Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / block / pktcdvd.c
diff --git a/kernel/drivers/block/pktcdvd.c b/kernel/drivers/block/pktcdvd.c
new file mode 100644 (file)
index 0000000..09e628d
--- /dev/null
@@ -0,0 +1,3029 @@
+/*
+ * Copyright (C) 2000 Jens Axboe <axboe@suse.de>
+ * Copyright (C) 2001-2004 Peter Osterlund <petero2@telia.com>
+ * Copyright (C) 2006 Thomas Maier <balagi@justmail.de>
+ *
+ * May be copied or modified under the terms of the GNU General Public
+ * License.  See linux/COPYING for more information.
+ *
+ * Packet writing layer for ATAPI and SCSI CD-RW, DVD+RW, DVD-RW and
+ * DVD-RAM devices.
+ *
+ * Theory of operation:
+ *
+ * At the lowest level, there is the standard driver for the CD/DVD device,
+ * typically ide-cd.c or sr.c. This driver can handle read and write requests,
+ * but it doesn't know anything about the special restrictions that apply to
+ * packet writing. One restriction is that write requests must be aligned to
+ * packet boundaries on the physical media, and the size of a write request
+ * must be equal to the packet size. Another restriction is that a
+ * GPCMD_FLUSH_CACHE command has to be issued to the drive before a read
+ * command, if the previous command was a write.
+ *
+ * The purpose of the packet writing driver is to hide these restrictions from
+ * higher layers, such as file systems, and present a block device that can be
+ * randomly read and written using 2kB-sized blocks.
+ *
+ * The lowest layer in the packet writing driver is the packet I/O scheduler.
+ * Its data is defined by the struct packet_iosched and includes two bio
+ * queues with pending read and write requests. These queues are processed
+ * by the pkt_iosched_process_queue() function. The write requests in this
+ * queue are already properly aligned and sized. This layer is responsible for
+ * issuing the flush cache commands and scheduling the I/O in a good order.
+ *
+ * The next layer transforms unaligned write requests to aligned writes. This
+ * transformation requires reading missing pieces of data from the underlying
+ * block device, assembling the pieces to full packets and queuing them to the
+ * packet I/O scheduler.
+ *
+ * At the top layer there is a custom make_request_fn function that forwards
+ * read requests directly to the iosched queue and puts write requests in the
+ * unaligned write queue. A kernel thread performs the necessary read
+ * gathering to convert the unaligned writes to aligned writes and then feeds
+ * them to the packet I/O scheduler.
+ *
+ *************************************************************************/
+
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
+
+#include <linux/pktcdvd.h>
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/compat.h>
+#include <linux/kthread.h>
+#include <linux/errno.h>
+#include <linux/spinlock.h>
+#include <linux/file.h>
+#include <linux/proc_fs.h>
+#include <linux/seq_file.h>
+#include <linux/miscdevice.h>
+#include <linux/freezer.h>
+#include <linux/mutex.h>
+#include <linux/slab.h>
+#include <scsi/scsi_cmnd.h>
+#include <scsi/scsi_ioctl.h>
+#include <scsi/scsi.h>
+#include <linux/debugfs.h>
+#include <linux/device.h>
+
+#include <asm/uaccess.h>
+
+#define DRIVER_NAME    "pktcdvd"
+
+#define pkt_err(pd, fmt, ...)                                          \
+       pr_err("%s: " fmt, pd->name, ##__VA_ARGS__)
+#define pkt_notice(pd, fmt, ...)                                       \
+       pr_notice("%s: " fmt, pd->name, ##__VA_ARGS__)
+#define pkt_info(pd, fmt, ...)                                         \
+       pr_info("%s: " fmt, pd->name, ##__VA_ARGS__)
+
+#define pkt_dbg(level, pd, fmt, ...)                                   \
+do {                                                                   \
+       if (level == 2 && PACKET_DEBUG >= 2)                            \
+               pr_notice("%s: %s():" fmt,                              \
+                         pd->name, __func__, ##__VA_ARGS__);           \
+       else if (level == 1 && PACKET_DEBUG >= 1)                       \
+               pr_notice("%s: " fmt, pd->name, ##__VA_ARGS__);         \
+} while (0)
+
+#define MAX_SPEED 0xffff
+
+static DEFINE_MUTEX(pktcdvd_mutex);
+static struct pktcdvd_device *pkt_devs[MAX_WRITERS];
+static struct proc_dir_entry *pkt_proc;
+static int pktdev_major;
+static int write_congestion_on  = PKT_WRITE_CONGESTION_ON;
+static int write_congestion_off = PKT_WRITE_CONGESTION_OFF;
+static struct mutex ctl_mutex; /* Serialize open/close/setup/teardown */
+static mempool_t *psd_pool;
+
+static struct class    *class_pktcdvd = NULL;    /* /sys/class/pktcdvd */
+static struct dentry   *pkt_debugfs_root = NULL; /* /sys/kernel/debug/pktcdvd */
+
+/* forward declaration */
+static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev);
+static int pkt_remove_dev(dev_t pkt_dev);
+static int pkt_seq_show(struct seq_file *m, void *p);
+
+static sector_t get_zone(sector_t sector, struct pktcdvd_device *pd)
+{
+       return (sector + pd->offset) & ~(sector_t)(pd->settings.size - 1);
+}
+
+/*
+ * create and register a pktcdvd kernel object.
+ */
+static struct pktcdvd_kobj* pkt_kobj_create(struct pktcdvd_device *pd,
+                                       const char* name,
+                                       struct kobject* parent,
+                                       struct kobj_type* ktype)
+{
+       struct pktcdvd_kobj *p;
+       int error;
+
+       p = kzalloc(sizeof(*p), GFP_KERNEL);
+       if (!p)
+               return NULL;
+       p->pd = pd;
+       error = kobject_init_and_add(&p->kobj, ktype, parent, "%s", name);
+       if (error) {
+               kobject_put(&p->kobj);
+               return NULL;
+       }
+       kobject_uevent(&p->kobj, KOBJ_ADD);
+       return p;
+}
+/*
+ * remove a pktcdvd kernel object.
+ */
+static void pkt_kobj_remove(struct pktcdvd_kobj *p)
+{
+       if (p)
+               kobject_put(&p->kobj);
+}
+/*
+ * default release function for pktcdvd kernel objects.
+ */
+static void pkt_kobj_release(struct kobject *kobj)
+{
+       kfree(to_pktcdvdkobj(kobj));
+}
+
+
+/**********************************************************
+ *
+ * sysfs interface for pktcdvd
+ * by (C) 2006  Thomas Maier <balagi@justmail.de>
+ *
+ **********************************************************/
+
+#define DEF_ATTR(_obj,_name,_mode) \
+       static struct attribute _obj = { .name = _name, .mode = _mode }
+
+/**********************************************************
+  /sys/class/pktcdvd/pktcdvd[0-7]/
+                     stat/reset
+                     stat/packets_started
+                     stat/packets_finished
+                     stat/kb_written
+                     stat/kb_read
+                     stat/kb_read_gather
+                     write_queue/size
+                     write_queue/congestion_off
+                     write_queue/congestion_on
+ **********************************************************/
+
+DEF_ATTR(kobj_pkt_attr_st1, "reset", 0200);
+DEF_ATTR(kobj_pkt_attr_st2, "packets_started", 0444);
+DEF_ATTR(kobj_pkt_attr_st3, "packets_finished", 0444);
+DEF_ATTR(kobj_pkt_attr_st4, "kb_written", 0444);
+DEF_ATTR(kobj_pkt_attr_st5, "kb_read", 0444);
+DEF_ATTR(kobj_pkt_attr_st6, "kb_read_gather", 0444);
+
+static struct attribute *kobj_pkt_attrs_stat[] = {
+       &kobj_pkt_attr_st1,
+       &kobj_pkt_attr_st2,
+       &kobj_pkt_attr_st3,
+       &kobj_pkt_attr_st4,
+       &kobj_pkt_attr_st5,
+       &kobj_pkt_attr_st6,
+       NULL
+};
+
+DEF_ATTR(kobj_pkt_attr_wq1, "size", 0444);
+DEF_ATTR(kobj_pkt_attr_wq2, "congestion_off", 0644);
+DEF_ATTR(kobj_pkt_attr_wq3, "congestion_on",  0644);
+
+static struct attribute *kobj_pkt_attrs_wqueue[] = {
+       &kobj_pkt_attr_wq1,
+       &kobj_pkt_attr_wq2,
+       &kobj_pkt_attr_wq3,
+       NULL
+};
+
+static ssize_t kobj_pkt_show(struct kobject *kobj,
+                       struct attribute *attr, char *data)
+{
+       struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
+       int n = 0;
+       int v;
+       if (strcmp(attr->name, "packets_started") == 0) {
+               n = sprintf(data, "%lu\n", pd->stats.pkt_started);
+
+       } else if (strcmp(attr->name, "packets_finished") == 0) {
+               n = sprintf(data, "%lu\n", pd->stats.pkt_ended);
+
+       } else if (strcmp(attr->name, "kb_written") == 0) {
+               n = sprintf(data, "%lu\n", pd->stats.secs_w >> 1);
+
+       } else if (strcmp(attr->name, "kb_read") == 0) {
+               n = sprintf(data, "%lu\n", pd->stats.secs_r >> 1);
+
+       } else if (strcmp(attr->name, "kb_read_gather") == 0) {
+               n = sprintf(data, "%lu\n", pd->stats.secs_rg >> 1);
+
+       } else if (strcmp(attr->name, "size") == 0) {
+               spin_lock(&pd->lock);
+               v = pd->bio_queue_size;
+               spin_unlock(&pd->lock);
+               n = sprintf(data, "%d\n", v);
+
+       } else if (strcmp(attr->name, "congestion_off") == 0) {
+               spin_lock(&pd->lock);
+               v = pd->write_congestion_off;
+               spin_unlock(&pd->lock);
+               n = sprintf(data, "%d\n", v);
+
+       } else if (strcmp(attr->name, "congestion_on") == 0) {
+               spin_lock(&pd->lock);
+               v = pd->write_congestion_on;
+               spin_unlock(&pd->lock);
+               n = sprintf(data, "%d\n", v);
+       }
+       return n;
+}
+
+static void init_write_congestion_marks(int* lo, int* hi)
+{
+       if (*hi > 0) {
+               *hi = max(*hi, 500);
+               *hi = min(*hi, 1000000);
+               if (*lo <= 0)
+                       *lo = *hi - 100;
+               else {
+                       *lo = min(*lo, *hi - 100);
+                       *lo = max(*lo, 100);
+               }
+       } else {
+               *hi = -1;
+               *lo = -1;
+       }
+}
+
+static ssize_t kobj_pkt_store(struct kobject *kobj,
+                       struct attribute *attr,
+                       const char *data, size_t len)
+{
+       struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
+       int val;
+
+       if (strcmp(attr->name, "reset") == 0 && len > 0) {
+               pd->stats.pkt_started = 0;
+               pd->stats.pkt_ended = 0;
+               pd->stats.secs_w = 0;
+               pd->stats.secs_rg = 0;
+               pd->stats.secs_r = 0;
+
+       } else if (strcmp(attr->name, "congestion_off") == 0
+                  && sscanf(data, "%d", &val) == 1) {
+               spin_lock(&pd->lock);
+               pd->write_congestion_off = val;
+               init_write_congestion_marks(&pd->write_congestion_off,
+                                       &pd->write_congestion_on);
+               spin_unlock(&pd->lock);
+
+       } else if (strcmp(attr->name, "congestion_on") == 0
+                  && sscanf(data, "%d", &val) == 1) {
+               spin_lock(&pd->lock);
+               pd->write_congestion_on = val;
+               init_write_congestion_marks(&pd->write_congestion_off,
+                                       &pd->write_congestion_on);
+               spin_unlock(&pd->lock);
+       }
+       return len;
+}
+
+static const struct sysfs_ops kobj_pkt_ops = {
+       .show = kobj_pkt_show,
+       .store = kobj_pkt_store
+};
+static struct kobj_type kobj_pkt_type_stat = {
+       .release = pkt_kobj_release,
+       .sysfs_ops = &kobj_pkt_ops,
+       .default_attrs = kobj_pkt_attrs_stat
+};
+static struct kobj_type kobj_pkt_type_wqueue = {
+       .release = pkt_kobj_release,
+       .sysfs_ops = &kobj_pkt_ops,
+       .default_attrs = kobj_pkt_attrs_wqueue
+};
+
+static void pkt_sysfs_dev_new(struct pktcdvd_device *pd)
+{
+       if (class_pktcdvd) {
+               pd->dev = device_create(class_pktcdvd, NULL, MKDEV(0, 0), NULL,
+                                       "%s", pd->name);
+               if (IS_ERR(pd->dev))
+                       pd->dev = NULL;
+       }
+       if (pd->dev) {
+               pd->kobj_stat = pkt_kobj_create(pd, "stat",
+                                       &pd->dev->kobj,
+                                       &kobj_pkt_type_stat);
+               pd->kobj_wqueue = pkt_kobj_create(pd, "write_queue",
+                                       &pd->dev->kobj,
+                                       &kobj_pkt_type_wqueue);
+       }
+}
+
+static void pkt_sysfs_dev_remove(struct pktcdvd_device *pd)
+{
+       pkt_kobj_remove(pd->kobj_stat);
+       pkt_kobj_remove(pd->kobj_wqueue);
+       if (class_pktcdvd)
+               device_unregister(pd->dev);
+}
+
+
+/********************************************************************
+  /sys/class/pktcdvd/
+                     add            map block device
+                     remove         unmap packet dev
+                     device_map     show mappings
+ *******************************************************************/
+
+static void class_pktcdvd_release(struct class *cls)
+{
+       kfree(cls);
+}
+static ssize_t class_pktcdvd_show_map(struct class *c,
+                                       struct class_attribute *attr,
+                                       char *data)
+{
+       int n = 0;
+       int idx;
+       mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
+       for (idx = 0; idx < MAX_WRITERS; idx++) {
+               struct pktcdvd_device *pd = pkt_devs[idx];
+               if (!pd)
+                       continue;
+               n += sprintf(data+n, "%s %u:%u %u:%u\n",
+                       pd->name,
+                       MAJOR(pd->pkt_dev), MINOR(pd->pkt_dev),
+                       MAJOR(pd->bdev->bd_dev),
+                       MINOR(pd->bdev->bd_dev));
+       }
+       mutex_unlock(&ctl_mutex);
+       return n;
+}
+
+static ssize_t class_pktcdvd_store_add(struct class *c,
+                                       struct class_attribute *attr,
+                                       const char *buf,
+                                       size_t count)
+{
+       unsigned int major, minor;
+
+       if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
+               /* pkt_setup_dev() expects caller to hold reference to self */
+               if (!try_module_get(THIS_MODULE))
+                       return -ENODEV;
+
+               pkt_setup_dev(MKDEV(major, minor), NULL);
+
+               module_put(THIS_MODULE);
+
+               return count;
+       }
+
+       return -EINVAL;
+}
+
+static ssize_t class_pktcdvd_store_remove(struct class *c,
+                                         struct class_attribute *attr,
+                                         const char *buf,
+                                       size_t count)
+{
+       unsigned int major, minor;
+       if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
+               pkt_remove_dev(MKDEV(major, minor));
+               return count;
+       }
+       return -EINVAL;
+}
+
+static struct class_attribute class_pktcdvd_attrs[] = {
+ __ATTR(add,            0200, NULL, class_pktcdvd_store_add),
+ __ATTR(remove,         0200, NULL, class_pktcdvd_store_remove),
+ __ATTR(device_map,     0444, class_pktcdvd_show_map, NULL),
+ __ATTR_NULL
+};
+
+
+static int pkt_sysfs_init(void)
+{
+       int ret = 0;
+
+       /*
+        * create control files in sysfs
+        * /sys/class/pktcdvd/...
