[PATCH] implement linux_sendfile support

[PATCH] implement linux_sendfile support

[Eric Wong]

This will also allow non-Linux users to use sendfile
if it is not available.
---
 ext/sleepy_penguin/extconf.rb |   1 +
 ext/sleepy_penguin/init.c     |   2 +
 ext/sleepy_penguin/sendfile.c | 120 ++++++++++++++++++++++++++++++++++++++++++
 lib/sleepy_penguin.rb         |  19 +++++++
 test/test_sendfile.rb         |  28 ++++++++++
 5 files changed, 170 insertions(+)
 create mode 100644 ext/sleepy_penguin/sendfile.c
 create mode 100644 test/test_sendfile.rb

diff --git a/ext/sleepy_penguin/extconf.rb b/ext/sleepy_penguin/extconf.rb
index b1a95c4..e07f5d2 100644
--- a/ext/sleepy_penguin/extconf.rb
+++ b/ext/sleepy_penguin/extconf.rb
@@ -6,6 +6,7 @@ if have_header('sys/event.h')
 end
 have_header('sys/mount.h')
 have_header('sys/eventfd.h')
+have_header('sys/sendfile.h')
 
 # it's impossible to use signalfd reliably with Ruby since Ruby currently
 # manages # (and overrides) all signal handling
diff --git a/ext/sleepy_penguin/init.c b/ext/sleepy_penguin/init.c
index 0a1458b..8295073 100644
--- a/ext/sleepy_penguin/init.c
+++ b/ext/sleepy_penguin/init.c
@@ -54,6 +54,7 @@ void sleepy_penguin_init_signalfd(void);
 
 void sleepy_penguin_init_splice(void);
 void sleepy_penguin_init_cfr(void);
+void sleepy_penguin_init_sendfile(void);
 
