UBI: R/O block driver on top of UBI volumes
authorEzequiel Garcia <ezequiel.garcia@free-electrons.com>
Tue, 25 Feb 2014 16:25:22 +0000 (13:25 -0300)
committerArtem Bityutskiy <artem.bityutskiy@linux.intel.com>
Fri, 28 Feb 2014 14:29:48 +0000 (16:29 +0200)
This commit introduces read-only block device emulation on top of UBI volumes.

Given UBI takes care of wear leveling and bad block management it's possible
to add a thin layer to enable block device access to UBI volumes.
This allows to use a block-oriented filesystem on a flash device.

The UBI block devices are meant to be used in conjunction with any
regular, block-oriented file system (e.g. ext4), although it's primarily
targeted at read-only file systems, such as squashfs.

Block devices are created upon user request through new ioctls:
UBI_IOCVOLATTBLK to attach and UBI_IOCVOLDETBLK to detach.
Also, a new UBI module parameter is added 'ubi.block'. This parameter is
needed in order to attach a block device on boot-up time, allowing to
mount the rootfs on a ubiblock device.
For instance, you could have these kernel parameters:

  ubi.mtd=5 ubi.block=0,0 root=/dev/ubiblock0_0

Or, if you compile ubi as a module:

  $ modprobe ubi mtd=/dev/mtd5 block=/dev/ubi0_0

Artem: amend commentaries and massage the patch a little bit.

Signed-off-by: Ezequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: Artem Bityutskiy <artem.bityutskiy@linux.intel.com>
drivers/mtd/ubi/Kconfig
drivers/mtd/ubi/Makefile
drivers/mtd/ubi/block.c [new file with mode: 0644]
drivers/mtd/ubi/build.c
drivers/mtd/ubi/cdev.c
drivers/mtd/ubi/ubi.h
include/uapi/mtd/ubi-user.h

index 36663af..783fb18 100644 (file)
@@ -87,4 +87,19 @@ config MTD_UBI_GLUEBI
           work on top of UBI. Do not enable this unless you use legacy
           software.
 
+config MTD_UBI_BLOCK
+       bool "Read-only block devices on top of UBI volumes"
+       default n
+       help
+          This option enables read-only UBI block devices support. UBI block
+          devices will be layered on top of UBI volumes, which means that the
+          UBI driver will transparently handle things like bad eraseblocks and
+          bit-flips. You can put any block-oriented file system on top of UBI
+          volumes in read-only mode (e.g., ext4), but it is probably most
+          practical for read-only file systems, like squashfs.
+
+          When selected, this feature will be built in the UBI driver.
+
+          If in doubt, say "N".
+
 endif # MTD_UBI
index b46b0c9..4e3c3d7 100644 (file)
@@ -3,5 +3,6 @@ obj-$(CONFIG_MTD_UBI) += ubi.o
 ubi-y += vtbl.o vmt.o upd.o build.o cdev.o kapi.o eba.o io.o wl.o attach.o
 ubi-y += misc.o debug.o
 ubi-$(CONFIG_MTD_UBI_FASTMAP) += fastmap.o
+ubi-$(CONFIG_MTD_UBI_BLOCK) += block.o
 
