diff mbox

UBI: new module ubiblk: block layer on top of UBI

Message ID 1308922482-14967-2-git-send-email-david.wagner@free-electrons.com
State New, archived
Headers show

Commit Message

David Wagner June 24, 2011, 1:34 p.m. UTC
From: David Wagner <david.wagner@free-electrons.com>

ubiblk is a read-only block layer on top of UBI.  It presents UBI volumes as
read-only block devices.

It is used by putting a block filesystem image on a UBI volume and then mounting
the corresponding device.

It uses the UBI API to register to UBI notifications (to dynamically create and
delete devices as volumes are added or removed) and to read from the volumes.

Known issues:
 * UBI_VOLUME_RESIZED notification hook isn't implemented yet ;

 * only squashfs is known to work. With ext2/3 and vfat, errors appear randomly
   in the middle of some files.  The reason for that is still unknown ;

 * the modules keeps a table of the devices which length is the maximum number
   of UBI volumes.  It should make use of a linked list.

A lot of code is taken from mtd_blkdevs and gluebi

Signed-off-by: David Wagner <david.wagner@free-electrons.com>
---
 drivers/mtd/ubi/Kconfig  |    9 +
 drivers/mtd/ubi/Makefile |    1 +
 drivers/mtd/ubi/ubiblk.c |  462 ++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 472 insertions(+), 0 deletions(-)
 create mode 100644 drivers/mtd/ubi/ubiblk.c

Comments

Artem Bityutskiy June 27, 2011, 7:26 p.m. UTC | #1
On Fri, 2011-06-24 at 15:34 +0200, david.wagner@free-electrons.com
wrote:
> +	/* Stolen from mtd_blkdevs.c */
> +	/* Create processing thread */
> +	dev->thread = kthread_run(ubi_ubiblk_thread, dev, "%s%d_%d",
> +				  "kubiblkd", dev->ubi_num, dev->vol_id);
> +	if (IS_ERR(dev->thread)) {
> +		ret = PTR_ERR(dev->thread);
> +		goto out_thread;
> +	}

Why we need a kernel thread? Could you please describe when exactly it
is needed and why we cannot avoid having it?

> +	ret = register_blkdev(0, "ubiblk");
> +	if (ret <= 0) {
> +		pr_err("UBIBLK: could not register_blkdev\n");
> +		return -ENODEV;
> +	}
> +	major = ret;
> +	pr_info("UBIBLK: device's major: %d\n", major);
> +
> +	mutex_init(&devtable_lock);
> +	ret = ubi_register_volume_notifier(&ubiblk_notifier, 0);
> +	if (ret < 0)
> +		return ret;

You should probably de-register the blkdev when you fail here.
David Wagner June 28, 2011, 11:35 a.m. UTC | #2
Hi,

On 06/27/2011 09:26 PM, Artem Bityutskiy wrote:
> On Fri, 2011-06-24 at 15:34 +0200, david.wagner@free-electrons.com
> wrote:
>> +	/* Stolen from mtd_blkdevs.c */
>> +	/* Create processing thread */
>> +	dev->thread = kthread_run(ubi_ubiblk_thread, dev, "%s%d_%d",
>> +				  "kubiblkd", dev->ubi_num, dev->vol_id);
>> +	if (IS_ERR(dev->thread)) {
>> +		ret = PTR_ERR(dev->thread);
>> +		goto out_thread;
>> +	}
> 
> Why we need a kernel thread? Could you please describe when exactly it
> is needed and why we cannot avoid having it?

Do you mean that there could be another/better way ?
I read that workqueues could be used for that but since they seem to
internally use kthreads, I don't see the advantage yet. Simpler API ?

I also tried without a kthread altogether (and call do_ubiblk_request
directly within the callback registered with blk_init_queue) but got
lost in locks/context debugging ...

It seems that do_ubiblk_request needs to be in process context because
there are thousands causes for blocking (locking, page fault, for
instance, are the one I encountered). And on the other hand,
blk_run_queue must not block ; So we need to wake the thread up and
return (what ubi_ubiblk_request does).
So, would this be a sufficient justification ?


