Message-Id: <20090225133929.f67163e3.akpm@linux-foundation.org>
Date:	Wed, 25 Feb 2009 13:39:29 -0800
From:	Andrew Morton <akpm@...ux-foundation.org>
To:	Steven Rostedt <rostedt@...dmis.org>
Cc:	linux-kernel@...r.kernel.org, mingo@...e.hu, peterz@...radead.org,
	fweisbec@...il.com, srostedt@...hat.com
Subject: Re: [PATCH 1/4] uaccess: add copy_word_from_user

On Wed, 25 Feb 2009 15:30:08 -0500
Steven Rostedt <rostedt@...dmis.org> wrote:

> From: Steven Rostedt <srostedt@...hat.com>
> 
> The ftrace utility reads space delimited words from user space.
> Andrew Morton did not like how ftrace open coded this. He had
> a good point since more than one location performed this feature.
> 
> This patch creates a copy_word_from_user function that can copy
> a space delimited word from user space. This puts the code in
> a new lib/uaccess.c file. This keeps the code in a single location
> and may be optimized in the future.
> 

Does your code actually still need this?  It is unacceptble to just
be more strict about userspace's write()s?

> diff --git a/include/linux/uaccess.h b/include/linux/uaccess.h
> index 6b58367..2d706d9 100644
> --- a/include/linux/uaccess.h
> +++ b/include/linux/uaccess.h
> @@ -106,4 +106,8 @@ extern long probe_kernel_read(void *dst, void *src, size_t size);
>   */
>  extern long probe_kernel_write(void *dst, void *src, size_t size);
>  
> +extern int copy_word_from_user(void *to, const void __user *from,
> +			       unsigned int copy, unsigned int read,
> +			       unsigned int *copied, int skip);
> +
>  #endif		/* __LINUX_UACCESS_H__ */
> diff --git a/lib/Makefile b/lib/Makefile
> index 32b0e64..46ce28c 100644
> --- a/lib/Makefile
> +++ b/lib/Makefile
> @@ -11,7 +11,8 @@ lib-y := ctype.o string.o vsprintf.o cmdline.o \
>  	 rbtree.o radix-tree.o dump_stack.o \
>  	 idr.o int_sqrt.o extable.o prio_tree.o \
>  	 sha1.o irq_regs.o reciprocal_div.o argv_split.o \
> -	 proportions.o prio_heap.o ratelimit.o show_mem.o is_single_threaded.o
> +	 proportions.o prio_heap.o ratelimit.o show_mem.o is_single_threaded.o \
> +	 uaccess.o
>  
>  lib-$(CONFIG_MMU) += ioremap.o
>  lib-$(CONFIG_SMP) += cpumask.o
> diff --git a/lib/uaccess.c b/lib/uaccess.c
> new file mode 100644
> index 0000000..5b9a4ac
> --- /dev/null
> +++ b/lib/uaccess.c
> @@ -0,0 +1,134 @@
> +/*
> + * lib/uaccess.c
> + * generic user access file.

That's a good place for it.  I wonder if we have other uaccess
functions which should be moved here sometime.

> + * started by Steven Rostedt
> + *
> + *  Copyright (C) 2009 Red Hat, Inc., Steven Rostedt <srostedt@...hat.com>
> + *
> + * This source code is licensed under the GNU General Public License,
> + * Version 2.  See the file COPYING for more details.
> + */
> +#include <linux/uaccess.h>
> +#include <linux/ctype.h>
> +
> +/**
> + * copy_word_from_user - copy a space delimited word from user space
> + * @to:	     The location to copy to
> + * @from:    The location to copy from
> + * @copy:    The number of bytes to copy
> + * @read:    The number of bytes to read
> + * @copied:  The number of bytes actually copied to @to
> + * @skip:    If other than zero, will skip leading white space
> + *
> + * This reads from a user buffer, a space delimited word.
> + * If skip is set, then it will trim all leading white space.
> + * Then it will copy all non white space until @copy bytes have
> + * been copied, @read bytes have been read from the user buffer,
> + * or more white space has been encountered.
> + *
> + * Note, if skip is not set, and white space exists at the beginning
> + *  it will return immediately.
> + *
> + * Returns:
> + *  The number of bytes read from user space

Confused.