 obj-$(CONFIG_MTD_UBI_GLUEBI) += gluebi.o
diff --git a/drivers/mtd/ubi/block.c b/drivers/mtd/ubi/block.c
new file mode 100644 (file)
index 0000000..cea7d1c
--- /dev/null
@@ -0,0 +1,646 @@
+/*
+ * Copyright (c) 2014 Ezequiel Garcia
+ * Copyright (c) 2011 Free Electrons
+ *
+ * Driver parameter handling strongly based on drivers/mtd/ubi/build.c
+ *   Copyright (c) International Business Machines Corp., 2006
+ *   Copyright (c) Nokia Corporation, 2007
+ *   Authors: Artem Bityutskiy, Frank Haverkamp
+ *
+ * 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, version 2.
+ *
+ * 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.
+ */
+
+/*
+ * Read-only block devices on top of UBI volumes
+ *
+ * A simple implementation to allow a block device to be layered on top of a
+ * UBI volume. The implementation is provided by creating a static 1-to-1
+ * mapping between the block device and the UBI volume.
+ *
+ * The addressed byte is obtained from the addressed block sector, which is
+ * mapped linearly into the corresponding LEB:
+ *
+ *   LEB number = addressed byte / LEB size
+ *
+ * This feature is compiled in the UBI core, and adds a new 'block' parameter
+ * to allow early block device attaching. Runtime  block attach/detach for UBI
+ * volumes is provided through two new UBI ioctls: UBI_IOCVOLATTBLK and
+ * UBI_IOCVOLDETBLK.
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/err.h>
+#include <linux/kernel.h>
+#include <linux/list.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mtd/ubi.h>
+#include <linux/workqueue.h>
+#include <linux/blkdev.h>
+#include <linux/hdreg.h>
+#include <asm/div64.h>
+
+#include "ubi-media.h"
+#include "ubi.h"
+
+/* Maximum number of supported devices */
+#define UBIBLOCK_MAX_DEVICES 32
+
+/* Maximum length of the 'block=' parameter */
+#define UBIBLOCK_PARAM_LEN 63
+
+/* Maximum number of comma-separated items in the 'block=' parameter */
+#define UBIBLOCK_PARAM_COUNT 2
+
+struct ubiblock_param {
+       int ubi_num;
+       int vol_id;
+       char name[UBIBLOCK_PARAM_LEN+1];
+};
+
+/* Numbers of elements set in the @ubiblock_param array */
+static int ubiblock_devs __initdata;
+
+/* MTD devices specification parameters */
+static struct ubiblock_param ubiblock_param[UBIBLOCK_MAX_DEVICES] __initdata;
+
+struct ubiblock {
+       struct ubi_volume_desc *desc;
+       int ubi_num;
+       int vol_id;
+       int refcnt;
+       int leb_size;
+
+       struct gendisk *gd;
+       struct request_queue *rq;
+
+       struct workqueue_struct *wq;
+       struct work_struct work;
+
+       struct mutex dev_mutex;
+       spinlock_t queue_lock;
+       struct list_head list;
+};
+
+/* Linked list of all ubiblock instances */
+static LIST_HEAD(ubiblock_devices);
+static DEFINE_MUTEX(devices_mutex);
+static int ubiblock_major;
+
+static int __init ubiblock_set_param(const char *val,
+                                    const struct kernel_param *kp)
+{
+       int i, ret;
+       size_t len;
+       struct ubiblock_param *param;
+       char buf[UBIBLOCK_PARAM_LEN];
+       char *pbuf = &buf[0];
+       char *tokens[UBIBLOCK_PARAM_COUNT];
+
+       if (!val)
+               return -EINVAL;
+
+       len = strnlen(val, UBIBLOCK_PARAM_LEN);
+       if (len == 0) {
+               ubi_warn("block: empty 'block=' parameter - ignored\n");
+               return 0;
+       }
+
+       if (len == UBIBLOCK_PARAM_LEN) {
+               ubi_err("block: parameter \"%s\" is too long, max. is %d\n",
+                       val, UBIBLOCK_PARAM_LEN);
+               return -EINVAL;
+       }
+
+       strcpy(buf, val);
+
+       /* Get rid of the final newline */
+       if (buf[len - 1] == '\n')
+               buf[len - 1] = '\0';
+
+       for (i = 0; i < UBIBLOCK_PARAM_COUNT; i++)
+               tokens[i] = strsep(&pbuf, ",");
+
+       param = &ubiblock_param[ubiblock_devs];
+       if (tokens[1]) {
+               /* Two parameters: can be 'ubi, vol_id' or 'ubi, vol_name' */
+               ret = kstrtoint(tokens[0], 10, &param->ubi_num);
+               if (ret < 0)
+                       return -EINVAL;
+
+               /* Second param can be a number or a name */
+               ret = kstrtoint(tokens[1], 10, &param->vol_id);
+               if (ret < 0) {
+                       param->vol_id = -1;
+                       strcpy(param->name, tokens[1]);
+               }
+
+       } else {
+               /* One parameter: must be device path */
+               strcpy(param->name, tokens[0]);
+               param->ubi_num = -1;
+               param->vol_id = -1;
+       }
+
+       ubiblock_devs++;
+
+       return 0;
+}
+
+static const struct kernel_param_ops ubiblock_param_ops = {
+       .set    = ubiblock_set_param,
+};
+module_param_cb(block, &ubiblock_param_ops, NULL, 0);
+MODULE_PARM_DESC(block, "Attach block devices to UBI volumes. Parameter format: block=<path|dev,num|dev,name>.\n"
