[PATCH 0/3] linux-user: Introducing support for 'recvmmsg_time64()'

[PATCH 0/3] linux-user: Introducing support for 'recvmmsg_time64()'

[Filip Bozuta]

This patch introduces functionality for 'recvmmsg_time64()' which
is a year 2038 safe variant of 'recvmmsg()'. This new time64 syscall
is introduced in series of three patches rather than in one patch
because a little modification was needed before the syscall could
be introduced properly.

The first patch in the series introduces a little modification
for already existing implementations for 'recvmmsg()' and
'sendmmsg()' to use the host syscall result instead of using
a loop over 'recvmsg()' and 'sendmsg()'.

The second patch in the series introduces a little fix in
the implementation of 'recvmmsg()' that introduces the timeout
argument which is of type 'struct timespec' for which the
separate time64 variant is added.

The third patch in the series introduces the implementation
of 'recvmmsg_time64()' itself.

Testing method:

    The implementation was tested using a modified LTP test suite
    where a timeout test case was added for 'recvmmsg()'. This
    modified test suite can be found at:
    https://github.com/bozutaf/ltp

    (A patch with this fix will be sent to LTP mailing list soon)

Filip Bozuta (3):
  linux-user: Modify 'sendmmsg()' and 'recvmmsg()' implementation
  linux-user: Fix 'recvmmsg()' implementation
  linux-user: Add support for 'recvmmsg_time64()'

 linux-user/syscall.c | 274 +++++++++++++++++++++++++++++--------------
 1 file changed, 184 insertions(+), 90 deletions(-)

-- 
2.25.1

[PATCH 1/3] linux-user: Modify 'sendmmsg()' and 'recvmmsg()' implementation

[Filip Bozuta]

Implementations of 'sendmmsg()' and 'recvmmsg()' in 'syscall.c' use
a loop over a host command of 'sendmsg()' and 'recvmsg()' respectively
to send/receive individual messages from a socket. This patch changes
these implementations to use the host commands 'sendmmsg()' and 'recvmmsg()'
to send all messages without looping over 'sendmsg()' and 'recvmsg()'.

Implementation notes:

    Parts of code from 'do_sendrecvmsg_locked()', that are used to transfer
    values of 'struct msghdr' between host and target, were moved to separate
    functions 'target_to_host_msghdr()' and 'host_to_target_msghdr()'. These
    functions are used in 'do_sendrecvmmsg()' to transfer the data of each
    individual 'struct msghdr' from the 'msgvec' argument. Memory allocation
    for the 'iovec' field is done outside of these functions as to ensure that
    the memory is freed after the syscall execution.

Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
---
 linux-user/syscall.c | 243 ++++++++++++++++++++++++++++---------------
 1 file changed, 159 insertions(+), 84 deletions(-)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 05f03919ff..8cbefdb561 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -798,6 +798,10 @@ safe_syscall6(ssize_t, recvfrom, int, fd, void *, buf, size_t, len,
               int, flags, struct sockaddr *, addr, socklen_t *, addrlen)
 safe_syscall3(ssize_t, sendmsg, int, fd, const struct msghdr *, msg, int, flags)
 safe_syscall3(ssize_t, recvmsg, int, fd, struct msghdr *, msg, int, flags)
+safe_syscall4(int, sendmmsg, int, fd, struct mmsghdr *, msgvec,
+              unsigned int, vlen, int, flags)
+safe_syscall5(int, recvmmsg, int, fd, struct mmsghdr *, msgvec,
+              unsigned int, vlen, int, flags, struct timespec *, timeout)
 safe_syscall2(int, flock, int, fd, int, operation)
 #ifdef TARGET_NR_rt_sigtimedwait
 safe_syscall4(int, rt_sigtimedwait, const sigset_t *, these, siginfo_t *, uinfo,
@@ -3064,41 +3068,94 @@ static abi_long do_connect(int sockfd, abi_ulong target_addr,
     return get_errno(safe_connect(sockfd, addr, addrlen));
 }
 
