From 900fab1e79ef6e158254803d859a44400d178536 Mon Sep 17 00:00:00 2001
From: Jason Volk <jason@zemos.net>
Date: Sat, 20 Apr 2019 18:23:52 -0700
Subject: [PATCH] ircd::fs: Propagate EAGAIN without exception for
 !opts.blocking operations.

---
 ircd/fs.cc     | 86 ++++++++++++++++++++++++++++++++++++++------------
 ircd/fs_aio.cc |  7 +++-
 2 files changed, 71 insertions(+), 22 deletions(-)

diff --git a/ircd/fs.cc b/ircd/fs.cc
index 0ad9b78ef..e5a854fcc 100644
--- a/ircd/fs.cc
+++ b/ircd/fs.cc
@@ -702,14 +702,23 @@ ircd::fs::_read(const fd &fd,
 		flags(opts)
 	};
 
-	const auto ret
+	ssize_t ret; do
 	{
-		opts.interruptible?
-			syscall(::preadv2, fd, iov.data(), iov.size(), opts.offset, flags_):
-			syscall_nointr(::preadv2, fd, iov.data(), iov.size(), opts.offset, flags_)
-	};
+		ret = ::preadv2(int(fd), iov.data(), iov.size(), opts.offset, flags_);
+	}
+	while(!opts.interruptible && unlikely(ret == -1 && errno == EINTR));
 
-	return size_t(ret);
+	static_assert(EAGAIN == EWOULDBLOCK);
+	if(!opts.blocking && ret == -1 && errno == EAGAIN)
+		return 0UL;
+
+	if(unlikely(ret == -1))
+		throw std::system_error
+		{
+			errno, std::system_category()
+		};
+
+	return ret;
 }
 #else
 size_t
@@ -717,14 +726,23 @@ ircd::fs::_read(const fd &fd,
                 const const_iovec_view &iov,
                 const read_opts &opts)
 {
-	const auto ret
+	ssize_t ret; do
 	{
-		opts.interruptible?
-			syscall(::preadv, fd, iov.data(), iov.size(), opts.offset):
-			syscall_nointr(::preadv, fd, iov.data(), iov.size(), opts.offset)
-	};
+		ret = ::preadv(int(fd), iov.data(), iov.size(), opts.offset);
+	}
+	while(!opts.interruptible && unlikely(ret == -1 && errno == EINTR));
 
-	return size_t(ret);
+	static_assert(EAGAIN == EWOULDBLOCK);
+	if(unlikely(!opts.blocking && ret == -1 && errno == EAGAIN))
+		return 0UL;
+
+	if(unlikely(ret == -1))
+		throw std::system_error
+		{
+			errno, std::system_category()
+		};
+
+	return ret;
 }
 #endif // HAVE_PREADV2
 
@@ -1050,7 +1068,7 @@ ircd::fs::write(const fd &fd,
 	while(opts.all && opts_.offset >= 0 && off < buffers::size(bufs));
 	assert(opts.offset >= opts_.offset);
 	assert(ssize_t(off) == opts.offset - opts_.offset);
-	assert(!opts.all || off == buffers::size(bufs));
+	assert(!opts.all || !opts.blocking || off == buffers::size(bufs));
 	return off;
 }
 #pragma GCC diagnostic pop
@@ -1099,10 +1117,23 @@ ircd::fs::_write__pwritev1(const fd &fd,
                            const const_iovec_view &iov,
                            const write_opts &opts)
 {
-	return
-		opts.interruptible?
-			syscall(::pwritev, fd, iov.data(), iov.size(), opts.offset):
-			syscall_nointr(::pwritev, fd, iov.data(), iov.size(), opts.offset);
+	ssize_t ret; do
+	{
+		ret = ::pwritev(int(fd), iov.data(), iov.size(), opts.offset);
+	}
+	while(!opts.interruptible && unlikely(ret == -1 && errno == EINTR));
+
+	static_assert(EAGAIN == EWOULDBLOCK);
+	if(unlikely(!opts.blocking && ret == -1 && errno == EAGAIN))
+		return 0UL;
+
+	if(unlikely(ret == -1))
+		throw std::system_error
+		{
+			errno, std::system_category()
+		};
+
+	return ret;
 }
 
 #ifdef HAVE_PWRITEV2
@@ -1124,10 +1155,23 @@ ircd::fs::_write__pwritev2(const fd &fd,
 		flags(opts)
 	};
 
-	return
-		opts.interruptible?
-			syscall(::pwritev2, fd, iov.data(), iov.size(), offset, flags_):
-			syscall_nointr(::pwritev2, fd, iov.data(), iov.size(), offset, flags_);
+	ssize_t ret; do
+	{
+		ret = ::pwritev2(int(fd), iov.data(), iov.size(), opts.offset, flags_);
+	}
+	while(!opts.interruptible && unlikely(ret == -1 && errno == EINTR));
+
+	static_assert(EAGAIN == EWOULDBLOCK);
+	if(!opts.blocking && ret == -1 && errno == EAGAIN)
+		return 0UL;
+
+	if(unlikely(ret == -1))
+		throw std::system_error
+		{
+			errno, std::system_category()
+		};
+
+	return ret;
 }
 #else
 size_t
diff --git a/ircd/fs_aio.cc b/ircd/fs_aio.cc
index 6a4a59bd6..27eea980c 100644
--- a/ircd/fs_aio.cc
+++ b/ircd/fs_aio.cc
@@ -279,7 +279,7 @@ ircd::fs::aio::write(const fd &fd,
 	};
 
 	// Does linux ever not complete all bytes for an AIO?
-	assert(bytes == req_bytes);
+	assert(!opts.blocking || bytes == req_bytes);
 
 	stats.bytes_write += bytes;
 	stats.writes++;
@@ -411,6 +411,11 @@ ircd::fs::aio::request::operator()()
 	if(likely(retval != -1))
 		return size_t(retval);
 
+	assert(opts);
+	static_assert(EAGAIN == EWOULDBLOCK);
+	if(!opts->blocking && retval == -1 && errcode == EAGAIN)
+		return 0UL;
+
 	stats.errors++;
 	stats.bytes_errors += submitted_bytes;
 	thread_local char errbuf[512]; fmt::sprintf