Skip to contentSkip to conversation

/dev/mem cleanups and bug fix

PATCH In review Wu Fengguang 3 participants 11 messages Sep 11, 2009 – Sep 12, 2009 Char and misc drivers

lore
Share & tools
lore
#0 lore raw

cover /dev/mem cleanups and bug fix

Hi Andrew,

Here are three patches for /dev/[k]mem.
They are mostly code cleanups, plus a minor bug fix:
if someone read high mem via /dev/kmem starting from an unaligned address, it's calculation of 'len' will go wrong.

I would recommend them for .32.

Thanks,
Fengguang

#1 lore raw

PATCH 1/3 devmem: remove redundant test on len

The len test in write_kmem() is always true, so can be reduced.

CC Marcelo Tosatti <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Mark Brown <[email protected]>
CC Johannes Berg <[email protected]>
CC Avi Kivity <[email protected]>
---
 drivers/char/mem.c |   14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)
drivers/char/mem.c +6 -8
@@ -580,18 +580,16 @@ static ssize_t write_kmem(struct file * 
 		while (count > 0) {
 			int len = count;
 
 			if (len > PAGE_SIZE)
 				len = PAGE_SIZE;
-			if (len) {
-				written = copy_from_user(kbuf, buf, len);
-				if (written) {
-					if (wrote + virtr)
-						break;
-					free_page((unsigned long)kbuf);
-					return -EFAULT;
-				}
+			written = copy_from_user(kbuf, buf, len);
+			if (written) {
+				if (wrote + virtr)
+					break;
+				free_page((unsigned long)kbuf);
+				return -EFAULT;
 			}
 			len = vwrite(kbuf, (char *)p, len);
 			count -= len;
 			buf += len;
 			virtr += len;
#2 lore raw

Re: [PATCH 1/3] devmem: remove redundant test on len

Quoted 29 lines · On Fri, 11 Sep 2009 10:23:34 +0800 Wu Fengguang <[email protected]> wrote:

The len test in write_kmem() is always true, so can be reduced.

CC Marcelo Tosatti <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Mark Brown <[email protected]>
CC Johannes Berg <[email protected]>
CC Avi Kivity <[email protected]>
---
 drivers/char/mem.c |   14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)
drivers/char/mem.c +1 -8
@@ -580,18 +580,16 @@ static ssize_t write_kmem(struct file * 
 		while (count > 0) {
 			int len = count;
 
 			if (len > PAGE_SIZE)
 				len = PAGE_SIZE;
-			if (len) {
-				written = copy_from_user(kbuf, buf, len);
-				if (written) {
-					if (wrote + virtr)
-						break;
-					free_page((unsigned long)kbuf);
-					return -EFAULT;
-				}
+			written = copy_from_user(kbuf, buf, len);

drivers/char/mem.c static ssize_t write_kmem(struct file * · line 595

Quoted 10 lines
drivers/char/mem.c +5 -0
+			if (written) {
+				if (wrote + virtr)
+					break;
+				free_page((unsigned long)kbuf);
+				return -EFAULT;
 			}
 			len = vwrite(kbuf, (char *)p, len);
 			count -= len;
 			buf += len;
 			virtr += len;

humpf. But take a closer look at what remains.

Local var `written' is unneeded here.

Which makes us look at what `written' _does_ do:

		if (written != wrote)
			return written;
		wrote = written;

lolwhowrotethat?

local var `written' can at least be made local to the first loop.

write_kmem() has a lot of typecasts which indicates that the choices of types were inappropriate.

#3 lore raw

Re: [PATCH 1/3] devmem: remove redundant test on len

Quoted 36 lines · Andrew Morton
Quoted 33 lines · On Fri, 11 Sep 2009 10:23:34 +0800 Wu Fengguang <[email protected]> wrote:

The len test in write_kmem() is always true, so can be reduced.