+        */
+       class_pktcdvd = kzalloc(sizeof(*class_pktcdvd), GFP_KERNEL);
+       if (!class_pktcdvd)
+               return -ENOMEM;
+       class_pktcdvd->name = DRIVER_NAME;
+       class_pktcdvd->owner = THIS_MODULE;
+       class_pktcdvd->class_release = class_pktcdvd_release;
+       class_pktcdvd->class_attrs = class_pktcdvd_attrs;
+       ret = class_register(class_pktcdvd);
+       if (ret) {
+               kfree(class_pktcdvd);
+               class_pktcdvd = NULL;
+               pr_err("failed to create class pktcdvd\n");
+               return ret;
+       }
+       return 0;
+}
+
+static void pkt_sysfs_cleanup(void)
+{
+       if (class_pktcdvd)
+               class_destroy(class_pktcdvd);
+       class_pktcdvd = NULL;
+}
+
+/********************************************************************
+  entries in debugfs
+
+  /sys/kernel/debug/pktcdvd[0-7]/
+                       info
+
+ *******************************************************************/
+
+static int pkt_debugfs_seq_show(struct seq_file *m, void *p)
+{
+       return pkt_seq_show(m, p);
+}
+
+static int pkt_debugfs_fops_open(struct inode *inode, struct file *file)
+{
+       return single_open(file, pkt_debugfs_seq_show, inode->i_private);
+}
+
+static const struct file_operations debug_fops = {
+       .open           = pkt_debugfs_fops_open,
+       .read           = seq_read,
+       .llseek         = seq_lseek,
+       .release        = single_release,
+       .owner          = THIS_MODULE,
+};
+
+static void pkt_debugfs_dev_new(struct pktcdvd_device *pd)
+{
+       if (!pkt_debugfs_root)
+               return;
+       pd->dfs_d_root = debugfs_create_dir(pd->name, pkt_debugfs_root);
+       if (!pd->dfs_d_root)
+               return;
+
+       pd->dfs_f_info = debugfs_create_file("info", S_IRUGO,
+                               pd->dfs_d_root, pd, &debug_fops);
+}
+
+static void pkt_debugfs_dev_remove(struct pktcdvd_device *pd)
+{
+       if (!pkt_debugfs_root)
+               return;
+       debugfs_remove(pd->dfs_f_info);
+       debugfs_remove(pd->dfs_d_root);
+       pd->dfs_f_info = NULL;
+       pd->dfs_d_root = NULL;
+}
+
+static void pkt_debugfs_init(void)
+{
+       pkt_debugfs_root = debugfs_create_dir(DRIVER_NAME, NULL);
+}
+
+static void pkt_debugfs_cleanup(void)
+{
+       debugfs_remove(pkt_debugfs_root);
+       pkt_debugfs_root = NULL;
+}
+
+/* ----------------------------------------------------------*/
+
+
+static void pkt_bio_finished(struct pktcdvd_device *pd)
+{
+       BUG_ON(atomic_read(&pd->cdrw.pending_bios) <= 0);
+       if (atomic_dec_and_test(&pd->cdrw.pending_bios)) {
+               pkt_dbg(2, pd, "queue empty\n");
+               atomic_set(&pd->iosched.attention, 1);
+               wake_up(&pd->wqueue);
+       }
+}
+
+/*
+ * Allocate a packet_data struct
+ */
+static struct packet_data *pkt_alloc_packet_data(int frames)
+{
+       int i;
+       struct packet_data *pkt;
+
+       pkt = kzalloc(sizeof(struct packet_data), GFP_KERNEL);
+       if (!pkt)
+               goto no_pkt;
+
+       pkt->frames = frames;
+       pkt->w_bio = bio_kmalloc(GFP_KERNEL, frames);
+       if (!pkt->w_bio)
+               goto no_bio;
+
+       for (i = 0; i < frames / FRAMES_PER_PAGE; i++) {
+               pkt->pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
+               if (!pkt->pages[i])
+                       goto no_page;
+       }
+
+       spin_lock_init(&pkt->lock);
+       bio_list_init(&pkt->orig_bios);
+
+       for (i = 0; i < frames; i++) {
+               struct bio *bio = bio_kmalloc(GFP_KERNEL, 1);
+               if (!bio)
+                       goto no_rd_bio;
+
+               pkt->r_bios[i] = bio;
+       }
+
+       return pkt;
+
+no_rd_bio:
+       for (i = 0; i < frames; i++) {
+               struct bio *bio = pkt->r_bios[i];
+               if (bio)
+                       bio_put(bio);
+       }
+
+no_page:
+       for (i = 0; i < frames / FRAMES_PER_PAGE; i++)
+               if (pkt->pages[i])
+                       __free_page(pkt->pages[i]);
+       bio_put(pkt->w_bio);
+no_bio:
+       kfree(pkt);
+no_pkt:
+       return NULL;
+}
+
+/*
+ * Free a packet_data struct
+ */
+static void pkt_free_packet_data(struct packet_data *pkt)
+{
+       int i;
+
+       for (i = 0; i < pkt->frames; i++) {
+               struct bio *bio = pkt->r_bios[i];
+               if (bio)
+                       bio_put(bio);
+       }
+       for (i = 0; i < pkt->frames / FRAMES_PER_PAGE; i++)
+               __free_page(pkt->pages[i]);
+       bio_put(pkt->w_bio);
+       kfree(pkt);
+}
+
+static void pkt_shrink_pktlist(struct pktcdvd_device *pd)
+{
+       struct packet_data *pkt, *next;
+
+       BUG_ON(!list_empty(&pd->cdrw.pkt_active_list));
+
+       list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_free_list, list) {
+               pkt_free_packet_data(pkt);
+       }
+       INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
+}
+
+static int pkt_grow_pktlist(struct pktcdvd_device *pd, int nr_packets)
+{
+       struct packet_data *pkt;
+
+       BUG_ON(!list_empty(&pd->cdrw.pkt_free_list));
+
+       while (nr_packets > 0) {
+               pkt = pkt_alloc_packet_data(pd->settings.size >> 2);
+               if (!pkt) {
+                       pkt_shrink_pktlist(pd);
+                       return 0;
+               }
+               pkt->id = nr_packets;
+               pkt->pd = pd;
+               list_add(&pkt->list, &pd->cdrw.pkt_free_list);
+               nr_packets--;
+       }
+       return 1;
+}
+
+static inline struct pkt_rb_node *pkt_rbtree_next(struct pkt_rb_node *node)
+{
+       struct rb_node *n = rb_next(&node->rb_node);
+       if (!n)
+               return NULL;
+       return rb_entry(n, struct pkt_rb_node, rb_node);
+}
+
+static void pkt_rbtree_erase(struct pktcdvd_device *pd, struct pkt_rb_node *node)
+{
+       rb_erase(&node->rb_node, &pd->bio_queue);
+       mempool_free(node, pd->rb_pool);
+       pd->bio_queue_size--;
+       BUG_ON(pd->bio_queue_size < 0);
+}
+
+/*
+ * Find the first node in the pd->bio_queue rb tree with a starting sector >= s.
+ */
+static struct pkt_rb_node *pkt_rbtree_find(struct pktcdvd_device *pd, sector_t s)
+{
+       struct rb_node *n = pd->bio_queue.rb_node;
+       struct rb_node *next;
+       struct pkt_rb_node *tmp;
+
+       if (!n) {
+               BUG_ON(pd->bio_queue_size > 0);
+               return NULL;
+       }
+
+       for (;;) {
+               tmp = rb_entry(n, struct pkt_rb_node, rb_node);
+               if (s <= tmp->bio->bi_iter.bi_sector)
+                       next = n->rb_left;
+               else
+                       next = n->rb_right;
+               if (!next)
+                       break;
+               n = next;
+       }
+
+       if (s > tmp->bio->bi_iter.bi_sector) {
+               tmp = pkt_rbtree_next(tmp);
+               if (!tmp)
+                       return NULL;
+       }
+       BUG_ON(s > tmp->bio->bi_iter.bi_sector);
+       return tmp;
+}
+
+/*
+ * Insert a node into the pd->bio_queue rb tree.
+ */
+static void pkt_rbtree_insert(struct pktcdvd_device *pd, struct pkt_rb_node *node)
+{
+       struct rb_node **p = &pd->bio_queue.rb_node;
+       struct rb_node *parent = NULL;
+       sector_t s = node->bio->bi_iter.bi_sector;
+       struct pkt_rb_node *tmp;
+
+       while (*p) {
+               parent = *p;
+               tmp = rb_entry(parent, struct pkt_rb_node, rb_node);
+               if (s < tmp->bio->bi_iter.bi_sector)
+                       p = &(*p)->rb_left;
+               else
+                       p = &(*p)->rb_right;
+       }
+       rb_link_node(&node->rb_node, parent, p);
+       rb_insert_color(&node->rb_node, &pd->bio_queue);
+       pd->bio_queue_size++;
+}
+
+/*
+ * Send a packet_command to the underlying block device and
+ * wait for completion.
+ */
+static int pkt_generic_packet(struct pktcdvd_device *pd, struct packet_command *cgc)
+{
+       struct request_queue *q = bdev_get_queue(pd->bdev);
+       struct request *rq;
+       int ret = 0;
+
+       rq = blk_get_request(q, (cgc->data_direction == CGC_DATA_WRITE) ?
+                            WRITE : READ, __GFP_WAIT);
+       if (IS_ERR(rq))
+               return PTR_ERR(rq);
+       blk_rq_set_block_pc(rq);
+
+       if (cgc->buflen) {
+               ret = blk_rq_map_kern(q, rq, cgc->buffer, cgc->buflen,
+                                     __GFP_WAIT);
+               if (ret)
+                       goto out;
+       }
+
+       rq->cmd_len = COMMAND_SIZE(cgc->cmd[0]);
+       memcpy(rq->cmd, cgc->cmd, CDROM_PACKET_SIZE);
+
+       rq->timeout = 60*HZ;
+       if (cgc->quiet)
+               rq->cmd_flags |= REQ_QUIET;
+
+       blk_execute_rq(rq->q, pd->bdev->bd_disk, rq, 0);
+       if (rq->errors)
+               ret = -EIO;
+out:
+       blk_put_request(rq);
+       return ret;
+}
+
+static const char *sense_key_string(__u8 index)
+{
+       static const char * const info[] = {
+               "No sense", "Recovered error", "Not ready",
+               "Medium error", "Hardware error", "Illegal request",
+               "Unit attention", "Data protect", "Blank check",
+       };
+
+       return index < ARRAY_SIZE(info) ? info[index] : "INVALID";
+}
+
+/*
+ * A generic sense dump / resolve mechanism should be implemented across
+ * all ATAPI + SCSI devices.
