From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755244Ab0CCVIs (ORCPT ); Wed, 3 Mar 2010 16:08:48 -0500 Received: from mail-bw0-f209.google.com ([209.85.218.209]:41411 "EHLO mail-bw0-f209.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753162Ab0CCVIn (ORCPT ); Wed, 3 Mar 2010 16:08:43 -0500 Subject: [PATCH] Forcing capacity sizes for block devices (RFC) From: Clemens Fruhwirth To: linux-kernel@vger.kernel.org Content-Type: text/plain; charset="ISO-8859-1" Date: Wed, 03 Mar 2010 22:08:25 +0100 Message-ID: <1267650505.29681.20.camel@ghanima> Mime-Version: 1.0 X-Mailer: Evolution 2.28.2 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org With this mail I like to restart a discussion about a patch I proposed on linux-scsi. Unfortunately my attempt to restart the discussion there failed. That is why I like to try it with a broader audience. My problem starts with my SD card reader, which is kind of broken in the sense that it fails to detect the SD card size properly. My first mail to linux-scsi http://marc.info/?l=linux-scsi&m=125555540311307&w=2 explains why it is broken (ignored additional device flags) and also contains a patch to force the block device to a certain size. After the kernel internal limit on device addresses is removed, my SD card -- despite not knowing the correct size -- is able to handle the write requests correctly. My initial attempt was to patch the SCSI layer to raise the kernel internal device address limit to the correct size (using a new ioctl call + a little user space program). Grant Grundler gave the good suggestion to make the "size" attribute of the block device in /sys writable instead of using a new interface. I rewrote the patch: http://marc.info/?l=linux-scsi&m=126098230006530&w=2 Bartlomiej Zolnier joined http://marc.info/?l=linux-scsi&m=126143291721837&w=2 and expressed his preference for an automatic solution instead of manual intervention. He favored to use the partition table to get the device sizes (iirc that's the behavior we see on Windows). I tried to explain why this is a bad idea http://marc.info/?l=linux-scsi&m=126166205514087&w=2 and why manual intervention is still necessary, i.e. for unformated cards. The problem that is still open in my opinion is that the p->nr_sects = val; line in the following patch should be replaced by something that properly informs the block layer about the size change. set_capacity from include/linux/genhd.h is a potential candidate, which also sets nr_sects but only after dereferencing a gendisk structure, so it is no drop-in replacement for that line above. Feedback is appreciated. Original patch below. Make the size attribute of block devices writable via sysfs. This allows userspace to force device boundaries in cases where the device driver is unable to detect the correct size, as with spec-incompliant SD card readers. diff -uprN -X linux-2.6.32-vanilla/Documentation/dontdiff linux-2.6.32-vanilla/block/genhd.c linux-2.6.32-patched/block/genhd.c --- linux-2.6.32-vanilla/block/genhd.c 2009-12-03 04:51:21.000000000 +0100 +++ linux-2.6.32-patched/block/genhd.c 2009-12-15 14:38:24.962463446 +0100 @@ -865,7 +865,7 @@ static DEVICE_ATTR(range, S_IRUGO, disk_ static DEVICE_ATTR(ext_range, S_IRUGO, disk_ext_range_show, NULL); static DEVICE_ATTR(removable, S_IRUGO, disk_removable_show, NULL); static DEVICE_ATTR(ro, S_IRUGO, disk_ro_show, NULL); -static DEVICE_ATTR(size, S_IRUGO, part_size_show, NULL); +static DEVICE_ATTR(size, S_IRUGO | S_IWUSR, part_size_show, part_size_set); static DEVICE_ATTR(alignment_offset, S_IRUGO, disk_alignment_offset_show, NULL); static DEVICE_ATTR(capability, S_IRUGO, disk_capability_show, NULL); static DEVICE_ATTR(stat, S_IRUGO, part_stat_show, NULL); diff -uprN -X linux-2.6.32-vanilla/Documentation/dontdiff linux-2.6.32-vanilla/fs/partitions/check.c linux-2.6.32-patched/fs/partitions/check.c --- linux-2.6.32-vanilla/fs/partitions/check.c 2009-12-03 04:51:21.000000000 +0100 +++ linux-2.6.32-patched/fs/partitions/check.c 2009-12-19 10:38:58.745294907 +0100 @@ -219,6 +219,18 @@ ssize_t part_size_show(struct device *de return sprintf(buf, "%llu\n",(unsigned long long)p->nr_sects); } +ssize_t part_size_set(struct device *dev, + struct device_attribute *attr, char *buf, size_t len) +{ + struct hd_struct *p = dev_to_part(dev); + unsigned long long val; + if (strict_strtoull(buf, 0, &val)) + return -EINVAL; + + p->nr_sects = val; + return len; +} + ssize_t part_alignment_offset_show(struct device *dev, struct device_attribute *attr, char *buf) { diff -uprN -X linux-2.6.32-vanilla/Documentation/dontdiff linux-2.6.32-vanilla/include/linux/genhd.h linux-2.6.32-patched/include/linux/genhd.h --- linux-2.6.32-vanilla/include/linux/genhd.h 2009-12-03 04:51:21.000000000 +0100 +++ linux-2.6.32-patched/include/linux/genhd.h 2009-12-19 10:40:30.900871273 +0100 @@ -549,6 +549,9 @@ extern void blk_unregister_region(dev_t extern ssize_t part_size_show(struct device *dev, struct device_attribute *attr, char *buf); +extern ssize_t part_size_set(struct device *dev, + struct device_attribute *attr, char *buf, + size_t len); extern ssize_t part_stat_show(struct device *dev, struct device_attribute *attr, char *buf); extern ssize_t part_inflight_show(struct device *dev, Signed-off-by: Clemens Fruhwirth Reviewed-by: Grant Grundler