id_mapping(Eterm, pid_t);
 static Eterm get_port_id(pid_t);
 static int forker_hash_init(void);
@@ -148,7 +185,7 @@ start_new_child(int pipes[])
 {
     struct sigaction sa;
     int errln = -1;
-    int size, res, i, pos = 0;
+    int size, i;
     char *buff, *o_buff;
 
     char *cmd, *cwd, *wd, **new_environ, **args = NULL;
@@ -166,12 +203,8 @@ start_new_child(int pipes[])
         perror(NULL);
         exit(1);
     }
-    
-    do {
-        res = read(pipes[0], (char*)&size, sizeof(size));
-    } while(res < 0 && (errno == EINTR || errno == ERRNO_BLOCK));
 
-    if (res <= 0) {
+    if (read_all(pipes[0], (char*)&size, sizeof(size)) <= 0) {
         errln = __LINE__;
         goto child_error;
     }
@@ -180,20 +213,10 @@ start_new_child(int pipes[])
 
     DEBUG_PRINT("size = %d", size);
 
-    do {
-        if ((res = read(pipes[0], buff + pos, size - pos)) < 0) {
-            if (errno == ERRNO_BLOCK || errno == EINTR)
-                continue;
-            errln = __LINE__;
-            goto child_error;
-        }
-        if (res == 0) {
-            errno = EPIPE;
-            errln = __LINE__;
-            goto child_error;
-        }
-        pos += res;
-    } while(size - pos != 0);
+    if (read_all(pipes[0], buff, size) <= 0) {
+        errln = __LINE__;
+        goto child_error;
+    }
 
     o_buff = buff;
 
@@ -252,18 +275,13 @@ start_new_child(int pipes[])
     }
 
     DEBUG_PRINT("read ack");
-    do {
+    {
         ErtsSysForkerProto proto;
-        res = read(pipes[0], &proto, sizeof(proto));
-        if (res > 0) {
-            ASSERT(proto.action == ErtsSysForkerProtoAction_Ack);
+        if (read_all(pipes[0], (char*)&proto, sizeof(proto)) <= 0) {
+            errln = __LINE__;
+            goto child_error;
         }
-    } while(res < 0 && (errno == EINTR || errno == ERRNO_BLOCK));
-
-    if (res < 1) {
-        errno = EPIPE;
-        errln = __LINE__;
-        goto child_error;
+        ASSERT(proto.action == ErtsSysForkerProtoAction_Ack);
     }
 
     DEBUG_PRINT("Set cwd to: '%s'",cwd);
@@ -373,15 +391,13 @@ child_error:
  * for posterity. */
 
 static void handle_sigchld(int sig) {
-    int buff[2], res, __preverrno = errno;
+    int buff[2], __preverrno = errno;
+    ssize_t res;
 
     sys_sigblock(SIGCHLD);
 
     while ((buff[0] = waitpid((pid_t)(-1), buff+1, WNOHANG)) > 0) {
-        do {
-            res = write(sigchld_pipe[1], buff, sizeof(buff));
-        } while (res < 0 && errno == EINTR);
-        if (res <= 0)
+        if ((res = write_all(sigchld_pipe[1], (char*)buff, sizeof(buff))) <= 0)
             ABORT("Failed to write to sigchld_pipe (%d): %d (%d)", sigchld_pipe[1], res, errno);
         DEBUG_PRINT("Reap child %d (%d)", buff[0], buff[1]);
     }
@@ -553,13 +569,11 @@ main(int argc, char *argv[])
             proto.action = ErtsSysForkerProtoAction_Go;
             proto.u.go.os_pid = os_pid;
             proto.u.go.error_number = errno;
-            while (write(pipes[1], &proto, sizeof(proto)) < 0 && errno == EINTR)
-                ; /* remove gcc warning */
+            write_all(pipes[1], (char *)&proto, sizeof(proto));
 
 #ifdef FORKER_PROTO_START_ACK
             proto.action = ErtsSysForkerProtoAction_StartAck;
-            while (write(uds_fd, &proto, sizeof(proto)) < 0 && errno == EINTR)
-                ; /* remove gcc warning */
+            write_all(uds_fd, (char *)&proto, sizeof(proto));
 #endif
 
             sys_sigrelease(SIGCHLD);
@@ -571,10 +585,8 @@ main(int argc, char *argv[])
         if (FD_ISSET(sigchld_pipe[0], &read_fds)) {
             int ibuff[2];
             ErtsSysForkerProto proto;
-            res = read(sigchld_pipe[0], ibuff, sizeof(ibuff));
+            res = read_all(sigchld_pipe[0], (char *)ibuff, sizeof(ibuff));
             if (res <= 0) {
-                if (errno == EINTR)
-                    continue;
                 ABORT("Failed to read from sigchld pipe: %d (%d)", res, errno);
             }
 
@@ -586,9 +598,7 @@ main(int argc, char *argv[])
             proto.action = ErtsSysForkerProtoAction_SigChld;
             proto.u.sigchld.error_number = ibuff[1];
             DEBUG_PRINT("send sigchld to %d (errno = %d)", uds_fd, ibuff[1]);
-            if (write(uds_fd, &proto, sizeof(proto)) < 0) {
-                if (errno == EINTR)
-                    continue;
+            if (write_all(uds_fd, (char *)&proto, sizeof(proto)) < 0) {
                 /* The uds was close, which most likely means that the VM
                    has exited. This will be detected when we try to read
                    from the uds_fd. */