+                       "Multiple \"block\" parameters may be specified.\n"
+                       "UBI volumes may be specified by their number, name, or path to the device node.\n"
+                       "Examples\n"
+                       "Using the UBI volume path:\n"
+                       "ubi.block=/dev/ubi0_0\n"
+                       "Using the UBI device, and the volume name:\n"
+                       "ubi.block=0,rootfs\n"
+                       "Using both UBI device number and UBI volume number:\n"
+                       "ubi.block=0,0\n");
+
+static struct ubiblock *find_dev_nolock(int ubi_num, int vol_id)
+{
+       struct ubiblock *dev;
+
+       list_for_each_entry(dev, &ubiblock_devices, list)
+               if (dev->ubi_num == ubi_num && dev->vol_id == vol_id)
+                       return dev;
+       return NULL;
+}
+
+static int ubiblock_read_to_buf(struct ubiblock *dev, char *buffer,
+                               int leb, int offset, int len)
+{
+       int ret;
+
+       ret = ubi_read(dev->desc, leb, buffer, offset, len);
+       if (ret) {
+               ubi_err("%s ubi_read error %d",
+                       dev->gd->disk_name, ret);
+               return ret;
+       }
+       return 0;
+}
+
+static int ubiblock_read(struct ubiblock *dev, char *buffer,
+                        sector_t sec, int len)
+{
+       int ret, leb, offset;
+       int bytes_left = len;
+       int to_read = len;
+       loff_t pos = sec << 9;
+
+       /* Get LEB:offset address to read from */
+       offset = do_div(pos, dev->leb_size);
+       leb = pos;
+
+       while (bytes_left) {
+               /*
+                * We can only read one LEB at a time. Therefore if the read
+                * length is larger than one LEB size, we split the operation.
+                */
+               if (offset + to_read > dev->leb_size)
+                       to_read = dev->leb_size - offset;
+
+               ret = ubiblock_read_to_buf(dev, buffer, leb, offset, to_read);
+               if (ret)
+                       return ret;
+
+               buffer += to_read;
+               bytes_left -= to_read;
+               to_read = bytes_left;
+               leb += 1;
+               offset = 0;
+       }
+       return 0;
+}
+
+static int do_ubiblock_request(struct ubiblock *dev, struct request *req)
+{
+       int len, ret;
+       sector_t sec;
+
+       if (req->cmd_type != REQ_TYPE_FS)
+               return -EIO;
+
+       if (blk_rq_pos(req) + blk_rq_cur_sectors(req) >
+           get_capacity(req->rq_disk))
+               return -EIO;
+
+       if (rq_data_dir(req) != READ)
+               return -ENOSYS; /* Write not implemented */
+
+       sec = blk_rq_pos(req);
+       len = blk_rq_cur_bytes(req);
+
+       /*
+        * Let's prevent the device from being removed while we're doing I/O
+        * work. Notice that this means we serialize all the I/O operations,
+        * but it's probably of no impact given the NAND core serializes
+        * flash access anyway.
+        */
+       mutex_lock(&dev->dev_mutex);
+       ret = ubiblock_read(dev, req->buffer, sec, len);
+       mutex_unlock(&dev->dev_mutex);
+
+       return ret;
+}
+
+static void ubiblock_do_work(struct work_struct *work)
+{
+       struct ubiblock *dev =
+               container_of(work, struct ubiblock, work);
+       struct request_queue *rq = dev->rq;
+       struct request *req;
+       int res;
+
+       spin_lock_irq(rq->queue_lock);
+
+       req = blk_fetch_request(rq);
+       while (req) {
+
+               spin_unlock_irq(rq->queue_lock);
+               res = do_ubiblock_request(dev, req);
+               spin_lock_irq(rq->queue_lock);
+
+               /*
+                * If we're done with this request,
+                * we need to fetch a new one
+                */
+               if (!__blk_end_request_cur(req, res))
+                       req = blk_fetch_request(rq);
+       }
+
+       spin_unlock_irq(rq->queue_lock);
+}
+
+static void ubiblock_request(struct request_queue *rq)
+{
+       struct ubiblock *dev;
+       struct request *req;
+
+       dev = rq->queuedata;
+
+       if (!dev)
+               while ((req = blk_fetch_request(rq)) != NULL)
+                       __blk_end_request_all(req, -ENODEV);
+       else
+               queue_work(dev->wq, &dev->work);
+}
+
+static int ubiblock_open(struct block_device *bdev, fmode_t mode)
+{
+       struct ubiblock *dev = bdev->bd_disk->private_data;
+       int ret;
+
+       mutex_lock(&dev->dev_mutex);
+       if (dev->refcnt > 0) {
+               /*
+                * The volume is already open, just increase the reference
+                * counter.
+                */
+               goto out_done;
+       }
+
+       /*
+        * We want users to be aware they should only mount us as read-only.