+ */
+static void pkt_dump_sense(struct pktcdvd_device *pd,
+                          struct packet_command *cgc)
+{
+       struct request_sense *sense = cgc->sense;
+
+       if (sense)
+               pkt_err(pd, "%*ph - sense %02x.%02x.%02x (%s)\n",
+                       CDROM_PACKET_SIZE, cgc->cmd,
+                       sense->sense_key, sense->asc, sense->ascq,
+                       sense_key_string(sense->sense_key));
+       else
+               pkt_err(pd, "%*ph - no sense\n", CDROM_PACKET_SIZE, cgc->cmd);
+}
+
+/*
+ * flush the drive cache to media
+ */
+static int pkt_flush_cache(struct pktcdvd_device *pd)
+{
+       struct packet_command cgc;
+
+       init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
+       cgc.cmd[0] = GPCMD_FLUSH_CACHE;
+       cgc.quiet = 1;
+
+       /*
+        * the IMMED bit -- we default to not setting it, although that
+        * would allow a much faster close, this is safer
+        */
+#if 0
+       cgc.cmd[1] = 1 << 1;
+#endif
+       return pkt_generic_packet(pd, &cgc);
+}
+
+/*
+ * speed is given as the normal factor, e.g. 4 for 4x
+ */
+static noinline_for_stack int pkt_set_speed(struct pktcdvd_device *pd,
+                               unsigned write_speed, unsigned read_speed)
+{
+       struct packet_command cgc;
+       struct request_sense sense;
+       int ret;
+
+       init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
+       cgc.sense = &sense;
+       cgc.cmd[0] = GPCMD_SET_SPEED;
+       cgc.cmd[2] = (read_speed >> 8) & 0xff;
+       cgc.cmd[3] = read_speed & 0xff;
+       cgc.cmd[4] = (write_speed >> 8) & 0xff;
+       cgc.cmd[5] = write_speed & 0xff;
+
+       if ((ret = pkt_generic_packet(pd, &cgc)))
+               pkt_dump_sense(pd, &cgc);
+
+       return ret;
+}
+
+/*
+ * Queue a bio for processing by the low-level CD device. Must be called
+ * from process context.
+ */
+static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio)
+{
+       spin_lock(&pd->iosched.lock);
+       if (bio_data_dir(bio) == READ)
+               bio_list_add(&pd->iosched.read_queue, bio);
+       else
+               bio_list_add(&pd->iosched.write_queue, bio);
+       spin_unlock(&pd->iosched.lock);
+
+       atomic_set(&pd->iosched.attention, 1);
+       wake_up(&pd->wqueue);
+}
+
+/*
+ * Process the queued read/write requests. This function handles special
+ * requirements for CDRW drives:
+ * - A cache flush command must be inserted before a read request if the
+ *   previous request was a write.
+ * - Switching between reading and writing is slow, so don't do it more often
+ *   than necessary.
+ * - Optimize for throughput at the expense of latency. This means that streaming
+ *   writes will never be interrupted by a read, but if the drive has to seek
+ *   before the next write, switch to reading instead if there are any pending
+ *   read requests.
+ * - Set the read speed according to current usage pattern. When only reading
+ *   from the device, it's best to use the highest possible read speed, but
+ *   when switching often between reading and writing, it's better to have the
+ *   same read and write speeds.
+ */
+static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
+{
+
+       if (atomic_read(&pd->iosched.attention) == 0)
+               return;
+       atomic_set(&pd->iosched.attention, 0);
+
+       for (;;) {
+               struct bio *bio;
+               int reads_queued, writes_queued;
+
+               spin_lock(&pd->iosched.lock);
+               reads_queued = !bio_list_empty(&pd->iosched.read_queue);
+               writes_queued = !bio_list_empty(&pd->iosched.write_queue);
+               spin_unlock(&pd->iosched.lock);
+
+               if (!reads_queued && !writes_queued)
+                       break;
+
+               if (pd->iosched.writing) {
+                       int need_write_seek = 1;
+                       spin_lock(&pd->iosched.lock);
+                       bio = bio_list_peek(&pd->iosched.write_queue);
+                       spin_unlock(&pd->iosched.lock);
+                       if (bio && (bio->bi_iter.bi_sector ==
+                                   pd->iosched.last_write))
+                               need_write_seek = 0;
+                       if (need_write_seek && reads_queued) {
+                               if (atomic_read(&pd->cdrw.pending_bios) > 0) {
+                                       pkt_dbg(2, pd, "write, waiting\n");
+                                       break;
+                               }
+                               pkt_flush_cache(pd);
+                               pd->iosched.writing = 0;
+                       }
+               } else {
+                       if (!reads_queued && writes_queued) {
+                               if (atomic_read(&pd->cdrw.pending_bios) > 0) {
+                                       pkt_dbg(2, pd, "read, waiting\n");
+                                       break;
+                               }
+                               pd->iosched.writing = 1;
+                       }
+               }
+
+               spin_lock(&pd->iosched.lock);
+               if (pd->iosched.writing)
+                       bio = bio_list_pop(&pd->iosched.write_queue);
+               else
+                       bio = bio_list_pop(&pd->iosched.read_queue);
+               spin_unlock(&pd->iosched.lock);
+
+               if (!bio)
+                       continue;
+
+               if (bio_data_dir(bio) == READ)
+                       pd->iosched.successive_reads +=
+                               bio->bi_iter.bi_size >> 10;
+               else {
+                       pd->iosched.successive_reads = 0;
+                       pd->iosched.last_write = bio_end_sector(bio);
+               }
+               if (pd->iosched.successive_reads >= HI_SPEED_SWITCH) {
+                       if (pd->read_speed == pd->write_speed) {
+                               pd->read_speed = MAX_SPEED;
+                               pkt_set_speed(pd, pd->write_speed, pd->read_speed);
+                       }
+               } else {
+                       if (pd->read_speed != pd->write_speed) {
+                               pd->read_speed = pd->write_speed;
+                               pkt_set_speed(pd, pd->write_speed, pd->read_speed);
+                       }
+               }
+
+               atomic_inc(&pd->cdrw.pending_bios);
+               generic_make_request(bio);
+       }
+}
+
+/*
+ * Special care is needed if the underlying block device has a small
+ * max_phys_segments value.
+ */
+static int pkt_set_segment_merging(struct pktcdvd_device *pd, struct request_queue *q)
+{
+       if ((pd->settings.size << 9) / CD_FRAMESIZE
+           <= queue_max_segments(q)) {
+               /*
+                * The cdrom device can handle one segment/frame
+                */
+               clear_bit(PACKET_MERGE_SEGS, &pd->flags);
+               return 0;
+       } else if ((pd->settings.size << 9) / PAGE_SIZE
+                  <= queue_max_segments(q)) {
+               /*
+                * We can handle this case at the expense of some extra memory
+                * copies during write operations
+                */
+               set_bit(PACKET_MERGE_SEGS, &pd->flags);
+               return 0;
+       } else {
+               pkt_err(pd, "cdrom max_phys_segments too small\n");
+               return -EIO;
+       }
+}
+
+/*
+ * Copy all data for this packet to pkt->pages[], so that
+ * a) The number of required segments for the write bio is minimized, which
+ *    is necessary for some scsi controllers.
+ * b) The data can be used as cache to avoid read requests if we receive a
+ *    new write request for the same zone.
+ */
+static void pkt_make_local_copy(struct packet_data *pkt, struct bio_vec *bvec)
+{
+       int f, p, offs;
+
+       /* Copy all data to pkt->pages[] */
+       p = 0;
+       offs = 0;
+       for (f = 0; f < pkt->frames; f++) {
+               if (bvec[f].bv_page != pkt->pages[p]) {
+                       void *vfrom = kmap_atomic(bvec[f].bv_page) + bvec[f].bv_offset;
+                       void *vto = page_address(pkt->pages[p]) + offs;
+                       memcpy(vto, vfrom, CD_FRAMESIZE);
+                       kunmap_atomic(vfrom);
+                       bvec[f].bv_page = pkt->pages[p];
+                       bvec[f].bv_offset = offs;
+               } else {
+                       BUG_ON(bvec[f].bv_offset != offs);
+               }
+               offs += CD_FRAMESIZE;
+               if (offs >= PAGE_SIZE) {
+                       offs = 0;
+                       p++;
+               }
+       }
+}
+
+static void pkt_end_io_read(struct bio *bio, int err)
+{
+       struct packet_data *pkt = bio->bi_private;
+       struct pktcdvd_device *pd = pkt->pd;
+       BUG_ON(!pd);
+
+       pkt_dbg(2, pd, "bio=%p sec0=%llx sec=%llx err=%d\n",
+               bio, (unsigned long long)pkt->sector,
+               (unsigned long long)bio->bi_iter.bi_sector, err);
+
+       if (err)
+               atomic_inc(&pkt->io_errors);
+       if (atomic_dec_and_test(&pkt->io_wait)) {
+               atomic_inc(&pkt->run_sm);
+               wake_up(&pd->wqueue);
+       }
+       pkt_bio_finished(pd);
+}
+
+static void pkt_end_io_packet_write(struct bio *bio, int err)
+{
+       struct packet_data *pkt = bio->bi_private;
+       struct pktcdvd_device *pd = pkt->pd;
+       BUG_ON(!pd);
+
+       pkt_dbg(2, pd, "id=%d, err=%d\n", pkt->id, err);
+
+       pd->stats.pkt_ended++;
+
+       pkt_bio_finished(pd);
+       atomic_dec(&pkt->io_wait);
+       atomic_inc(&pkt->run_sm);
+       wake_up(&pd->wqueue);
+}
+
+/*
+ * Schedule reads for the holes in a packet
+ */
+static void pkt_gather_data(struct pktcdvd_device *pd, struct packet_data *pkt)
+{
+       int frames_read = 0;
+       struct bio *bio;
+       int f;
+       char written[PACKET_MAX_SIZE];
+
+       BUG_ON(bio_list_empty(&pkt->orig_bios));
+
+       atomic_set(&pkt->io_wait, 0);
+       atomic_set(&pkt->io_errors, 0);
+
+       /*
+        * Figure out which frames we need to read before we can write.
+        */
+       memset(written, 0, sizeof(written));
+       spin_lock(&pkt->lock);
+       bio_list_for_each(bio, &pkt->orig_bios) {
+               int first_frame = (bio->bi_iter.bi_sector - pkt->sector) /
+                       (CD_FRAMESIZE >> 9);
+               int num_frames = bio->bi_iter.bi_size / CD_FRAMESIZE;
+               pd->stats.secs_w += num_frames * (CD_FRAMESIZE >> 9);
+               BUG_ON(first_frame < 0);
+               BUG_ON(first_frame + num_frames > pkt->frames);
+               for (f = first_frame; f < first_frame + num_frames; f++)
+                       written[f] = 1;
+       }
+       spin_unlock(&pkt->lock);
+
+       if (pkt->cache_valid) {
+               pkt_dbg(2, pd, "zone %llx cached\n",
+                       (unsigned long long)pkt->sector);
+               goto out_account;
+       }
+
+       /*
+        * Schedule reads for missing parts of the packet.
+        */
+       for (f = 0; f < pkt->frames; f++) {
+               int p, offset;
+
+               if (written[f])
+                       continue;
+
+               bio = pkt->r_bios[f];
+               bio_reset(bio);
+               bio->bi_iter.bi_sector = pkt->sector + f * (CD_FRAMESIZE >> 9);
+               bio->bi_bdev = pd->bdev;
+               bio->bi_end_io = pkt_end_io_read;
+               bio->bi_private = pkt;
+
+               p = (f * CD_FRAMESIZE) / PAGE_SIZE;
+               offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
+               pkt_dbg(2, pd, "Adding frame %d, page:%p offs:%d\n",
+                       f, pkt->pages[p], offset);
+               if (!bio_add_page(bio, pkt->pages[p], CD_FRAMESIZE, offset))
+                       BUG();
+
+               atomic_inc(&pkt->io_wait);
+               bio->bi_rw = READ;
+               pkt_queue_bio(pd, bio);
+               frames_read++;
+       }
+
+out_account:
+       pkt_dbg(2, pd, "need %d frames for zone %llx\n",
+               frames_read, (unsigned long long)pkt->sector);
+       pd->stats.pkt_started++;
+       pd->stats.secs_rg += frames_read * (CD_FRAMESIZE >> 9);
+}
+
+/*
+ * Find a packet matching zone, or the least recently used packet if
+ * there is no match.
+ */
+static struct packet_data *pkt_get_packet_data(struct pktcdvd_device *pd, int zone)
+{
+       struct packet_data *pkt;
+
+       list_for_each_entry(pkt, &pd->cdrw.pkt_free_list, list) {
+               if (pkt->sector == zone || pkt->list.next == &pd->cdrw.pkt_free_list) {
+                       list_del_init(&pkt->list);
+                       if (pkt->sector != zone)
+                               pkt->cache_valid = 0;
+                       return pkt;
+               }
+       }
+       BUG();
+       return NULL;
+}
+
+static void pkt_put_packet_data(struct pktcdvd_device *pd, struct packet_data *pkt)
+{
+       if (pkt->cache_valid) {
+               list_add(&pkt->list, &pd->cdrw.pkt_free_list);
+       } else {
+               list_add_tail(&pkt->list, &pd->cdrw.pkt_free_list);
+       }
+}
+
+/*
+ * recover a failed write, query for relocation if possible
+ *
+ * returns 1 if recovery is possible, or 0 if not
+ *
+ */
+static int pkt_start_recovery(struct packet_data *pkt)
+{
+       /*
+        * FIXME. We need help from the file system to implement
+        * recovery handling.
