Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / char / raw.c
diff --git a/kernel/drivers/char/raw.c b/kernel/drivers/char/raw.c
new file mode 100644 (file)
index 0000000..5fc291c
--- /dev/null
@@ -0,0 +1,369 @@
+/*
+ * linux/drivers/char/raw.c
+ *
+ * Front-end raw character devices.  These can be bound to any block
+ * devices to provide genuine Unix raw character device semantics.
+ *
+ * We reserve minor number 0 for a control interface.  ioctl()s on this
+ * device are used to bind the other minor numbers to block devices.
+ */
+
+#include <linux/init.h>
+#include <linux/fs.h>
+#include <linux/major.h>
+#include <linux/blkdev.h>
+#include <linux/module.h>
+#include <linux/raw.h>
+#include <linux/capability.h>
+#include <linux/uio.h>
+#include <linux/cdev.h>
+#include <linux/device.h>
+#include <linux/mutex.h>
+#include <linux/gfp.h>
+#include <linux/compat.h>
+#include <linux/vmalloc.h>
+
+#include <asm/uaccess.h>
+
+struct raw_device_data {
+       struct block_device *binding;
+       int inuse;
+};
+
+static struct class *raw_class;
+static struct raw_device_data *raw_devices;
+static DEFINE_MUTEX(raw_mutex);
+static const struct file_operations raw_ctl_fops; /* forward declaration */
+
+static int max_raw_minors = MAX_RAW_MINORS;
+
+module_param(max_raw_minors, int, 0);
+MODULE_PARM_DESC(max_raw_minors, "Maximum number of raw devices (1-65536)");
+
+/*
+ * Open/close code for raw IO.
+ *
+ * We just rewrite the i_mapping for the /dev/raw/rawN file descriptor to
+ * point at the blockdev's address_space and set the file handle to use
+ * O_DIRECT.
+ *
+ * Set the device's soft blocksize to the minimum possible.  This gives the
+ * finest possible alignment and has no adverse impact on performance.
+ */
+static int raw_open(struct inode *inode, struct file *filp)
+{
+       const int minor = iminor(inode);
+       struct block_device *bdev;
+       int err;
+
+       if (minor == 0) {       /* It is the control device */
+               filp->f_op = &raw_ctl_fops;
+               return 0;
+       }
+
+       mutex_lock(&raw_mutex);
+
+       /*
+        * All we need to do on open is check that the device is bound.
+        */
+       bdev = raw_devices[minor].binding;
+       err = -ENODEV;
+       if (!bdev)
+               goto out;
+       igrab(bdev->bd_inode);
+       err = blkdev_get(bdev, filp->f_mode | FMODE_EXCL, raw_open);
+       if (err)
+               goto out;
+       err = set_blocksize(bdev, bdev_logical_block_size(bdev));
+       if (err)
+               goto out1;
+       filp->f_flags |= O_DIRECT;
+       filp->f_mapping = bdev->bd_inode->i_mapping;
+       if (++raw_devices[minor].inuse == 1)
+               file_inode(filp)->i_mapping =
+                       bdev->bd_inode->i_mapping;
+       filp->private_data = bdev;
+       mutex_unlock(&raw_mutex);
+       return 0;
+
+out1:
+       blkdev_put(bdev, filp->f_mode | FMODE_EXCL);
+out:
+       mutex_unlock(&raw_mutex);
+       return err;
+}
+
+/*
+ * When the final fd which refers to this character-special node is closed, we
+ * make its ->mapping point back at its own i_data.
+ */
+static int raw_release(struct inode *inode, struct file *filp)
+{
+       const int minor= iminor(inode);
+       struct block_device *bdev;
+
+       mutex_lock(&raw_mutex);
+       bdev = raw_devices[minor].binding;
+       if (--raw_devices[minor].inuse == 0)
+               /* Here  inode->i_mapping == bdev->bd_inode->i_mapping  */
+               inode->i_mapping = &inode->i_data;
+       mutex_unlock(&raw_mutex);
+
+       blkdev_put(bdev, filp->f_mode | FMODE_EXCL);
+       return 0;
+}
+
+/*
+ * Forward ioctls to the underlying block device.
