[PATCH 2/2] fs: Support compiling out sendfile

[Pieter Smith]

Many embedded systems will not need this syscall, and omitting it
saves space. Add a new EXPERT config option CONFIG_SENDFILE_SYSCALL
(default y) to support compiling it out.

bloat-o-meter:
add/remove: 0/4 grow/shrink: 5/0 up/down: 23/-751 (-728)
function old new delta
sys_pwritev 115 122 +7
sys_preadv 115 122 +7
fdput_pos 29 36 +7
sys_pwrite64 115 116 +1
sys_pread64 115 116 +1
fdput 11 - -11
sys_sendfile 122 - -122
sys_sendfile64 126 - -126
do_sendfile 492 - -492

Signed-off-by: Pieter Smith <pieter-qeJ+1H9vRZbz+***@public.gmane.org>
---
fs/Makefile | 3 ++-
init/Kconfig | 10 ++++++++++
kernel/sys_ni.c | 4 ++++
3 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/fs/Makefile b/fs/Makefile
index 1e3423f..1bbfea7 100644
--- a/fs/Makefile
+++ b/fs/Makefile
@@ -5,7 +5,7 @@
# Rewritten to use lists instead of if-statements.
#

-obj-y := open.o read_write.o sendfile.o file_table.o super.o \
+obj-y := open.o read_write.o file_table.o super.o \
char_dev.o stat.o exec.o pipe.o namei.o fcntl.o \
ioctl.o readdir.o select.o dcache.o inode.o \
attr.o bad_inode.o file.o filesystems.o namespace.o \
@@ -38,6 +38,7 @@ obj-$(CONFIG_COMPAT_BINFMT_ELF) += compat_binfmt_elf.o
obj-$(CONFIG_BINFMT_ELF_FDPIC) += binfmt_elf_fdpic.o
obj-$(CONFIG_BINFMT_SOM) += binfmt_som.o
obj-$(CONFIG_BINFMT_FLAT) += binfmt_flat.o
+obj-$(CONFIG_SENDFILE_SYSCALL) += sendfile.o

obj-$(CONFIG_FS_MBCACHE) += mbcache.o
obj-$(CONFIG_FS_POSIX_ACL) += posix_acl.o
diff --git a/init/Kconfig b/init/Kconfig
index 782a65b..df6785c 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -1547,6 +1547,16 @@ config ADVISE_SYSCALLS
applications use these syscalls, you can disable this option to save
space.

+config SENDFILE_SYSCALL
+ bool "Enable sendfile syscall" if EXPERT
+ default y
+ help
+ This option enables the sendfile syscall, used by applications to copy
+ data between file descriptors. Because sendfile performs the copying
+ within the kernel, it is more efficient than the combination of read
+ and write. If building an embedded system where no applications use
+ the sendfile syscall, you can disable this option to save space.
+
config PCI_QUIRKS
default y
bool "Enable PCI quirk workarounds" if EXPERT
diff --git a/kernel/sys_ni.c b/kernel/sys_ni.c
index d4709d4..b068de7 100644
--- a/kernel/sys_ni.c
+++ b/kernel/sys_ni.c
@@ -159,6 +159,10 @@ cond_syscall(sys_uselib);
cond_syscall(sys_fadvise64);
cond_syscall(sys_fadvise64_64);
cond_syscall(sys_madvise);
+cond_syscall(sys_sendfile);
+cond_syscall(sys_sendfile64);
+cond_syscall(compat_sys_sendfile);
+cond_syscall(compat_sys_sendfile64);

/* arch-specific weak syscall entries */
cond_syscall(sys_pciconfig_read);

[j***@joshtriplett.org]

Post by Pieter Smith
Many embedded systems will not need this syscall, and omitting it
saves space. Add a new EXPERT config option CONFIG_SENDFILE_SYSCALL
(default y) to support compiling it out.

Nice work, thanks!

If there are no objections, and nobody has a tree they'd rather carry
this through, I'll take the series through the tiny tree when it's ready
to merge.

Post by Pieter Smith
add/remove: 0/4 grow/shrink: 5/0 up/down: 23/-751 (-728)
function old new delta
sys_pwritev 115 122 +7
sys_preadv 115 122 +7
fdput_pos 29 36 +7
sys_pwrite64 115 116 +1
sys_pread64 115 116 +1
fdput 11 - -11
sys_sendfile 122 - -122
sys_sendfile64 126 - -126
do_sendfile 492 - -492

Interesting inlining decisions by GCC here. Got a bloat-o-meter for the
two-patch series, by any chance? (Also, is this with tinyconfig? In
particular, with OPTIMIZE_INLINING and OPTIMIZE_FOR_SIZE?) I'm
wondering if moving sendfile to a separate file made GCC put fdput
out-of-line, and compiling it out reversed that again.

