[PATCH v2 18/36] fs: add ksys_write() helper; remove in-kernel calls to sys_write()

[PATCH v2 18/36] fs: add ksys_write() helper; remove in-kernel calls to sys_write()

[Dominik Brodowski]

Using this helper allows us to avoid the in-kernel calls to the sys_write()
syscall.

Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: linux-s390@vger.kernel.org
Signed-off-by: Dominik Brodowski <linux@dominikbrodowski.net>
---
 arch/s390/kernel/compat_linux.c | 2 +-
 fs/read_write.c                 | 9 +++++++--
 include/linux/syscalls.h        | 1 +
 init/do_mounts_rd.c             | 4 ++--
 init/initramfs.c                | 2 +-
 5 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/arch/s390/kernel/compat_linux.c b/arch/s390/kernel/compat_linux.c
index 79b7a3438d54..5a9cfde5fc28 100644
--- a/arch/s390/kernel/compat_linux.c
+++ b/arch/s390/kernel/compat_linux.c
@@ -468,7 +468,7 @@ COMPAT_SYSCALL_DEFINE3(s390_write, unsigned int, fd, const char __user *, buf, c
 	if ((compat_ssize_t) count < 0)
 		return -EINVAL; 
 
-	return sys_write(fd, buf, count);
+	return ksys_write(fd, buf, count);
 }
 
 /*
diff --git a/fs/read_write.c b/fs/read_write.c
index f8547b82dfb3..8e8f0b4f52e2 100644
--- a/fs/read_write.c
+++ b/fs/read_write.c
@@ -578,8 +578,7 @@ SYSCALL_DEFINE3(read, unsigned int, fd, char __user *, buf, size_t, count)
 	return ret;
 }
 
-SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
-		size_t, count)
+ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count)
 {
 	struct fd f = fdget_pos(fd);
 	ssize_t ret = -EBADF;
@@ -595,6 +594,12 @@ SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
 	return ret;
 }
 
+SYSCALL_DEFINE3(write, unsigned int, fd, const char __user *, buf,
+		size_t, count)
+{
+	return ksys_write(fd, buf, count);
+}
+
 SYSCALL_DEFINE4(pread64, unsigned int, fd, char __user *, buf,
 			size_t, count, loff_t, pos)
 {
diff --git a/include/linux/syscalls.h b/include/linux/syscalls.h
index 13c7bc43b6ef..c6aa44b6a0a2 100644
--- a/include/linux/syscalls.h
+++ b/include/linux/syscalls.h
@@ -953,5 +953,6 @@ int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type,
 int ksys_umount(char __user *name, int flags);
 int ksys_dup(unsigned int fildes);
 int ksys_chroot(const char __user *filename);
+ssize_t ksys_write(unsigned int fd, const char __user *buf, size_t count);
 
 #endif
diff --git a/init/do_mounts_rd.c b/init/do_mounts_rd.c
index 99e0b649fc0e..2d365c398ccc 100644
--- a/init/do_mounts_rd.c
+++ b/init/do_mounts_rd.c
@@ -270,7 +270,7 @@ int __init rd_load_image(char *from)
 			printk("Loading disk #%d... ", disk);
 		}
 		sys_read(in_fd, buf, BLOCK_SIZE);
-		sys_write(out_fd, buf, BLOCK_SIZE);
+		ksys_write(out_fd, buf, BLOCK_SIZE);
 #if !defined(CONFIG_S390)
 		if (!(i % 16)) {
 			pr_cont("%c\b", rotator[rotate & 0x3]);
@@ -317,7 +317,7 @@ static long __init compr_fill(void *buf, unsigned long len)
 
 static long __init compr_flush(void *window, unsigned long outcnt)
 {
-	long written = sys_write(crd_outfd, window, outcnt);
+	long written = ksys_write(crd_outfd, window, outcnt);
 	if (written != outcnt) {
 		if (decompress_error == 0)
 			printk(KERN_ERR
diff --git a/init/initramfs.c b/init/initramfs.c
index 7e99a0038942..6f972df15bf2 100644
--- a/init/initramfs.c
+++ b/init/initramfs.c
@@ -27,7 +27,7 @@ static ssize_t __init xwrite(int fd, const char *p, size_t count)
 
 	/* sys_write only can write MAX_RW_COUNT aka 2G-4K bytes at most */
 	while (count) {
-		ssize_t rv = sys_write(fd, p, count);
+		ssize_t rv = ksys_write(fd, p, count);
 
 		if (rv < 0) {
 			if (rv == -EINTR || rv == -EAGAIN)
-- 
2.16.2

Re: [PATCH v2 18/36] fs: add ksys_write() helper; remove in-kernel calls to sys_write()

[Christoph Hellwig]

I really don't like this, as this is the wrong level of abstraction.

