wasip2 support for close, poll, pselect

This enables `wasm32-wasip2` support for `close`, `poll`, and `pselect`.  I
cheated a bit for the latter by re-implementing `pselect` in terms of `poll` to
avoid having to implement wasip2 versions of both.

Signed-off-by: Joel Dice <[email protected]>

Makefile

@@ -91,7 +91,8 @@ LIBC_BOTTOM_HALF_OMIT_SOURCES := \
 	$(LIBC_BOTTOM_HALF_SOURCES)/listen.c \
 	$(LIBC_BOTTOM_HALF_SOURCES)/accept-wasip2.c \
 	$(LIBC_BOTTOM_HALF_SOURCES)/shutdown.c \
-	$(LIBC_BOTTOM_HALF_SOURCES)/sockopt.c
+	$(LIBC_BOTTOM_HALF_SOURCES)/sockopt.c \
+	$(LIBC_BOTTOM_HALF_SOURCES)/poll-wasip2.c
 LIBC_BOTTOM_HALF_ALL_SOURCES := $(filter-out $(LIBC_BOTTOM_HALF_OMIT_SOURCES),$(LIBC_BOTTOM_HALF_ALL_SOURCES))
 # Omit p2-specific headers from include-all.c test.
 INCLUDE_ALL_CLAUSES := -not -name wasip2.h -not -name descriptor_table.h

expected/wasm32-wasip2/defined-symbols.txt

@@ -1012,6 +1012,7 @@ poll_method_pollable_ready
 poll_poll
 poll_pollable_drop_borrow
 poll_pollable_drop_own
+poll_wasip2
 posix_close
 posix_fadvise
 posix_fallocate

libc-bottom-half/cloudlibc/src/libc/poll/poll.c

@@ -7,7 +7,7 @@
 #include <poll.h>
 #include <stdbool.h>
 
-int poll(struct pollfd *fds, size_t nfds, int timeout) {
+static int poll_wasip1(struct pollfd *fds, size_t nfds, int timeout) {
   // Construct events for poll().
   size_t maxevents = 2 * nfds + 1;
   __wasi_subscription_t subscriptions[maxevents];
@@ -127,3 +127,48 @@ int poll(struct pollfd *fds, size_t nfds, int timeout) {
   }
   return retval;
 }
+
+#ifdef __wasilibc_use_wasip2
+#include <wasi/descriptor_table.h>
+
+int poll_wasip2(struct pollfd *fds, size_t nfds, int timeout);
+
+int poll(struct pollfd* fds, nfds_t nfds, int timeout)
+{
+    bool found_socket = false;
+    bool found_non_socket = false;
+    for (size_t i = 0; i < nfds; ++i) {
+        descriptor_table_entry_t* entry;
+        if (descriptor_table_get_ref(fds[i].fd, &entry)) {
+            found_socket = true;
+        } else {
+            found_non_socket = true;
+        }
+    }
+
+    if (found_socket) {
+        if (found_non_socket) {
+            // We currently don't support polling a mix of non-sockets and
+            // sockets here (though you can do it by using the host APIs
+            // directly), and we probably won't until we've migrated entirely to
+            // WASI 0.2.
+            errno = ENOTSUP;
+            return -1;
+        }
+
+        return poll_wasip2(fds, nfds, timeout);
+    } else if (found_non_socket) {
+        return poll_wasip1(fds, nfds, timeout);
+    } else if (timeout >= 0) {
+        return poll_wasip2(fds, nfds, timeout);
+    } else {
+        errno = ENOTSUP;
+        return -1;
+    }
+}
+#else // not __wasilibc_use_wasip2
+int poll(struct pollfd* fds, nfds_t nfds, int timeout)
+{
+    return poll_wasip1(fds, nfds, timeout);
+}
+#endif // not __wasilibc_use_wasip2

libc-bottom-half/cloudlibc/src/libc/sys/select/pselect.c

@@ -8,6 +8,7 @@
 
 #include <wasi/api.h>
 #include <errno.h>
+#include <poll.h>
 
 int pselect(int nfds, fd_set *restrict readfds, fd_set *restrict writefds,
             fd_set *restrict errorfds, const struct timespec *restrict timeout,
@@ -33,93 +34,66 @@ int pselect(int nfds, fd_set *restrict readfds, fd_set *restrict writefds,
   if (writefds == NULL)
     writefds = &empty;
 