It's probably possible, however, to have only one thread for the whole
module instead of having one for each volume ; but that seemed good
enough on first approach.


I fixed the read errors issue with filesystems != SquashFS, so they
should all work, now.
I'll send the next iteration, probably later today.
Matthieu CASTET June 28, 2011, 2:50 p.m. UTC | #3
Hi,

Artem Bityutskiy a écrit :
> On Fri, 2011-06-24 at 15:34 +0200, david.wagner@free-electrons.com
> wrote:
>> +	/* Stolen from mtd_blkdevs.c */
>> +	/* Create processing thread */
>> +	dev->thread = kthread_run(ubi_ubiblk_thread, dev, "%s%d_%d",
>> +				  "kubiblkd", dev->ubi_num, dev->vol_id);
>> +	if (IS_ERR(dev->thread)) {
>> +		ret = PTR_ERR(dev->thread);
>> +		goto out_thread;
>> +	}
> 
> Why we need a kernel thread? Could you please describe when exactly it
> is needed and why we cannot avoid having it?
Also what are the advantage against gluebi + mtdblock_ro ?
David Wagner June 28, 2011, 3:32 p.m. UTC | #4
On 06/28/2011 04:50 PM, Matthieu CASTET wrote:
> Hi,
>
[...]
> Also what are the advantage against gluebi + mtdblock_ro ?

The main advantage is a reduced number of layers ; I must say I cannot
see much more for now.  I could add that the Kconfig help of gluebi
advises not to use it except when needed by legacy software.
Artem Bityutskiy June 29, 2011, 6:25 a.m. UTC | #5
On Tue, 2011-06-28 at 17:32 +0200, David Wagner wrote:
> On 06/28/2011 04:50 PM, Matthieu CASTET wrote:
> > Hi,
> >
> [...]
> > Also what are the advantage against gluebi + mtdblock_ro ?
> 
> The main advantage is a reduced number of layers ; I must say I cannot
> see much more for now.  I could add that the Kconfig help of gluebi
> advises not to use it except when needed by legacy software.

Well, I think Matthieu has a valid poit, you should try to come up with
a set of advantages, otherwise why would this drivers be needed? Why
would people spend time reviewing it? May be less memcpy's? Do we do an
extra memcpy in gluebi? If yes, can we avoid doing this. Anyway, please,
try to sell this driver a bit better.
Artem Bityutskiy June 29, 2011, 6:52 a.m. UTC | #6
On Tue, 2011-06-28 at 13:35 +0200, David Wagner wrote:
> Hi,
> 
> On 06/27/2011 09:26 PM, Artem Bityutskiy wrote:
> > On Fri, 2011-06-24 at 15:34 +0200, david.wagner@free-electrons.com
> > wrote:
> >> +	/* Stolen from mtd_blkdevs.c */
> >> +	/* Create processing thread */
> >> +	dev->thread = kthread_run(ubi_ubiblk_thread, dev, "%s%d_%d",
> >> +				  "kubiblkd", dev->ubi_num, dev->vol_id);
> >> +	if (IS_ERR(dev->thread)) {
> >> +		ret = PTR_ERR(dev->thread);
> >> +		goto out_thread;
> >> +	}
> > 
> > Why we need a kernel thread? Could you please describe when exactly it
> > is needed and why we cannot avoid having it?
> 
> Do you mean that there could be another/better way ?

No, I just do not understand why it is there. I think this is juts block
layer's design, but I wanted you to explain this - the design, how block
requests are handled, and where exactly the thread is needed. I expected
you just have the explanation.
diff mbox

Patch

diff --git a/drivers/mtd/ubi/Kconfig b/drivers/mtd/ubi/Kconfig
index 4dcc752..389a996 100644
--- a/drivers/mtd/ubi/Kconfig
+++ b/drivers/mtd/ubi/Kconfig
@@ -60,4 +60,13 @@  config MTD_UBI_DEBUG
 	help
 	  This option enables UBI debugging.
 
