ubd_kern.c

来自「Linux Kernel 2.6.9 for OMAP1710」· C语言 代码 · 共 1,258 行 · 第 1/2 页

	int n, err;

	str = uml_strdup(str);
	if(str == NULL){
		printk(KERN_ERR "ubd_config failed to strdup string\n");
		return(1);
	}
	err = ubd_setup_common(str, &n);
	if(err){
		kfree(str);
		return(-1);
	}
	if(n == -1) return(0);

 	spin_lock(&ubd_lock);
	err = ubd_add(n);
	if(err)
		ubd_dev[n].file = NULL;
 	spin_unlock(&ubd_lock);

	return(err);
}

static int ubd_get_config(char *name, char *str, int size, char **error_out)
{
	struct ubd *dev;
	char *end;
	int n, len = 0;

	n = simple_strtoul(name, &end, 0);
	if((*end != '\0') || (end == name)){
		*error_out = "ubd_get_config : didn't parse device number";
		return(-1);
	}

	if((n >= MAX_DEV) || (n < 0)){
		*error_out = "ubd_get_config : device number out of range";
		return(-1);
	}

	dev = &ubd_dev[n];
	spin_lock(&ubd_lock);

	if(dev->file == NULL){
		CONFIG_CHUNK(str, size, len, "", 1);
		goto out;
	}

	CONFIG_CHUNK(str, size, len, dev->file, 0);

	if(dev->cow.file != NULL){
		CONFIG_CHUNK(str, size, len, ",", 0);
		CONFIG_CHUNK(str, size, len, dev->cow.file, 1);
	}
	else CONFIG_CHUNK(str, size, len, "", 1);

 out:
	spin_unlock(&ubd_lock);
	return(len);
}

static int ubd_remove(char *str)
{
	struct ubd *dev;
	int n, err = -ENODEV;

	n = parse_unit(&str);

	if((n < 0) || (n >= MAX_DEV))
		return(err);

	dev = &ubd_dev[n];
	if(dev->count > 0)
		return(-EBUSY);	/* you cannot remove a open disk */

	err = 0;
 	spin_lock(&ubd_lock);

	if(ubd_gendisk[n] == NULL)
		goto out;

	del_gendisk(ubd_gendisk[n]);
	put_disk(ubd_gendisk[n]);
	ubd_gendisk[n] = NULL;

	if(fake_gendisk[n] != NULL){
		del_gendisk(fake_gendisk[n]);
		put_disk(fake_gendisk[n]);
		fake_gendisk[n] = NULL;
	}

	*dev = ((struct ubd) DEFAULT_UBD);
	err = 0;
 out:
 	spin_unlock(&ubd_lock);
	return(err);
}

static struct mc_device ubd_mc = {
	.name		= "ubd",
	.config		= ubd_config,
 	.get_config	= ubd_get_config,
	.remove		= ubd_remove,
};

static int ubd_mc_init(void)
{
	mconsole_register_dev(&ubd_mc);
	return 0;
}

__initcall(ubd_mc_init);

int ubd_init(void)
{
        int i;

	devfs_mk_dir("ubd");
	if (register_blkdev(MAJOR_NR, "ubd"))
		return -1;

	ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
	if (!ubd_queue) {
		unregister_blkdev(MAJOR_NR, "ubd");
		return -1;
	}
		
	if (fake_major != MAJOR_NR) {
		char name[sizeof("ubd_nnn\0")];

		snprintf(name, sizeof(name), "ubd_%d", fake_major);
		devfs_mk_dir(name);
		if (register_blkdev(fake_major, "ubd"))
			return -1;
	}
	for (i = 0; i < MAX_DEV; i++) 
		ubd_add(i);
	return 0;
}

late_initcall(ubd_init);

int ubd_driver_init(void){
	unsigned long stack;
	int err;

	/* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
	if(global_openflags.s){
		printk(KERN_INFO "ubd: Synchronous mode\n");
		/* Letting ubd=sync be like using ubd#s= instead of ubd#= is
		 * enough. So use anyway the io thread. */
	}
	stack = alloc_stack(0, 0);
	io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *), 
				 &thread_fd);
	if(io_pid < 0){
		printk(KERN_ERR 
		       "ubd : Failed to start I/O thread (errno = %d) - "
		       "falling back to synchronous I/O\n", -io_pid);
		io_pid = -1;
		return(0);
	}
	err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr, 
			     SA_INTERRUPT, "ubd", ubd_dev);
	if(err != 0)
		printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
	return(err);
}

device_initcall(ubd_driver_init);