+        */
+       return 0;
+#if 0
+       struct request *rq = pkt->rq;
+       struct pktcdvd_device *pd = rq->rq_disk->private_data;
+       struct block_device *pkt_bdev;
+       struct super_block *sb = NULL;
+       unsigned long old_block, new_block;
+       sector_t new_sector;
+
+       pkt_bdev = bdget(kdev_t_to_nr(pd->pkt_dev));
+       if (pkt_bdev) {
+               sb = get_super(pkt_bdev);
+               bdput(pkt_bdev);
+       }
+
+       if (!sb)
+               return 0;
+
+       if (!sb->s_op->relocate_blocks)
+               goto out;
+
+       old_block = pkt->sector / (CD_FRAMESIZE >> 9);
+       if (sb->s_op->relocate_blocks(sb, old_block, &new_block))
+               goto out;
+
+       new_sector = new_block * (CD_FRAMESIZE >> 9);
+       pkt->sector = new_sector;
+
+       bio_reset(pkt->bio);
+       pkt->bio->bi_bdev = pd->bdev;
+       pkt->bio->bi_rw = REQ_WRITE;
+       pkt->bio->bi_iter.bi_sector = new_sector;
+       pkt->bio->bi_iter.bi_size = pkt->frames * CD_FRAMESIZE;
+       pkt->bio->bi_vcnt = pkt->frames;
+
+       pkt->bio->bi_end_io = pkt_end_io_packet_write;
+       pkt->bio->bi_private = pkt;
+
+       drop_super(sb);
+       return 1;
+
+out:
+       drop_super(sb);
+       return 0;
+#endif
+}
+
+static inline void pkt_set_state(struct packet_data *pkt, enum packet_data_state state)
+{
+#if PACKET_DEBUG > 1
+       static const char *state_name[] = {
+               "IDLE", "WAITING", "READ_WAIT", "WRITE_WAIT", "RECOVERY", "FINISHED"
+       };
+       enum packet_data_state old_state = pkt->state;
+       pkt_dbg(2, pd, "pkt %2d : s=%6llx %s -> %s\n",
+               pkt->id, (unsigned long long)pkt->sector,
+               state_name[old_state], state_name[state]);
+#endif
+       pkt->state = state;
+}
+
+/*
+ * Scan the work queue to see if we can start a new packet.
+ * returns non-zero if any work was done.
+ */
+static int pkt_handle_queue(struct pktcdvd_device *pd)
+{
+       struct packet_data *pkt, *p;
+       struct bio *bio = NULL;
+       sector_t zone = 0; /* Suppress gcc warning */
+       struct pkt_rb_node *node, *first_node;
+       struct rb_node *n;
+       int wakeup;
+
+       atomic_set(&pd->scan_queue, 0);
+
+       if (list_empty(&pd->cdrw.pkt_free_list)) {
+               pkt_dbg(2, pd, "no pkt\n");
+               return 0;
+       }
+
+       /*
+        * Try to find a zone we are not already working on.
+        */
+       spin_lock(&pd->lock);
+       first_node = pkt_rbtree_find(pd, pd->current_sector);
+       if (!first_node) {
+               n = rb_first(&pd->bio_queue);
+               if (n)
+                       first_node = rb_entry(n, struct pkt_rb_node, rb_node);
+       }
+       node = first_node;
+       while (node) {
+               bio = node->bio;
+               zone = get_zone(bio->bi_iter.bi_sector, pd);
+               list_for_each_entry(p, &pd->cdrw.pkt_active_list, list) {
+                       if (p->sector == zone) {
+                               bio = NULL;
+                               goto try_next_bio;
+                       }
+               }
+               break;
+try_next_bio:
+               node = pkt_rbtree_next(node);
+               if (!node) {
+                       n = rb_first(&pd->bio_queue);
+                       if (n)
+                               node = rb_entry(n, struct pkt_rb_node, rb_node);
+               }
+               if (node == first_node)
+                       node = NULL;
+       }
+       spin_unlock(&pd->lock);
+       if (!bio) {
+               pkt_dbg(2, pd, "no bio\n");
+               return 0;
+       }
+
+       pkt = pkt_get_packet_data(pd, zone);
+
+       pd->current_sector = zone + pd->settings.size;
+       pkt->sector = zone;
+       BUG_ON(pkt->frames != pd->settings.size >> 2);
+       pkt->write_size = 0;
+
+       /*
+        * Scan work queue for bios in the same zone and link them
+        * to this packet.
+        */
+       spin_lock(&pd->lock);
+       pkt_dbg(2, pd, "looking for zone %llx\n", (unsigned long long)zone);
+       while ((node = pkt_rbtree_find(pd, zone)) != NULL) {
+               bio = node->bio;
+               pkt_dbg(2, pd, "found zone=%llx\n", (unsigned long long)
+                       get_zone(bio->bi_iter.bi_sector, pd));
+               if (get_zone(bio->bi_iter.bi_sector, pd) != zone)
+                       break;
+               pkt_rbtree_erase(pd, node);
+               spin_lock(&pkt->lock);
+               bio_list_add(&pkt->orig_bios, bio);
+               pkt->write_size += bio->bi_iter.bi_size / CD_FRAMESIZE;
+               spin_unlock(&pkt->lock);
+       }
+       /* check write congestion marks, and if bio_queue_size is
+          below, wake up any waiters */
+       wakeup = (pd->write_congestion_on > 0
+                       && pd->bio_queue_size <= pd->write_congestion_off);
+       spin_unlock(&pd->lock);
+       if (wakeup) {
+               clear_bdi_congested(&pd->disk->queue->backing_dev_info,
+                                       BLK_RW_ASYNC);
+       }
+
+       pkt->sleep_time = max(PACKET_WAIT_TIME, 1);
+       pkt_set_state(pkt, PACKET_WAITING_STATE);
+       atomic_set(&pkt->run_sm, 1);
+
+       spin_lock(&pd->cdrw.active_list_lock);
+       list_add(&pkt->list, &pd->cdrw.pkt_active_list);
+       spin_unlock(&pd->cdrw.active_list_lock);
+
+       return 1;
+}
+
+/*
+ * Assemble a bio to write one packet and queue the bio for processing
+ * by the underlying block device.
+ */
+static void pkt_start_write(struct pktcdvd_device *pd, struct packet_data *pkt)
+{
+       int f;
+       struct bio_vec *bvec = pkt->w_bio->bi_io_vec;
+
+       bio_reset(pkt->w_bio);
+       pkt->w_bio->bi_iter.bi_sector = pkt->sector;
+       pkt->w_bio->bi_bdev = pd->bdev;
+       pkt->w_bio->bi_end_io = pkt_end_io_packet_write;
+       pkt->w_bio->bi_private = pkt;
+
+       /* XXX: locking? */
+       for (f = 0; f < pkt->frames; f++) {
+               bvec[f].bv_page = pkt->pages[(f * CD_FRAMESIZE) / PAGE_SIZE];
+               bvec[f].bv_offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
+               if (!bio_add_page(pkt->w_bio, bvec[f].bv_page, CD_FRAMESIZE, bvec[f].bv_offset))
+                       BUG();
+       }
+       pkt_dbg(2, pd, "vcnt=%d\n", pkt->w_bio->bi_vcnt);
+
+       /*
+        * Fill-in bvec with data from orig_bios.
+        */
+       spin_lock(&pkt->lock);
+       bio_copy_data(pkt->w_bio, pkt->orig_bios.head);
+
+       pkt_set_state(pkt, PACKET_WRITE_WAIT_STATE);
+       spin_unlock(&pkt->lock);
+
+       pkt_dbg(2, pd, "Writing %d frames for zone %llx\n",
+               pkt->write_size, (unsigned long long)pkt->sector);
+
+       if (test_bit(PACKET_MERGE_SEGS, &pd->flags) || (pkt->write_size < pkt->frames)) {
+               pkt_make_local_copy(pkt, bvec);
+               pkt->cache_valid = 1;
+       } else {
+               pkt->cache_valid = 0;
+       }
+
+       /* Start the write request */
+       atomic_set(&pkt->io_wait, 1);
+       pkt->w_bio->bi_rw = WRITE;
+       pkt_queue_bio(pd, pkt->w_bio);
+}
+
+static void pkt_finish_packet(struct packet_data *pkt, int uptodate)
+{
+       struct bio *bio;
+
+       if (!uptodate)
+               pkt->cache_valid = 0;
+
+       /* Finish all bios corresponding to this packet */
+       while ((bio = bio_list_pop(&pkt->orig_bios)))
+               bio_endio(bio, uptodate ? 0 : -EIO);
+}
+
+static void pkt_run_state_machine(struct pktcdvd_device *pd, struct packet_data *pkt)
+{
+       int uptodate;
+
+       pkt_dbg(2, pd, "pkt %d\n", pkt->id);
+
+       for (;;) {
+               switch (pkt->state) {
+               case PACKET_WAITING_STATE:
+                       if ((pkt->write_size < pkt->frames) && (pkt->sleep_time > 0))
+                               return;
+
+                       pkt->sleep_time = 0;
+                       pkt_gather_data(pd, pkt);
+                       pkt_set_state(pkt, PACKET_READ_WAIT_STATE);
+                       break;
+
+               case PACKET_READ_WAIT_STATE:
+                       if (atomic_read(&pkt->io_wait) > 0)
+                               return;
+
+                       if (atomic_read(&pkt->io_errors) > 0) {
+                               pkt_set_state(pkt, PACKET_RECOVERY_STATE);
+                       } else {
+                               pkt_start_write(pd, pkt);
+                       }
+                       break;
+
+               case PACKET_WRITE_WAIT_STATE:
+                       if (atomic_read(&pkt->io_wait) > 0)
+                               return;
+
+                       if (test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags)) {
+                               pkt_set_state(pkt, PACKET_FINISHED_STATE);
+                       } else {
+                               pkt_set_state(pkt, PACKET_RECOVERY_STATE);
+                       }
+                       break;
+
+               case PACKET_RECOVERY_STATE:
+                       if (pkt_start_recovery(pkt)) {
+                               pkt_start_write(pd, pkt);
+                       } else {
+                               pkt_dbg(2, pd, "No recovery possible\n");
+                               pkt_set_state(pkt, PACKET_FINISHED_STATE);
+                       }
+                       break;
+
+               case PACKET_FINISHED_STATE:
+                       uptodate = test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags);
+                       pkt_finish_packet(pkt, uptodate);
+                       return;
+
+               default:
+                       BUG();
+                       break;
+               }
+       }
+}
+
+static void pkt_handle_packets(struct pktcdvd_device *pd)
+{
+       struct packet_data *pkt, *next;
+
+       /*
+        * Run state machine for active packets
+        */
+       list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
+               if (atomic_read(&pkt->run_sm) > 0) {
+                       atomic_set(&pkt->run_sm, 0);
+                       pkt_run_state_machine(pd, pkt);
+               }
+       }
+
+       /*
+        * Move no longer active packets to the free list
+        */
+       spin_lock(&pd->cdrw.active_list_lock);
+       list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_active_list, list) {
+               if (pkt->state == PACKET_FINISHED_STATE) {
+                       list_del(&pkt->list);
+                       pkt_put_packet_data(pd, pkt);
+                       pkt_set_state(pkt, PACKET_IDLE_STATE);
+                       atomic_set(&pd->scan_queue, 1);
+               }
+       }
+       spin_unlock(&pd->cdrw.active_list_lock);
+}
+
+static void pkt_count_states(struct pktcdvd_device *pd, int *states)
+{
+       struct packet_data *pkt;
+       int i;
+
+       for (i = 0; i < PACKET_NUM_STATES; i++)
+               states[i] = 0;
+
+       spin_lock(&pd->cdrw.active_list_lock);
+       list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
+               states[pkt->state]++;
+       }
+       spin_unlock(&pd->cdrw.active_list_lock);
+}
+
+/*
+ * kcdrwd is woken up when writes have been queued for one of our
+ * registered devices
+ */
+static int kcdrwd(void *foobar)
+{
+       struct pktcdvd_device *pd = foobar;
+       struct packet_data *pkt;
+       long min_sleep_time, residue;
+
+       set_user_nice(current, MIN_NICE);
+       set_freezable();
+
+       for (;;) {
+               DECLARE_WAITQUEUE(wait, current);
+
+               /*
+                * Wait until there is something to do
+                */
+               add_wait_queue(&pd->wqueue, &wait);
+               for (;;) {
+                       set_current_state(TASK_INTERRUPTIBLE);
+
+                       /* Check if we need to run pkt_handle_queue */
+                       if (atomic_read(&pd->scan_queue) > 0)
+                               goto work_to_do;
+
+                       /* Check if we need to run the state machine for some packet */
+                       list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
+                               if (atomic_read(&pkt->run_sm) > 0)
+                                       goto work_to_do;
+                       }
+
+                       /* Check if we need to process the iosched queues */
+                       if (atomic_read(&pd->iosched.attention) != 0)
+                               goto work_to_do;
+
+                       /* Otherwise, go to sleep */
+                       if (PACKET_DEBUG > 1) {
+                               int states[PACKET_NUM_STATES];
+                               pkt_count_states(pd, states);
+                               pkt_dbg(2, pd, "i:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
+                                       states[0], states[1], states[2],
+                                       states[3], states[4], states[5]);
+                       }
+
+                       min_sleep_time = MAX_SCHEDULE_TIMEOUT;
+                       list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
+                               if (pkt->sleep_time && pkt->sleep_time < min_sleep_time)
+                                       min_sleep_time = pkt->sleep_time;
+                       }
+
+                       pkt_dbg(2, pd, "sleeping\n");
+                       residue = schedule_timeout(min_sleep_time);
+                       pkt_dbg(2, pd, "wake up\n");
+
+                       /* make swsusp happy with our thread */
+                       try_to_freeze();
+
+                       list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
+                               if (!pkt->sleep_time)
+                                       continue;
+                               pkt->sleep_time -= min_sleep_time - residue;
+                               if (pkt->sleep_time <= 0) {
+                                       pkt->sleep_time = 0;
+                                       atomic_inc(&pkt->run_sm);
+                               }
+                       }
+
+                       if (kthread_should_stop())
+                               break;
+               }
+work_to_do:
+               set_current_state(TASK_RUNNING);
+               remove_wait_queue(&pd->wqueue, &wait);
+
+               if (kthread_should_stop())
+                       break;
+
+               /*
+                * if pkt_handle_queue returns true, we can queue
+                * another request.