CC Marcelo Tosatti <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Mark Brown <[email protected]>
CC Johannes Berg <[email protected]>
CC Avi Kivity <[email protected]>
---
 drivers/char/mem.c |   14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)
drivers/char/mem.c +5 -8
@@ -580,18 +580,16 @@ static ssize_t write_kmem(struct file * 
 		while (count > 0) {
 			int len = count;
 
 			if (len > PAGE_SIZE)
 				len = PAGE_SIZE;
-			if (len) {
-				written = copy_from_user(kbuf, buf, len);
-				if (written) {
-					if (wrote + virtr)
-						break;
-					free_page((unsigned long)kbuf);
-					return -EFAULT;
-				}
+			written = copy_from_user(kbuf, buf, len);
+			if (written) {
+				if (wrote + virtr)
+					break;
+				free_page((unsigned long)kbuf);
Quoted 8 lines

drivers/char/mem.c static ssize_t write_kmem(struct file * · line 595

Quoted 6 lines
drivers/char/mem.c +1 -0
+				return -EFAULT;
 			}
 			len = vwrite(kbuf, (char *)p, len);
 			count -= len;
 			buf += len;
 			virtr += len;

humpf. But take a closer look at what remains.

Yeah, it asks for more cleanup patches..

Quoted 6 lines

Local var `written' is unneeded here.

Which makes us look at what `written' _does_ do:

Quoted 7 lines
		if (written != wrote)
			return written;
		wrote = written;

lolwhowrotethat?

local var `written' can at least be made local to the first loop.

Right, I'll make this straight.

Quoted 2 lines

write_kmem() has a lot of typecasts which indicates that the choices of types were inappropriate.

It seems hard to improve because of a fundamental issue: the same value is taken as both virtual and physical address, and also be compared and calculated as numbers..

Thanks,
Fengguang

#4 lore raw

PATCH 2/3 devmem: introduce size_inside_page()

In reply to Wu Fengguang

Introduce size_inside_page() to replace duplicate /dev/mem code.

Also apply it to /dev/kmem, whose alignment logic was buggy.

CC Marcelo Tosatti <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Mark Brown <[email protected]>
CC Johannes Berg <[email protected]>
CC Avi Kivity <[email protected]>
---
 drivers/char/mem.c |   60 +++++++++++++------------------------------
 1 file changed, 19 insertions(+), 41 deletions(-)
drivers/char/mem.c +19 -41
@@ -35,6 +35,19 @@
 # include <linux/efi.h>
 #endif
 
+static inline unsigned long size_inside_page(unsigned long start,
+					     unsigned long size)
+{
+	unsigned long sz;
+
+	if (-start & (PAGE_SIZE - 1))
+		sz = -start & (PAGE_SIZE - 1);
+	else
+		sz = PAGE_SIZE;
+
+	return min_t(unsigned long, sz, size);
+}
+
 /*
  * Architectures vary in how they handle caching for addresses
  * outside of main memory.
@@ -142,15 +155,7 @@ static ssize_t read_mem(struct file * fi
 #endif
 
 	while (count > 0) {
-		/*
-		 * Handle first page in case it's not aligned
-		 */
-		if (-p & (PAGE_SIZE - 1))
-			sz = -p & (PAGE_SIZE - 1);
-		else
-			sz = PAGE_SIZE;
-
-		sz = min_t(unsigned long, sz, count);
+		sz = size_inside_page(p, count);
 
 		if (!range_is_allowed(p >> PAGE_SHIFT, count))
 			return -EPERM;
@@ -209,15 +214,7 @@ static ssize_t write_mem(struct file * f
 #endif
 
 	while (count > 0) {
-		/*
-		 * Handle first page in case it's not aligned
-		 */
-		if (-p & (PAGE_SIZE - 1))
-			sz = -p & (PAGE_SIZE - 1);
-		else
-			sz = PAGE_SIZE;
-
-		sz = min_t(unsigned long, sz, count);
+		sz = size_inside_page(p, count);
 
 		if (!range_is_allowed(p >> PAGE_SHIFT, sz))
 			return -EPERM;
@@ -430,15 +427,7 @@ static ssize_t read_kmem(struct file *fi
 		}
 #endif
 		while (low_count > 0) {
-			/*
-			 * Handle first page in case it's not aligned
-			 */
-			if (-p & (PAGE_SIZE - 1))
-				sz = -p & (PAGE_SIZE - 1);
-			else
-				sz = PAGE_SIZE;
-
-			sz = min_t(unsigned long, sz, low_count);
+			sz = size_inside_page(p, low_count);
 
