[RFC / BUG] mtd: provide proper 32/64-bit compat_ioctl() support for BLKPG

[RFC / BUG] mtd: provide proper 32/64-bit compat_ioctl() support for BLKPG

[Brian Norris]

After a bit of poking around wondering why my 32-bit user-space can't
seem to send a proper ioctl(BLKPG) to an MTD on my 64-bit kernel
(ARM64), I noticed that struct blkpg_ioctl_arg is actually pretty
unsuitable for use in the ioctl() ABI, due to its use of raw pointers,
and its lack of alignment/packing restrictions (32-bit arch'es tend to
pack the 4 fields into 4 32-bit words, whereas 64-bit arch'es would add
padding after the third int, and make this 6 32-bit words).

Anyway, this means BLKPG deserves some special compat_ioctl handling. Do
the conversion in a small shim for MTD.

The same bug applies to block/ioctl.c, but I wanted to get some comments
first. I can send a non-RFC with the same approach for the block
subsystem. But then: which tree should it go in?

Tested only on MTD, with an ARM32 user space on an ARM64 kernel.

Signed-off-by: Brian Norris <computersforpeace@gmail.com>
---
 drivers/mtd/mtdchar.c      | 42 +++++++++++++++++++++++++++++++++---------
 include/uapi/linux/blkpg.h | 10 ++++++++++
 2 files changed, 43 insertions(+), 9 deletions(-)

diff --git a/drivers/mtd/mtdchar.c b/drivers/mtd/mtdchar.c
index 55fa27ecf4e1..bf966be09e79 100644
--- a/drivers/mtd/mtdchar.c
+++ b/drivers/mtd/mtdchar.c
@@ -498,21 +498,17 @@ static int shrink_ecclayout(const struct nand_ecclayout *from,
 }
 
 static int mtdchar_blkpg_ioctl(struct mtd_info *mtd,
-			   struct blkpg_ioctl_arg __user *arg)
+			       struct blkpg_ioctl_arg *arg)
 {
-	struct blkpg_ioctl_arg a;
 	struct blkpg_partition p;
 
 	if (!capable(CAP_SYS_ADMIN))
 		return -EPERM;
 
-	if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg)))
+	if (copy_from_user(&p, arg->data, sizeof(p)))
 		return -EFAULT;
 
-	if (copy_from_user(&p, a.data, sizeof(struct blkpg_partition)))
-		return -EFAULT;
-
-	switch (a.op) {
+	switch (arg->op) {
 	case BLKPG_ADD_PARTITION:
 
 		/* Only master mtd device must be used to add partitions */
@@ -966,8 +962,13 @@ static int mtdchar_ioctl(struct file *file, u_int cmd, u_long arg)
 
 	case BLKPG:
 	{
-		ret = mtdchar_blkpg_ioctl(mtd,
-		      (struct blkpg_ioctl_arg __user *)arg);
+		struct blkpg_ioctl_arg __user *blk_arg = argp;
+		struct blkpg_ioctl_arg a;
+
+		if (copy_from_user(&a, blk_arg, sizeof(a)))
+			ret = -EFAULT;
+		else
+			ret = mtdchar_blkpg_ioctl(mtd, &a);
 		break;
 	}
 
@@ -1046,6 +1047,29 @@ static long mtdchar_compat_ioctl(struct file *file, unsigned int cmd,
 				&buf_user->start);
 		break;
 	}
+
+	case BLKPG:
+	{
+		/* Convert from blkpg_compat_ioctl_arg to blkpg_ioctl_arg */
+		struct blkpg_compat_ioctl_arg __user *uarg = argp;
+		struct blkpg_compat_ioctl_arg arg;
+		struct blkpg_ioctl_arg a;
+
+		if (copy_from_user(&arg, uarg, sizeof(arg))) {
+			ret = -EFAULT;
+			break;
+		}
+
+		memset(&a, 0, sizeof(a));
+		a.op = arg.op;
+		a.flags = arg.flags;
+		a.datalen = arg.datalen;
+		a.data = compat_ptr(arg.data);
+
+		ret = mtdchar_blkpg_ioctl(mtd, &a);
+		break;
+	}
+
 	default:
 		ret = mtdchar_ioctl(file, cmd, (unsigned long)argp);
 	}
diff --git a/include/uapi/linux/blkpg.h b/include/uapi/linux/blkpg.h
index a8519446c111..0574147f4490 100644
--- a/include/uapi/linux/blkpg.h
+++ b/include/uapi/linux/blkpg.h
@@ -26,6 +26,7 @@
  */
 #include <linux/compiler.h>
 #include <linux/ioctl.h>
+#include <linux/compat.h>
 
 #define BLKPG      _IO(0x12,105)
 
@@ -37,6 +38,15 @@ struct blkpg_ioctl_arg {
         void __user *data;
 };
 