-/* do_sendrecvmsg_locked() Must return target values and target errnos. */
-static abi_long do_sendrecvmsg_locked(int fd, struct target_msghdr *msgp,
-                                      int flags, int send)
+static abi_long target_to_host_msghdr(int fd, struct msghdr *host_msg,
+                                      struct target_msghdr *target_msg,
+                                      int send)
 {
-    abi_long ret, len;
-    struct msghdr msg;
-    abi_ulong count;
-    struct iovec *vec;
-    abi_ulong target_vec;
-
-    if (msgp->msg_name) {
-        msg.msg_namelen = tswap32(msgp->msg_namelen);
-        msg.msg_name = alloca(msg.msg_namelen+1);
-        ret = target_to_host_sockaddr(fd, msg.msg_name,
-                                      tswapal(msgp->msg_name),
-                                      msg.msg_namelen);
+    abi_long ret = 0;
+    if (target_msg->msg_name) {
+        host_msg->msg_namelen = tswap32(target_msg->msg_namelen);
+        host_msg->msg_name = alloca(host_msg->msg_namelen + 1);
+        ret = target_to_host_sockaddr(fd, host_msg->msg_name,
+                                      tswapal(target_msg->msg_name),
+                                      host_msg->msg_namelen);
         if (ret == -TARGET_EFAULT) {
-            /* For connected sockets msg_name and msg_namelen must
+            /*
+             * For connected sockets msg_name and msg_namelen must
              * be ignored, so returning EFAULT immediately is wrong.
              * Instead, pass a bad msg_name to the host kernel, and
              * let it decide whether to return EFAULT or not.
              */
-            msg.msg_name = (void *)-1;
+            host_msg->msg_name = (void *)-1;
+            ret = 0;
         } else if (ret) {
-            goto out2;
+            return ret;
         }
     } else {
-        msg.msg_name = NULL;
-        msg.msg_namelen = 0;
+        host_msg->msg_name = NULL;
+        host_msg->msg_namelen = 0;
+    }
+    host_msg->msg_controllen = 2 * tswapal(target_msg->msg_controllen);
+    host_msg->msg_control = alloca(host_msg->msg_controllen);
+    memset(host_msg->msg_control, 0, host_msg->msg_controllen);
+    host_msg->msg_flags = tswap32(target_msg->msg_flags);
+    if (send) {
+        if (fd_trans_target_to_host_data(fd)) {
+            void *host_iov_base;
+
+            host_iov_base = g_malloc(host_msg->msg_iov->iov_len);
+            memcpy(host_iov_base, host_msg->msg_iov->iov_base,
+                   host_msg->msg_iov->iov_len);
+            ret = fd_trans_target_to_host_data(fd)(host_iov_base,
+                                                   host_msg->msg_iov->iov_len);
+            if (ret >= 0) {
+                host_msg->msg_iov->iov_base = host_iov_base;
+                ret = 0;
+            }
+            g_free(host_iov_base);
+        } else {
+            ret = target_to_host_cmsg(host_msg, target_msg);
+        }
+    }
+    return ret;
+}
+
+static abi_long host_to_target_msghdr(int fd, struct target_msghdr *target_msg,
+                                      struct msghdr *host_msg, int len)
+{
+    abi_long ret;
+
+    if (fd_trans_host_to_target_data(fd)) {
+        ret = fd_trans_host_to_target_data(fd)(host_msg->msg_iov->iov_base,
+                                      MIN(host_msg->msg_iov->iov_len, len));
+    } else {
+        ret = host_to_target_cmsg(target_msg, host_msg);
+    }
+    if (!is_error(ret)) {
+        target_msg->msg_namelen = tswap32(host_msg->msg_namelen);
+        target_msg->msg_flags = tswap32(host_msg->msg_flags);
+        if (host_msg->msg_name != NULL && host_msg->msg_name != (void *)-1) {
+            ret = host_to_target_sockaddr(tswapal(target_msg->msg_name),
+                              host_msg->msg_name, host_msg->msg_namelen);
+            if (ret) {
+                return ret;
+            }
+        }
+        ret = len;
     }
-    msg.msg_controllen = 2 * tswapal(msgp->msg_controllen);
-    msg.msg_control = alloca(msg.msg_controllen);
-    memset(msg.msg_control, 0, msg.msg_controllen);
 