 			/*
 			 * On ia64 if a page has been mapped somewhere as
@@ -462,10 +451,8 @@ static ssize_t read_kmem(struct file *fi
 		if (!kbuf)
 			return -ENOMEM;
 		while (count > 0) {
-			int len = count;
+			int len = size_inside_page(p, count);
 
-			if (len > PAGE_SIZE)
-				len = PAGE_SIZE;
 			len = vread(kbuf, (char *)p, len);
 			if (!len)
 				break;
@@ -510,15 +497,8 @@ do_write_kmem(void *p, unsigned long rea
 
 	while (count > 0) {
 		char *ptr;
-		/*
-		 * Handle first page in case it's not aligned
-		 */
-		if (-realp & (PAGE_SIZE - 1))
-			sz = -realp & (PAGE_SIZE - 1);
-		else
-			sz = PAGE_SIZE;
 
-		sz = min_t(unsigned long, sz, count);
+		sz = size_inside_page(realp, count);
 
 		/*
 		 * On ia64 if a page has been mapped somewhere as
@@ -578,10 +558,8 @@ static ssize_t write_kmem(struct file * 
 		if (!kbuf)
 			return wrote ? wrote : -ENOMEM;
 		while (count > 0) {
-			int len = count;
+			int len = size_inside_page(p, count);
 
-			if (len > PAGE_SIZE)
-				len = PAGE_SIZE;
 			written = copy_from_user(kbuf, buf, len);
 			if (written) {
 				if (wrote + virtr)
#5 lore raw

Re: [PATCH 2/3] devmem: introduce size_inside_page()

Quoted 18 lines · On Fri, 11 Sep 2009 10:23:35 +0800 Wu Fengguang <[email protected]> wrote:

Introduce size_inside_page() to replace duplicate /dev/mem code.

Also apply it to /dev/kmem, whose alignment logic was buggy.

CC Marcelo Tosatti <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Mark Brown <[email protected]>
CC Johannes Berg <[email protected]>
CC Avi Kivity <[email protected]>
---
 drivers/char/mem.c |   60 +++++++++++++------------------------------
 1 file changed, 19 insertions(+), 41 deletions(-)

drivers/char/mem.c line 44

Quoted 10 lines
drivers/char/mem.c +7 -0
@@ -35,6 +35,19 @@
 # include <linux/efi.h>
 #endif
 
+static inline unsigned long size_inside_page(unsigned long start,
+					     unsigned long size)
+{
+	unsigned long sz;
+
+	if (-start & (PAGE_SIZE - 1))
+		sz = -start & (PAGE_SIZE - 1);

What on earth is this doing? Negating an unsigned number?

Can we get rid of these party tricks and use something more conventional here? In a separate patch I guess.

drivers/char/mem.c line 48

Quoted 4 lines
drivers/char/mem.c +4 -0
+	else
+		sz = PAGE_SIZE;
+
+	return min_t(unsigned long, sz, size);

Can use min() here.

drivers/char/mem.c line 49

Quoted 1 line
drivers/char/mem.c +1 -0
+}

Please have a think about the types. Should we be using unsigned long, or size_t? Which makes more sense? Which maps better onto reality?