> diff --git a/arch/s390/kernel/compat_linux.c b/arch/s390/kernel/compat_linux.c
> index 79b7a3438d54..5a9cfde5fc28 100644
> --- a/arch/s390/kernel/compat_linux.c
> +++ b/arch/s390/kernel/compat_linux.c
> @@ -468,7 +468,7 @@ COMPAT_SYSCALL_DEFINE3(s390_write, unsigned int, fd, const char __user *, buf, c
>  	if ((compat_ssize_t) count < 0)
>  		return -EINVAL; 
>  
> -	return sys_write(fd, buf, count);
> +	return ksys_write(fd, buf, count);
>  }

This looks bogus to me.  Why does s390 have its own compat version of
write but not any of the other read and write familty calls?

> diff --git a/init/do_mounts_rd.c b/init/do_mounts_rd.c
> index 99e0b649fc0e..2d365c398ccc 100644
> --- a/init/do_mounts_rd.c
> +++ b/init/do_mounts_rd.c
> @@ -270,7 +270,7 @@ int __init rd_load_image(char *from)
>  			printk("Loading disk #%d... ", disk);
>  		}
>  		sys_read(in_fd, buf, BLOCK_SIZE);
> -		sys_write(out_fd, buf, BLOCK_SIZE);
> +		ksys_write(out_fd, buf, BLOCK_SIZE);
>  #if !defined(CONFIG_S390)
>  		if (!(i % 16)) {
>  			pr_cont("%c\b", rotator[rotate & 0x3]);

All the do_mounts / initramfs code should be rewritten to use filp_open
and vfs_read/vfs_write instead of adding hacks like this.

Re: [PATCH v2 18/36] fs: add ksys_write() helper; remove in-kernel calls to sys_write()

[Dominik Brodowski]

On Fri, Mar 16, 2018 at 01:52:40AM -0700, Christoph Hellwig wrote:
> I really don't like this, as this is the wrong level of abstraction.
> 
> > diff --git a/arch/s390/kernel/compat_linux.c b/arch/s390/kernel/compat_linux.c
> > index 79b7a3438d54..5a9cfde5fc28 100644
> > --- a/arch/s390/kernel/compat_linux.c
> > +++ b/arch/s390/kernel/compat_linux.c
> > @@ -468,7 +468,7 @@ COMPAT_SYSCALL_DEFINE3(s390_write, unsigned int, fd, const char __user *, buf, c
> >  	if ((compat_ssize_t) count < 0)
> >  		return -EINVAL; 
> >  
> > -	return sys_write(fd, buf, count);
> > +	return ksys_write(fd, buf, count);
> >  }
> 
> This looks bogus to me.  Why does s390 have its own compat version of
> write but not any of the other read and write familty calls?
> 
> > diff --git a/init/do_mounts_rd.c b/init/do_mounts_rd.c
> > index 99e0b649fc0e..2d365c398ccc 100644
> > --- a/init/do_mounts_rd.c
> > +++ b/init/do_mounts_rd.c
> > @@ -270,7 +270,7 @@ int __init rd_load_image(char *from)
> >  			printk("Loading disk #%d... ", disk);
> >  		}
> >  		sys_read(in_fd, buf, BLOCK_SIZE);
> > -		sys_write(out_fd, buf, BLOCK_SIZE);
> > +		ksys_write(out_fd, buf, BLOCK_SIZE);
> >  #if !defined(CONFIG_S390)
> >  		if (!(i % 16)) {
> >  			pr_cont("%c\b", rotator[rotate & 0x3]);
> 
> All the do_mounts / initramfs code should be rewritten to use filp_open
> and vfs_read/vfs_write instead of adding hacks like this.

In line with the other patches, I have added the following paragraph to the
commit message:

	In the near future, the do_mounts / initramfs callers of ksys_write()
	should be converted to use filp_open() and vfs_write() instead.

Thanks,
	Dominik