Is this "the number of bytes which I copied into *to", or is it "the
number of userspace bytes over which I advanced"?

Hopefully the latter, because callers of copy_word_from_user() should
be able to call this function multiple times to be able to parse "foo
bar zot\0" into three separate words with three separate calls to
copy_word_from_user().  It might be worth mentioning how callers should
do this in the covering comment?

> + *  -EAGAIN, if we copied a word successfully, but never hit
> + *    ending white space. The number of bytes copied will be the same
> + *    as @read. Note, if skip is set, and all we hit was white space
> + *    then we will also returne -EAGAIN with @copied = 0.
> + *
> + *  @copied will contain the number of bytes copied into @to
> + *
> + *  -EFAULT, if we faulted during any part of the copy.
> + *     @copied will be undefined.
> + *
> + *  -EINVAL, if we fill up @from before hitting white space.
> + *    @copy must be bigger than the expected word to read.
> + */
> +int copy_word_from_user(void *to, const void __user *from,
> +			unsigned int copy, unsigned int read,
> +			unsigned int *copied, int skip)
> +{

The uaccess functions are a bit confused about whether the `size' args
are unsigned, unsigned long, etc.  They should be size_t.  unsigned is
OK here.

> +	unsigned int have_read = 0;
> +	unsigned int have_copied = 0;
> +	const char __user *user = from;
> +	char *kern = to;
> +	int ret;
> +	char ch;
> +
> +	/* get the first character */
> +	ret = get_user(ch, user++);
> +	if (ret)
> +		return ret;
> +	have_read++;
> +
> +	/*
> +	 * If skip is set, and the first character is white space
> +	 * then we will continue to read until we find non white space.
> +	 */
> +	if (skip) {
> +		while (have_read < read && isspace(ch)) {
> +			ret = get_user(ch, user++);
> +			if (ret)
> +				return ret;
> +			have_read++;
> +		}
> +
> +		/*
> +		 * If ch is still white space, then have_read == read.
> +		 * We successfully copied zero bytes. But this is
> +		 * still valid. Just let the caller try again.
> +		 */
> +		if (isspace(ch)) {
> +			ret = -EAGAIN;
> +			goto out;
> +		}
> +	} else if (isspace(ch)) {
> +		/*
> +		 * If skip was not set and the first character was
> +		 * white space, then we return immediately.
> +		 */
> +		ret = have_read;
> +		goto out;
> +	}
> +
> +
> +	/* Now read the actual word */
> +	while (have_read < read &&
> +	       have_copied < copy && !isspace(ch)) {
> +
> +		kern[have_copied++] = ch;
> +
> +		ret = get_user(ch, user++);
> +		if (ret)
> +			return ret;
> +
> +		have_read++;
> +	}
> +
> +	/*
> +	 * If we ended with white space then we have successfully
> +	 * read in a full word.
> +	 *
> +	 * If ch is not white space, and we have filled up @from,
> +	 * then this was an invalid word.
> +	 *
> +	 * If ch is not white space, and we still have room in @from
> +	 * then we let the caller know we have split a word.
> +	 *  (have_read == read)
> +	 */
> +	if (isspace(ch))
> +		ret = have_read;
> +	else if (have_copied == copy)
> +		ret = -EINVAL;
> +	else {
> +		WARN_ON(have_read != read);
> +		ret = -EAGAIN;
> +	}
> +
> + out:
> +	*copied = have_copied;
> +
> +	return ret;
> +}

Sheer madness ;)

Someone is going to want to extend the "isspace" to include other
tokens.  We can fall off that bridge when we come to it.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists