On Wed, Oct 29, 2008 at 08:13:19AM +0100, Borislav Petkov wrote:
> Hi Tejun,
>
> recent changes at 0762b8bde9729f10f8e6249809660ff2ec3ad735 and around
> break ide-floppy. Since it is a removable media drive and the partition
> scan during boot returns empty (no media in the drive), when you later
> put in a disk and try to mount it, mount returns saying
>
> /dev/hdc4 is not a valid block device.
>
> Which brings me to the other possible issue: Since having a hdc4
> partition as a single FAT16 partition on a ZIP drive is the "factory
> default" you could fabricate a case where you have a partition number> 1
> as the only partition on a hard drive too, i.e. no continuous
> partition numbering and the mount would theoretically fail there too
> since, for example, there's a check in disk_get_part() which does:
>
> if (likely(partno < ptbl->len)) {
>
> and in this case the check will fail if partno >= 1 while you have only
> one partition on the disk with a number higher than the partition table
> length and the above described failure will happen too. Anyways, this is
> just a hypothesis, but it happens with the ZIP drive here so other block
> devices should behave similarly.
>
> Here's a patch that fixes the ide-floppy case a bit clumsily, I admit.
>
> ---
> diff --git a/fs/block_dev.c b/fs/block_dev.c
> index 88a776f..b798ea0 100644
> --- a/fs/block_dev.c
> +++ b/fs/block_dev.c
> @@ -1011,6 +1011,23 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
> disk = get_gendisk(bdev->bd_dev, &partno);
> if (!disk)
> goto out_unlock_kernel;
> +
> + part = disk_get_part(disk, partno);
> + if (!part) {
> + struct block_device *whole;
> +
> + mutex_lock_nested(&bdev->bd_mutex, for_part);
> + whole = bdget_disk(disk, 0);
> + ret = -ENOMEM;
> + if (!whole)
> + goto out_clear;
> + ret = __blkdev_get(whole, mode, 1);
> + if (ret)
> + goto out_clear;
> + bdev->bd_contains = whole;
> + mutex_unlock(&bdev->bd_mutex);
> + }
> +
> part = disk_get_part(disk, partno);
> if (!part)
> goto out_unlock_kernel;
Tejun, Jens,
can you guys please ACK/NACK this so that we can have an agreed upon solution
and so that I can stop applying it ontop of current git before testing any
further.
Thanks.
--
Regards/Gruss,
Boris.
Borislav Petkov wrote:
> can you guys please ACK/NACK this so that we can have an agreed upon solution
> and so that I can stop applying it ontop of current git before testing any
> further.
Ah... right. I missed removable devices. I think it would be better to
just expand the table rather than recursing into __blkdev_get() again.
I'll try to come up with cleaner solution. Also, I don't think the
discontinuous partition would be a problem. ptbl is expanded to the
highest numbered partition not the number of partitions. I'll test that
too. Thanks.
--
tejun
Hi
On Mon, Nov 3, 2008 at 4:37 PM, Tejun Heo <[email protected]> wrote:
> Borislav Petkov wrote:
>> can you guys please ACK/NACK this so that we can have an agreed upon solution
>> and so that I can stop applying it ontop of current git before testing any
>> further.
>
> Ah... right. I missed removable devices. I think it would be better to
> just expand the table rather than recursing into __blkdev_get() again.
> I'll try to come up with cleaner solution.
Sounds good. It should simply rescan partitions upon media change before it does
disk_get_part so that the ptbl is updated.
> Also, I don't think the
> discontinuous partition would be a problem. ptbl is expanded to the
> highest numbered partition not the number of partitions. I'll test that
> too. Thanks.
You don't have to. I already did that with a harddrive in a external case over
usb and it works since upon connection the usb core rescans partitions and ptbl
is valid then.
Thanks.
--
Regards/Gruss,
Boris
Commit 0762b8bde9729f10f8e6249809660ff2ec3ad735 moved disk_get_part()
in front of recursive get on the whole disk, which caused removable
devices to try disk_get_part() before rescanning after a new media is
inserted, which might fail legit open attempts or give the old
partition.
This patch fixes the problem by moving disk_get_part() after
__blkdev_get() on the whole disk.
This problem was spotted by Borislav Petkov.
Signed-off-by: Tejun Heo <[email protected]>
Cc: <Borislav Petkov> [email protected]
---
Borislav, can you please verify this patch? Thanks.
fs/block_dev.c | 23 +++++++++++------------
1 file changed, 11 insertions(+), 12 deletions(-)
diff --git a/fs/block_dev.c b/fs/block_dev.c
index 88a776f..db831ef 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -986,7 +986,6 @@ static int __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part);
static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
{
struct gendisk *disk;
- struct hd_struct *part = NULL;
int ret;
int partno;
int perm = 0;
@@ -1004,24 +1003,25 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
return ret;
}
- ret = -ENXIO;
-
lock_kernel();
+ ret = -ENXIO;
disk = get_gendisk(bdev->bd_dev, &partno);
if (!disk)
goto out_unlock_kernel;
- part = disk_get_part(disk, partno);
- if (!part)
- goto out_unlock_kernel;
mutex_lock_nested(&bdev->bd_mutex, for_part);
if (!bdev->bd_openers) {
bdev->bd_disk = disk;
- bdev->bd_part = part;
bdev->bd_contains = bdev;
if (!partno) {
struct backing_dev_info *bdi;
+
+ ret = -ENXIO;
+ bdev->bd_part = disk_get_part(disk, partno);
+ if (!bdev->bd_part)
+ goto out_clear;
+
if (disk->fops->open) {
ret = disk->fops->open(bdev, mode);
if (ret)
@@ -1049,18 +1049,17 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
bdev->bd_contains = whole;
bdev->bd_inode->i_data.backing_dev_info =
whole->bd_inode->i_data.backing_dev_info;
+ bdev->bd_part = disk_get_part(disk, partno);
if (!(disk->flags & GENHD_FL_UP) ||
- !part || !part->nr_sects) {
+ !bdev->bd_part || !bdev->bd_part->nr_sects) {
ret = -ENXIO;
goto out_clear;
}
- bd_set_size(bdev, (loff_t)part->nr_sects << 9);
+ bd_set_size(bdev, (loff_t)bdev->bd_part->nr_sects << 9);
}
} else {
- disk_put_part(part);
put_disk(disk);
module_put(disk->fops->owner);
- part = NULL;
disk = NULL;
if (bdev->bd_contains == bdev) {
if (bdev->bd_disk->fops->open) {
@@ -1080,6 +1079,7 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
return 0;
out_clear:
+ disk_put_part(bdev->bd_part);
bdev->bd_disk = NULL;
bdev->bd_part = NULL;
bdev->bd_inode->i_data.backing_dev_info = &default_backing_dev_info;
@@ -1091,7 +1091,6 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode, int for_part)
out_unlock_kernel:
unlock_kernel();
- disk_put_part(part);
if (disk)
module_put(disk->fops->owner);
put_disk(disk);
On Tue, Nov 04, 2008 at 01:16:07PM +0900, Tejun Heo wrote:
> Commit 0762b8bde9729f10f8e6249809660ff2ec3ad735 moved disk_get_part()
> in front of recursive get on the whole disk, which caused removable
> devices to try disk_get_part() before rescanning after a new media is
> inserted, which might fail legit open attempts or give the old
> partition.
>
> This patch fixes the problem by moving disk_get_part() after
> __blkdev_get() on the whole disk.
>
> This problem was spotted by Borislav Petkov.
>
> Signed-off-by: Tejun Heo <[email protected]>
> Cc: <Borislav Petkov> [email protected]
> ---
> Borislav, can you please verify this patch? Thanks.
Yep, it works. Thanks.
Tested-by: Borislav Petkov <[email protected]>
--
Regards/Gruss,
Boris.
On Wed, Nov 05 2008, Borislav Petkov wrote:
> On Tue, Nov 04, 2008 at 01:16:07PM +0900, Tejun Heo wrote:
> > Commit 0762b8bde9729f10f8e6249809660ff2ec3ad735 moved disk_get_part()
> > in front of recursive get on the whole disk, which caused removable
> > devices to try disk_get_part() before rescanning after a new media is
> > inserted, which might fail legit open attempts or give the old
> > partition.
> >
> > This patch fixes the problem by moving disk_get_part() after
> > __blkdev_get() on the whole disk.
> >
> > This problem was spotted by Borislav Petkov.
> >
> > Signed-off-by: Tejun Heo <[email protected]>
> > Cc: <Borislav Petkov> [email protected]
> > ---
> > Borislav, can you please verify this patch? Thanks.
>
> Yep, it works. Thanks.
>
> Tested-by: Borislav Petkov <[email protected]>
I merged it and added your tested-by.
--
Jens Axboe