+                */
+               while (pkt_handle_queue(pd))
+                       ;
+
+               /*
+                * Handle packet state machine
+                */
+               pkt_handle_packets(pd);
+
+               /*
+                * Handle iosched queues
+                */
+               pkt_iosched_process_queue(pd);
+       }
+
+       return 0;
+}
+
+static void pkt_print_settings(struct pktcdvd_device *pd)
+{
+       pkt_info(pd, "%s packets, %u blocks, Mode-%c disc\n",
+                pd->settings.fp ? "Fixed" : "Variable",
+                pd->settings.size >> 2,
+                pd->settings.block_mode == 8 ? '1' : '2');
+}
+
+static int pkt_mode_sense(struct pktcdvd_device *pd, struct packet_command *cgc, int page_code, int page_control)
+{
+       memset(cgc->cmd, 0, sizeof(cgc->cmd));
+
+       cgc->cmd[0] = GPCMD_MODE_SENSE_10;
+       cgc->cmd[2] = page_code | (page_control << 6);
+       cgc->cmd[7] = cgc->buflen >> 8;
+       cgc->cmd[8] = cgc->buflen & 0xff;
+       cgc->data_direction = CGC_DATA_READ;
+       return pkt_generic_packet(pd, cgc);
+}
+
+static int pkt_mode_select(struct pktcdvd_device *pd, struct packet_command *cgc)
+{
+       memset(cgc->cmd, 0, sizeof(cgc->cmd));
+       memset(cgc->buffer, 0, 2);
+       cgc->cmd[0] = GPCMD_MODE_SELECT_10;
+       cgc->cmd[1] = 0x10;             /* PF */
+       cgc->cmd[7] = cgc->buflen >> 8;
+       cgc->cmd[8] = cgc->buflen & 0xff;
+       cgc->data_direction = CGC_DATA_WRITE;
+       return pkt_generic_packet(pd, cgc);
+}
+
+static int pkt_get_disc_info(struct pktcdvd_device *pd, disc_information *di)
+{
+       struct packet_command cgc;
+       int ret;
+
+       /* set up command and get the disc info */
+       init_cdrom_command(&cgc, di, sizeof(*di), CGC_DATA_READ);
+       cgc.cmd[0] = GPCMD_READ_DISC_INFO;
+       cgc.cmd[8] = cgc.buflen = 2;
+       cgc.quiet = 1;
+
+       if ((ret = pkt_generic_packet(pd, &cgc)))
+               return ret;
+
+       /* not all drives have the same disc_info length, so requeue
+        * packet with the length the drive tells us it can supply
+        */
+       cgc.buflen = be16_to_cpu(di->disc_information_length) +
+                    sizeof(di->disc_information_length);
+
+       if (cgc.buflen > sizeof(disc_information))
+               cgc.buflen = sizeof(disc_information);
+
+       cgc.cmd[8] = cgc.buflen;
+       return pkt_generic_packet(pd, &cgc);
+}
+
+static int pkt_get_track_info(struct pktcdvd_device *pd, __u16 track, __u8 type, track_information *ti)
+{
+       struct packet_command cgc;
+       int ret;
+
+       init_cdrom_command(&cgc, ti, 8, CGC_DATA_READ);
+       cgc.cmd[0] = GPCMD_READ_TRACK_RZONE_INFO;
+       cgc.cmd[1] = type & 3;
+       cgc.cmd[4] = (track & 0xff00) >> 8;
+       cgc.cmd[5] = track & 0xff;
+       cgc.cmd[8] = 8;
+       cgc.quiet = 1;
+
+       if ((ret = pkt_generic_packet(pd, &cgc)))
+               return ret;
+
+       cgc.buflen = be16_to_cpu(ti->track_information_length) +
+                    sizeof(ti->track_information_length);
+
+       if (cgc.buflen > sizeof(track_information))
+               cgc.buflen = sizeof(track_information);
+
+       cgc.cmd[8] = cgc.buflen;
+       return pkt_generic_packet(pd, &cgc);
+}
+
+static noinline_for_stack int pkt_get_last_written(struct pktcdvd_device *pd,
+                                               long *last_written)
+{
+       disc_information di;
+       track_information ti;
+       __u32 last_track;
+       int ret = -1;
+
+       if ((ret = pkt_get_disc_info(pd, &di)))
+               return ret;
+
+       last_track = (di.last_track_msb << 8) | di.last_track_lsb;
+       if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
+               return ret;
+
+       /* if this track is blank, try the previous. */
+       if (ti.blank) {
+               last_track--;
+               if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
+                       return ret;
+       }
+
+       /* if last recorded field is valid, return it. */
+       if (ti.lra_v) {
+               *last_written = be32_to_cpu(ti.last_rec_address);
+       } else {
+               /* make it up instead */
+               *last_written = be32_to_cpu(ti.track_start) +
+                               be32_to_cpu(ti.track_size);
+               if (ti.free_blocks)
+                       *last_written -= (be32_to_cpu(ti.free_blocks) + 7);
+       }
+       return 0;
+}
+
+/*
+ * write mode select package based on pd->settings
+ */
+static noinline_for_stack int pkt_set_write_settings(struct pktcdvd_device *pd)
+{
+       struct packet_command cgc;
+       struct request_sense sense;
+       write_param_page *wp;
+       char buffer[128];
+       int ret, size;
+
+       /* doesn't apply to DVD+RW or DVD-RAM */
+       if ((pd->mmc3_profile == 0x1a) || (pd->mmc3_profile == 0x12))
+               return 0;
+
+       memset(buffer, 0, sizeof(buffer));
+       init_cdrom_command(&cgc, buffer, sizeof(*wp), CGC_DATA_READ);
+       cgc.sense = &sense;
+       if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
+               pkt_dump_sense(pd, &cgc);
+               return ret;
+       }
+
+       size = 2 + ((buffer[0] << 8) | (buffer[1] & 0xff));
+       pd->mode_offset = (buffer[6] << 8) | (buffer[7] & 0xff);
+       if (size > sizeof(buffer))
+               size = sizeof(buffer);
+
+       /*
+        * now get it all
+        */
+       init_cdrom_command(&cgc, buffer, size, CGC_DATA_READ);
+       cgc.sense = &sense;
+       if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
+               pkt_dump_sense(pd, &cgc);
+               return ret;
+       }
+
+       /*
+        * write page is offset header + block descriptor length
+        */
+       wp = (write_param_page *) &buffer[sizeof(struct mode_page_header) + pd->mode_offset];
+
+       wp->fp = pd->settings.fp;
+       wp->track_mode = pd->settings.track_mode;
+       wp->write_type = pd->settings.write_type;
+       wp->data_block_type = pd->settings.block_mode;
+
+       wp->multi_session = 0;
+
+#ifdef PACKET_USE_LS
+       wp->link_size = 7;
+       wp->ls_v = 1;
+#endif
+
+       if (wp->data_block_type == PACKET_BLOCK_MODE1) {
+               wp->session_format = 0;
+               wp->subhdr2 = 0x20;
+       } else if (wp->data_block_type == PACKET_BLOCK_MODE2) {
+               wp->session_format = 0x20;
+               wp->subhdr2 = 8;
+#if 0
+               wp->mcn[0] = 0x80;
+               memcpy(&wp->mcn[1], PACKET_MCN, sizeof(wp->mcn) - 1);
+#endif
+       } else {
+               /*
+                * paranoia
+                */
+               pkt_err(pd, "write mode wrong %d\n", wp->data_block_type);
+               return 1;
+       }
+       wp->packet_size = cpu_to_be32(pd->settings.size >> 2);
+
+       cgc.buflen = cgc.cmd[8] = size;
+       if ((ret = pkt_mode_select(pd, &cgc))) {
+               pkt_dump_sense(pd, &cgc);
+               return ret;
+       }
+
+       pkt_print_settings(pd);
+       return 0;
+}
+
+/*
+ * 1 -- we can write to this track, 0 -- we can't
+ */
+static int pkt_writable_track(struct pktcdvd_device *pd, track_information *ti)
+{
+       switch (pd->mmc3_profile) {
+               case 0x1a: /* DVD+RW */
+               case 0x12: /* DVD-RAM */
+                       /* The track is always writable on DVD+RW/DVD-RAM */
+                       return 1;
+               default:
+                       break;
+       }
+
+       if (!ti->packet || !ti->fp)
+               return 0;
+
+       /*
+        * "good" settings as per Mt Fuji.
+        */
+       if (ti->rt == 0 && ti->blank == 0)
+               return 1;
+
+       if (ti->rt == 0 && ti->blank == 1)
+               return 1;
+
+       if (ti->rt == 1 && ti->blank == 0)
+               return 1;
+
+       pkt_err(pd, "bad state %d-%d-%d\n", ti->rt, ti->blank, ti->packet);
+       return 0;
+}
+
+/*
+ * 1 -- we can write to this disc, 0 -- we can't
+ */
+static int pkt_writable_disc(struct pktcdvd_device *pd, disc_information *di)
+{
+       switch (pd->mmc3_profile) {
+               case 0x0a: /* CD-RW */
+               case 0xffff: /* MMC3 not supported */
+                       break;
+               case 0x1a: /* DVD+RW */
+               case 0x13: /* DVD-RW */
+               case 0x12: /* DVD-RAM */
+                       return 1;
+               default:
+                       pkt_dbg(2, pd, "Wrong disc profile (%x)\n",
+                               pd->mmc3_profile);
+                       return 0;
+       }
+
+       /*
+        * for disc type 0xff we should probably reserve a new track.
+        * but i'm not sure, should we leave this to user apps? probably.
+        */
+       if (di->disc_type == 0xff) {
+               pkt_notice(pd, "unknown disc - no track?\n");
+               return 0;
+       }
+
+       if (di->disc_type != 0x20 && di->disc_type != 0) {
+               pkt_err(pd, "wrong disc type (%x)\n", di->disc_type);
+               return 0;
+       }
+
+       if (di->erasable == 0) {
+               pkt_notice(pd, "disc not erasable\n");
+               return 0;
+       }
+
+       if (di->border_status == PACKET_SESSION_RESERVED) {
+               pkt_err(pd, "can't write to last track (reserved)\n");
+               return 0;
+       }
+
+       return 1;
+}
+
+static noinline_for_stack int pkt_probe_settings(struct pktcdvd_device *pd)
+{
+       struct packet_command cgc;
+       unsigned char buf[12];
+       disc_information di;
+       track_information ti;
+       int ret, track;
+
+       init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
+       cgc.cmd[0] = GPCMD_GET_CONFIGURATION;
+       cgc.cmd[8] = 8;
+       ret = pkt_generic_packet(pd, &cgc);
+       pd->mmc3_profile = ret ? 0xffff : buf[6] << 8 | buf[7];
+
+       memset(&di, 0, sizeof(disc_information));
+       memset(&ti, 0, sizeof(track_information));
+
+       if ((ret = pkt_get_disc_info(pd, &di))) {
+               pkt_err(pd, "failed get_disc\n");
+               return ret;
+       }
+
+       if (!pkt_writable_disc(pd, &di))
+               return -EROFS;
+
+       pd->type = di.erasable ? PACKET_CDRW : PACKET_CDR;
+
+       track = 1; /* (di.last_track_msb << 8) | di.last_track_lsb; */
+       if ((ret = pkt_get_track_info(pd, track, 1, &ti))) {
+               pkt_err(pd, "failed get_track\n");
+               return ret;
+       }
+
+       if (!pkt_writable_track(pd, &ti)) {
+               pkt_err(pd, "can't write to this track\n");
+               return -EROFS;
+       }
+
+       /*
+        * we keep packet size in 512 byte units, makes it easier to
+        * deal with request calculations.