-  // Determine the maximum number of events.
-  size_t maxevents = readfds->__nfds + writefds->__nfds + 1;
-  __wasi_subscription_t subscriptions[maxevents];
-  size_t nsubscriptions = 0;
-
-  // Convert the readfds set.
+  struct pollfd poll_fds[readfds->__nfds + writefds->__nfds];
+  size_t poll_nfds = 0;
+
   for (size_t i = 0; i < readfds->__nfds; ++i) {
     int fd = readfds->__fds[i];
     if (fd < nfds) {
-      __wasi_subscription_t *subscription = &subscriptions[nsubscriptions++];
-      *subscription = (__wasi_subscription_t){
-          .userdata = fd,
-          .u.tag = __WASI_EVENTTYPE_FD_READ,
-          .u.u.fd_read.file_descriptor = fd,
-      };
+        poll_fds[poll_nfds++] = (struct pollfd){
+            .fd = fd,
+            .events = POLLRDNORM,
+            .revents = 0
+        };
     }
   }
-
-  // Convert the writefds set.
+
   for (size_t i = 0; i < writefds->__nfds; ++i) {
     int fd = writefds->__fds[i];
     if (fd < nfds) {
-      __wasi_subscription_t *subscription = &subscriptions[nsubscriptions++];
-      *subscription = (__wasi_subscription_t){
-          .userdata = fd,
-          .u.tag = __WASI_EVENTTYPE_FD_WRITE,
-          .u.u.fd_write.file_descriptor = fd,
+      poll_fds[poll_nfds++] = (struct pollfd){
+          .fd = fd,
+          .events = POLLWRNORM,
+          .revents = 0
       };
     }
   }
 
-  // Create extra event for the timeout.
-  if (timeout != NULL) {
-    __wasi_subscription_t *subscription = &subscriptions[nsubscriptions++];
-    *subscription = (__wasi_subscription_t){
-        .u.tag = __WASI_EVENTTYPE_CLOCK,
-        .u.u.clock.id = __WASI_CLOCKID_REALTIME,
-    };
-    if (!timespec_to_timestamp_clamp(timeout, &subscription->u.u.clock.timeout)) {
+  int poll_timeout;
+  if (timeout) {
+    uint64_t timeout_u64;
+    if (!timespec_to_timestamp_clamp(timeout, &timeout_u64) ) {
       errno = EINVAL;
       return -1;
     }
-  }
 
-  // Execute poll().
-  size_t nevents;
-  __wasi_event_t events[nsubscriptions];
-  __wasi_errno_t error =
-      __wasi_poll_oneoff(subscriptions, events, nsubscriptions, &nevents);
-  if (error != 0) {
-    // WASI's poll requires at least one subscription, or else it returns
-    // `EINVAL`. Since a `pselect` with nothing to wait for is valid in POSIX,
-    // return `ENOTSUP` to indicate that we don't support that case.
-    //
-    // Wasm has no signal handling, so if none of the user-provided `pollfd`
-    // elements, nor the timeout, led us to producing even one subscription
-    // to wait for, there would be no way for the poll to wake up. WASI
-    // returns `EINVAL` in this case, but for users of `poll`, `ENOTSUP` is
-    // more likely to be understood.
-    if (nsubscriptions == 0)
-      errno = ENOTSUP;
-    else
-      errno = error;
-    return -1;
-  }
+    // Convert nanoseconds to milliseconds:
+    timeout_u64 /= 1000000;
 
-  // Test for EBADF.
-  for (size_t i = 0; i < nevents; ++i) {
-    const __wasi_event_t *event = &events[i];
-    if ((event->type == __WASI_EVENTTYPE_FD_READ ||
-         event->type == __WASI_EVENTTYPE_FD_WRITE) &&
-        event->error == __WASI_ERRNO_BADF) {
-      errno = EBADF;
-      return -1;
+    if (timeout_u64 > INT_MAX) {
+      timeout_u64 = INT_MAX;
     }
+
+    poll_timeout = (int) timeout_u64;
+  } else {
+    poll_timeout = -1;
+  };
+
+  if (poll(poll_fds, poll_nfds, poll_timeout) < 0) {
+    return -1;
   }
 