-    msg.msg_flags = tswap32(msgp->msg_flags);
+    return ret;
+}
+
+/* do_sendrecvmsg_locked() Must return target values and target errnos. */
+static abi_long do_sendrecvmsg_locked(int fd, struct target_msghdr *msgp,
+                                      int flags, int send)
+{
+    abi_long ret;
+    struct msghdr msg;
+    abi_ulong count;
+    struct iovec *vec;
+    abi_ulong target_vec;
 
     count = tswapal(msgp->msg_iovlen);
     target_vec = tswapal(msgp->msg_iov);
@@ -3120,48 +3177,18 @@ static abi_long do_sendrecvmsg_locked(int fd, struct target_msghdr *msgp,
     msg.msg_iovlen = count;
     msg.msg_iov = vec;
 
-    if (send) {
-        if (fd_trans_target_to_host_data(fd)) {
-            void *host_msg;
+    ret = target_to_host_msghdr(fd, &msg, msgp, send);
 
-            host_msg = g_malloc(msg.msg_iov->iov_len);
-            memcpy(host_msg, msg.msg_iov->iov_base, msg.msg_iov->iov_len);
-            ret = fd_trans_target_to_host_data(fd)(host_msg,
-                                                   msg.msg_iov->iov_len);
-            if (ret >= 0) {
-                msg.msg_iov->iov_base = host_msg;
-                ret = get_errno(safe_sendmsg(fd, &msg, flags));
-            }
-            g_free(host_msg);
-        } else {
-            ret = target_to_host_cmsg(&msg, msgp);
-            if (ret == 0) {
-                ret = get_errno(safe_sendmsg(fd, &msg, flags));
-            }
-        }
+    if (ret) {
+        goto out;
+    }
+
+    if (send) {
+        ret = get_errno(safe_sendmsg(fd, &msg, flags));
     } else {
         ret = get_errno(safe_recvmsg(fd, &msg, flags));
         if (!is_error(ret)) {
-            len = ret;
-            if (fd_trans_host_to_target_data(fd)) {
-                ret = fd_trans_host_to_target_data(fd)(msg.msg_iov->iov_base,
-                                               MIN(msg.msg_iov->iov_len, len));
-            } else {
-                ret = host_to_target_cmsg(msgp, &msg);
-            }
-            if (!is_error(ret)) {
-                msgp->msg_namelen = tswap32(msg.msg_namelen);
-                msgp->msg_flags = tswap32(msg.msg_flags);
-                if (msg.msg_name != NULL && msg.msg_name != (void *)-1) {
-                    ret = host_to_target_sockaddr(tswapal(msgp->msg_name),
-                                    msg.msg_name, msg.msg_namelen);
-                    if (ret) {
-                        goto out;
-                    }
-                }
-
-                ret = len;
-            }
+            ret = host_to_target_msghdr(fd, msgp, &msg, ret);
         }
     }
 
@@ -3188,50 +3215,98 @@ static abi_long do_sendrecvmsg(int fd, abi_ulong target_msg,
     return ret;
 }
 