+        * It's just a paranoid check, as write requests will get rejected
+        * in any case.
+        */
+       if (mode & FMODE_WRITE) {
+               ret = -EPERM;
+               goto out_unlock;
+       }
+
+       dev->desc = ubi_open_volume(dev->ubi_num, dev->vol_id, UBI_READONLY);
+       if (IS_ERR(dev->desc)) {
+               ubi_err("%s failed to open ubi volume %d_%d",
+                       dev->gd->disk_name, dev->ubi_num, dev->vol_id);
+               ret = PTR_ERR(dev->desc);
+               dev->desc = NULL;
+               goto out_unlock;
+       }
+
+out_done:
+       dev->refcnt++;
+       mutex_unlock(&dev->dev_mutex);
+       return 0;
+
+out_unlock:
+       mutex_unlock(&dev->dev_mutex);
+       return ret;
+}
+
+static void ubiblock_release(struct gendisk *gd, fmode_t mode)
+{
+       struct ubiblock *dev = gd->private_data;
+
+       mutex_lock(&dev->dev_mutex);
+       dev->refcnt--;
+       if (dev->refcnt == 0) {
+               ubi_close_volume(dev->desc);
+               dev->desc = NULL;
+       }
+       mutex_unlock(&dev->dev_mutex);
+}
+
+static int ubiblock_getgeo(struct block_device *bdev, struct hd_geometry *geo)
+{
+       /* Some tools might require this information */
+       geo->heads = 1;
+       geo->cylinders = 1;
+       geo->sectors = get_capacity(bdev->bd_disk);
+       geo->start = 0;
+       return 0;
+}
+
+static const struct block_device_operations ubiblock_ops = {
+       .owner = THIS_MODULE,
+       .open = ubiblock_open,
+       .release = ubiblock_release,
+       .getgeo = ubiblock_getgeo,
+};
+
+int ubiblock_add(struct ubi_volume_info *vi)
+{
+       struct ubiblock *dev;
+       struct gendisk *gd;
+       int disk_capacity;
+       int ret;
+
+       /* Check that the volume isn't already handled */
+       mutex_lock(&devices_mutex);
+       if (find_dev_nolock(vi->ubi_num, vi->vol_id)) {
+               mutex_unlock(&devices_mutex);
+               return -EEXIST;
+       }
+       mutex_unlock(&devices_mutex);
+
+       dev = kzalloc(sizeof(struct ubiblock), GFP_KERNEL);
+       if (!dev)
+               return -ENOMEM;
+
+       mutex_init(&dev->dev_mutex);
+
+       dev->ubi_num = vi->ubi_num;
+       dev->vol_id = vi->vol_id;
+       dev->leb_size = vi->usable_leb_size;
+
+       /* Initialize the gendisk of this ubiblock device */
+       gd = alloc_disk(1);
+       if (!gd) {
+               ubi_err("block: alloc_disk failed");
+               ret = -ENODEV;
+               goto out_free_dev;
+       }
+
+       gd->fops = &ubiblock_ops;
+       gd->major = ubiblock_major;
+       gd->first_minor = dev->ubi_num * UBI_MAX_VOLUMES + dev->vol_id;
+       gd->private_data = dev;
+       sprintf(gd->disk_name, "ubiblock%d_%d", dev->ubi_num, dev->vol_id);
+       disk_capacity = (vi->size * vi->usable_leb_size) >> 9;
+       set_capacity(gd, disk_capacity);
+       dev->gd = gd;
+
+       spin_lock_init(&dev->queue_lock);
+       dev->rq = blk_init_queue(ubiblock_request, &dev->queue_lock);
+       if (!dev->rq) {
+               ubi_err("block: blk_init_queue failed");
+               ret = -ENODEV;
+               goto out_put_disk;
+       }
+
+       dev->rq->queuedata = dev;
+       dev->gd->queue = dev->rq;
+
+       /*
+        * Create one workqueue per volume (per registered block device).
+        * Rembember workqueues are cheap, they're not threads.
+        */
+       dev->wq = alloc_workqueue(gd->disk_name, 0, 0);
+       if (!dev->wq)
+               goto out_free_queue;
+       INIT_WORK(&dev->work, ubiblock_do_work);
+
+       mutex_lock(&devices_mutex);
+       list_add_tail(&dev->list, &ubiblock_devices);
+       mutex_unlock(&devices_mutex);
+
+       /* Must be the last step: anyone can call file ops from now on */
+       add_disk(dev->gd);
+       ubi_msg("%s created from ubi%d:%d(%s)",
+               dev->gd->disk_name, dev->ubi_num, dev->vol_id, vi->name);
+       return 0;
+
+out_free_queue:
+       blk_cleanup_queue(dev->rq);
+out_put_disk:
+       put_disk(dev->gd);
+out_free_dev:
+       kfree(dev);
+
+       return ret;
+}
+
+static void ubiblock_cleanup(struct ubiblock *dev)
+{
+       del_gendisk(dev->gd);
+       blk_cleanup_queue(dev->rq);
+       ubi_msg("%s released", dev->gd->disk_name);
+       put_disk(dev->gd);
+}
+
+int ubiblock_del(struct ubi_volume_info *vi)
+{
+       struct ubiblock *dev;
+
+       mutex_lock(&devices_mutex);
+       dev = find_dev_nolock(vi->ubi_num, vi->vol_id);
+       if (!dev) {
+               mutex_unlock(&devices_mutex);
+               return -ENODEV;
+       }
+
+       /* Found a device, let's lock it so we can check if it's busy */
+       mutex_lock(&dev->dev_mutex);
+       if (dev->refcnt > 0) {
+               mutex_unlock(&dev->dev_mutex);
+               mutex_unlock(&devices_mutex);