-  // Clear and set entries in the result sets.
   FD_ZERO(readfds);
   FD_ZERO(writefds);
-  for (size_t i = 0; i < nevents; ++i) {
-    const __wasi_event_t *event = &events[i];
-    if (event->type == __WASI_EVENTTYPE_FD_READ) {
-      readfds->__fds[readfds->__nfds++] = event->userdata;
-    } else if (event->type == __WASI_EVENTTYPE_FD_WRITE) {
-      writefds->__fds[writefds->__nfds++] = event->userdata;
+  for (size_t i = 0; i < poll_nfds; ++i) {
+    struct pollfd* pollfd = poll_fds + i;
+    if ((pollfd->revents & POLLRDNORM) != 0) {
+      readfds->__fds[readfds->__nfds++] = pollfd->fd;
+    }
+    if ((pollfd->revents & POLLWRNORM) != 0) {
+      writefds->__fds[writefds->__nfds++] = pollfd->fd;
     }
   }
+
   return readfds->__nfds + writefds->__nfds;
 }

libc-bottom-half/sources/__wasilibc_fd_renumber.c

@@ -15,10 +15,85 @@ int __wasilibc_fd_renumber(int fd, int newfd) {
     return 0;
 }
 
+#ifdef __wasilibc_use_wasip2
+#include <wasi/descriptor_table.h>
+
+void drop_tcp_socket(tcp_socket_t socket) {
+    switch (socket.state.tag) {
+    case TCP_SOCKET_STATE_UNBOUND:
+    case TCP_SOCKET_STATE_BOUND:
+    case TCP_SOCKET_STATE_CONNECTING:
+    case TCP_SOCKET_STATE_LISTENING:
+    case TCP_SOCKET_STATE_CONNECT_FAILED:
+        // No additional resources to drop.
+        break;
+    case TCP_SOCKET_STATE_CONNECTED: {
+        tcp_socket_state_connected_t connection = socket.state.connected;
+
+        poll_pollable_drop_own(connection.input_pollable);
+        poll_pollable_drop_own(connection.output_pollable);
+        streams_input_stream_drop_own(connection.input);
+        streams_output_stream_drop_own(connection.output);
+        break;
+    }
+    default: /* unreachable */ abort();
+    }
+
+    poll_pollable_drop_own(socket.socket_pollable);
+    tcp_tcp_socket_drop_own(socket.socket);
+}
+
+void drop_udp_socket_streams(udp_socket_streams_t streams) {
+    poll_pollable_drop_own(streams.incoming_pollable);
+    poll_pollable_drop_own(streams.outgoing_pollable);
+    udp_incoming_datagram_stream_drop_own(streams.incoming);
+    udp_outgoing_datagram_stream_drop_own(streams.outgoing);
+}
+
+void drop_udp_socket(udp_socket_t socket) {
+    switch (socket.state.tag) {
+    case UDP_SOCKET_STATE_UNBOUND:
+    case UDP_SOCKET_STATE_BOUND_NOSTREAMS:
+        // No additional resources to drop.
+        break;
+    case UDP_SOCKET_STATE_BOUND_STREAMING:
+        drop_udp_socket_streams(socket.state.bound_streaming.streams);
+        break;
+    case UDP_SOCKET_STATE_CONNECTED: {
+        drop_udp_socket_streams(socket.state.connected.streams);
+        break;
+    }
+    default: /* unreachable */ abort();
+    }
+
+    poll_pollable_drop_own(socket.socket_pollable);
+    udp_udp_socket_drop_own(socket.socket);
+}
+#endif // __wasilibc_use_wasip2
+
 int close(int fd) {
     // Scan the preopen fds before making any changes.
     __wasilibc_populate_preopens();
 
+#ifdef __wasilibc_use_wasip2
+    descriptor_table_entry_t entry;
+    if (descriptor_table_remove(fd, &entry)) {
+
+        switch (entry.tag)
+        {
+        case DESCRIPTOR_TABLE_ENTRY_TCP_SOCKET:
+            drop_tcp_socket(entry.tcp_socket);
+            break;
+        case DESCRIPTOR_TABLE_ENTRY_UDP_SOCKET:
+            drop_udp_socket(entry.udp_socket);
+            break;
+        default: /* unreachable */ abort();
+        }
+
+        return 0;
+    }
+#endif // __wasilibc_use_wasip2
+
     __wasi_errno_t error = __wasi_fd_close(fd);
     if (error != 0) {
         errno = error;