I suspect that the min_t which you inherited was added somewhere because someone didn't get the types right: int-vs-size_t or something. If we actually get the types right, this sort of thing goes away.

drivers/char/mem.c static ssize_t read_kmem(struct file *fi · line 454

Quoted 6 lines
drivers/char/mem.c +1 -1
@@ -462,10 +451,8 @@ static ssize_t read_kmem(struct file *fi
 		if (!kbuf)
 			return -ENOMEM;
 		while (count > 0) {
-			int len = count;
+			int len = size_inside_page(p, count);

int?

#6 lore raw

Re: [PATCH 2/3] devmem: introduce size_inside_page()

Quoted 34 lines · Andrew Morton
Quoted 18 lines · On Fri, 11 Sep 2009 10:23:35 +0800 Wu Fengguang <[email protected]> wrote:

Introduce size_inside_page() to replace duplicate /dev/mem code.

Also apply it to /dev/kmem, whose alignment logic was buggy.

CC Marcelo Tosatti <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Mark Brown <[email protected]>
CC Johannes Berg <[email protected]>
CC Avi Kivity <[email protected]>
---
 drivers/char/mem.c |   60 +++++++++++++------------------------------
 1 file changed, 19 insertions(+), 41 deletions(-)

drivers/char/mem.c line 44

Quoted 10 lines
drivers/char/mem.c +7 -0
@@ -35,6 +35,19 @@
 # include <linux/efi.h>
 #endif
 
+static inline unsigned long size_inside_page(unsigned long start,
+					     unsigned long size)
+{
+	unsigned long sz;
+
+	if (-start & (PAGE_SIZE - 1))
+		sz = -start & (PAGE_SIZE - 1);

What on earth is this doing? Negating an unsigned number?

Quoted 2 lines

Can we get rid of these party tricks and use something more conventional here? In a separate patch I guess.

OK. See the followed patches.

Quoted 6 lines

drivers/char/mem.c line 48

Quoted 4 lines
drivers/char/mem.c +4 -0
+	else
+		sz = PAGE_SIZE;
+
+	return min_t(unsigned long, sz, size);

Can use min() here.

Done.

Quoted 8 lines

drivers/char/mem.c line 49

Quoted 1 line
drivers/char/mem.c +1 -0
+}

Please have a think about the types. Should we be using unsigned long, or size_t? Which makes more sense? Which maps better onto reality?

I suspect that the min_t which you inherited was added somewhere because someone didn't get the types right: int-vs-size_t or something. If we actually get the types right, this sort of thing goes away.

I tend to just use unsigned long because even though the value itself is small, it will be elevated to unsigned long in majority use cases.

Does that make sense?

Quoted 9 lines

drivers/char/mem.c static ssize_t read_kmem(struct file *fi · line 454

Quoted 7 lines
drivers/char/mem.c +1 -1
@@ -462,10 +451,8 @@ static ssize_t read_kmem(struct file *fi
 		if (!kbuf)
 			return -ENOMEM;
 		while (count > 0) {
-			int len = count;
+			int len = size_inside_page(p, count);

int?

Err, changed it to unsigned long sz.

Thanks,
Fengguang

#7 lore raw

PATCH 3/3 devmem: cleanup unxlate_dev_mem_ptr() calls

In reply to Wu Fengguang

No behavior change.

CC Marcelo Tosatti <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Mark Brown <[email protected]>
CC Johannes Berg <[email protected]>
CC Avi Kivity <[email protected]>
---
 drivers/char/mem.c |   13 +++++--------
 1 file changed, 5 insertions(+), 8 deletions(-)
drivers/char/mem.c +5 -8
@@ -131,6 +131,7 @@ static ssize_t read_mem(struct file * fi
 			size_t count, loff_t *ppos)
 {
 	unsigned long p = *ppos;
+	unsigned long ret;
 	ssize_t read, sz;
 	char *ptr;
 
@@ -169,12 +170,10 @@ static ssize_t read_mem(struct file * fi
 		if (!ptr)
 			return -EFAULT;
 
