config: tegra3: enable /dev mount with ACL
[linux-2.6.git] / block / bsg.c
index c85d961..702f131 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * bsg.c - block layer implementation of the sg v3 interface
+ * bsg.c - block layer implementation of the sg v4 interface
  *
  * Copyright (C) 2004 Jens Axboe <axboe@suse.de> SUSE Labs
  * Copyright (C) 2004 Peter M. Jones <pjones@redhat.com>
@@ -9,79 +9,68 @@
  *  archive for more details.
  *
  */
-/*
- * TODO
- *     - Should this get merged, block/scsi_ioctl.c will be migrated into
- *       this file. To keep maintenance down, it's easier to have them
- *       seperated right now.
- *
- */
 #include <linux/module.h>
 #include <linux/init.h>
 #include <linux/file.h>
 #include <linux/blkdev.h>
 #include <linux/poll.h>
 #include <linux/cdev.h>
+#include <linux/jiffies.h>
 #include <linux/percpu.h>
 #include <linux/uio.h>
+#include <linux/idr.h>
 #include <linux/bsg.h>
+#include <linux/slab.h>
 
 #include <scsi/scsi.h>
 #include <scsi/scsi_ioctl.h>
 #include <scsi/scsi_cmnd.h>
+#include <scsi/scsi_device.h>
+#include <scsi/scsi_driver.h>
 #include <scsi/sg.h>
 
-static char bsg_version[] = "block layer sg (bsg) 0.4";
+#define BSG_DESCRIPTION        "Block layer SCSI generic (bsg) driver"
+#define BSG_VERSION    "0.4"
 
 struct bsg_device {
-       struct gendisk *disk;
-       request_queue_t *queue;
+       struct request_queue *queue;
        spinlock_t lock;
        struct list_head busy_list;
        struct list_head done_list;
        struct hlist_node dev_list;
        atomic_t ref_count;
-       int minor;
        int queued_cmds;
        int done_cmds;
        wait_queue_head_t wq_done;
        wait_queue_head_t wq_free;
-       char name[BDEVNAME_SIZE];
+       char name[20];
        int max_queue;
        unsigned long flags;
 };
 
 enum {
        BSG_F_BLOCK             = 1,
-       BSG_F_WRITE_PERM        = 2,
 };
 
 #define BSG_DEFAULT_CMDS       64
+#define BSG_MAX_DEVS           32768
 
 #undef BSG_DEBUG
 
 #ifdef BSG_DEBUG
-#define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##args)
+#define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __func__, ##args)
 #else
 #define dprintk(fmt, args...)
 #endif
 
-#define list_entry_bc(entry)   list_entry((entry), struct bsg_command, list)
-
-/*
- * just for testing
- */
-#define BSG_MAJOR      (240)
-
 static DEFINE_MUTEX(bsg_mutex);
-static int bsg_device_nr;
+static DEFINE_IDR(bsg_minor_idr);
 
-#define BSG_LIST_SIZE  (8)
-#define bsg_list_idx(minor)    ((minor) & (BSG_LIST_SIZE - 1))
-static struct hlist_head bsg_device_list[BSG_LIST_SIZE];
+#define BSG_LIST_ARRAY_SIZE    8
+static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
 
 static struct class *bsg_class;
-static LIST_HEAD(bsg_class_list);
+static int bsg_major;
 
 static struct kmem_cache *bsg_cmd_cachep;
 
@@ -93,9 +82,9 @@ struct bsg_command {
        struct list_head list;
        struct request *rq;
        struct bio *bio;
+       struct bio *bidi_bio;
        int err;
        struct sg_io_v4 hdr;
-       struct sg_io_v4 __user *uhdr;
        char sense[SCSI_SENSE_BUFFERSIZE];
 };
 
@@ -113,9 +102,9 @@ static void bsg_free_command(struct bsg_command *bc)
        wake_up(&bd->wq_free);
 }
 
-static struct bsg_command *__bsg_alloc_command(struct bsg_device *bd)
+static struct bsg_command *bsg_alloc_command(struct bsg_device *bd)
 {
-       struct bsg_command *bc = NULL;
+       struct bsg_command *bc = ERR_PTR(-EINVAL);
 
        spin_lock_irq(&bd->lock);
 
@@ -125,14 +114,14 @@ static struct bsg_command *__bsg_alloc_command(struct bsg_device *bd)
        bd->queued_cmds++;
        spin_unlock_irq(&bd->lock);
 
-       bc = kmem_cache_alloc(bsg_cmd_cachep, GFP_USER);
+       bc = kmem_cache_zalloc(bsg_cmd_cachep, GFP_KERNEL);
        if (unlikely(!bc)) {
                spin_lock_irq(&bd->lock);
                bd->queued_cmds--;
+               bc = ERR_PTR(-ENOMEM);
                goto out;
        }
 
-       memset(bc, 0, sizeof(*bc));
        bc->bd = bd;
        INIT_LIST_HEAD(&bc->list);
        dprintk("%s: returning free cmd %p\n", bd->name, bc);
@@ -142,22 +131,12 @@ out:
        return bc;
 }
 