+ */
+static long
+raw_ioctl(struct file *filp, unsigned int command, unsigned long arg)
+{
+       struct block_device *bdev = filp->private_data;
+       return blkdev_ioctl(bdev, 0, command, arg);
+}
+
+static int bind_set(int number, u64 major, u64 minor)
+{
+       dev_t dev = MKDEV(major, minor);
+       struct raw_device_data *rawdev;
+       int err = 0;
+
+       if (number <= 0 || number >= max_raw_minors)
+               return -EINVAL;
+
+       if (MAJOR(dev) != major || MINOR(dev) != minor)
+               return -EINVAL;
+
+       rawdev = &raw_devices[number];
+
+       /*
+        * This is like making block devices, so demand the
+        * same capability
+        */
+       if (!capable(CAP_SYS_ADMIN))
+               return -EPERM;
+
+       /*
+        * For now, we don't need to check that the underlying
+        * block device is present or not: we can do that when
+        * the raw device is opened.  Just check that the
+        * major/minor numbers make sense.
+        */
+
+       if (MAJOR(dev) == 0 && dev != 0)
+               return -EINVAL;
+
+       mutex_lock(&raw_mutex);
+       if (rawdev->inuse) {
+               mutex_unlock(&raw_mutex);
+               return -EBUSY;
+       }
+       if (rawdev->binding) {
+               bdput(rawdev->binding);
+               module_put(THIS_MODULE);
+       }
+       if (!dev) {
+               /* unbind */
+               rawdev->binding = NULL;
+               device_destroy(raw_class, MKDEV(RAW_MAJOR, number));
+       } else {
+               rawdev->binding = bdget(dev);
+               if (rawdev->binding == NULL) {
+                       err = -ENOMEM;
+               } else {
+                       dev_t raw = MKDEV(RAW_MAJOR, number);
+                       __module_get(THIS_MODULE);
+                       device_destroy(raw_class, raw);
+                       device_create(raw_class, NULL, raw, NULL,
+                                     "raw%d", number);
+               }
+       }
+       mutex_unlock(&raw_mutex);
+       return err;
+}
+
+static int bind_get(int number, dev_t *dev)
+{
+       struct raw_device_data *rawdev;
+       struct block_device *bdev;
+
+       if (number <= 0 || number >= max_raw_minors)
+               return -EINVAL;
+
+       rawdev = &raw_devices[number];
+
+       mutex_lock(&raw_mutex);
+       bdev = rawdev->binding;
+       *dev = bdev ? bdev->bd_dev : 0;
+       mutex_unlock(&raw_mutex);
+       return 0;
+}
+
+/*
+ * Deal with ioctls against the raw-device control interface, to bind
+ * and unbind other raw devices.
+ */
+static long raw_ctl_ioctl(struct file *filp, unsigned int command,
+                         unsigned long arg)
+{
+       struct raw_config_request rq;
+       dev_t dev;
+       int err;
+
+       switch (command) {
+       case RAW_SETBIND:
+               if (copy_from_user(&rq, (void __user *) arg, sizeof(rq)))
+                       return -EFAULT;
+
+               return bind_set(rq.raw_minor, rq.block_major, rq.block_minor);
+
+       case RAW_GETBIND:
+               if (copy_from_user(&rq, (void __user *) arg, sizeof(rq)))
+                       return -EFAULT;
+
+               err = bind_get(rq.raw_minor, &dev);
+               if (err)
+                       return err;
+
+               rq.block_major = MAJOR(dev);
+               rq.block_minor = MINOR(dev);
+
+               if (copy_to_user((void __user *)arg, &rq, sizeof(rq)))
+                       return -EFAULT;
+
+               return 0;
+       }
+
+       return -EINVAL;
+}
+
+#ifdef CONFIG_COMPAT
+struct raw32_config_request {
+       compat_int_t    raw_minor;
+       compat_u64      block_major;