-/* We don't rely on the C library to have sendmmsg/recvmmsg support,
- * so it might not have this *mmsg-specific flag either.
- */
-#ifndef MSG_WAITFORONE
-#define MSG_WAITFORONE 0x10000
-#endif
-
-static abi_long do_sendrecvmmsg(int fd, abi_ulong target_msgvec,
+static abi_long do_sendrecvmmsg(int fd, abi_ulong target_msgvec_addr,
                                 unsigned int vlen, unsigned int flags,
                                 int send)
 {
-    struct target_mmsghdr *mmsgp;
+    struct mmsghdr *host_msgvec;
+    struct target_mmsghdr *target_msgvec;
     abi_long ret = 0;
+    int num_sentrecv = 0;
+    int num_iovec = 0;
+    abi_ulong iovlen = 0;
+    struct iovec *host_iovec;
+    abi_ulong target_iovec;
     int i;
 
     if (vlen > UIO_MAXIOV) {
         vlen = UIO_MAXIOV;
     }
 
-    mmsgp = lock_user(VERIFY_WRITE, target_msgvec, sizeof(*mmsgp) * vlen, 1);
-    if (!mmsgp) {
+    target_msgvec = lock_user(VERIFY_WRITE, target_msgvec_addr,
+                              sizeof(target_msgvec) * vlen, 1);
+    if (!target_msgvec) {
         return -TARGET_EFAULT;
     }
 
+    host_msgvec = alloca(sizeof(*host_msgvec) * vlen);
+
     for (i = 0; i < vlen; i++) {
-        ret = do_sendrecvmsg_locked(fd, &mmsgp[i].msg_hdr, flags, send);
+
+        iovlen = tswapal(target_msgvec[i].msg_hdr.msg_iovlen);
+        target_iovec = tswapal(target_msgvec[i].msg_hdr.msg_iov);
+
+        if (iovlen > IOV_MAX) {
+            ret = -TARGET_EMSGSIZE;
+            break;
+         }
+         host_iovec = lock_iovec(send ? VERIFY_READ : VERIFY_WRITE,
+                                 target_iovec, iovlen, send);
+
+        if (host_iovec == NULL) {
+            ret = -host_to_target_errno(errno);
+            break;
+        }
+
+        num_iovec++;
+
+        host_msgvec[i].msg_hdr.msg_iovlen = iovlen;
+        host_msgvec[i].msg_hdr.msg_iov = host_iovec;
+
+        ret = target_to_host_msghdr(fd, &host_msgvec[i].msg_hdr,
+                                    &target_msgvec[i].msg_hdr, send);
+
         if (is_error(ret)) {
             break;
         }
-        mmsgp[i].msg_len = tswap32(ret);
-        /* MSG_WAITFORONE turns on MSG_DONTWAIT after one packet */
-        if (flags & MSG_WAITFORONE) {
-            flags |= MSG_DONTWAIT;
+    }
+
+    if (i > 0) {
+        if (send) {
+            ret = get_errno(safe_sendmmsg(fd, host_msgvec, i, flags));
+        } else {
+            ret = get_errno(safe_recvmmsg(fd, host_msgvec, i, flags, NULL));
+        }
+    }
+
+    if (!is_error(ret)) {
+        num_sentrecv = ret;
+        for (i = 0; i < num_sentrecv; i++) {
+            if (!send) {
+                ret = host_to_target_msghdr(fd, &target_msgvec[i].msg_hdr,
+                                            &host_msgvec[i].msg_hdr,
+                                            host_msgvec[i].msg_len);
+                if (is_error(ret)) {
+                    num_sentrecv = i;
+                    break;
+                }
+            }
+            target_msgvec[i].msg_len = tswap32(host_msgvec[i].msg_len);
         }
     }
 
-    unlock_user(mmsgp, target_msgvec, sizeof(*mmsgp) * i);
+    for (i = 0; i < num_iovec; i++) {
+        target_iovec = tswapal(target_msgvec[i].msg_hdr.msg_iov);
+        unlock_iovec(host_msgvec[i].msg_hdr.msg_iov,
+                     target_iovec, iovlen, !send);
+    }
+
+    unlock_user(target_msgvec, target_msgvec_addr, sizeof(*target_msgvec) * i);
 
