Re: [PATCH v6 0/7] vfs: Non-blockling buffered fs read (page cache only)

Dave Chinner [off-list ref] writes:

On Mon, Nov 10, 2014 at 11:40:23AM -0500, Milosz Tanski wrote:

quoted

This patcheset introduces an ability to perform a non-blocking read from
regular files in buffered IO mode. This works by only for those filesystems
that have data in the page cache.

It does this by introducing new syscalls new syscalls preadv2/pwritev2. These
new syscalls behave like the network sendmsg, recvmsg syscalls that accept an
extra flag argument (RWF_NONBLOCK).

It's a very common patern today (samba, libuv, etc..) use a large threadpool to
perform buffered IO operations. They submit the work form another thread
that performs network IO and epoll or other threads that perform CPU work. This
leads to increased latency for processing, esp. in the case of data that's
already cached in the page cache.

With the new interface the applications will now be able to fetch the data in
their network / cpu bound thread(s) and only defer to a threadpool if it's not
there. In our own application (VLDB) we've observed a decrease in latency for
"fast" request by avoiding unnecessary queuing and having to swap out current
tasks in IO bound work threads.

Can you write a test (or set of) for fstests that exercises this new
functionality? I'm not worried about performance, just
correctness....

On the subject of testing, I added support to trinity (attached,
untested).  That did raise one question.  Do we expect applications to
#include <linux/fs.h> to get the RWF_NONBLOCK definition?

Cheers,
Jeff

diff --git a/include/syscalls-i386.h b/include/syscalls-i386.h
index 767be6e..3125064 100644
--- a/include/syscalls-i386.h
+++ b/include/syscalls-i386.h

@@ -365,4 +365,6 @@ struct syscalltable syscalls_i386[] = {
 	{ .entry = &syscall_getrandom },
 	{ .entry = &syscall_memfd_create },
 	{ .entry = &syscall_bpf },
+	{ .entry = &syscall_preadv2 },
+	{ .entry = &syscall_pwritev2 },
 };

diff --git a/include/syscalls-x86_64.h b/include/syscalls-x86_64.h
index cb609ad..8d32571 100644
--- a/include/syscalls-x86_64.h
+++ b/include/syscalls-x86_64.h

@@ -329,4 +329,6 @@ struct syscalltable syscalls_x86_64[] = {
 	{ .entry = &syscall_memfd_create },
 	{ .entry = &syscall_kexec_file_load },
 	{ .entry = &syscall_bpf },
+	{ .entry = &syscall_preadv2 },
+	{ .entry = &syscall_pwritev2 },
 };

diff --git a/syscalls/read.c b/syscalls/read.c
index e0948a2..adbf146 100644
--- a/syscalls/read.c
+++ b/syscalls/read.c

@@ -3,6 +3,7 @@
  */
 #include <stdlib.h>
 #include <string.h>
+#include <linux/fs.h>
 #include "arch.h"
 #include "maps.h"
 #include "random.h"

@@ -94,3 +95,29 @@ struct syscallentry syscall_preadv = {
 	.arg5name = "pos_h",
 	.flags = NEED_ALARM,
 };
+
+/*
+ * SYSCALL_DEFINE6(preadv2, unsigned long, fd, const struct iovec __user *, vec,
+		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
+		int, flags)
+ */
+
+struct syscallentry syscall_preadv2 = {
+	.name = "preadv2",
+	.num_args = 5,
+	.arg1name = "fd",
+	.arg1type = ARG_FD,
+	.arg2name = "vec",
+	.arg2type = ARG_IOVEC,
+	.arg3name = "vlen",
+	.arg3type = ARG_IOVECLEN,
+	.arg4name = "pos_l",
+	.arg5name = "pos_h",
+	.arg6name = "flags",
+	.arg6type = ARG_OP,
+	.arg6list = {
+		.num = 1,
+		.values = { RWF_NONBLOCK, },
+	},
+	.flags = NEED_ALARM,
+};

diff --git a/syscalls/syscalls.h b/syscalls/syscalls.h
index 5a7748b..04400dd 100644
--- a/syscalls/syscalls.h
+++ b/syscalls/syscalls.h

@@ -375,5 +375,7 @@ extern struct syscallentry syscall_seccomp;
 extern struct syscallentry syscall_memfd_create;
 extern struct syscallentry syscall_kexec_file_load;
 extern struct syscallentry syscall_bpf;
+extern struct syscallentry syscall_preadv2;
+extern struct syscallentry syscall_pwritev2;
 
 unsigned int random_fcntl_setfl_flags(void);

diff --git a/syscalls/write.c b/syscalls/write.c
index f37e760..4218ccc 100644
--- a/syscalls/write.c
+++ b/syscalls/write.c

@@ -2,6 +2,7 @@
  * SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf, size_t, count)
  */
 #include <stdlib.h>
+#include <linux/fs.h>
 #include "arch.h"	// page_size
 #include "maps.h"
 #include "random.h"

@@ -95,3 +96,30 @@ struct syscallentry syscall_pwritev = {
 	.arg5name = "pos_h",
 	.flags = NEED_ALARM,
 };
+
+
+/*
+ * SYSCALL_DEFINE6(pwritev2, unsigned long, fd, const struct iovec __user *, vec,
+		unsigned long, vlen, unsigned long, pos_l, unsigned long, pos_h,
+		int, flags)
+ */
+
+struct syscallentry syscall_pwritev2 = {
+	.name = "pwritev2",
+	.num_args = 6,
+	.arg1name = "fd",
+	.arg1type = ARG_FD,
+	.arg2name = "vec",
+	.arg2type = ARG_IOVEC,
+	.arg3name = "vlen",
+	.arg3type = ARG_IOVECLEN,
+	.arg4name = "pos_l",
+	.arg5name = "pos_h",
+	.arg6name = "flags",
+	.arg6type = ARG_OP,
+	.arg6list = {
+		.num = 1,
+		.values = { RWF_NONBLOCK, },
+	},
+	.flags = NEED_ALARM,
+};

--

To unsubscribe, send a message with 'unsubscribe linux-aio' in
the body to majordomo@kvack.org.  For more info on Linux AIO,
see: http://www.kvack.org/aio/
Don't email: <a href=mailto:"aart@kvack.org">aart@kvack.org</a>

On Fri, Nov 14, 2014 at 11:32:53AM -0500, Jeff Moyer wrote:

 > > Can you write a test (or set of) for fstests that exercises this new
 > > functionality? I'm not worried about performance, just
 > > correctness....
 > 
 > On the subject of testing, I added support to trinity (attached,
 > untested).  That did raise one question.  Do we expect applications to
 > #include <linux/fs.h> to get the RWF_NONBLOCK definition?
 
Trinity will at least need an addition to include/compat.h for
older headers that won't have the definition.  Looks ok otherwise.

Also, I usually sit on stuff like this until the syscall numbers are
in Linus tree. This is 3.19 stuff I presume ?
istr akpm picked up execveat recently, so if that goes in first, we'll
need to respin this anyway..

	Dave

--
To unsubscribe, send a message with 'unsubscribe linux-aio' in
the body to majordomo@kvack.org.  For more info on Linux AIO,
see: http://www.kvack.org/aio/
Don't email: <a href=mailto:"aart@kvack.org">aart@kvack.org</a>