+config MTD_UBI_UBIBLK
+	tristate "Read-only block transition layer on top of UBI"
+	help
+	   Read-only block interface on top of UBI.
+
+	   Creates block devices that can be used to mount read-only block
+	   filesystems. Caches reads on a raw UBI volume. Filesystems images
+	   can be flashed using ubiupdatevol or put in a UBI image with
+	   ubinize
 endif # MTD_UBI
diff --git a/drivers/mtd/ubi/Makefile b/drivers/mtd/ubi/Makefile
index c9302a5..354b2df 100644
--- a/drivers/mtd/ubi/Makefile
+++ b/drivers/mtd/ubi/Makefile
@@ -5,3 +5,4 @@  ubi-y += misc.o
 
 ubi-$(CONFIG_MTD_UBI_DEBUG) += debug.o
 obj-$(CONFIG_MTD_UBI_GLUEBI) += gluebi.o
+obj-$(CONFIG_MTD_UBI_UBIBLK) += ubiblk.o
diff --git a/drivers/mtd/ubi/ubiblk.c b/drivers/mtd/ubi/ubiblk.c
new file mode 100644
index 0000000..aab940a
--- /dev/null
+++ b/drivers/mtd/ubi/ubiblk.c
@@ -0,0 +1,462 @@ 
+/*
+ * Copyright (c) Free Electrons, 2011
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
+ * the GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ * Author: David Wagner
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/vmalloc.h>
+#include <linux/mtd/ubi.h>
+#include <linux/blkdev.h>
+#include <linux/kthread.h>
+#include <linux/mutex.h>
+#include "ubi.h"
+
+#define BLK_SIZE 512
+
+#define UBIBLK_MAX_DEVS (UBI_MAX_DEVICES * UBI_MAX_VOLUMES)
+
+/*
+ * Structure representing a ubiblk device, proxying a UBI volume
+ */
+struct ubiblk_dev {
+	struct ubi_volume_desc *vol_desc;
+	struct ubi_volume_info *vol_info;
+	int ubi_num;
+	int vol_id;
+
+	/* Block stuff */
+	struct gendisk *gd;
+	struct request_queue *rq;
+	struct task_struct *thread;
+
+	/* Protects the access to the UBI volume */
+	struct mutex lock;
+
+	/* Avoids concurrent accesses to the request queue */
+	spinlock_t queue_lock;
+};
+
+/*
+ * Contains the pointers to all ubiblk_dev instances
+ * TODO: use a linked list
+ */
+static struct ubiblk_dev *ubiblk_devs[UBIBLK_MAX_DEVS];
+static struct mutex devtable_lock;
+
+int major;
+static const struct block_device_operations ubiblk_ops;
+
+/*
+ * Read a LEB and fill the request buffer with the requested sector
+ */
+static int do_ubiblk_request(struct request *req, struct ubiblk_dev *dev)
+{
+	unsigned long start, len;
+	int offset;
+	int leb;
+	int ret;
+
+	start = blk_rq_pos(req) << 9;
+	len = blk_rq_cur_bytes(req);
+
+	/* We are always reading. No need to handle writing for now */
+
+	leb = start / dev->vol_info->usable_leb_size;
+	offset = start % dev->vol_info->usable_leb_size;
+
+	if (offset + len > dev->vol_info->usable_leb_size)
+		len = dev->vol_info->usable_leb_size - offset;
+
+	pr_debug("%s(%s) of sector %llu (LEB %d). start=%lu, len=%lu\n",
+		 __func__, rq_data_dir(req) ? "Write" : "Read",
+		 blk_rq_pos(req), leb, start, len);
+