+#ifdef CONFIG_COMPAT
+struct blkpg_compat_ioctl_arg {
+	compat_int_t op;
+	compat_int_t flags;
+	compat_int_t datalen;
+	compat_uptr_t data;
+};
+#endif
+
 /* The subfunctions (for the op field) */
 #define BLKPG_ADD_PARTITION	1
 #define BLKPG_DEL_PARTITION	2
-- 
2.5.0.457.gab17608

Re: [RFC / BUG] mtd: provide proper 32/64-bit compat_ioctl() support for BLKPG

[Brian Norris]

On Fri, Aug 28, 2015 at 07:12:36PM -0700, Brian Norris wrote:
> The same bug applies to block/ioctl.c

Actually I'll correct myself: it looks like block/ already has OK
compat_ioctl support for this. Block devices do a copy to/from user
space to marshal a new struct. Personally, I find my approach a little
clearer. 

> But then: which tree should it go in?

I guess since this is only an MTD bug, then it'd be fair to take via
MTD. I'll plan to do that eventually if there are no objections.

> Tested only on MTD, with an ARM32 user space on an ARM64 kernel.
> 
> Signed-off-by: Brian Norris <computersforpeace@gmail.com>
> ---
>  drivers/mtd/mtdchar.c      | 42 +++++++++++++++++++++++++++++++++---------
>  include/uapi/linux/blkpg.h | 10 ++++++++++
>  2 files changed, 43 insertions(+), 9 deletions(-)

I also suspect it might make more sense to move the compat definitions
to a new non-UAPI include/linux/blkpg.h. I'll probably put that in v2.

Brian

[PATCH] mtd: provide proper 32/64-bit compat_ioctl() support for BLKPG

[Brian Norris]

After a bit of poking around wondering why my 32-bit user-space can't
seem to send a proper ioctl(BLKPG) to an MTD on my 64-bit kernel
(ARM64), I noticed that struct blkpg_ioctl_arg is actually pretty
unsuitable for use in the ioctl() ABI, due to its use of raw pointers,
and its lack of alignment/packing restrictions (32-bit arch'es tend to
pack the 4 fields into 4 32-bit words, whereas 64-bit arch'es would add
padding after the third int, and make this 6 32-bit words).

Anyway, this means BLKPG deserves some special compat_ioctl handling. Do
the conversion in a small shim for MTD.

block/compat_ioctl.c already has compat support for the block subsystem,
but it does so by a re-marshalling data to/from user-space (see
compat_blkpg_ioctl()). Personally, I think this approach is cleaner.

Tested only on MTD, with an ARM32 user space on an ARM64 kernel.

Signed-off-by: Brian Norris <computersforpeace@gmail.com>
---
You can find the initial bug report / patch here:

  http://article.gmane.org/gmane.linux.kernel/2028927
  https://lkml.org/lkml/2015/8/28/594

Changes since RFC:
 * create new non-UAPI header, instead of cluttering the UAPI header with
   compat stuff
 * remove mention of "same bug in block/", since block/ solves this problem
   already, just in a slightly different way

 drivers/mtd/mtdchar.c      | 42 +++++++++++++++++++++++++++++++++---------
 include/linux/blkpg.h      | 21 +++++++++++++++++++++
 include/uapi/linux/blkpg.h |  6 +++---
 3 files changed, 57 insertions(+), 12 deletions(-)
 create mode 100644 include/linux/blkpg.h

diff --git a/drivers/mtd/mtdchar.c b/drivers/mtd/mtdchar.c
index 55fa27ecf4e1..bf966be09e79 100644
--- a/drivers/mtd/mtdchar.c
+++ b/drivers/mtd/mtdchar.c
@@ -498,21 +498,17 @@ static int shrink_ecclayout(const struct nand_ecclayout *from,
 }
 
 static int mtdchar_blkpg_ioctl(struct mtd_info *mtd,
-			   struct blkpg_ioctl_arg __user *arg)
+			       struct blkpg_ioctl_arg *arg)
 {
-	struct blkpg_ioctl_arg a;
 	struct blkpg_partition p;
 
 	if (!capable(CAP_SYS_ADMIN))
 		return -EPERM;
 
-	if (copy_from_user(&a, arg, sizeof(struct blkpg_ioctl_arg)))
+	if (copy_from_user(&p, arg->data, sizeof(p)))
 		return -EFAULT;
 
-	if (copy_from_user(&p, a.data, sizeof(struct blkpg_partition)))
-		return -EFAULT;
-
-	switch (a.op) {
+	switch (arg->op) {
 	case BLKPG_ADD_PARTITION:
 