+               return -EBUSY;
+       }
+
+       /* Remove from device list */
+       list_del(&dev->list);
+       mutex_unlock(&devices_mutex);
+
+       /* Flush pending work and stop this workqueue */
+       destroy_workqueue(dev->wq);
+
+       ubiblock_cleanup(dev);
+       mutex_unlock(&dev->dev_mutex);
+       kfree(dev);
+       return 0;
+}
+
+static void ubiblock_resize(struct ubi_volume_info *vi)
+{
+       struct ubiblock *dev;
+       int disk_capacity;
+
+       /*
+        * Need to lock the device list until we stop using the device,
+        * otherwise the device struct might get released in 'ubiblock_del()'.
+        */
+       mutex_lock(&devices_mutex);
+       dev = find_dev_nolock(vi->ubi_num, vi->vol_id);
+       if (!dev) {
+               mutex_unlock(&devices_mutex);
+               return;
+       }
+
+       mutex_lock(&dev->dev_mutex);
+       disk_capacity = (vi->size * vi->usable_leb_size) >> 9;
+       set_capacity(dev->gd, disk_capacity);
+       ubi_msg("%s resized to %d LEBs", dev->gd->disk_name, vi->size);
+       mutex_unlock(&dev->dev_mutex);
+       mutex_unlock(&devices_mutex);
+}
+
+static int ubiblock_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:
+               /*
+                * We want to enforce explicit block device attaching for
+                * volumes, so when a volume is added we do nothing.
+                */
+               break;
+       case UBI_VOLUME_REMOVED:
+               ubiblock_del(&nt->vi);
+               break;
+       case UBI_VOLUME_RESIZED:
+               ubiblock_resize(&nt->vi);
+               break;
+       default:
+               break;
+       }
+       return NOTIFY_OK;
+}
+
+static struct notifier_block ubiblock_notifier = {
+       .notifier_call = ubiblock_notify,
+};
+
+static struct ubi_volume_desc * __init
+open_volume_desc(const char *name, int ubi_num, int vol_id)
+{
+       if (ubi_num == -1)
+               /* No ubi num, name must be a vol device path */
+               return ubi_open_volume_path(name, UBI_READONLY);
+       else if (vol_id == -1)
+               /* No vol_id, must be vol_name */
+               return ubi_open_volume_nm(ubi_num, name, UBI_READONLY);
+       else
+               return ubi_open_volume(ubi_num, vol_id, UBI_READONLY);
+}
+
+static int __init ubiblock_attach_from_param(void)
+{
+       int i, ret;
+       struct ubiblock_param *p;
+       struct ubi_volume_desc *desc;
+       struct ubi_volume_info vi;
+
+       for (i = 0; i < ubiblock_devs; i++) {
+               p = &ubiblock_param[i];
+
+               desc = open_volume_desc(p->name, p->ubi_num, p->vol_id);
+               if (IS_ERR(desc)) {
+                       ubi_err("block: can't open volume, err=%ld\n",
+                               PTR_ERR(desc));
+                       ret = PTR_ERR(desc);
+                       break;
+               }
+
+               ubi_get_volume_info(desc, &vi);
+               ubi_close_volume(desc);
+
+               ret = ubiblock_add(&vi);
+               if (ret) {
+                       ubi_err("block: can't add '%s' volume, err=%d\n",
+                               vi.name, ret);
+                       break;
+               }
+       }
+       return ret;
+}
+
+static void ubiblock_detach_all(void)
+{
+       struct ubiblock *next;
+       struct ubiblock *dev;
+
+       list_for_each_entry_safe(dev, next, &ubiblock_devices, list) {
+               /* Flush pending work and stop workqueue */
+               destroy_workqueue(dev->wq);
+               /* The module is being forcefully removed */
+               WARN_ON(dev->desc);
+               /* Remove from device list */
+               list_del(&dev->list);
+               ubiblock_cleanup(dev);
+               kfree(dev);
+       }
+}
+
+int __init ubiblock_init(void)
+{
+       int ret;
+
+       ubiblock_major = register_blkdev(0, "ubiblock");
+       if (ubiblock_major < 0)
+               return ubiblock_major;
+
+       /* Attach block devices from 'block=' module param */
+       ret = ubiblock_attach_from_param();
+       if (ret)
+               goto err_detach;
+
+       /*
+        * Block devices needs to be attached to volumes explicitly
+        * upon user request. So we ignore existing volumes.
+        */
+       ret = ubi_register_volume_notifier(&ubiblock_notifier, 1);
+       if (ret)
+               goto err_unreg;
+       return 0;
+
+err_unreg:
+       unregister_blkdev(ubiblock_major, "ubiblock");
+err_detach:
+       ubiblock_detach_all();
+       return ret;
+}
+
+void __exit ubiblock_exit(void)
+{
+       ubi_unregister_volume_notifier(&ubiblock_notifier);
+       ubiblock_detach_all();
+       unregister_blkdev(ubiblock_major, "ubiblock");
+}
index 57deae9..6e30a3c 100644 (file)
@@ -1298,6 +1298,15 @@ static int __init ubi_init(void)
                }
        }
 