-static inline void
-bsg_del_done_cmd(struct bsg_device *bd, struct bsg_command *bc)
-{
-       bd->done_cmds--;
-       list_del(&bc->list);
-}
-
-static inline void
-bsg_add_done_cmd(struct bsg_device *bd, struct bsg_command *bc)
+static inline struct hlist_head *bsg_dev_idx_hash(int index)
 {
-       bd->done_cmds++;
-       list_add_tail(&bc->list, &bd->done_list);
-       wake_up(&bd->wq_done);
+       return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
 }
 
-static inline int bsg_io_schedule(struct bsg_device *bd, int state)
+static int bsg_io_schedule(struct bsg_device *bd)
 {
        DEFINE_WAIT(wait);
        int ret = 0;
@@ -182,53 +161,35 @@ static inline int bsg_io_schedule(struct bsg_device *bd, int state)
                goto unlock;
        }
 
-       prepare_to_wait(&bd->wq_done, &wait, state);
+       prepare_to_wait(&bd->wq_done, &wait, TASK_UNINTERRUPTIBLE);
        spin_unlock_irq(&bd->lock);
        io_schedule();
        finish_wait(&bd->wq_done, &wait);
 
-       if ((state == TASK_INTERRUPTIBLE) && signal_pending(current))
-               ret = -ERESTARTSYS;
-
        return ret;
 unlock:
        spin_unlock_irq(&bd->lock);
        return ret;
 }
 
-/*
- * get a new free command, blocking if needed and specified
- */
-static struct bsg_command *bsg_get_command(struct bsg_device *bd)
+static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
+                               struct sg_io_v4 *hdr, struct bsg_device *bd,
+                               fmode_t has_write_perm)
 {
-       struct bsg_command *bc;
-       int ret;
-
-       do {
-               bc = __bsg_alloc_command(bd);
-               if (bc)
-                       break;
-
-               ret = bsg_io_schedule(bd, TASK_INTERRUPTIBLE);
-               if (ret) {
-                       bc = ERR_PTR(ret);
-                       break;
-               }
-
-       } while (1);
-
-       return bc;
-}
-
-static int blk_fill_sgv4_hdr_rq(request_queue_t *q, struct request *rq,
-                               struct sg_io_v4 *hdr, int has_write_perm)
-{
-       memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
+       if (hdr->request_len > BLK_MAX_CDB) {
+               rq->cmd = kzalloc(hdr->request_len, GFP_KERNEL);
+               if (!rq->cmd)
+                       return -ENOMEM;
+       }
 
-       if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
+       if (copy_from_user(rq->cmd, (void __user *)(unsigned long)hdr->request,
                           hdr->request_len))
                return -EFAULT;
-       if (blk_verify_command(rq->cmd, has_write_perm))
+
+       if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
+               if (blk_verify_command(rq->cmd, has_write_perm))
+                       return -EPERM;
+       } else if (!capable(CAP_SYS_RAWIO))
                return -EPERM;
 
        /*
@@ -237,11 +198,13 @@ static int blk_fill_sgv4_hdr_rq(request_queue_t *q, struct request *rq,
        rq->cmd_len = hdr->request_len;
        rq->cmd_type = REQ_TYPE_BLOCK_PC;
 
-       rq->timeout = (hdr->timeout * HZ) / 1000;
+       rq->timeout = msecs_to_jiffies(hdr->timeout);
        if (!rq->timeout)
                rq->timeout = q->sg_timeout;
        if (!rq->timeout)
                rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
+       if (rq->timeout < BLK_MIN_SG_TIMEOUT)
+               rq->timeout = BLK_MIN_SG_TIMEOUT;
 
        return 0;
 }
@@ -250,46 +213,51 @@ static int blk_fill_sgv4_hdr_rq(request_queue_t *q, struct request *rq,
  * Check if sg_io_v4 from user is allowed and valid
  */
 static int
-bsg_validate_sgv4_hdr(request_queue_t *q, struct sg_io_v4 *hdr, int *rw)
+bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
 {
-       if (hdr->guard != 'Q')
-               return -EINVAL;
-       if (hdr->request_len > BLK_MAX_CDB)
-               return -EINVAL;
-       if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
-           hdr->din_xfer_len > (q->max_sectors << 9))
-               return -EIO;
+       int ret = 0;
 
-       /* not supported currently */
-       if (hdr->protocol || hdr->subprotocol)
+       if (hdr->guard != 'Q')
                return -EINVAL;
 
-       /*
-        * looks sane, if no data then it should be fine from our POV
-        */
-       if (!hdr->dout_xfer_len && !hdr->din_xfer_len)
-               return 0;
-
-       /* not supported currently */
-       if (hdr->dout_xfer_len && hdr->din_xfer_len)
-               return -EINVAL;
+       switch (hdr->protocol) {
+       case BSG_PROTOCOL_SCSI:
+               switch (hdr->subprotocol) {
+               case BSG_SUB_PROTOCOL_SCSI_CMD:
+               case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
+                       break;
+               default:
+                       ret = -EINVAL;
+               }
+               break;
+       default:
+               ret = -EINVAL;
+       }
 