 static size_t l1_cache_line_size_detect(void)
 {
@@ -132,4 +133,5 @@ void Init_sleepy_penguin_ext(void)
 	sleepy_penguin_init_signalfd();
 	sleepy_penguin_init_splice();
 	sleepy_penguin_init_cfr();
+	sleepy_penguin_init_sendfile();
 }
diff --git a/ext/sleepy_penguin/sendfile.c b/ext/sleepy_penguin/sendfile.c
new file mode 100644
index 0000000..887dfb6
--- /dev/null
+++ b/ext/sleepy_penguin/sendfile.c
@@ -0,0 +1,120 @@
+#include "sleepy_penguin.h"
+#include <sys/types.h>
+#include <sys/socket.h>
+#include <sys/uio.h>
+
+#if defined(HAVE_SYS_SENDFILE_H) && !defined(HAVE_BSD_SENDFILE)
+#  include <sys/sendfile.h>
+#endif
+
+#if defined(__linux__) && defined(HAVE_SENDFILE)
+#  define linux_sendfile(in_fd, out_fd, offset, count) \
+		sendfile((in_fd),(out_fd),(offset),(count))
+
+/* all good */
+#elif defined(HAVE_SENDFILE) && \
+      (defined(__FreeBSD__) || defined(__DragonFly__))
+/*
+ * make BSD sendfile look like Linux for now...
+ * we can support SF_NODISKIO later
+ */
+static ssize_t linux_sendfile(int sockfd, int filefd, off_t *off, size_t count)
+{
+	off_t sbytes = 0;
+        off_t offset = off ? *off : lseek(filefd, 0, SEEK_CUR);
+
+	int rc = sendfile(filefd, sockfd, offset, count, NULL, &sbytes, 0);
+	if (sbytes > 0) {
+		if (off)
+			*off += sbytes;
+		else
+			lseek(filefd, sbytes, SEEK_CUR);
+		return (ssize_t)sbytes;
+	}
+
+	return (ssize_t)rc;
+}
+#else /* emulate sendfile using (read|pread) + write */
+static ssize_t pread_sendfile(int sockfd, int filefd, off_t *off, size_t count)
+{
+	size_t max_read = 16384;
+	void *buf;
+	ssize_t r;
+	ssize_t w;
+
+	max_read = count > max_read ? max_read : count;
+	buf = xmalloc(max_read);
+
+	do {
+		r = off ? pread(filefd, buf, max_read, *off) :
+			  read(filefd, buf, max_read);
+	} while (r < 0 && errno == EINTR);
+
+	if (r <= 0) {
+		int err = errno;
+		xfree(buf);
+		errno = err;
+		return r;
+	}
+	w = write(sockfd, buf, r);
+	if (w > 0 && off)
+		*off += w;
+	xfree(buf);
+	return w;
+}
+#    define linux_sendfile(out_fd, in_fd, offset, count) \
+            pread_sendfile((out_fd),(in_fd),(offset),(count))
+#endif
+
+struct sf_args {
+	int dst_fd;
+	int src_fd;
+	off_t *off;
+	size_t count;
+};
+
+static VALUE sym_wait_writable;
+
+static VALUE nogvl_sf(void *ptr)
+{
+	struct sf_args *a = ptr;
+
+	return (VALUE)linux_sendfile(a->dst_fd, a->src_fd, a->off, a->count);
+}
+
+static VALUE lsf(VALUE mod, VALUE dst, VALUE src, VALUE src_off, VALUE count)
+{
+	off_t off = 0;
+	struct sf_args a;
+	ssize_t bytes;
+	int retried = 0;
+
+	a.off = NIL_P(src_off) ? NULL : (off = NUM2OFFT(src_off), &off);
+	a.count = NUM2SIZET(count);
+again:
+	a.src_fd = rb_sp_fileno(src);
+	a.dst_fd = rb_sp_fileno(dst);
+	bytes = (ssize_t)rb_sp_fd_region(nogvl_sf, &a, a.dst_fd);
+	if (bytes < 0) {
+		switch (errno) {
+		case EAGAIN:
+			return sym_wait_writable;
+		case ENOMEM:
+		case ENOBUFS:
+			if (!retried) {
+				rb_gc();
+				retried = 1;
+				goto again;
+			}
+		}
+		rb_sys_fail("sendfile");
+	}
+	return SSIZET2NUM(bytes);
+}
+
+void sleepy_penguin_init_sendfile(void)
+{
+	VALUE m = rb_define_module("SleepyPenguin");
+	rb_define_singleton_method(m, "__lsf", lsf, 4);
+	sym_wait_writable = ID2SYM(rb_intern("wait_writable"));
+}
diff --git a/lib/sleepy_penguin.rb b/lib/sleepy_penguin.rb
index 07c431b..5722f2a 100644
--- a/lib/sleepy_penguin.rb
+++ b/lib/sleepy_penguin.rb
@@ -19,4 +19,23 @@ end
 module SleepyPenguin
   require_relative 'sleepy_penguin/splice' if respond_to?(:__splice)
   require_relative 'sleepy_penguin/cfr' if respond_to?(:__cfr)
+
+  # Copies +len+ bytes from +src+ to +dst+, where +src+ refers to
+  # an open, mmap(2)-able File and +dst+ refers to a Socket.
+  # An optional +offset+ keyword may be specified for the +src+ File.
+  # Using +offset+ will not adjust the offset of the underlying file
+  # handle itself; in other words: this allows concurrent threads to
+  # use linux_sendfile to write data from one open file to multiple
+  # sockets.
+  #
+  # Returns the number of bytes written on success, or :wait_writable
+  # if the +dst+ Socket is non-blocking and the operation would block.
+  # A return value of zero bytes indicates EOF is reached on the +src+
+  # file.
+  #
+  # Newer OSes may be more flexible in whether or not +dst+ or +src+
+  # is a regular file or socket, respectively.
+  def self.linux_sendfile(dst, src, len, offset: nil)
+    __lsf(dst, src, offset, len)
+  end
 end
diff --git a/test/test_sendfile.rb b/test/test_sendfile.rb
new file mode 100644
index 0000000..a25b711
--- /dev/null
+++ b/test/test_sendfile.rb
@@ -0,0 +1,28 @@
+# -*- encoding: binary -*-
+require 'test/unit'
+require 'tempfile'
+require 'socket'
+$-w = true
+require 'sleepy_penguin'
+
+class TestSendfile < Test::Unit::TestCase
+  def test_linux_sendfile
+    rd, wr = UNIXSocket.pair
+    size = 5
+    src = Tempfile.new('ruby_sf_src')
+    assert_equal 0, SleepyPenguin.linux_sendfile(wr, src, size)
+    str = 'abcde'.freeze
+    assert_equal str.bytesize, src.syswrite(str)
+    assert_equal 0, SleepyPenguin.linux_sendfile(wr, src, size)
+    src.sysseek(0, IO::SEEK_SET)
+    assert_equal str.bytesize,
+                 SleepyPenguin.linux_sendfile(wr, src, size, offset: 0)
+    assert_equal str, rd.read(size)
+    assert_equal 0, src.sysseek(0, IO::SEEK_CUR), 'handle offset not changed'
+    assert_equal 3, SleepyPenguin.linux_sendfile(wr, src, 3)
+    assert_equal 3, src.sysseek(0, IO::SEEK_CUR), 'handle offset changed'
+  ensure
+    [ rd, wr ].compact.each(&:close)
+    src.close! if src
+  end
+end
-- 
EW

[PATCH 2/1] portability fixes for systems w/o splice, copy_file_range

[Eric Wong]