static int ubd_open(struct inode *inode, struct file *filp)
{
	struct gendisk *disk = inode->i_bdev->bd_disk;
	struct ubd *dev = disk->private_data;
	int err = 0;

	if(dev->count == 0){
		err = ubd_open_dev(dev);
		if(err){
			printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
			       disk->disk_name, dev->file, -err);
			goto out;
		}
	}
	dev->count++;
	if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
	        if(--dev->count == 0) ubd_close(dev);
	        err = -EROFS;
	}
 out:
	return(err);
}

static int ubd_release(struct inode * inode, struct file * file)
{
	struct gendisk *disk = inode->i_bdev->bd_disk;
	struct ubd *dev = disk->private_data;

	if(--dev->count == 0)
		ubd_close(dev);
	return(0);
}

static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
			  __u64 *cow_offset, unsigned long *bitmap,
			  __u64 bitmap_offset, unsigned long *bitmap_words,
			  __u64 bitmap_len)
{
	__u64 sector = io_offset >> 9;
	int i, update_bitmap = 0;

	for(i = 0; i < length >> 9; i++){
		if(cow_mask != NULL)
			ubd_set_bit(i, (unsigned char *) cow_mask);
		if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
			continue;

		update_bitmap = 1;
		ubd_set_bit(sector + i, (unsigned char *) bitmap);
	}

	if(!update_bitmap)
		return;

	*cow_offset = sector / (sizeof(unsigned long) * 8);

	/* This takes care of the case where we're exactly at the end of the
	 * device, and *cow_offset + 1 is off the end.  So, just back it up
	 * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
	 * for the original diagnosis.
	 */
	if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
			   sizeof(unsigned long) - 1))
		(*cow_offset)--;

	bitmap_words[0] = bitmap[*cow_offset];
	bitmap_words[1] = bitmap[*cow_offset + 1];

	*cow_offset *= sizeof(unsigned long);
	*cow_offset += bitmap_offset;
}

static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
		       __u64 bitmap_offset, __u64 bitmap_len)
{
	__u64 sector = req->offset >> 9;
	int i;

	if(req->length > (sizeof(req->sector_mask) * 8) << 9)
		panic("Operation too long");

	if(req->op == UBD_READ) {
		for(i = 0; i < req->length >> 9; i++){
			if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
				ubd_set_bit(i, (unsigned char *) 
					    &req->sector_mask);
                }
	}
	else cowify_bitmap(req->offset, req->length, &req->sector_mask,
			   &req->cow_offset, bitmap, bitmap_offset,
			   req->bitmap_words, bitmap_len);
}

static int mmap_fd(struct request *req, struct ubd *dev, __u64 offset)
{
	__u64 sector;
	unsigned char *bitmap;
	int bit, i;

	/* mmap must have been requested on the command line */
	if(!ubd_do_mmap)
		return(-1);

	/* The buffer must be page aligned */
	if(((unsigned long) req->buffer % UBD_MMAP_BLOCK_SIZE) != 0)
		return(-1);

	/* The request must be a page long */
	if((req->current_nr_sectors << 9) != PAGE_SIZE)
		return(-1);

	if(dev->cow.file == NULL)
		return(dev->fd);

	sector = offset >> 9;
	bitmap = (unsigned char *) dev->cow.bitmap;
	bit = ubd_test_bit(sector, bitmap);

	for(i = 1; i < req->current_nr_sectors; i++){
		if(ubd_test_bit(sector + i, bitmap) != bit)
			return(-1);
	}

	if(bit || (rq_data_dir(req) == WRITE))
		offset += dev->cow.data_offset;

	/* The data on disk must be page aligned */
	if((offset % UBD_MMAP_BLOCK_SIZE) != 0)
		return(-1);

	return(bit ? dev->fd : dev->cow.fd);
}

static int prepare_mmap_request(struct ubd *dev, int fd, __u64 offset,
				struct request *req,
				struct io_thread_req *io_req)
{
	int err;

	if(rq_data_dir(req) == WRITE){
		/* Writes are almost no-ops since the new data is already in the
		 * host page cache
		 */
		dev->map_writes++;
		if(dev->cow.file != NULL)
			cowify_bitmap(io_req->offset, io_req->length,
				      &io_req->sector_mask, &io_req->cow_offset,
				      dev->cow.bitmap, dev->cow.bitmap_offset,
				      io_req->bitmap_words,
				      dev->cow.bitmap_len);
	}
	else {
		int w;

		if((dev->cow.file != NULL) && (fd == dev->cow.fd))
			w = 0;
		else w = dev->openflags.w;

		if((dev->cow.file != NULL) && (fd == dev->fd))
			offset += dev->cow.data_offset;

		err = physmem_subst_mapping(req->buffer, fd, offset, w);
		if(err){
			printk("physmem_subst_mapping failed, err = %d\n",
			       -err);
			return(1);
		}
		dev->map_reads++;
	}
	io_req->op = UBD_MMAP;
	io_req->buffer = req->buffer;
	return(0);
}