-    /* Return number of datagrams sent if we sent any at all;
-     * otherwise return the error.
-     */
-    if (i) {
-        return i;
+    if (num_sentrecv) {
+        return num_sentrecv;
     }
+
     return ret;
 }
 
-- 
2.25.1

[PATCH 2/3] linux-user: Fix 'recvmmsg()' implementation

[Filip Bozuta]

Syscall 'recvmmsg()' takes an argument which is of type 'struct timespec'
that represents a timeout for the receive operation. Before changes from
this patch, the timeout argument was ignored. This was probably due to
the way 'recvmmsg()' was implemented with looping over 'recvmsg()' which
doesn't have the timeout argument. This was changed with the previous
patch in this series and which is why the timeout argument should be
added accordingly.

Implementation notes:

    Function 'do_sendrecvmmsg()' was changed with the addition of a new
    argument which represents the timeout. This argument is only passed
    in case of 'TARGET_NR_recvmmsg' and for 'TARGENT_NR_sendmmsg' 0 is
    passed. Function 'do_sendrecvmmsg()' was also updated accordingly in
    'do_socketcall()' for 'TARGET_SYS_recvmmsg' and 'TARGET_SYS_sendmmsg'.

Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
---
 linux-user/syscall.c | 22 +++++++++++++++-------
 1 file changed, 15 insertions(+), 7 deletions(-)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 8cbefdb561..420d7e7334 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -3217,10 +3217,11 @@ static abi_long do_sendrecvmsg(int fd, abi_ulong target_msg,
 
 static abi_long do_sendrecvmmsg(int fd, abi_ulong target_msgvec_addr,
                                 unsigned int vlen, unsigned int flags,
-                                int send)
+                                abi_long timeout, int send)
 {
     struct mmsghdr *host_msgvec;
     struct target_mmsghdr *target_msgvec;
+    struct timespec ts;
     abi_long ret = 0;
     int num_sentrecv = 0;
     int num_iovec = 0;
@@ -3275,7 +3276,14 @@ static abi_long do_sendrecvmmsg(int fd, abi_ulong target_msgvec_addr,
         if (send) {
             ret = get_errno(safe_sendmmsg(fd, host_msgvec, i, flags));
         } else {
-            ret = get_errno(safe_recvmmsg(fd, host_msgvec, i, flags, NULL));
+            if (timeout) {
+                if (target_to_host_timespec(&ts, timeout)) {
+                    return -TARGET_EFAULT;
+                }
+                ret = get_errno(safe_recvmmsg(fd, host_msgvec, i, flags, &ts));
+            } else {
+                ret = get_errno(safe_recvmmsg(fd, host_msgvec, i, flags, NULL));
+            }
         }
     }
 
@@ -3611,10 +3619,10 @@ static abi_long do_socketcall(int num, abi_ulong vptr)
         return do_sendrecvmsg(a[0], a[1], a[2], 0);
     case TARGET_SYS_ACCEPT4: /* sockfd, addr, addrlen, flags */
         return do_accept4(a[0], a[1], a[2], a[3]);
-    case TARGET_SYS_RECVMMSG: /* sockfd, msgvec, vlen, flags */
-        return do_sendrecvmmsg(a[0], a[1], a[2], a[3], 0);
+    case TARGET_SYS_RECVMMSG: /* sockfd, msgvec, vlen, timeout, flags */
+        return do_sendrecvmmsg(a[0], a[1], a[2], a[3], a[4], 0);
     case TARGET_SYS_SENDMMSG: /* sockfd, msgvec, vlen, flags */
-        return do_sendrecvmmsg(a[0], a[1], a[2], a[3], 1);
+        return do_sendrecvmmsg(a[0], a[1], a[2], a[3], 0, 1);
     default:
         qemu_log_mask(LOG_UNIMP, "Unsupported socketcall: %d\n", num);
         return -TARGET_EINVAL;
@@ -9418,11 +9426,11 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
 #endif
 #ifdef TARGET_NR_sendmmsg
     case TARGET_NR_sendmmsg:
-        return do_sendrecvmmsg(arg1, arg2, arg3, arg4, 1);
+        return do_sendrecvmmsg(arg1, arg2, arg3, arg4, 0, 1);
 #endif
 #ifdef TARGET_NR_recvmmsg
     case TARGET_NR_recvmmsg:
-        return do_sendrecvmmsg(arg1, arg2, arg3, arg4, 0);
+        return do_sendrecvmmsg(arg1, arg2, arg3, arg4, arg5, 0);
 #endif
 #ifdef TARGET_NR_sendto
     case TARGET_NR_sendto:
-- 
2.25.1