Post by Pieter Smith
---
fs/Makefile | 3 ++-
init/Kconfig | 10 ++++++++++
kernel/sys_ni.c | 4 ++++
3 files changed, 16 insertions(+), 1 deletion(-)
diff --git a/fs/Makefile b/fs/Makefile
index 1e3423f..1bbfea7 100644
--- a/fs/Makefile
+++ b/fs/Makefile
@@ -5,7 +5,7 @@
# Rewritten to use lists instead of if-statements.
#
-obj-y := open.o read_write.o sendfile.o file_table.o super.o \
+obj-y := open.o read_write.o file_table.o super.o \
char_dev.o stat.o exec.o pipe.o namei.o fcntl.o \
ioctl.o readdir.o select.o dcache.o inode.o \
attr.o bad_inode.o file.o filesystems.o namespace.o \
@@ -38,6 +38,7 @@ obj-$(CONFIG_COMPAT_BINFMT_ELF) += compat_binfmt_elf.o
obj-$(CONFIG_BINFMT_ELF_FDPIC) += binfmt_elf_fdpic.o
obj-$(CONFIG_BINFMT_SOM) += binfmt_som.o
obj-$(CONFIG_BINFMT_FLAT) += binfmt_flat.o
+obj-$(CONFIG_SENDFILE_SYSCALL) += sendfile.o
obj-$(CONFIG_FS_MBCACHE) += mbcache.o
obj-$(CONFIG_FS_POSIX_ACL) += posix_acl.o
diff --git a/init/Kconfig b/init/Kconfig
index 782a65b..df6785c 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -1547,6 +1547,16 @@ config ADVISE_SYSCALLS
applications use these syscalls, you can disable this option to save
space.
+config SENDFILE_SYSCALL
+ bool "Enable sendfile syscall" if EXPERT
+ default y
+ help
+ This option enables the sendfile syscall, used by applications to copy
+ data between file descriptors. Because sendfile performs the copying
+ within the kernel, it is more efficient than the combination of read
+ and write. If building an embedded system where no applications use
+ the sendfile syscall, you can disable this option to save space.
+

I'm thinking of adding a submenu to group config FOO_SYSCALL options. :)
I'll probably push that as part of the 3.19 merge window, as a patch on
top of all of the individual tinification options.

Post by Pieter Smith
config PCI_QUIRKS
default y
bool "Enable PCI quirk workarounds" if EXPERT
diff --git a/kernel/sys_ni.c b/kernel/sys_ni.c
index d4709d4..b068de7 100644
--- a/kernel/sys_ni.c
+++ b/kernel/sys_ni.c
@@ -159,6 +159,10 @@ cond_syscall(sys_uselib);
cond_syscall(sys_fadvise64);
cond_syscall(sys_fadvise64_64);
cond_syscall(sys_madvise);
+cond_syscall(sys_sendfile);
+cond_syscall(sys_sendfile64);
+cond_syscall(compat_sys_sendfile);
+cond_syscall(compat_sys_sendfile64);
/* arch-specific weak syscall entries */
cond_syscall(sys_pciconfig_read);
--
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

[Christoph Hellwig]

Post by j***@joshtriplett.org

Post by Pieter Smith
Many embedded systems will not need this syscall, and omitting it
saves space. Add a new EXPERT config option CONFIG_SENDFILE_SYSCALL
(default y) to support compiling it out.

Nice work, thanks!
If there are no objections, and nobody has a tree they'd rather carry
this through, I'll take the series through the tiny tree when it's ready
to merge.

I think it's rather pointless - there is very little sendfile code,
so you'd rather want to disable splice.

[Josh Triplett]

Post by Christoph Hellwig

Post by j***@joshtriplett.org

Post by Pieter Smith
Many embedded systems will not need this syscall, and omitting it
saves space. Add a new EXPERT config option CONFIG_SENDFILE_SYSCALL
(default y) to support compiling it out.

Nice work, thanks!
If there are no objections, and nobody has a tree they'd rather carry
this through, I'll take the series through the tiny tree when it's ready
to merge.

I think it's rather pointless - there is very little sendfile code,
so you'd rather want to disable splice.

That's the plan, but since sendfile depends on some of the splice bits,
sendfile needs to be optional as well; SENDFILE_SYSCALL will then select
SPLICE_SYSCALLS.

- Josh Triplett

[Christoph Hellwig]

Post by Josh Triplett
That's the plan, but since sendfile depends on some of the splice bits,
sendfile needs to be optional as well; SENDFILE_SYSCALL will then select
SPLICE_SYSCALLS.

Just include sendfile with the splice syscalls - we don't really need a
config option for every obscure syscall.

[Josh Triplett]

Post by Christoph Hellwig

Post by Josh Triplett
That's the plan, but since sendfile depends on some of the splice bits,
sendfile needs to be optional as well; SENDFILE_SYSCALL will then select
SPLICE_SYSCALLS.

Just include sendfile with the splice syscalls - we don't really need a
config option for every obscure syscall.

No objection here. Pieter, since you're planning to remove splice
anyway, can you just fold the two together under the same Kconfig
option? That should simplify the patch series, since you won't need to
split the two.

- Josh Triplett
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

[Eric Paris]

Post by Pieter Smith
Many embedded systems will not need this syscall, and omitting it
saves space. Add a new EXPERT config option CONFIG_SENDFILE_SYSCALL
(default y) to support compiling it out.

<bikeshed>
I believe these options ought to be CONFIG_SYSCALL_*
</bikeshed>

I agree. I think people started using CONFIG_*_SYSCALL because of
things like AUDITSYSCALL

AUDITSYSCALL audits syscalls. It doesn't actually implement any
syscalls. You are right about SYSFS_SYSCALL though...