+	ret = ubi_read(dev->vol_desc, leb, req->buffer, offset, len);
+
+	if (ret) {
+		pr_err("ubi_read error\n");
+		return ret;
+	}
+
+	pr_debug("ubi_read done.\n");
+
+	return 0;
+}
+
+static void ubi_ubiblk_request(struct request_queue *rq)
+{
+	struct ubiblk_dev *dev;
+	struct request *req = NULL;
+
+	dev = rq->queuedata;
+
+	if (!dev)
+		while ((req = blk_fetch_request(rq)) != NULL)
+			__blk_end_request_all(req, -ENODEV);
+	else
+		wake_up_process(dev->thread);
+}
+
+/*
+ * Open a UBI volume (get the volume descriptor)
+ */
+static int ubiblk_open(struct block_device *bdev, fmode_t mode)
+{
+	struct ubiblk_dev *dev = bdev->bd_disk->private_data;
+	pr_debug("%s() disk_name=%s, mode=%d\n", __func__,
+		 bdev->bd_disk->disk_name, mode);
+
+	dev->vol_desc = ubi_open_volume(dev->ubi_num, dev->vol_id,
+					UBI_READONLY);
+	if (!dev->vol_desc) {
+		pr_err("open_volume failed");
+		return -EINVAL;
+	}
+
+	dev->vol_info = kzalloc(sizeof(struct ubi_volume_info), GFP_KERNEL);
+	if (!dev->vol_info) {
+		ubi_close_volume(dev->vol_desc);
+		dev->vol_desc = NULL;
+		return -ENOMEM;
+	}
+	ubi_get_volume_info(dev->vol_desc, dev->vol_info);
+
+	return 0;
+}
+
+/*
+ * Close a UBI volume (close the volume descriptor)
+ */
+static int ubiblk_release(struct gendisk *gd, fmode_t mode)
+{
+	struct ubiblk_dev *dev = gd->private_data;
+	pr_debug("%s() disk_name=%s, mode=%d\n", __func__, gd->disk_name,
+		 mode);
+
+	kfree(dev->vol_info);
+	dev->vol_info = NULL;
+	if (dev->vol_desc) {
+		ubi_close_volume(dev->vol_desc);
+		dev->vol_desc = NULL;
+	}
+
+	return 0;
+}
+
+/*
+ * Loop on the block request queue and wait for new requests ; run them with
+ * do_ubiblk_request()
+ *
+ * Mostly stolen from mtd_blkdevs.c
+ */
+static int ubi_ubiblk_thread(void *arg)
+{
+	struct ubiblk_dev *dev = arg;
+	struct request_queue *rq = dev->rq;
+	struct request *req = NULL;
+
+	spin_lock_irq(rq->queue_lock);
+
+	while (!kthread_should_stop()) {
+		int res;
+
+		if (!req && !(req = blk_fetch_request(rq))) {
+			set_current_state(TASK_INTERRUPTIBLE);
+
+			if (kthread_should_stop())
+				set_current_state(TASK_RUNNING);
+
+			spin_unlock_irq(rq->queue_lock);
+			schedule();
+			spin_lock_irq(rq->queue_lock);
+			continue;
+		}
+
+		spin_unlock_irq(rq->queue_lock);
+
+		mutex_lock(&dev->lock);
+		res = do_ubiblk_request(req, dev);
+		pr_debug("return from request: %d\n", res);
+		mutex_unlock(&dev->lock);
+
+		spin_lock_irq(rq->queue_lock);
+
+		if (!__blk_end_request_cur(req, res))
+			req = NULL;
+	}
+
+	if (req)
+		__blk_end_request_all(req, -EIO);
+
+	spin_unlock_irq(rq->queue_lock);
+
+	return 0;
+}
+
+static int ubiblk_create(struct ubi_device_info *dev_info,
+			 struct ubi_volume_info *vol_info)
+{
+	struct ubiblk_dev *dev;
+	struct gendisk *gd;
+	int i;