+       err = ubiblock_init();
+       if (err) {
+               ubi_err("block: cannot initialize, error %d", err);
+
+               /* See comment above re-ubi_is_module(). */
+               if (ubi_is_module())
+                       goto out_detach;
+       }
+
        return 0;
 
 out_detach:
@@ -1326,6 +1335,8 @@ static void __exit ubi_exit(void)
 {
        int i;
 
+       ubiblock_exit();
+
        for (i = 0; i < UBI_MAX_DEVICES; i++)
                if (ubi_devices[i]) {
                        mutex_lock(&ubi_devices_mutex);
index 8ca49f2..39d3774 100644 (file)
@@ -561,6 +561,26 @@ static long vol_cdev_ioctl(struct file *file, unsigned int cmd,
                break;
        }
 
+       /* Attach a block device to an UBI volume */
+       case UBI_IOCVOLATTBLK:
+       {
+               struct ubi_volume_info vi;
+
+               ubi_get_volume_info(desc, &vi);
+               err = ubiblock_add(&vi);
+               break;
+       }
+
+       /* Dettach a block device from an UBI volume */
+       case UBI_IOCVOLDETBLK:
+       {
+               struct ubi_volume_info vi;
+
+               ubi_get_volume_info(desc, &vi);
+               err = ubiblock_del(&vi);
+               break;
+       }
+
        default:
                err = -ENOTTY;
                break;
index 8ea6297..e76ff98 100644 (file)
@@ -864,6 +864,20 @@ int ubi_update_fastmap(struct ubi_device *ubi);
 int ubi_scan_fastmap(struct ubi_device *ubi, struct ubi_attach_info *ai,
                     int fm_anchor);
 