+        */
+       pd->settings.size = be32_to_cpu(ti.fixed_packet_size) << 2;
+       if (pd->settings.size == 0) {
+               pkt_notice(pd, "detected zero packet size!\n");
+               return -ENXIO;
+       }
+       if (pd->settings.size > PACKET_MAX_SECTORS) {
+               pkt_err(pd, "packet size is too big\n");
+               return -EROFS;
+       }
+       pd->settings.fp = ti.fp;
+       pd->offset = (be32_to_cpu(ti.track_start) << 2) & (pd->settings.size - 1);
+
+       if (ti.nwa_v) {
+               pd->nwa = be32_to_cpu(ti.next_writable);
+               set_bit(PACKET_NWA_VALID, &pd->flags);
+       }
+
+       /*
+        * in theory we could use lra on -RW media as well and just zero
+        * blocks that haven't been written yet, but in practice that
+        * is just a no-go. we'll use that for -R, naturally.
+        */
+       if (ti.lra_v) {
+               pd->lra = be32_to_cpu(ti.last_rec_address);
+               set_bit(PACKET_LRA_VALID, &pd->flags);
+       } else {
+               pd->lra = 0xffffffff;
+               set_bit(PACKET_LRA_VALID, &pd->flags);
+       }
+
+       /*
+        * fine for now
+        */
+       pd->settings.link_loss = 7;
+       pd->settings.write_type = 0;    /* packet */
+       pd->settings.track_mode = ti.track_mode;
+
+       /*
+        * mode1 or mode2 disc
+        */
+       switch (ti.data_mode) {
+               case PACKET_MODE1:
+                       pd->settings.block_mode = PACKET_BLOCK_MODE1;
+                       break;
+               case PACKET_MODE2:
+                       pd->settings.block_mode = PACKET_BLOCK_MODE2;
+                       break;
+               default:
+                       pkt_err(pd, "unknown data mode\n");
+                       return -EROFS;
+       }
+       return 0;
+}
+
+/*
+ * enable/disable write caching on drive
+ */
+static noinline_for_stack int pkt_write_caching(struct pktcdvd_device *pd,
+                                               int set)
+{
+       struct packet_command cgc;
+       struct request_sense sense;
+       unsigned char buf[64];
+       int ret;
+
+       init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
+       cgc.sense = &sense;
+       cgc.buflen = pd->mode_offset + 12;
+
+       /*
+        * caching mode page might not be there, so quiet this command
+        */
+       cgc.quiet = 1;
+
+       if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WCACHING_PAGE, 0)))
+               return ret;
+
+       buf[pd->mode_offset + 10] |= (!!set << 2);
+
+       cgc.buflen = cgc.cmd[8] = 2 + ((buf[0] << 8) | (buf[1] & 0xff));
+       ret = pkt_mode_select(pd, &cgc);
+       if (ret) {
+               pkt_err(pd, "write caching control failed\n");
+               pkt_dump_sense(pd, &cgc);
+       } else if (!ret && set)
+               pkt_notice(pd, "enabled write caching\n");
+       return ret;
+}
+
+static int pkt_lock_door(struct pktcdvd_device *pd, int lockflag)
+{
+       struct packet_command cgc;
+
+       init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
+       cgc.cmd[0] = GPCMD_PREVENT_ALLOW_MEDIUM_REMOVAL;
+       cgc.cmd[4] = lockflag ? 1 : 0;
+       return pkt_generic_packet(pd, &cgc);
+}
+
+/*
+ * Returns drive maximum write speed
+ */
+static noinline_for_stack int pkt_get_max_speed(struct pktcdvd_device *pd,
+                                               unsigned *write_speed)
+{
+       struct packet_command cgc;
+       struct request_sense sense;
+       unsigned char buf[256+18];
+       unsigned char *cap_buf;
+       int ret, offset;
+
+       cap_buf = &buf[sizeof(struct mode_page_header) + pd->mode_offset];
+       init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_UNKNOWN);
+       cgc.sense = &sense;
+
+       ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
+       if (ret) {
+               cgc.buflen = pd->mode_offset + cap_buf[1] + 2 +
+                            sizeof(struct mode_page_header);
+               ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
+               if (ret) {
+                       pkt_dump_sense(pd, &cgc);
+                       return ret;
+               }
+       }
+
+       offset = 20;                        /* Obsoleted field, used by older drives */
+       if (cap_buf[1] >= 28)
+               offset = 28;                /* Current write speed selected */
+       if (cap_buf[1] >= 30) {
+               /* If the drive reports at least one "Logical Unit Write
+                * Speed Performance Descriptor Block", use the information
+                * in the first block. (contains the highest speed)
+                */
+               int num_spdb = (cap_buf[30] << 8) + cap_buf[31];
+               if (num_spdb > 0)
+                       offset = 34;
+       }
+
+       *write_speed = (cap_buf[offset] << 8) | cap_buf[offset + 1];
+       return 0;
+}
+
+/* These tables from cdrecord - I don't have orange book */
+/* standard speed CD-RW (1-4x) */
+static char clv_to_speed[16] = {
+       /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
+          0, 2, 4, 6, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
+};
+/* high speed CD-RW (-10x) */
+static char hs_clv_to_speed[16] = {
+       /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
+          0, 2, 4, 6, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
+};
+/* ultra high speed CD-RW */
+static char us_clv_to_speed[16] = {
+       /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
+          0, 2, 4, 8, 0, 0,16, 0,24,32,40,48, 0, 0, 0, 0
+};
+
+/*
+ * reads the maximum media speed from ATIP
+ */
+static noinline_for_stack int pkt_media_speed(struct pktcdvd_device *pd,
+                                               unsigned *speed)
+{
+       struct packet_command cgc;
+       struct request_sense sense;
+       unsigned char buf[64];
+       unsigned int size, st, sp;
+       int ret;
+
+       init_cdrom_command(&cgc, buf, 2, CGC_DATA_READ);
+       cgc.sense = &sense;
+       cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
+       cgc.cmd[1] = 2;
+       cgc.cmd[2] = 4; /* READ ATIP */
+       cgc.cmd[8] = 2;
+       ret = pkt_generic_packet(pd, &cgc);
+       if (ret) {
+               pkt_dump_sense(pd, &cgc);
+               return ret;
+       }
+       size = ((unsigned int) buf[0]<<8) + buf[1] + 2;
+       if (size > sizeof(buf))
+               size = sizeof(buf);
+
+       init_cdrom_command(&cgc, buf, size, CGC_DATA_READ);
+       cgc.sense = &sense;
+       cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
+       cgc.cmd[1] = 2;
+       cgc.cmd[2] = 4;
+       cgc.cmd[8] = size;
+       ret = pkt_generic_packet(pd, &cgc);
+       if (ret) {
+               pkt_dump_sense(pd, &cgc);
+               return ret;
+       }
+
+       if (!(buf[6] & 0x40)) {
+               pkt_notice(pd, "disc type is not CD-RW\n");
+               return 1;
+       }
+       if (!(buf[6] & 0x4)) {
+               pkt_notice(pd, "A1 values on media are not valid, maybe not CDRW?\n");
+               return 1;
+       }
+
+       st = (buf[6] >> 3) & 0x7; /* disc sub-type */
+
+       sp = buf[16] & 0xf; /* max speed from ATIP A1 field */
+
+       /* Info from cdrecord */
+       switch (st) {
+               case 0: /* standard speed */
+                       *speed = clv_to_speed[sp];
+                       break;
+               case 1: /* high speed */
+                       *speed = hs_clv_to_speed[sp];
+                       break;
+               case 2: /* ultra high speed */
+                       *speed = us_clv_to_speed[sp];
+                       break;
+               default:
+                       pkt_notice(pd, "unknown disc sub-type %d\n", st);
+                       return 1;
+       }
+       if (*speed) {
+               pkt_info(pd, "maximum media speed: %d\n", *speed);
+               return 0;
+       } else {
+               pkt_notice(pd, "unknown speed %d for sub-type %d\n", sp, st);
+               return 1;
+       }
+}
+
+static noinline_for_stack int pkt_perform_opc(struct pktcdvd_device *pd)
+{
+       struct packet_command cgc;
+       struct request_sense sense;
+       int ret;
+
+       pkt_dbg(2, pd, "Performing OPC\n");
+
+       init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
+       cgc.sense = &sense;
+       cgc.timeout = 60*HZ;
+       cgc.cmd[0] = GPCMD_SEND_OPC;
+       cgc.cmd[1] = 1;
+       if ((ret = pkt_generic_packet(pd, &cgc)))
+               pkt_dump_sense(pd, &cgc);
+       return ret;
+}
+
+static int pkt_open_write(struct pktcdvd_device *pd)
+{
+       int ret;
+       unsigned int write_speed, media_write_speed, read_speed;
+
+       if ((ret = pkt_probe_settings(pd))) {
+               pkt_dbg(2, pd, "failed probe\n");
+               return ret;
+       }
+
+       if ((ret = pkt_set_write_settings(pd))) {
+               pkt_dbg(1, pd, "failed saving write settings\n");
+               return -EIO;
+       }
+
+       pkt_write_caching(pd, USE_WCACHING);
+
+       if ((ret = pkt_get_max_speed(pd, &write_speed)))
+               write_speed = 16 * 177;
+       switch (pd->mmc3_profile) {
+               case 0x13: /* DVD-RW */
+               case 0x1a: /* DVD+RW */
+               case 0x12: /* DVD-RAM */
+                       pkt_dbg(1, pd, "write speed %ukB/s\n", write_speed);
+                       break;
+               default:
+                       if ((ret = pkt_media_speed(pd, &media_write_speed)))
+                               media_write_speed = 16;
+                       write_speed = min(write_speed, media_write_speed * 177);
+                       pkt_dbg(1, pd, "write speed %ux\n", write_speed / 176);
+                       break;
+       }
+       read_speed = write_speed;
+
+       if ((ret = pkt_set_speed(pd, write_speed, read_speed))) {
+               pkt_dbg(1, pd, "couldn't set write speed\n");
+               return -EIO;
+       }
+       pd->write_speed = write_speed;
+       pd->read_speed = read_speed;
+
+       if ((ret = pkt_perform_opc(pd))) {
+               pkt_dbg(1, pd, "Optimum Power Calibration failed\n");
+       }
+
+       return 0;
+}
+
+/*
+ * called at open time.
+ */
+static int pkt_open_dev(struct pktcdvd_device *pd, fmode_t write)
+{
+       int ret;
+       long lba;
+       struct request_queue *q;
+
+       /*
+        * We need to re-open the cdrom device without O_NONBLOCK to be able
+        * to read/write from/to it. It is already opened in O_NONBLOCK mode
+        * so bdget() can't fail.
+        */
+       bdget(pd->bdev->bd_dev);
+       if ((ret = blkdev_get(pd->bdev, FMODE_READ | FMODE_EXCL, pd)))
+               goto out;
+
+       if ((ret = pkt_get_last_written(pd, &lba))) {
+               pkt_err(pd, "pkt_get_last_written failed\n");
+               goto out_putdev;
+       }
+
+       set_capacity(pd->disk, lba << 2);
+       set_capacity(pd->bdev->bd_disk, lba << 2);
+       bd_set_size(pd->bdev, (loff_t)lba << 11);
+
+       q = bdev_get_queue(pd->bdev);
+       if (write) {
+               if ((ret = pkt_open_write(pd)))
+                       goto out_putdev;
+               /*
+                * Some CDRW drives can not handle writes larger than one packet,
+                * even if the size is a multiple of the packet size.
+                */
+               spin_lock_irq(q->queue_lock);
+               blk_queue_max_hw_sectors(q, pd->settings.size);
+               spin_unlock_irq(q->queue_lock);
+               set_bit(PACKET_WRITABLE, &pd->flags);
+       } else {
+               pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
+               clear_bit(PACKET_WRITABLE, &pd->flags);
+       }
+
+       if ((ret = pkt_set_segment_merging(pd, q)))
+               goto out_putdev;
+
+       if (write) {
+               if (!pkt_grow_pktlist(pd, CONFIG_CDROM_PKTCDVD_BUFFERS)) {
+                       pkt_err(pd, "not enough memory for buffers\n");
+                       ret = -ENOMEM;
+                       goto out_putdev;
+               }
+               pkt_info(pd, "%lukB available on disc\n", lba << 1);
+       }
+
+       return 0;
+
+out_putdev:
+       blkdev_put(pd->bdev, FMODE_READ | FMODE_EXCL);
+out:
+       return ret;
+}
+
+/*
+ * called when the device is closed. makes sure that the device flushes
+ * the internal cache before we close.