+	int ret = 0;
+
+	mutex_lock(&devtable_lock);
+	for (i = 0; i < UBIBLK_MAX_DEVS; i++)
+		if (!ubiblk_devs[i])
+			break;
+
+	if (i == UBIBLK_MAX_DEVS) {
+		/* Shouldn't happen: UBI can't make more volumes than that */
+		pr_err("no slot left for a new ubiblk device.\n");
+		mutex_unlock(&devtable_lock);
+		return -ENOMEM;
+	}
+
+	dev = kzalloc(sizeof(struct ubiblk_dev), GFP_KERNEL);
+	if (!dev) {
+		pr_err("UBIBLK: ENOMEM when trying to create a new"
+		       "ubiblk dev\n");
+		mutex_unlock(&devtable_lock);
+		return -ENOMEM;
+	}
+	ubiblk_devs[i] = dev;
+	mutex_unlock(&devtable_lock);
+
+	mutex_init(&dev->lock);
+	mutex_lock(&dev->lock);
+
+	dev->ubi_num = vol_info->ubi_num;
+	dev->vol_id = vol_info->vol_id;
+
+	dev->vol_desc = ubi_open_volume(dev->ubi_num, dev->vol_id,
+					UBI_READONLY);
+	if (IS_ERR(dev->vol_desc)) {
+		pr_err("open_volume failed\n");
+		ret = PTR_ERR(dev->vol_desc);
+		goto out_vol;
+	}
+
+	dev->vol_info = kzalloc(sizeof(struct ubi_volume_info), GFP_KERNEL);
+	if (!dev->vol_info) {
+		ret = -ENOMEM;
+		goto out_info;
+	}
+	ubi_get_volume_info(dev->vol_desc, dev->vol_info);
+
+	pr_info("Got volume %s: device %d/volume %d of size %d\n",
+		dev->vol_info->name, dev->ubi_num, dev->vol_id,
+		dev->vol_info->size);
+
+	/* Initialize the gendisk of this ubiblk device */
+	gd = alloc_disk(1);
+	if (!gd) {
+		pr_err("alloc_disk failed\n");
+		ret = -ENODEV;
+		goto out_disk;
+	}
+
+	gd->fops = &ubiblk_ops;
+	gd->major = major;
+	gd->first_minor = dev->ubi_num * UBI_MAX_VOLUMES + dev->vol_id;
+	gd->private_data = dev;
+	sprintf(gd->disk_name, "ubiblk%d_%d", dev->ubi_num, dev->vol_id);
+	pr_debug("creating a gd '%s'\n", gd->disk_name);
+	set_capacity(gd,
+		     (dev->vol_info->size *
+		      dev->vol_info->usable_leb_size) >> 9);
+	set_disk_ro(gd, 1);
+	dev->gd = gd;
+
+	spin_lock_init(&dev->queue_lock);
+	dev->rq = blk_init_queue(ubi_ubiblk_request, &dev->queue_lock);
+	if (!dev->rq) {
+		pr_err("init_queue failed\n");
+		ret = -ENODEV;
+		goto out_queue;
+	}
+	dev->rq->queuedata = dev;
+	blk_queue_logical_block_size(dev->rq, BLK_SIZE);
+	dev->gd->queue = dev->rq;
+
+	/* Stolen from mtd_blkdevs.c */
+	/* Create processing thread */
+	dev->thread = kthread_run(ubi_ubiblk_thread, dev, "%s%d_%d",
+				  "kubiblkd", dev->ubi_num, dev->vol_id);
+	if (IS_ERR(dev->thread)) {
+		ret = PTR_ERR(dev->thread);
+		goto out_thread;
+	}
+
+	add_disk(dev->gd);
+	kfree(dev->vol_info);
+	dev->vol_info = NULL;
+	ubi_close_volume(dev->vol_desc);
+	dev->vol_desc = NULL;
+	mutex_unlock(&dev->lock);
+
+	return 0;
+
+out_thread:
+	blk_cleanup_queue(dev->rq);
+out_queue:
+	put_disk(dev->gd);
+out_disk:
+	kfree(dev->vol_info);
+	dev->vol_info = NULL;
+out_info:
+	ubi_close_volume(dev->vol_desc);
+	dev->vol_desc = NULL;