-		if (copy_to_user(buf, ptr, sz)) {
-			unxlate_dev_mem_ptr(p, ptr);
-			return -EFAULT;
-		}
-
+		ret = copy_to_user(buf, ptr, sz);
 		unxlate_dev_mem_ptr(p, ptr);
+		if (ret)
+			return -EFAULT;
 
 		buf += sz;
 		p += sz;
@@ -232,16 +231,14 @@ static ssize_t write_mem(struct file * f
 		}
 
 		copied = copy_from_user(ptr, buf, sz);
+		unxlate_dev_mem_ptr(p, ptr);
 		if (copied) {
 			written += sz - copied;
-			unxlate_dev_mem_ptr(p, ptr);
 			if (written)
 				break;
 			return -EFAULT;
 		}
 
-		unxlate_dev_mem_ptr(p, ptr);
-
 		buf += sz;
 		p += sz;
 		count -= sz;
#8 lore raw

Re: [PATCH 3/3] devmem: cleanup unxlate_dev_mem_ptr() calls

Quoted 28 lines · On Fri, 11 Sep 2009 10:23:36 +0800 Wu Fengguang <[email protected]> wrote:

No behavior change.

CC Marcelo Tosatti <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Mark Brown <[email protected]>
CC Johannes Berg <[email protected]>
CC Avi Kivity <[email protected]>
---
 drivers/char/mem.c |   13 +++++--------
 1 file changed, 5 insertions(+), 8 deletions(-)
drivers/char/mem.c +1 -2
@@ -131,6 +131,7 @@ static ssize_t read_mem(struct file * fi
 			size_t count, loff_t *ppos)
 {
 	unsigned long p = *ppos;
+	unsigned long ret;
 	ssize_t read, sz;
 	char *ptr;
 
@@ -169,12 +170,10 @@ static ssize_t read_mem(struct file * fi
 		if (!ptr)
 			return -EFAULT;
 
-		if (copy_to_user(buf, ptr, sz)) {
-			unxlate_dev_mem_ptr(p, ptr);

drivers/char/mem.c static ssize_t read_mem(struct file * fi · line 179

Quoted 10 lines
drivers/char/mem.c +3 -3
-			return -EFAULT;
-		}
-
+		ret = copy_to_user(buf, ptr, sz);
 		unxlate_dev_mem_ptr(p, ptr);
+		if (ret)
+			return -EFAULT;
 
 		buf += sz;
 		p += sz;

- local var `ret' didn't need function-wide scope. I think it's better to reduce its scope if poss.

- conventionally the identifier ret' refers to "the value which this function will return". Ditto retval' and `rc'.

But that's not what ret' does here so let's call it something else? remaining' is rather verbose and formal, but accurate.

drivers/char/mem.c renamed from drivers/char/mem.c~dev-mem-cleanup-unxlate_dev_mem_ptr-calls-fix +4 -3
@@ -131,7 +131,6 @@ static ssize_t read_mem(struct file * fi
 			size_t count, loff_t *ppos)
 {
 	unsigned long p = *ppos;
-	unsigned long ret;
 	ssize_t read, sz;
 	char *ptr;
 
@@ -156,6 +155,8 @@ static ssize_t read_mem(struct file * fi
 #endif
 
 	while (count > 0) {
+		unsigned long remaining;
+
 		sz = size_inside_page(p, count);
 
 		if (!range_is_allowed(p >> PAGE_SHIFT, count))
@@ -170,9 +171,9 @@ static ssize_t read_mem(struct file * fi
 		if (!ptr)
 			return -EFAULT;
 
-		ret = copy_to_user(buf, ptr, sz);
+		remaining = copy_to_user(buf, ptr, sz);
 		unxlate_dev_mem_ptr(p, ptr);
-		if (ret)
+		if (remaining)
 			return -EFAULT;
 
 		buf += sz;

_

#9 lore raw

Re: [PATCH 3/3] devmem: cleanup unxlate_dev_mem_ptr() calls

Quoted 39 lines · Andrew Morton
Quoted 36 lines · On Fri, 11 Sep 2009 10:23:36 +0800 Wu Fengguang <[email protected]> wrote:

No behavior change.