+ */
+static void pkt_release_dev(struct pktcdvd_device *pd, int flush)
+{
+       if (flush && pkt_flush_cache(pd))
+               pkt_dbg(1, pd, "not flushing cache\n");
+
+       pkt_lock_door(pd, 0);
+
+       pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
+       blkdev_put(pd->bdev, FMODE_READ | FMODE_EXCL);
+
+       pkt_shrink_pktlist(pd);
+}
+
+static struct pktcdvd_device *pkt_find_dev_from_minor(unsigned int dev_minor)
+{
+       if (dev_minor >= MAX_WRITERS)
+               return NULL;
+       return pkt_devs[dev_minor];
+}
+
+static int pkt_open(struct block_device *bdev, fmode_t mode)
+{
+       struct pktcdvd_device *pd = NULL;
+       int ret;
+
+       mutex_lock(&pktcdvd_mutex);
+       mutex_lock(&ctl_mutex);
+       pd = pkt_find_dev_from_minor(MINOR(bdev->bd_dev));
+       if (!pd) {
+               ret = -ENODEV;
+               goto out;
+       }
+       BUG_ON(pd->refcnt < 0);
+
+       pd->refcnt++;
+       if (pd->refcnt > 1) {
+               if ((mode & FMODE_WRITE) &&
+                   !test_bit(PACKET_WRITABLE, &pd->flags)) {
+                       ret = -EBUSY;
+                       goto out_dec;
+               }
+       } else {
+               ret = pkt_open_dev(pd, mode & FMODE_WRITE);
+               if (ret)
+                       goto out_dec;
+               /*
+                * needed here as well, since ext2 (among others) may change
+                * the blocksize at mount time
+                */
+               set_blocksize(bdev, CD_FRAMESIZE);
+       }
+
+       mutex_unlock(&ctl_mutex);
+       mutex_unlock(&pktcdvd_mutex);
+       return 0;
+
+out_dec:
+       pd->refcnt--;
+out:
+       mutex_unlock(&ctl_mutex);
+       mutex_unlock(&pktcdvd_mutex);
+       return ret;
+}
+
+static void pkt_close(struct gendisk *disk, fmode_t mode)
+{
+       struct pktcdvd_device *pd = disk->private_data;
+
+       mutex_lock(&pktcdvd_mutex);
+       mutex_lock(&ctl_mutex);
+       pd->refcnt--;
+       BUG_ON(pd->refcnt < 0);
+       if (pd->refcnt == 0) {
+               int flush = test_bit(PACKET_WRITABLE, &pd->flags);
+               pkt_release_dev(pd, flush);
+       }
+       mutex_unlock(&ctl_mutex);
+       mutex_unlock(&pktcdvd_mutex);
+}
+
+
+static void pkt_end_io_read_cloned(struct bio *bio, int err)
+{
+       struct packet_stacked_data *psd = bio->bi_private;
+       struct pktcdvd_device *pd = psd->pd;
+
+       bio_put(bio);
+       bio_endio(psd->bio, err);
+       mempool_free(psd, psd_pool);
+       pkt_bio_finished(pd);
+}
+
+static void pkt_make_request_read(struct pktcdvd_device *pd, struct bio *bio)
+{
+       struct bio *cloned_bio = bio_clone(bio, GFP_NOIO);
+       struct packet_stacked_data *psd = mempool_alloc(psd_pool, GFP_NOIO);
+
+       psd->pd = pd;
+       psd->bio = bio;
+       cloned_bio->bi_bdev = pd->bdev;
+       cloned_bio->bi_private = psd;
+       cloned_bio->bi_end_io = pkt_end_io_read_cloned;
+       pd->stats.secs_r += bio_sectors(bio);
+       pkt_queue_bio(pd, cloned_bio);
+}
+
+static void pkt_make_request_write(struct request_queue *q, struct bio *bio)
+{
+       struct pktcdvd_device *pd = q->queuedata;
+       sector_t zone;
+       struct packet_data *pkt;
+       int was_empty, blocked_bio;
+       struct pkt_rb_node *node;
+
+       zone = get_zone(bio->bi_iter.bi_sector, pd);
+
+       /*
+        * If we find a matching packet in state WAITING or READ_WAIT, we can
+        * just append this bio to that packet.
+        */
+       spin_lock(&pd->cdrw.active_list_lock);
+       blocked_bio = 0;
+       list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
+               if (pkt->sector == zone) {
+                       spin_lock(&pkt->lock);
+                       if ((pkt->state == PACKET_WAITING_STATE) ||
+                           (pkt->state == PACKET_READ_WAIT_STATE)) {
+                               bio_list_add(&pkt->orig_bios, bio);
+                               pkt->write_size +=
+                                       bio->bi_iter.bi_size / CD_FRAMESIZE;
+                               if ((pkt->write_size >= pkt->frames) &&
+                                   (pkt->state == PACKET_WAITING_STATE)) {
+                                       atomic_inc(&pkt->run_sm);
+                                       wake_up(&pd->wqueue);
+                               }
+                               spin_unlock(&pkt->lock);
+                               spin_unlock(&pd->cdrw.active_list_lock);
+                               return;
+                       } else {
+                               blocked_bio = 1;
+                       }
+                       spin_unlock(&pkt->lock);
+               }
+       }
+       spin_unlock(&pd->cdrw.active_list_lock);
+
+       /*
+        * Test if there is enough room left in the bio work queue
+        * (queue size >= congestion on mark).
+        * If not, wait till the work queue size is below the congestion off mark.
+        */
+       spin_lock(&pd->lock);
+       if (pd->write_congestion_on > 0
+           && pd->bio_queue_size >= pd->write_congestion_on) {
+               set_bdi_congested(&q->backing_dev_info, BLK_RW_ASYNC);
+               do {
+                       spin_unlock(&pd->lock);
+                       congestion_wait(BLK_RW_ASYNC, HZ);
+                       spin_lock(&pd->lock);
+               } while(pd->bio_queue_size > pd->write_congestion_off);
+       }
+       spin_unlock(&pd->lock);
+
+       /*
+        * No matching packet found. Store the bio in the work queue.
+        */
+       node = mempool_alloc(pd->rb_pool, GFP_NOIO);
+       node->bio = bio;
+       spin_lock(&pd->lock);
+       BUG_ON(pd->bio_queue_size < 0);
+       was_empty = (pd->bio_queue_size == 0);
+       pkt_rbtree_insert(pd, node);
+       spin_unlock(&pd->lock);
+
+       /*
+        * Wake up the worker thread.
+        */
+       atomic_set(&pd->scan_queue, 1);
+       if (was_empty) {
+               /* This wake_up is required for correct operation */
+               wake_up(&pd->wqueue);
+       } else if (!list_empty(&pd->cdrw.pkt_free_list) && !blocked_bio) {
+               /*
+                * This wake up is not required for correct operation,
+                * but improves performance in some cases.
+                */
+               wake_up(&pd->wqueue);
+       }
+}
+
+static void pkt_make_request(struct request_queue *q, struct bio *bio)
+{
+       struct pktcdvd_device *pd;
+       char b[BDEVNAME_SIZE];
+       struct bio *split;
+
+       pd = q->queuedata;
+       if (!pd) {
+               pr_err("%s incorrect request queue\n",
+                      bdevname(bio->bi_bdev, b));
+               goto end_io;
+       }
+
+       pkt_dbg(2, pd, "start = %6llx stop = %6llx\n",
+               (unsigned long long)bio->bi_iter.bi_sector,
+               (unsigned long long)bio_end_sector(bio));
+
+       /*
+        * Clone READ bios so we can have our own bi_end_io callback.
+        */
+       if (bio_data_dir(bio) == READ) {
+               pkt_make_request_read(pd, bio);
+               return;
+       }
+
+       if (!test_bit(PACKET_WRITABLE, &pd->flags)) {
+               pkt_notice(pd, "WRITE for ro device (%llu)\n",
+                          (unsigned long long)bio->bi_iter.bi_sector);
+               goto end_io;
+       }
+
+       if (!bio->bi_iter.bi_size || (bio->bi_iter.bi_size % CD_FRAMESIZE)) {
+               pkt_err(pd, "wrong bio size\n");
+               goto end_io;
+       }
+
+       blk_queue_bounce(q, &bio);
+
+       do {
+               sector_t zone = get_zone(bio->bi_iter.bi_sector, pd);
+               sector_t last_zone = get_zone(bio_end_sector(bio) - 1, pd);
+
+               if (last_zone != zone) {
+                       BUG_ON(last_zone != zone + pd->settings.size);
+
+                       split = bio_split(bio, last_zone -
+                                         bio->bi_iter.bi_sector,
+                                         GFP_NOIO, fs_bio_set);
+                       bio_chain(split, bio);
+               } else {
+                       split = bio;
+               }
+
+               pkt_make_request_write(q, split);
+       } while (split != bio);
+
+       return;
+end_io:
+       bio_io_error(bio);
+}
+
+
+
+static int pkt_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
+                         struct bio_vec *bvec)
+{
+       struct pktcdvd_device *pd = q->queuedata;
+       sector_t zone = get_zone(bmd->bi_sector, pd);
+       int used = ((bmd->bi_sector - zone) << 9) + bmd->bi_size;
+       int remaining = (pd->settings.size << 9) - used;
+       int remaining2;
+
+       /*
+        * A bio <= PAGE_SIZE must be allowed. If it crosses a packet
+        * boundary, pkt_make_request() will split the bio.
+        */
+       remaining2 = PAGE_SIZE - bmd->bi_size;
+       remaining = max(remaining, remaining2);
+
+       BUG_ON(remaining < 0);
+       return remaining;
+}
+
+static void pkt_init_queue(struct pktcdvd_device *pd)
+{
+       struct request_queue *q = pd->disk->queue;
+
+       blk_queue_make_request(q, pkt_make_request);
+       blk_queue_logical_block_size(q, CD_FRAMESIZE);
+       blk_queue_max_hw_sectors(q, PACKET_MAX_SECTORS);
+       blk_queue_merge_bvec(q, pkt_merge_bvec);
+       q->queuedata = pd;
+}
+
+static int pkt_seq_show(struct seq_file *m, void *p)
+{
+       struct pktcdvd_device *pd = m->private;
+       char *msg;
+       char bdev_buf[BDEVNAME_SIZE];
+       int states[PACKET_NUM_STATES];
+
+       seq_printf(m, "Writer %s mapped to %s:\n", pd->name,
+                  bdevname(pd->bdev, bdev_buf));
+
+       seq_printf(m, "\nSettings:\n");
+       seq_printf(m, "\tpacket size:\t\t%dkB\n", pd->settings.size / 2);
+
+       if (pd->settings.write_type == 0)
+               msg = "Packet";
+       else
+               msg = "Unknown";
+       seq_printf(m, "\twrite type:\t\t%s\n", msg);
+
+       seq_printf(m, "\tpacket type:\t\t%s\n", pd->settings.fp ? "Fixed" : "Variable");
+       seq_printf(m, "\tlink loss:\t\t%d\n", pd->settings.link_loss);
+
+       seq_printf(m, "\ttrack mode:\t\t%d\n", pd->settings.track_mode);
+
+       if (pd->settings.block_mode == PACKET_BLOCK_MODE1)
+               msg = "Mode 1";
+       else if (pd->settings.block_mode == PACKET_BLOCK_MODE2)
+               msg = "Mode 2";
+       else
+               msg = "Unknown";
+       seq_printf(m, "\tblock mode:\t\t%s\n", msg);
+
+       seq_printf(m, "\nStatistics:\n");
+       seq_printf(m, "\tpackets started:\t%lu\n", pd->stats.pkt_started);
+       seq_printf(m, "\tpackets ended:\t\t%lu\n", pd->stats.pkt_ended);
+       seq_printf(m, "\twritten:\t\t%lukB\n", pd->stats.secs_w >> 1);
+       seq_printf(m, "\tread gather:\t\t%lukB\n", pd->stats.secs_rg >> 1);
+       seq_printf(m, "\tread:\t\t\t%lukB\n", pd->stats.secs_r >> 1);
+
+       seq_printf(m, "\nMisc:\n");
+       seq_printf(m, "\treference count:\t%d\n", pd->refcnt);
+       seq_printf(m, "\tflags:\t\t\t0x%lx\n", pd->flags);
+       seq_printf(m, "\tread speed:\t\t%ukB/s\n", pd->read_speed);
+       seq_printf(m, "\twrite speed:\t\t%ukB/s\n", pd->write_speed);
+       seq_printf(m, "\tstart offset:\t\t%lu\n", pd->offset);
+       seq_printf(m, "\tmode page offset:\t%u\n", pd->mode_offset);
+
+       seq_printf(m, "\nQueue state:\n");
+       seq_printf(m, "\tbios queued:\t\t%d\n", pd->bio_queue_size);
+       seq_printf(m, "\tbios pending:\t\t%d\n", atomic_read(&pd->cdrw.pending_bios));
+       seq_printf(m, "\tcurrent sector:\t\t0x%llx\n", (unsigned long long)pd->current_sector);
+
+       pkt_count_states(pd, states);
+       seq_printf(m, "\tstate:\t\t\ti:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
+                  states[0], states[1], states[2], states[3], states[4], states[5]);
+
+       seq_printf(m, "\twrite congestion marks:\toff=%d on=%d\n",
+                       pd->write_congestion_off,
+                       pd->write_congestion_on);
+       return 0;
+}
+
+static int pkt_seq_open(struct inode *inode, struct file *file)
+{
+       return single_open(file, pkt_seq_show, PDE_DATA(inode));
+}
+
+static const struct file_operations pkt_proc_fops = {
+       .open   = pkt_seq_open,
+       .read   = seq_read,
+       .llseek = seq_lseek,
+       .release = single_release
+};
+
+static int pkt_new_dev(struct pktcdvd_device *pd, dev_t dev)
+{
+       int i;
+       int ret = 0;
+       char b[BDEVNAME_SIZE];
+       struct block_device *bdev;
+
+       if (pd->pkt_dev == dev) {
+               pkt_err(pd, "recursive setup not allowed\n");
+               return -EBUSY;
+       }
+       for (i = 0; i < MAX_WRITERS; i++) {
+               struct pktcdvd_device *pd2 = pkt_devs[i];
+               if (!pd2)
+                       continue;
+               if (pd2->bdev->bd_dev == dev) {
+                       pkt_err(pd, "%s already setup\n",
+                               bdevname(pd2->bdev, b));
+                       return -EBUSY;
+               }
+               if (pd2->pkt_dev == dev) {
+                       pkt_err(pd, "can't chain pktcdvd devices\n");
+                       return -EBUSY;
+               }
+       }
+
+       bdev = bdget(dev);
+       if (!bdev)
+               return -ENOMEM;
+       ret = blkdev_get(bdev, FMODE_READ | FMODE_NDELAY, NULL);
+       if (ret)
+               return ret;
+
+       /* This is safe, since we have a reference from open(). */
+       __module_get(THIS_MODULE);
+
+       pd->bdev = bdev;
+       set_blocksize(bdev, CD_FRAMESIZE);
+
+       pkt_init_queue(pd);
+
+       atomic_set(&pd->cdrw.pending_bios, 0);
+       pd->cdrw.thread = kthread_run(kcdrwd, pd, "%s", pd->name);
+       if (IS_ERR(pd->cdrw.thread)) {
+               pkt_err(pd, "can't start kernel thread\n");
+               ret = -ENOMEM;
+               goto out_mem;
+       }
+
+       proc_create_data(pd->name, 0, pkt_proc, &pkt_proc_fops, pd);
+       pkt_dbg(1, pd, "writer mapped to %s\n", bdevname(bdev, b));
+       return 0;
+
+out_mem:
+       blkdev_put(bdev, FMODE_READ | FMODE_NDELAY);
+       /* This is safe: open() is still holding a reference. */
+       module_put(THIS_MODULE);
+       return ret;
+}
+
+static int pkt_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, unsigned long arg)
+{
+       struct pktcdvd_device *pd = bdev->bd_disk->private_data;
+       int ret;
+
+       pkt_dbg(2, pd, "cmd %x, dev %d:%d\n",
+               cmd, MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev));
+
+       mutex_lock(&pktcdvd_mutex);
+       switch (cmd) {
+       case CDROMEJECT:
+               /*
+                * The door gets locked when the device is opened, so we
+                * have to unlock it or else the eject command fails.