+out_vol:
+	mutex_unlock(&dev->lock);
+
+	return ret;
+}
+
+static int ubiblk_remove(struct ubi_volume_info *vol_info)
+{
+	int i;
+	struct ubiblk_dev *dev;
+
+	for (i = 0; i < UBIBLK_MAX_DEVS; i++) {
+		dev = ubiblk_devs[i];
+		if (dev && dev->ubi_num == vol_info->ubi_num &&
+		    dev->vol_id == vol_info->vol_id)
+			break;
+	}
+	if (i == UBIBLK_MAX_DEVS) {
+		pr_warn("Trying to remove %s, which is unknown from ubiblk\n",
+			vol_info->name);
+		return -ENODEV;
+	}
+
+	pr_info("Removing %s\n", vol_info->name);
+
+	if (dev->vol_desc)
+		ubi_close_volume(dev->vol_desc);
+
+	del_gendisk(dev->gd);
+	blk_cleanup_queue(dev->rq);
+	kthread_stop(dev->thread);
+	put_disk(dev->gd);
+
+	kfree(dev->vol_info);
+
+	kfree(ubiblk_devs[i]);
+	ubiblk_devs[i] = NULL;
+
+	return 0;
+}
+
+static int ubiblk_notify(struct notifier_block *nb,
+			 unsigned long notification_type, void *ns_ptr)
+{
+	struct ubi_notification *nt = ns_ptr;
+
+	switch (notification_type) {
+	case UBI_VOLUME_ADDED:
+		ubiblk_create(&nt->di, &nt->vi);
+		break;
+	case UBI_VOLUME_REMOVED:
+		/* TODO */
+		ubiblk_remove(&nt->vi);
+		break;
+	case UBI_VOLUME_RESIZED:
+		/* TODO. needed ? */
+		/* gluebi_resized(&nt->vi); */
+		break;
+	case UBI_VOLUME_UPDATED:
+		break;
+	case UBI_VOLUME_RENAMED:
+		break;
+	default:
+		break;
+	}
+	return NOTIFY_OK;
+}
+
+static const struct block_device_operations ubiblk_ops = {
+	.owner = THIS_MODULE,
+	.open = ubiblk_open,
+	.release = ubiblk_release,
+};
+
+static struct notifier_block ubiblk_notifier = {
+	.notifier_call = ubiblk_notify,
+};
+
+/*
+ * Initialize the module
+ */
+static int __init ubi_ubiblk_init(void)
+{
+	int ret;
+
+	pr_info("UBIBLK starting\n");
+
+	ret = register_blkdev(0, "ubiblk");
+	if (ret <= 0) {
+		pr_err("UBIBLK: could not register_blkdev\n");
+		return -ENODEV;
+	}
+	major = ret;
+	pr_info("UBIBLK: device's major: %d\n", major);
+
+	mutex_init(&devtable_lock);
+	ret = ubi_register_volume_notifier(&ubiblk_notifier, 0);
+	if (ret < 0)
+		return ret;
+
+	return 0;
+}
+
+static void __exit ubi_ubiblk_exit(void)
+{
+	int i;
+
+	pr_info("UBIBLK: going to exit\n");
+
+	ubi_unregister_volume_notifier(&ubiblk_notifier);
+
+	for (i = 0; i < UBIBLK_MAX_DEVS; i++) {
+		struct ubiblk_dev *dev = ubiblk_devs[i];
+		if (!dev)
+			continue;
+
+		if (dev->vol_desc)
+			ubi_close_volume(dev->vol_desc);
+
+		del_gendisk(dev->gd);
+		blk_cleanup_queue(dev->rq);
+		kthread_stop(dev->thread);
+		put_disk(dev->gd);
+
+		kfree(dev->vol_info);
+		kfree(dev);
+	}
+
+	unregister_blkdev(major, "ubiblk");
+	pr_info("UBIBLK: The End\n");
+}
+
+module_init(ubi_ubiblk_init);
+module_exit(ubi_ubiblk_exit);
+MODULE_DESCRIPTION("Read-only block transition layer on top of UBI");
+MODULE_AUTHOR("David Wagner");
+MODULE_LICENSE("GPL");