+       compat_u64      block_minor;
+};
+
+static long raw_ctl_compat_ioctl(struct file *file, unsigned int cmd,
+                               unsigned long arg)
+{
+       struct raw32_config_request __user *user_req = compat_ptr(arg);
+       struct raw32_config_request rq;
+       dev_t dev;
+       int err = 0;
+
+       switch (cmd) {
+       case RAW_SETBIND:
+               if (copy_from_user(&rq, user_req, sizeof(rq)))
+                       return -EFAULT;
+
+               return bind_set(rq.raw_minor, rq.block_major, rq.block_minor);
+
+       case RAW_GETBIND:
+               if (copy_from_user(&rq, user_req, sizeof(rq)))
+                       return -EFAULT;
+
+               err = bind_get(rq.raw_minor, &dev);
+               if (err)
+                       return err;
+
+               rq.block_major = MAJOR(dev);
+               rq.block_minor = MINOR(dev);
+
+               if (copy_to_user(user_req, &rq, sizeof(rq)))
+                       return -EFAULT;
+
+               return 0;
+       }
+
+       return -EINVAL;
+}
+#endif
+
+static const struct file_operations raw_fops = {
+       .read_iter      = blkdev_read_iter,
+       .write_iter     = blkdev_write_iter,
+       .fsync          = blkdev_fsync,
+       .open           = raw_open,
+       .release        = raw_release,
+       .unlocked_ioctl = raw_ioctl,
+       .llseek         = default_llseek,
+       .owner          = THIS_MODULE,
+};
+
+static const struct file_operations raw_ctl_fops = {
+       .unlocked_ioctl = raw_ctl_ioctl,
+#ifdef CONFIG_COMPAT
+       .compat_ioctl   = raw_ctl_compat_ioctl,
+#endif
+       .open           = raw_open,
+       .owner          = THIS_MODULE,
+       .llseek         = noop_llseek,
+};
+
+static struct cdev raw_cdev;
+
+static char *raw_devnode(struct device *dev, umode_t *mode)
+{
+       return kasprintf(GFP_KERNEL, "raw/%s", dev_name(dev));
+}
+
+static int __init raw_init(void)
+{
+       dev_t dev = MKDEV(RAW_MAJOR, 0);
+       int ret;
+
+       if (max_raw_minors < 1 || max_raw_minors > 65536) {
+               printk(KERN_WARNING "raw: invalid max_raw_minors (must be"
+                       " between 1 and 65536), using %d\n", MAX_RAW_MINORS);
+               max_raw_minors = MAX_RAW_MINORS;
+       }
+
+       raw_devices = vzalloc(sizeof(struct raw_device_data) * max_raw_minors);
+       if (!raw_devices) {
+               printk(KERN_ERR "Not enough memory for raw device structures\n");
+               ret = -ENOMEM;
+               goto error;
+       }
+
+       ret = register_chrdev_region(dev, max_raw_minors, "raw");
+       if (ret)
+               goto error;
+
+       cdev_init(&raw_cdev, &raw_fops);
+       ret = cdev_add(&raw_cdev, dev, max_raw_minors);
+       if (ret) {
+               goto error_region;
+       }
+
+       raw_class = class_create(THIS_MODULE, "raw");
+       if (IS_ERR(raw_class)) {
+               printk(KERN_ERR "Error creating raw class.\n");
+               cdev_del(&raw_cdev);
+               ret = PTR_ERR(raw_class);
+               goto error_region;
+       }
+       raw_class->devnode = raw_devnode;
+       device_create(raw_class, NULL, MKDEV(RAW_MAJOR, 0), NULL, "rawctl");
+
+       return 0;
+
+error_region:
+       unregister_chrdev_region(dev, max_raw_minors);
+error:
+       vfree(raw_devices);
+       return ret;
+}
+
+static void __exit raw_exit(void)
+{
+       device_destroy(raw_class, MKDEV(RAW_MAJOR, 0));
+       class_destroy(raw_class);
+       cdev_del(&raw_cdev);
+       unregister_chrdev_region(MKDEV(RAW_MAJOR, 0), max_raw_minors);
+}
+
+module_init(raw_init);
+module_exit(raw_exit);
+MODULE_LICENSE("GPL");