Re: [PATCH 03/11] fs: add new read_uptr and write_uptr file operations

From: Matthew Wilcox
Date: Wed Jun 24 2020 - 13:57:06 EST


On Wed, Jun 24, 2020 at 10:19:16AM -0700, Linus Torvalds wrote:
> On Wed, Jun 24, 2020 at 9:29 AM Christoph Hellwig <hch@xxxxxx> wrote:
> >
> > Add two new file operations that are identical to ->read and ->write
> > except that they can also safely take kernel pointers using the uptr_t
> > type.
>
> Honestly, I think this is the wrong way to go.
>
> All of this new complexity and messiness, just to remove a few
> unimportant final cases?
>
> If somebody can't be bothered to convert a driver to
> iter_read/iter_write, why would they be bothered to convert it to
> read_uptr/write_uptr?
>
> And this messiness will stay around for decades.
>
> So let's not go down that path.
>
> If you want to do "splice() and kernel_read() requires read_iter"
> (with a warning so that we find any cases), then that's fine. But
> let's not add yet _another_ read type.
>
> Why did you care so much about sysctl, and why couldn't they use the iter ops?

Heh, when I saw patch 4, I started working on that. It doesn't seem all
that bad, except I've never used the iov_iter before, so I have no idea
if I did this right. Also, this fixes a bug if 'count' is too large,
which I should split out and send separately.

diff --git a/fs/proc/proc_sysctl.c b/fs/proc/proc_sysctl.c
index 42c5128c7d1c..7a8c474bc196 100644
--- a/fs/proc/proc_sysctl.c
+++ b/fs/proc/proc_sysctl.c
@@ -12,6 +12,7 @@
#include <linux/cred.h>
#include <linux/namei.h>
#include <linux/mm.h>
+#include <linux/uio.h>
#include <linux/module.h>
#include <linux/bpf-cgroup.h>
#include <linux/mount.h>
@@ -540,12 +541,13 @@ static struct dentry *proc_sys_lookup(struct inode *dir, struct dentry *dentry,
return err;
}

-static ssize_t proc_sys_call_handler(struct file *filp, void __user *ubuf,
- size_t count, loff_t *ppos, int write)
+static ssize_t proc_sys_call_handler(struct kiocb *iocb, struct iov_iter *iter,
+ int write)
{
- struct inode *inode = file_inode(filp);
+ struct inode *inode = file_inode(iocb->ki_filp);
struct ctl_table_header *head = grab_header(inode);
struct ctl_table *table = PROC_I(inode)->sysctl_entry;
+ size_t count = iov_iter_count(iter);
void *kbuf;
ssize_t error;

@@ -566,35 +568,32 @@ static ssize_t proc_sys_call_handler(struct file *filp, void __user *ubuf,
goto out;

/* don't even try if the size is too large */
+ error = -ENOMEM;
if (count > KMALLOC_MAX_SIZE)
- return -ENOMEM;
+ goto out;
+ kbuf = kzalloc(count, GFP_KERNEL);
+ if (!kbuf)
+ goto out;

if (write) {
- kbuf = memdup_user_nul(ubuf, count);
- if (IS_ERR(kbuf)) {
- error = PTR_ERR(kbuf);
- goto out;
- }
- } else {
- error = -ENOMEM;
- kbuf = kzalloc(count, GFP_KERNEL);
- if (!kbuf)
+ error = -EFAULT;
+ if (!copy_from_iter_full(kbuf, count, iter))
goto out;
}

error = BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, &kbuf, &count,
- ppos);
+ &iocb->ki_pos);
if (error)
goto out_free_buf;

/* careful: calling conventions are nasty here */
- error = table->proc_handler(table, write, kbuf, &count, ppos);
+ error = table->proc_handler(table, write, kbuf, &count, &iocb->ki_pos);
if (error)
goto out_free_buf;

if (!write) {
error = -EFAULT;
- if (copy_to_user(ubuf, kbuf, count))
+ if (copy_to_iter(kbuf, count, iter) < count)
goto out_free_buf;
}

@@ -607,16 +606,14 @@ static ssize_t proc_sys_call_handler(struct file *filp, void __user *ubuf,
return error;
}

-static ssize_t proc_sys_read(struct file *filp, char __user *buf,
- size_t count, loff_t *ppos)
+static ssize_t proc_sys_read(struct kiocb *iocb, struct iov_iter *iter)
{
- return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 0);
+ return proc_sys_call_handler(iocb, iter, 0);
}

-static ssize_t proc_sys_write(struct file *filp, const char __user *buf,
- size_t count, loff_t *ppos)
+static ssize_t proc_sys_write(struct kiocb *iocb, struct iov_iter *iter)
{
- return proc_sys_call_handler(filp, (void __user *)buf, count, ppos, 1);
+ return proc_sys_call_handler(iocb, iter, 1);
}

static int proc_sys_open(struct inode *inode, struct file *filp)
@@ -853,8 +850,8 @@ static int proc_sys_getattr(const struct path *path, struct kstat *stat,
static const struct file_operations proc_sys_file_operations = {
.open = proc_sys_open,
.poll = proc_sys_poll,
- .read = proc_sys_read,
- .write = proc_sys_write,
+ .read_iter = proc_sys_read,
+ .write_iter = proc_sys_write,
.llseek = default_llseek,
};