 		/* Only master mtd device must be used to add partitions */
@@ -966,8 +962,13 @@ static int mtdchar_ioctl(struct file *file, u_int cmd, u_long arg)
 
 	case BLKPG:
 	{
-		ret = mtdchar_blkpg_ioctl(mtd,
-		      (struct blkpg_ioctl_arg __user *)arg);
+		struct blkpg_ioctl_arg __user *blk_arg = argp;
+		struct blkpg_ioctl_arg a;
+
+		if (copy_from_user(&a, blk_arg, sizeof(a)))
+			ret = -EFAULT;
+		else
+			ret = mtdchar_blkpg_ioctl(mtd, &a);
 		break;
 	}
 
@@ -1046,6 +1047,29 @@ static long mtdchar_compat_ioctl(struct file *file, unsigned int cmd,
 				&buf_user->start);
 		break;
 	}
+
+	case BLKPG:
+	{
+		/* Convert from blkpg_compat_ioctl_arg to blkpg_ioctl_arg */
+		struct blkpg_compat_ioctl_arg __user *uarg = argp;
+		struct blkpg_compat_ioctl_arg arg;
+		struct blkpg_ioctl_arg a;
+
+		if (copy_from_user(&arg, uarg, sizeof(arg))) {
+			ret = -EFAULT;
+			break;
+		}
+
+		memset(&a, 0, sizeof(a));
+		a.op = arg.op;
+		a.flags = arg.flags;
+		a.datalen = arg.datalen;
+		a.data = compat_ptr(arg.data);
+
+		ret = mtdchar_blkpg_ioctl(mtd, &a);
+		break;
+	}
+
 	default:
 		ret = mtdchar_ioctl(file, cmd, (unsigned long)argp);
 	}
diff --git a/include/linux/blkpg.h b/include/linux/blkpg.h
new file mode 100644
index 000000000000..bef124fde61e
--- /dev/null
+++ b/include/linux/blkpg.h
@@ -0,0 +1,21 @@
+#ifndef _LINUX_BLKPG_H
+#define _LINUX_BLKPG_H
+
+/*
+ * Partition table and disk geometry handling
+ */
+
+#include <linux/compat.h>
+#include <uapi/linux/blkpg.h>
+
+#ifdef CONFIG_COMPAT
+/* For 32-bit/64-bit compatibility of struct blkpg_ioctl_arg */
+struct blkpg_compat_ioctl_arg {
+	compat_int_t op;
+	compat_int_t flags;
+	compat_int_t datalen;
+	compat_uptr_t data;
+};
+#endif
+
+#endif /* _LINUX_BLKPG_H */
diff --git a/include/uapi/linux/blkpg.h b/include/uapi/linux/blkpg.h
index a8519446c111..63739a035085 100644
--- a/include/uapi/linux/blkpg.h
+++ b/include/uapi/linux/blkpg.h
@@ -1,5 +1,5 @@
-#ifndef _LINUX_BLKPG_H
-#define _LINUX_BLKPG_H
+#ifndef _UAPI__LINUX_BLKPG_H
+#define _UAPI__LINUX_BLKPG_H
 
 /*
  * Partition table and disk geometry handling
@@ -56,4 +56,4 @@ struct blkpg_partition {
 	char volname[BLKPG_VOLNAMELTH];	/* volume label */
 };
 
-#endif /* _LINUX_BLKPG_H */
+#endif /* _UAPI__LINUX_BLKPG_H */
-- 
2.6.0.rc0.131.gf624c3d

Re: [PATCH] mtd: provide proper 32/64-bit compat_ioctl() support for BLKPG

[Brian Norris]

On Mon, Sep 21, 2015 at 01:26:59PM -0700, Brian Norris wrote:
> After a bit of poking around wondering why my 32-bit user-space can't
> seem to send a proper ioctl(BLKPG) to an MTD on my 64-bit kernel
> (ARM64), I noticed that struct blkpg_ioctl_arg is actually pretty
> unsuitable for use in the ioctl() ABI, due to its use of raw pointers,
> and its lack of alignment/packing restrictions (32-bit arch'es tend to
> pack the 4 fields into 4 32-bit words, whereas 64-bit arch'es would add
> padding after the third int, and make this 6 32-bit words).
> 
> Anyway, this means BLKPG deserves some special compat_ioctl handling. Do
> the conversion in a small shim for MTD.
> 
> block/compat_ioctl.c already has compat support for the block subsystem,
> but it does so by a re-marshalling data to/from user-space (see
> compat_blkpg_ioctl()). Personally, I think this approach is cleaner.
> 
> Tested only on MTD, with an ARM32 user space on an ARM64 kernel.
> 
> Signed-off-by: Brian Norris <computersforpeace@gmail.com>
> ---
> You can find the initial bug report / patch here:
> 
>   http://article.gmane.org/gmane.linux.kernel/2028927
>   https://lkml.org/lkml/2015/8/28/594
> 
> Changes since RFC:
>  * create new non-UAPI header, instead of cluttering the UAPI header with
>    compat stuff
>  * remove mention of "same bug in block/", since block/ solves this problem
>    already, just in a slightly different way

Pushed to l2-mtd.git