[PATCH 3/3] linux-user: Add support for 'recvmmsg_time64()'

[Filip Bozuta]

This patch introduces functionality for syscall:

*recvmmsg_time64

    This syscall is a 2038 safe variant for syscall:

    int recvmmsg(int sockfd, struct mmsghdr *msgvec, unsigned int vlen,
                 int flags, struct timespec *timeout)
    --receive multiple messages on a socket--
    man page: https://man7.org/linux/man-pages/man2/recvmmsg.2.html

Implementation notes:

    Function 'do_sendrecvmmsg()' in 'syscall.c' was changed with the
    addition of a new argument 'time64' which represents a flag by
    which the function knows what kind of 'struct timespec' converting
    function to call ('target_to_host_timespec() or
    'target_to_host_timespec64()'). It is 0 in case of 'TARGET_NR_recvmmsg'
    and 1 in case of 'TARGET_NR_recvmmsg_time64'.

    In 'do_socketcall()', the 'TARGET_ABI_BITS' was checked to know
    what value for 'time64' argument to pass (0 if 'TARGET_ABI_BITS == 32'
    and 1 otherwise).

Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
---
 linux-user/syscall.c | 25 ++++++++++++++++++-------
 1 file changed, 18 insertions(+), 7 deletions(-)

diff --git a/linux-user/syscall.c b/linux-user/syscall.c
index 420d7e7334..ed9c2cb7ab 100644
--- a/linux-user/syscall.c
+++ b/linux-user/syscall.c
@@ -3217,7 +3217,7 @@ static abi_long do_sendrecvmsg(int fd, abi_ulong target_msg,
 