CC Marcelo Tosatti <[email protected]>
CC Greg Kroah-Hartman <[email protected]>
CC Mark Brown <[email protected]>
CC Johannes Berg <[email protected]>
CC Avi Kivity <[email protected]>
---
 drivers/char/mem.c |   13 +++++--------
 1 file changed, 5 insertions(+), 8 deletions(-)
drivers/char/mem.c +4 -5
@@ -131,6 +131,7 @@ static ssize_t read_mem(struct file * fi
 			size_t count, loff_t *ppos)
 {
 	unsigned long p = *ppos;
+	unsigned long ret;
 	ssize_t read, sz;
 	char *ptr;
 
@@ -169,12 +170,10 @@ static ssize_t read_mem(struct file * fi
 		if (!ptr)
 			return -EFAULT;
 
-		if (copy_to_user(buf, ptr, sz)) {
-			unxlate_dev_mem_ptr(p, ptr);
-			return -EFAULT;
-		}
-
+		ret = copy_to_user(buf, ptr, sz);
 		unxlate_dev_mem_ptr(p, ptr);
+		if (ret)
+			return -EFAULT;
 
Quoted 5 lines

drivers/char/mem.c static ssize_t read_mem(struct file * fi · line 179

Quoted 2 lines
drivers/char/mem.c +0 -0
 		buf += sz;
 		p += sz;

- local var `ret' didn't need function-wide scope. I think it's better to reduce its scope if poss.

OK.

Quoted 2 lines

- conventionally the identifier ret' refers to "the value which this function will return". Ditto retval' and `rc'.

Good to know that, thanks!

drivers/char/mem.c static ssize_t read_mem(struct file * fi · line 134

Quoted 11 lines

But that's not what ret' does here so let's call it something else? remaining' is rather verbose and formal, but accurate.

drivers/char/mem.c renamed from drivers/char/mem.c~dev-mem-cleanup-unxlate_dev_mem_ptr-calls-fix +0 -1
@@ -131,7 +131,6 @@ static ssize_t read_mem(struct file * fi
 			size_t count, loff_t *ppos)
 {
 	unsigned long p = *ppos;
-	unsigned long ret;

I got compile error because another place in read_mem() used ret. I'll send you an updated fix.

Thanks,
Fengguang

Quoted 25 lines
drivers/char/mem.c renamed from drivers/char/mem.c~dev-mem-cleanup-unxlate_dev_mem_ptr-calls-fix +4 -2
 	ssize_t read, sz;
 	char *ptr;
 
@@ -156,6 +155,8 @@ static ssize_t read_mem(struct file * fi
 #endif
 
 	while (count > 0) {
+		unsigned long remaining;
+
 		sz = size_inside_page(p, count);
 
 		if (!range_is_allowed(p >> PAGE_SHIFT, count))
@@ -170,9 +171,9 @@ static ssize_t read_mem(struct file * fi
 		if (!ptr)
 			return -EFAULT;
 
-		ret = copy_to_user(buf, ptr, sz);
+		remaining = copy_to_user(buf, ptr, sz);
 		unxlate_dev_mem_ptr(p, ptr);
-		if (ret)
+		if (remaining)
 			return -EFAULT;
 
 		buf += sz;

_

Acked-by on the v1 series
#10 lore raw
In reply to Wu Fengguang
Quoted 9 lines · Wu Fengguang

Hi Andrew,

Here are three patches for /dev/[k]mem.
They are mostly code cleanups, plus a minor bug fix:
if someone read high mem via /dev/kmem starting from an unaligned address, it's calculation of 'len' will go wrong.

I would recommend them for .32.

I reviewed the three patches and they all look good to me.

-Andi

Signature · 1 line
[email protected] -- Speaking for myself only.