Introduce the new nonblock API.

This patch is an attempt to create a better API for reads from
non-blocking file descriptors. It adds readv_nonblock() and implements
read_nonblock() as a simple wrapper for readv_nonblock(). Both new
functions check the given file descriptor for readability and read
as much as possible until an error occurs or the buffer is full.

Two additional parameters are introduced: An fd_set and a result
pointer for the number of bytes that have been read successfully. The
optional fd_set pointer is used to have the fixup code for unreliable
returns of FD_ISSET() at one place only.

Having an extra parameter for storing the number of bytes read allows
to treat EOF as an error condition but EAGAIN as a normal condition.
This will simplify callers (dccp_recv and upd_recv) a bit.

diff --git a/error.h b/error.h
index 3ad88cd4fab44cf7a2527bdcbf905b7d2eaf82da..e30e8a284c35c6d8b16b5fc5c8d7702540c9d989 100644 (file)
--- a/error.h
+++ b/error.h
@@ -374,6 +374,7 @@ extern const char **para_errlist[];
 
 #define FD_ERRORS \
        PARA_ERROR(FGETS, "fgets error"), \
+       PARA_ERROR(EOF, "end of file"), \
 
 
 #define WRITE_ERRORS \

diff --git a/fd.c b/fd.c
index 46be22891e93f2f094f3118daef5780c067b4cea..d56a31101ade30bd8790d8082f8a3be5f5b84004 100644 (file)
--- a/fd.c
+++ b/fd.c
@@ -11,12 +11,12 @@
 #include <dirent.h>
 #include <sys/mman.h>
 #include <fcntl.h>
-#include <sys/select.h>
 #include <sys/uio.h>
 
 #include "para.h"
 #include "error.h"
 #include "string.h"
+#include "fd.h"
 
 /**
  * Write a buffer to a file descriptor, re-write on short writes.
@@ -82,6 +82,101 @@ int write_nonblock(int fd, const char *buf, size_t len,
        return written;
 }
 
+/**
+ * Read from a non-blocking file descriptor into multiple buffers.
+ *
+ * \param fd The file descriptor to read from.
+ * \param iov Scatter/gather array used in readv().
+ * \param iovcnt Number of elements in \a iov.
+ * \param rfds An optional fd set pointer.
+ * \param num_bytes Result pointer. Contains the number of bytes read from \a fd.
+ *
+ * If \a rfds is not \p NULL and the (non-blocking) file descriptor \a fd is
+ * not set in \a rfds, this function returns early without doing anything.
+ * Otherwise The function tries to read up to \a sz bytes from \a fd. As for
+ * write_nonblock(), EAGAIN is not considered an error condition. However, EOF
+ * is.
+ *
+ * \return Zero or a negative error code. If the underlying call to readv(2)
+ * returned zero (indicating an end of file condition) or failed for some
+ * reason other than \p EAGAIN, a negative return value is returned.
+ *
+ * In any case, \a num_bytes contains the number of bytes that have been
+ * successfully read from \a fd (zero if the first readv() call failed with
+ * EAGAIN). Note that even if the function returns negative, some data might
+ * have been read before the error occured. In this case \a num_bytes is
+ * positive.
+ *
+ * \sa \ref write_nonblock(), read(2), readv(2).
+ */
+int readv_nonblock(int fd, struct iovec *iov, int iovcnt, fd_set *rfds,
+               size_t *num_bytes)
+{
+       int ret, i, j;
+
+       *num_bytes = 0;
+       /*
+        * Avoid a shortcoming of select(): Reads from a non-blocking fd might
+        * return EAGAIN even if FD_ISSET() returns true. However, FD_ISSET()
+        * returning false definitely means that no data can currently be read.
+        * This is the common case, so it is worth to avoid the overhead of the
+        * read() system call in this case.
+        */
+       if (rfds && !FD_ISSET(fd, rfds))
+               return 0;
+
+       for (i = 0, j = 0; i < iovcnt;) {
+
+               /* fix up the first iov */
+               assert(j < iov[i].iov_len);
+               iov[i].iov_base += j;
+               iov[i].iov_len -= j;
+               ret = readv(fd, iov + i, iovcnt - i);
+               iov[i].iov_base -= j;
+               iov[i].iov_len += j;
+
+               if (ret == 0)
+                       return -E_EOF;
+               if (ret < 0) {
+                       if (errno == EAGAIN)
+                               return 0;
+                       return -ERRNO_TO_PARA_ERROR(errno);
+               }
+               *num_bytes += ret;
+               while (ret > 0) {
+                       if (ret < iov[i].iov_len - j) {
+                               j += ret;
+                               break;
+                       }
+                       ret -= iov[i].iov_len - j;
+                       j = 0;
+                       if (++i >= iovcnt)
+                               break;
+               }
+       }
+       return 0;
+}
+
+/**
+ * Read from a non-blocking file descriptor into a single buffer.
+ *
+ * \param fd The file descriptor to read from.
+ * \param buf The buffer to read data to.
+ * \param sz The size of \a buf.
+ * \param rfds \see \ref readv_nonblock().
+ * \param num_bytes \see \ref readv_nonblock().
+ *
+ * This is a simple wrapper for readv_nonblock() which uses an iovec with a single
+ * buffer.
+ *
+ * \return The return value of the underlying call to readv_nonblock().
+ */
+int read_nonblock(int fd, void *buf, size_t sz, fd_set *rfds, size_t *num_bytes)
+{
+       struct iovec iov = {.iov_base = buf, .iov_len = sz};
+       return readv_nonblock(fd, &iov, 1, rfds, num_bytes);
+}
+
 /**
  * Simple wrapper for readv().
  *

diff --git a/fd.h b/fd.h
index 680923087e5acbf94838e82c1576c10d62af9242..b95588207d4c69e303409184e7a91e52df643e00 100644 (file)
--- a/fd.h
+++ b/fd.h
@@ -27,6 +27,9 @@ int mmap_full_file(const char *filename, int open_mode, void **map,
 int para_munmap(void *start, size_t length);
 int write_ok(int fd);
 void valid_fd_012(void);
+int readv_nonblock(int fd, struct iovec *iov, int iovcnt, fd_set *rfds,
+               size_t *num_bytes);
+int read_nonblock(int fd, void *buf, size_t sz, fd_set *rfds, size_t *num_bytes);
 int write_nonblock(int fd, const char *buf, size_t len,
                size_t max_bytes_per_write);
 int for_each_file_in_dir(const char *dirname,