 static abi_long do_sendrecvmmsg(int fd, abi_ulong target_msgvec_addr,
                                 unsigned int vlen, unsigned int flags,
-                                abi_long timeout, int send)
+                                abi_long timeout, int time64, int send)
 {
     struct mmsghdr *host_msgvec;
     struct target_mmsghdr *target_msgvec;
@@ -3277,7 +3277,10 @@ static abi_long do_sendrecvmmsg(int fd, abi_ulong target_msgvec_addr,
             ret = get_errno(safe_sendmmsg(fd, host_msgvec, i, flags));
         } else {
             if (timeout) {
-                if (target_to_host_timespec(&ts, timeout)) {
+                if (!time64 && target_to_host_timespec(&ts, timeout)) {
+                    return -TARGET_EFAULT;
+                }
+                if (time64 && target_to_host_timespec64(&ts, timeout)) {
                     return -TARGET_EFAULT;
                 }
                 ret = get_errno(safe_recvmmsg(fd, host_msgvec, i, flags, &ts));
@@ -3619,10 +3622,14 @@ static abi_long do_socketcall(int num, abi_ulong vptr)
         return do_sendrecvmsg(a[0], a[1], a[2], 0);
     case TARGET_SYS_ACCEPT4: /* sockfd, addr, addrlen, flags */
         return do_accept4(a[0], a[1], a[2], a[3]);
-    case TARGET_SYS_RECVMMSG: /* sockfd, msgvec, vlen, timeout, flags */
-        return do_sendrecvmmsg(a[0], a[1], a[2], a[3], a[4], 0);
+    case TARGET_SYS_RECVMMSG: /* sockfd, msgvec, vlen, flags, timeout, */
+#if TARGET_ABI_BITS == 32
+        return do_sendrecvmmsg(a[0], a[1], a[2], a[3], a[4], 0, 0);
+#else
+        return do_sendrecvmmsg(a[0], a[1], a[2], a[3], a[4], 1, 0);
+#endif
     case TARGET_SYS_SENDMMSG: /* sockfd, msgvec, vlen, flags */
-        return do_sendrecvmmsg(a[0], a[1], a[2], a[3], 0, 1);
+        return do_sendrecvmmsg(a[0], a[1], a[2], a[3], 0, 0, 1);
     default:
         qemu_log_mask(LOG_UNIMP, "Unsupported socketcall: %d\n", num);
         return -TARGET_EINVAL;
@@ -9426,11 +9433,15 @@ static abi_long do_syscall1(void *cpu_env, int num, abi_long arg1,
 #endif
 #ifdef TARGET_NR_sendmmsg
     case TARGET_NR_sendmmsg:
-        return do_sendrecvmmsg(arg1, arg2, arg3, arg4, 0, 1);
+        return do_sendrecvmmsg(arg1, arg2, arg3, arg4, 0, 0, 1);
 #endif
 #ifdef TARGET_NR_recvmmsg
     case TARGET_NR_recvmmsg:
-        return do_sendrecvmmsg(arg1, arg2, arg3, arg4, arg5, 0);
+        return do_sendrecvmmsg(arg1, arg2, arg3, arg4, arg5, 0, 0);
+#endif
+#ifdef TARGET_NR_recvmmsg_time64
+    case TARGET_NR_recvmmsg_time64:
+        return do_sendrecvmmsg(arg1, arg2, arg3, arg4, arg5, 1, 0);
 #endif
 #ifdef TARGET_NR_sendto
     case TARGET_NR_sendto:
-- 
2.25.1

Re: [PATCH 1/3] linux-user: Modify 'sendmmsg()' and 'recvmmsg()' implementation

[Laurent Vivier]

Le 31/07/2020 à 21:06, Filip Bozuta a écrit :
> Implementations of 'sendmmsg()' and 'recvmmsg()' in 'syscall.c' use
> a loop over a host command of 'sendmsg()' and 'recvmsg()' respectively
> to send/receive individual messages from a socket. This patch changes
> these implementations to use the host commands 'sendmmsg()' and 'recvmmsg()'
> to send all messages without looping over 'sendmsg()' and 'recvmsg()'.
> 
> Implementation notes:
> 
>     Parts of code from 'do_sendrecvmsg_locked()', that are used to transfer
>     values of 'struct msghdr' between host and target, were moved to separate
>     functions 'target_to_host_msghdr()' and 'host_to_target_msghdr()'. These
>     functions are used in 'do_sendrecvmmsg()' to transfer the data of each
>     individual 'struct msghdr' from the 'msgvec' argument. Memory allocation
>     for the 'iovec' field is done outside of these functions as to ensure that
>     the memory is freed after the syscall execution.
> 
> Signed-off-by: Filip Bozuta <Filip.Bozuta@syrmia.com>
> ---
>  linux-user/syscall.c | 243 ++++++++++++++++++++++++++++---------------
>  1 file changed, 159 insertions(+), 84 deletions(-)
> 

I'm sorry but after studying the changes needed I think it's better to
keep the existing code.

For instance, if we have a EFAULT (or EMSGSIZE) while reading the iovec
we must stop to send  data and exit with the error code.

Your code correctly manages the detection of the error and stops the
conversion of the iovec. Then it uses the converted iovecs with
sendmmsg() and recvmmsg(), but "ret" is overwritten and the error is lost.

So, in the end, and as it is done in kernel, the best is to loop around
sendmsg()... and this is what the existing code does.

Thanks,
Laurent