We need to support FreeBSD, at least.
---
 ext/sleepy_penguin/cfr.c            | 14 +++-----------
 ext/sleepy_penguin/init.c           | 12 ++++++++++++
 ext/sleepy_penguin/sleepy_penguin.h | 11 +++++++++++
 test/test_splice.rb                 |  2 +-
 test/test_splice_eintr.rb           |  2 +-
 5 files changed, 28 insertions(+), 13 deletions(-)

diff --git a/ext/sleepy_penguin/cfr.c b/ext/sleepy_penguin/cfr.c
index acece4f..860c5f5 100644
--- a/ext/sleepy_penguin/cfr.c
+++ b/ext/sleepy_penguin/cfr.c
@@ -2,17 +2,6 @@
 #include "sp_copy.h"
 #include <unistd.h>
 
-#ifndef HAVE_COPY_FILE_RANGE
-#  include <sys/syscall.h>
-#  if !defined(__NR_copy_file_range)
-#    if defined(__x86_64__)
-#      define __NR_copy_file_range 326
-#    elif defined(__i386__)
-#      define __NR_copy_file_range 377
-#    endif /* supported arches */
-#  endif /* __NR_copy_file_range */
-#endif
-
 #ifdef __NR_copy_file_range
 static ssize_t my_cfr(int fd_in, off_t *off_in, int fd_out, off_t *off_out,
 		       size_t len, unsigned int flags)
@@ -26,6 +15,8 @@ static ssize_t my_cfr(int fd_in, off_t *off_in, int fd_out, off_t *off_out,
 		my_cfr((fd_in),(off_in),(fd_out),(off_out),(len),(flags))
 #endif
 
+#if defined(HAVE_COPY_FILE_RANGE) || \
+    (defined(__linux__) && defined(__NR_copy_file_range))
 static void *nogvl_cfr(void *ptr)
 {
 	struct copy_args *a = ptr;
@@ -68,3 +59,4 @@ void sleepy_penguin_init_cfr(void)
 
 	rb_define_singleton_method(mod, "__cfr", rb_sp_cfr, 6);
 }
+#endif /* !HAVE_COPY_FILE_RANGE */
diff --git a/ext/sleepy_penguin/init.c b/ext/sleepy_penguin/init.c
index 8295073..01bb52f 100644
--- a/ext/sleepy_penguin/init.c
+++ b/ext/sleepy_penguin/init.c
@@ -52,8 +52,20 @@ void sleepy_penguin_init_signalfd(void);
 #  define sleepy_penguin_init_signalfd() for(;0;)
 #endif
 
+#ifdef HAVE_SPLICE
 void sleepy_penguin_init_splice(void);
+#else
+#  define sleepy_penguin_init_splice() for(;0;)
+#endif
+
+#if defined(HAVE_COPY_FILE_RANGE) || \
+    (defined(__linux__) && defined(__NR_copy_file_range))
 void sleepy_penguin_init_cfr(void);
+#else
+#  define sleepy_penguin_init_cfr() for (;0;)
+#endif
+
+/* everyone */
 void sleepy_penguin_init_sendfile(void);
 
 static size_t l1_cache_line_size_detect(void)
diff --git a/ext/sleepy_penguin/sleepy_penguin.h b/ext/sleepy_penguin/sleepy_penguin.h
index 8aa514a..99ad0b7 100644
--- a/ext/sleepy_penguin/sleepy_penguin.h
+++ b/ext/sleepy_penguin/sleepy_penguin.h
@@ -92,4 +92,15 @@ void *rb_sp_gettlsbuf(size_t *size);
 
 int rb_sp_gc_for_fd(int err);
 
+#ifndef HAVE_COPY_FILE_RANGE
+#  include <sys/syscall.h>
+#  if !defined(__NR_copy_file_range) && defined(__linux__)
+#    if defined(__x86_64__)
+#      define __NR_copy_file_range 326
+#    elif defined(__i386__)
+#      define __NR_copy_file_range 377
+#    endif /* supported arches */
+#  endif /* __NR_copy_file_range */
+#endif
+
 #endif /* SLEEPY_PENGUIN_H */
diff --git a/test/test_splice.rb b/test/test_splice.rb
index 3650ee3..266f52b 100644
--- a/test/test_splice.rb
+++ b/test/test_splice.rb
@@ -249,4 +249,4 @@ class TestSplice < Test::Unit::TestCase
       assert Integer === SleepyPenguin.const_get("F_#{x.upcase}")
     }
   end
-end
+end if SleepyPenguin.respond_to?(:splice)
diff --git a/test/test_splice_eintr.rb b/test/test_splice_eintr.rb
index 41b6dd0..fb2dbbc 100644
--- a/test/test_splice_eintr.rb
+++ b/test/test_splice_eintr.rb
@@ -31,4 +31,4 @@ class Test_Splice_EINTR < Test::Unit::TestCase
     assert_equal 2, nr
     assert_equal 1, @usr1
   end
-end if defined?(RUBY_ENGINE)
+end if SleepyPenguin.respond_to?(:splice)
-- 
EW