+                */
+               if (pd->refcnt == 1)
+                       pkt_lock_door(pd, 0);
+               /* fallthru */
+       /*
+        * forward selected CDROM ioctls to CD-ROM, for UDF
+        */
+       case CDROMMULTISESSION:
+       case CDROMREADTOCENTRY:
+       case CDROM_LAST_WRITTEN:
+       case CDROM_SEND_PACKET:
+       case SCSI_IOCTL_SEND_COMMAND:
+               ret = __blkdev_driver_ioctl(pd->bdev, mode, cmd, arg);
+               break;
+
+       default:
+               pkt_dbg(2, pd, "Unknown ioctl (%x)\n", cmd);
+               ret = -ENOTTY;
+       }
+       mutex_unlock(&pktcdvd_mutex);
+
+       return ret;
+}
+
+static unsigned int pkt_check_events(struct gendisk *disk,
+                                    unsigned int clearing)
+{
+       struct pktcdvd_device *pd = disk->private_data;
+       struct gendisk *attached_disk;
+
+       if (!pd)
+               return 0;
+       if (!pd->bdev)
+               return 0;
+       attached_disk = pd->bdev->bd_disk;
+       if (!attached_disk || !attached_disk->fops->check_events)
+               return 0;
+       return attached_disk->fops->check_events(attached_disk, clearing);
+}
+
+static const struct block_device_operations pktcdvd_ops = {
+       .owner =                THIS_MODULE,
+       .open =                 pkt_open,
+       .release =              pkt_close,
+       .ioctl =                pkt_ioctl,
+       .check_events =         pkt_check_events,
+};
+
+static char *pktcdvd_devnode(struct gendisk *gd, umode_t *mode)
+{
+       return kasprintf(GFP_KERNEL, "pktcdvd/%s", gd->disk_name);
+}
+
+/*
+ * Set up mapping from pktcdvd device to CD-ROM device.
+ */
+static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev)
+{
+       int idx;
+       int ret = -ENOMEM;
+       struct pktcdvd_device *pd;
+       struct gendisk *disk;
+
+       mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
+
+       for (idx = 0; idx < MAX_WRITERS; idx++)
+               if (!pkt_devs[idx])
+                       break;
+       if (idx == MAX_WRITERS) {
+               pr_err("max %d writers supported\n", MAX_WRITERS);
+               ret = -EBUSY;
+               goto out_mutex;
+       }
+
+       pd = kzalloc(sizeof(struct pktcdvd_device), GFP_KERNEL);
+       if (!pd)
+               goto out_mutex;
+
+       pd->rb_pool = mempool_create_kmalloc_pool(PKT_RB_POOL_SIZE,
+                                                 sizeof(struct pkt_rb_node));
+       if (!pd->rb_pool)
+               goto out_mem;
+
+       INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
+       INIT_LIST_HEAD(&pd->cdrw.pkt_active_list);
+       spin_lock_init(&pd->cdrw.active_list_lock);
+
+       spin_lock_init(&pd->lock);
+       spin_lock_init(&pd->iosched.lock);
+       bio_list_init(&pd->iosched.read_queue);
+       bio_list_init(&pd->iosched.write_queue);
+       sprintf(pd->name, DRIVER_NAME"%d", idx);
+       init_waitqueue_head(&pd->wqueue);
+       pd->bio_queue = RB_ROOT;
+
+       pd->write_congestion_on  = write_congestion_on;
+       pd->write_congestion_off = write_congestion_off;
+
+       disk = alloc_disk(1);
+       if (!disk)
+               goto out_mem;
+       pd->disk = disk;
+       disk->major = pktdev_major;
+       disk->first_minor = idx;
+       disk->fops = &pktcdvd_ops;
+       disk->flags = GENHD_FL_REMOVABLE;
+       strcpy(disk->disk_name, pd->name);
+       disk->devnode = pktcdvd_devnode;
+       disk->private_data = pd;
+       disk->queue = blk_alloc_queue(GFP_KERNEL);
+       if (!disk->queue)
+               goto out_mem2;
+
+       pd->pkt_dev = MKDEV(pktdev_major, idx);
+       ret = pkt_new_dev(pd, dev);
+       if (ret)
+               goto out_new_dev;
+
+       /* inherit events of the host device */
+       disk->events = pd->bdev->bd_disk->events;
+       disk->async_events = pd->bdev->bd_disk->async_events;
+
+       add_disk(disk);
+
+       pkt_sysfs_dev_new(pd);
+       pkt_debugfs_dev_new(pd);
+
+       pkt_devs[idx] = pd;
+       if (pkt_dev)
+               *pkt_dev = pd->pkt_dev;
+
+       mutex_unlock(&ctl_mutex);
+       return 0;
+
+out_new_dev:
+       blk_cleanup_queue(disk->queue);
+out_mem2:
+       put_disk(disk);
+out_mem:
+       if (pd->rb_pool)
+               mempool_destroy(pd->rb_pool);
+       kfree(pd);
+out_mutex:
+       mutex_unlock(&ctl_mutex);
+       pr_err("setup of pktcdvd device failed\n");
+       return ret;
+}
+
+/*
+ * Tear down mapping from pktcdvd device to CD-ROM device.
+ */
+static int pkt_remove_dev(dev_t pkt_dev)
+{
+       struct pktcdvd_device *pd;
+       int idx;
+       int ret = 0;
+
+       mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
+
+       for (idx = 0; idx < MAX_WRITERS; idx++) {
+               pd = pkt_devs[idx];
+               if (pd && (pd->pkt_dev == pkt_dev))
+                       break;
+       }
+       if (idx == MAX_WRITERS) {
+               pr_debug("dev not setup\n");
+               ret = -ENXIO;
+               goto out;
+       }
+
+       if (pd->refcnt > 0) {
+               ret = -EBUSY;
+               goto out;
+       }
+       if (!IS_ERR(pd->cdrw.thread))
+               kthread_stop(pd->cdrw.thread);
+
+       pkt_devs[idx] = NULL;
+
+       pkt_debugfs_dev_remove(pd);
+       pkt_sysfs_dev_remove(pd);
+
+       blkdev_put(pd->bdev, FMODE_READ | FMODE_NDELAY);
+
+       remove_proc_entry(pd->name, pkt_proc);
+       pkt_dbg(1, pd, "writer unmapped\n");
+
+       del_gendisk(pd->disk);
+       blk_cleanup_queue(pd->disk->queue);
+       put_disk(pd->disk);
+
+       mempool_destroy(pd->rb_pool);
+       kfree(pd);
+
+       /* This is safe: open() is still holding a reference. */
+       module_put(THIS_MODULE);
+
+out:
+       mutex_unlock(&ctl_mutex);
+       return ret;
+}
+
+static void pkt_get_status(struct pkt_ctrl_command *ctrl_cmd)
+{
+       struct pktcdvd_device *pd;
+
+       mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
+
+       pd = pkt_find_dev_from_minor(ctrl_cmd->dev_index);
+       if (pd) {
+               ctrl_cmd->dev = new_encode_dev(pd->bdev->bd_dev);
+               ctrl_cmd->pkt_dev = new_encode_dev(pd->pkt_dev);
+       } else {
+               ctrl_cmd->dev = 0;
+               ctrl_cmd->pkt_dev = 0;
+       }
+       ctrl_cmd->num_devices = MAX_WRITERS;
+
+       mutex_unlock(&ctl_mutex);
+}
+
+static long pkt_ctl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+       void __user *argp = (void __user *)arg;
+       struct pkt_ctrl_command ctrl_cmd;
+       int ret = 0;
+       dev_t pkt_dev = 0;
+
+       if (cmd != PACKET_CTRL_CMD)
+               return -ENOTTY;
+
+       if (copy_from_user(&ctrl_cmd, argp, sizeof(struct pkt_ctrl_command)))
+               return -EFAULT;
+
+       switch (ctrl_cmd.command) {
+       case PKT_CTRL_CMD_SETUP:
+               if (!capable(CAP_SYS_ADMIN))
+                       return -EPERM;
+               ret = pkt_setup_dev(new_decode_dev(ctrl_cmd.dev), &pkt_dev);
+               ctrl_cmd.pkt_dev = new_encode_dev(pkt_dev);
+               break;
+       case PKT_CTRL_CMD_TEARDOWN:
+               if (!capable(CAP_SYS_ADMIN))
+                       return -EPERM;
+               ret = pkt_remove_dev(new_decode_dev(ctrl_cmd.pkt_dev));
+               break;
+       case PKT_CTRL_CMD_STATUS:
+               pkt_get_status(&ctrl_cmd);
+               break;
+       default:
+               return -ENOTTY;
+       }
+
+       if (copy_to_user(argp, &ctrl_cmd, sizeof(struct pkt_ctrl_command)))
+               return -EFAULT;
+       return ret;
+}
+
+#ifdef CONFIG_COMPAT
+static long pkt_ctl_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
+{
+       return pkt_ctl_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
+}
+#endif
+
+static const struct file_operations pkt_ctl_fops = {
+       .open           = nonseekable_open,
+       .unlocked_ioctl = pkt_ctl_ioctl,
+#ifdef CONFIG_COMPAT
+       .compat_ioctl   = pkt_ctl_compat_ioctl,
+#endif
+       .owner          = THIS_MODULE,
+       .llseek         = no_llseek,
+};
+
+static struct miscdevice pkt_misc = {
+       .minor          = MISC_DYNAMIC_MINOR,
+       .name           = DRIVER_NAME,
+       .nodename       = "pktcdvd/control",
+       .fops           = &pkt_ctl_fops
+};
+
+static int __init pkt_init(void)
+{
+       int ret;
+
+       mutex_init(&ctl_mutex);
+
+       psd_pool = mempool_create_kmalloc_pool(PSD_POOL_SIZE,
+                                       sizeof(struct packet_stacked_data));
+       if (!psd_pool)
+               return -ENOMEM;
+
+       ret = register_blkdev(pktdev_major, DRIVER_NAME);
+       if (ret < 0) {
+               pr_err("unable to register block device\n");
+               goto out2;
+       }
+       if (!pktdev_major)
+               pktdev_major = ret;
+
+       ret = pkt_sysfs_init();
+       if (ret)
+               goto out;
+
+       pkt_debugfs_init();
+
+       ret = misc_register(&pkt_misc);
+       if (ret) {
+               pr_err("unable to register misc device\n");
+               goto out_misc;
+       }
+
+       pkt_proc = proc_mkdir("driver/"DRIVER_NAME, NULL);
+
+       return 0;
+
+out_misc:
+       pkt_debugfs_cleanup();
+       pkt_sysfs_cleanup();
+out:
+       unregister_blkdev(pktdev_major, DRIVER_NAME);
+out2:
+       mempool_destroy(psd_pool);
+       return ret;
+}
+
+static void __exit pkt_exit(void)
+{
+       remove_proc_entry("driver/"DRIVER_NAME, NULL);
+       misc_deregister(&pkt_misc);
+
+       pkt_debugfs_cleanup();
+       pkt_sysfs_cleanup();
+
+       unregister_blkdev(pktdev_major, DRIVER_NAME);
+       mempool_destroy(psd_pool);
+}
+
+MODULE_DESCRIPTION("Packet writing layer for CD/DVD drives");
+MODULE_AUTHOR("Jens Axboe <axboe@suse.de>");
+MODULE_LICENSE("GPL");
+
+module_init(pkt_init);
+module_exit(pkt_exit);