/* Called with ubd_io_lock held */
static int prepare_request(struct request *req, struct io_thread_req *io_req)
{
	struct gendisk *disk = req->rq_disk;
	struct ubd *dev = disk->private_data;
	__u64 offset;
	int len, fd;

	if(req->rq_status == RQ_INACTIVE) return(1);

	if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
		printk("Write attempted on readonly ubd device %s\n", 
		       disk->disk_name);
		end_request(req, 0);
		return(1);
	}

	offset = ((__u64) req->sector) << 9;
	len = req->current_nr_sectors << 9;

	io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
	io_req->fds[1] = dev->fd;
	io_req->map_fd = -1;
	io_req->cow_offset = -1;
	io_req->offset = offset;
	io_req->length = len;
	io_req->error = 0;
	io_req->sector_mask = 0;

	fd = mmap_fd(req, dev, io_req->offset);
	if(fd > 0){
		/* If mmapping is otherwise OK, but the first access to the
		 * page is a write, then it's not mapped in yet.  So we have
		 * to write the data to disk first, then we can map the disk
		 * page in and continue normally from there.
		 */
		if((rq_data_dir(req) == WRITE) && !is_remapped(req->buffer)){
			io_req->map_fd = dev->fd;
			io_req->map_offset = io_req->offset +
				dev->cow.data_offset;
			dev->write_maps++;
		}
		else return(prepare_mmap_request(dev, fd, io_req->offset, req,
						 io_req));
	}

	if(rq_data_dir(req) == READ)
		dev->nomap_reads++;
	else dev->nomap_writes++;

	io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
	io_req->offsets[0] = 0;
	io_req->offsets[1] = dev->cow.data_offset;
	io_req->buffer = req->buffer;
	io_req->sectorsize = 1 << 9;

	if(dev->cow.file != NULL)
		cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset,
			   dev->cow.bitmap_len);

	return(0);
}

/* Called with ubd_io_lock held */
static void do_ubd_request(request_queue_t *q)
{
	struct io_thread_req io_req;
	struct request *req;
	int err, n;

	if(thread_fd == -1){
		while((req = elv_next_request(q)) != NULL){
			err = prepare_request(req, &io_req);
			if(!err){
				do_io(&io_req);
				__ubd_finish(req, io_req.error);
			}
		}
	}
	else {
		if(do_ubd || (req = elv_next_request(q)) == NULL)
			return;
		err = prepare_request(req, &io_req);
		if(!err){
			do_ubd = ubd_handler;
			n = write_ubd_fs(thread_fd, (char *) &io_req, 
					 sizeof(io_req));
			if(n != sizeof(io_req))
				printk("write to io thread failed, "
				       "errno = %d\n", -n);
		}
	}
}

static int ubd_ioctl(struct inode * inode, struct file * file,
		     unsigned int cmd, unsigned long arg)
{
	struct hd_geometry *loc = (struct hd_geometry *) arg;
	struct ubd *dev = inode->i_bdev->bd_disk->private_data;
	struct hd_driveid ubd_id = {
		.cyls		= 0,
		.heads		= 128,
		.sectors	= 32,
	};

	switch (cmd) {
	        struct hd_geometry g;
		struct cdrom_volctrl volume;
	case HDIO_GETGEO:
		if(!loc) return(-EINVAL);
		g.heads = 128;
		g.sectors = 32;
		g.cylinders = dev->size / (128 * 32 * 512);
		g.start = get_start_sect(inode->i_bdev);
		return(copy_to_user(loc, &g, sizeof(g)) ? -EFAULT : 0);

	case HDIO_GET_IDENTITY:
		ubd_id.cyls = dev->size / (128 * 32 * 512);
		if(copy_to_user((char *) arg, (char *) &ubd_id, 
				 sizeof(ubd_id)))
			return(-EFAULT);
		return(0);
		
	case CDROMVOLREAD:
		if(copy_from_user(&volume, (char *) arg, sizeof(volume)))
			return(-EFAULT);
		volume.channel0 = 255;
		volume.channel1 = 255;
		volume.channel2 = 255;
		volume.channel3 = 255;
		if(copy_to_user((char *) arg, &volume, sizeof(volume)))
			return(-EFAULT);
		return(0);
	}
	return(-EINVAL);
}

static int ubd_check_remapped(int fd, unsigned long address, int is_write,
			      __u64 offset)
{
	__u64 bitmap_offset;
	unsigned long new_bitmap[2];
	int i, err, n;