+/* block.c */
+#ifdef CONFIG_MTD_UBI_BLOCK
+int ubiblock_init(void);
+void ubiblock_exit(void);
+int ubiblock_add(struct ubi_volume_info *vi);
+int ubiblock_del(struct ubi_volume_info *vi);
+#else
+static inline int ubiblock_init(void) { return 0; }
+static inline void ubiblock_exit(void) {}
+static inline int ubiblock_add(struct ubi_volume_info *vi) { return -ENOTTY; }
+static inline int ubiblock_del(struct ubi_volume_info *vi) { return -ENOTTY; }
+#endif
+
+
 /*
  * ubi_rb_for_each_entry - walk an RB-tree.
  * @rb: a pointer to type 'struct rb_node' to use as a loop counter
index 723c324..b98585a 100644 (file)
  * used. A pointer to a &struct ubi_set_vol_prop_req object is expected to be
  * passed. The object describes which property should be set, and to which value
  * it should be set.
+ *
+ * Block devices on UBI volumes
+ * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+ *
+ * To attach or detach a block device from an UBI volume the %UBI_IOCVOLATTBLK
+ * and %UBI_IOCVOLDETBLK ioctl commands should be used, respectively.
+ * These commands take no arguments.
  */
 
 /*
 /* Set an UBI volume property */
 #define UBI_IOCSETVOLPROP _IOW(UBI_VOL_IOC_MAGIC, 6, \
                               struct ubi_set_vol_prop_req)
+/* Attach a block device to an UBI volume */
+#define UBI_IOCVOLATTBLK _IO(UBI_VOL_IOC_MAGIC, 7)
+/* Detach a block device from an UBI volume */
+#define UBI_IOCVOLDETBLK _IO(UBI_VOL_IOC_MAGIC, 8)
 
 /* Maximum MTD device name length supported by UBI */
 #define MAX_UBI_MTD_NAME_LEN 127