	/* If it's not a write access, we can't do anything about it */
	if(!is_write)
		return(0);

	/* We have a write */
	for(i = 0; i < sizeof(ubd_dev) / sizeof(ubd_dev[0]); i++){
		struct ubd *dev = &ubd_dev[i];

		if((dev->fd != fd) && (dev->cow.fd != fd))
			continue;

		/* It's a write to a ubd device */

		if(!dev->openflags.w){
			/* It's a write access on a read-only device - probably
			 * shouldn't happen.  If the kernel is trying to change
			 * something with no intention of writing it back out,
			 * then this message will clue us in that this needs
			 * fixing
			 */
			printk("Write access to mapped page from readonly ubd "
			       "device %d\n", i);
			return(0);
		}

		/* It's a write to a writeable ubd device - it must be COWed
		 * because, otherwise, the page would have been mapped in
		 * writeable
		 */

		if(!dev->cow.file)
			panic("Write fault on writeable non-COW ubd device %d",
			      i);

		/* It should also be an access to the backing file since the
		 * COW pages should be mapped in read-write
		 */

		if(fd == dev->fd)
			panic("Write fault on a backing page of ubd "
			      "device %d\n", i);

		/* So, we do the write, copying the backing data to the COW
		 * file...
		 */

		err = os_seek_file(dev->fd, offset + dev->cow.data_offset);
		if(err < 0)
			panic("Couldn't seek to %lld in COW file of ubd "
			      "device %d, err = %d",
			      offset + dev->cow.data_offset, i, -err);

		n = os_write_file(dev->fd, (void *) address, PAGE_SIZE);
		if(n != PAGE_SIZE)
			panic("Couldn't copy data to COW file of ubd "
			      "device %d, err = %d", i, -n);

		/* ... updating the COW bitmap... */

		cowify_bitmap(offset, PAGE_SIZE, NULL, &bitmap_offset,
			      dev->cow.bitmap, dev->cow.bitmap_offset,
			      new_bitmap, dev->cow.bitmap_len);

		err = os_seek_file(dev->fd, bitmap_offset);
		if(err < 0)
			panic("Couldn't seek to %lld in COW file of ubd "
			      "device %d, err = %d", bitmap_offset, i, -err);

		n = os_write_file(dev->fd, new_bitmap, sizeof(new_bitmap));
		if(n != sizeof(new_bitmap))
			panic("Couldn't update bitmap  of ubd device %d, "
			      "err = %d", i, -n);

		/* Maybe we can map the COW page in, and maybe we can't.  If
		 * it is a pre-V3 COW file, we can't, since the alignment will
		 * be wrong.  If it is a V3 or later COW file which has been
		 * moved to a system with a larger page size, then maybe we
		 * can't, depending on the exact location of the page.
		 */

		offset += dev->cow.data_offset;

		/* Remove the remapping, putting the original anonymous page
		 * back.  If the COW file can be mapped in, that is done.
		 * Otherwise, the COW page is read in.
		 */

		if(!physmem_remove_mapping((void *) address))
			panic("Address 0x%lx not remapped by ubd device %d",
			      address, i);
		if((offset % UBD_MMAP_BLOCK_SIZE) == 0)
			physmem_subst_mapping((void *) address, dev->fd,
					      offset, 1);
		else {
			err = os_seek_file(dev->fd, offset);
			if(err < 0)
				panic("Couldn't seek to %lld in COW file of "
				      "ubd device %d, err = %d", offset, i,
				      -err);

			n = os_read_file(dev->fd, (void *) address, PAGE_SIZE);
			if(n != PAGE_SIZE)
				panic("Failed to read page from offset %llx of "
				      "COW file of ubd device %d, err = %d",
				      offset, i, -n);
		}

		return(1);
	}

	/* It's not a write on a ubd device */
	return(0);
}

static struct remapper ubd_remapper = {
	.list	= LIST_HEAD_INIT(ubd_remapper.list),
	.proc	= ubd_check_remapped,
};

static int ubd_remapper_setup(void)
{
	if(ubd_do_mmap)
		register_remapper(&ubd_remapper);

	return(0);
}

__initcall(ubd_remapper_setup);

/*
 * Overrides for Emacs so that we follow Linus's tabbing style.
 * Emacs will notice this stuff at the end of the file and automatically
 * adjust the settings for this buffer only.  This must remain at the end
 * of the file.
 * ---------------------------------------------------------------------------
 * Local variables:
 * c-file-style: "linux"
 * End:
 */