        *rw = hdr->dout_xfer_len ? WRITE : READ;
-
-       return 0;
+       return ret;
 }
 
 /*
  * map sg_io_v4 to a request.
  */
 static struct request *
-bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
+bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr, fmode_t has_write_perm,
+           u8 *sense)
 {
-       request_queue_t *q = bd->queue;
-       struct request *rq;
-       int ret, rw = 0; /* shut up gcc */
+       struct request_queue *q = bd->queue;
+       struct request *rq, *next_rq = NULL;
+       int ret, rw;
        unsigned int dxfer_len;
-       void *dxferp = NULL;
+       void __user *dxferp = NULL;
+       struct bsg_class_device *bcd = &q->bsg_dev;
+
+       /* if the LLD has been removed then the bsg_unregister_queue will
+        * eventually be called and the class_dev was freed, so we can no
+        * longer use this request_queue. Return no such address.
+        */
+       if (!bcd->class_dev)
+               return ERR_PTR(-ENXIO);
 
        dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
                hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
@@ -300,35 +268,65 @@ bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
                return ERR_PTR(ret);
 
        /*
-        * map scatter-gather elements seperately and string them to request
+        * map scatter-gather elements separately and string them to request
         */
        rq = blk_get_request(q, rw, GFP_KERNEL);
-       ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM,
-                                                      &bd->flags));
-       if (ret) {
-               blk_put_request(rq);
-               return ERR_PTR(ret);
+       if (!rq)
+               return ERR_PTR(-ENOMEM);
+       ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, bd, has_write_perm);
+       if (ret)
+               goto out;
+
+       if (rw == WRITE && hdr->din_xfer_len) {
+               if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
+                       ret = -EOPNOTSUPP;
+                       goto out;
+               }
+
+               next_rq = blk_get_request(q, READ, GFP_KERNEL);
+               if (!next_rq) {
+                       ret = -ENOMEM;
+                       goto out;
+               }
+               rq->next_rq = next_rq;
+               next_rq->cmd_type = rq->cmd_type;
+
+               dxferp = (void __user *)(unsigned long)hdr->din_xferp;
+               ret =  blk_rq_map_user(q, next_rq, NULL, dxferp,
+                                      hdr->din_xfer_len, GFP_KERNEL);
+               if (ret)
+                       goto out;
        }
 
        if (hdr->dout_xfer_len) {
                dxfer_len = hdr->dout_xfer_len;
-               dxferp = (void*)(unsigned long)hdr->dout_xferp;
+               dxferp = (void __user *)(unsigned long)hdr->dout_xferp;
        } else if (hdr->din_xfer_len) {
                dxfer_len = hdr->din_xfer_len;
-               dxferp = (void*)(unsigned long)hdr->din_xferp;
+               dxferp = (void __user *)(unsigned long)hdr->din_xferp;
        } else
                dxfer_len = 0;
 
        if (dxfer_len) {
-               ret = blk_rq_map_user(q, rq, dxferp, dxfer_len);
-               if (ret) {
-                       dprintk("failed map at %d\n", ret);
-                       blk_put_request(rq);
-                       rq = ERR_PTR(ret);
-               }
+               ret = blk_rq_map_user(q, rq, NULL, dxferp, dxfer_len,
+                                     GFP_KERNEL);
+               if (ret)
+                       goto out;
        }
 
+       rq->sense = sense;
+       rq->sense_len = 0;
+
        return rq;
+out:
+       if (rq->cmd != rq->__cmd)
+               kfree(rq->cmd);
+       blk_put_request(rq);
+       if (next_rq) {
+               blk_rq_unmap_user(next_rq->bio);
+               blk_put_request(next_rq);
+       }
+       return ERR_PTR(ret);
 }
 
 /*
@@ -347,26 +345,29 @@ static void bsg_rq_end_io(struct request *rq, int uptodate)
        bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
 
        spin_lock_irqsave(&bd->lock, flags);
-       list_del(&bc->list);
-       bsg_add_done_cmd(bd, bc);
+       list_move_tail(&bc->list, &bd->done_list);
+       bd->done_cmds++;
        spin_unlock_irqrestore(&bd->lock, flags);
+
+       wake_up(&bd->wq_done);
 }
 
 /*
  * do final setup of a 'bc' and submit the matching 'rq' to the block
  * layer for io
  */
-static void bsg_add_command(struct bsg_device *bd, request_queue_t *q,
+static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
                            struct bsg_command *bc, struct request *rq)
 {
-       rq->sense = bc->sense;
-       rq->sense_len = 0;
+       int at_head = (0 == (bc->hdr.flags & BSG_FLAG_Q_AT_TAIL));
 
        /*
         * add bc command to busy queue and submit rq for io
         */
        bc->rq = rq;
        bc->bio = rq->bio;
+       if (rq->next_rq)
+               bc->bidi_bio = rq->next_rq->bio;
        bc->hdr.duration = jiffies;
        spin_lock_irq(&bd->lock);
        list_add_tail(&bc->list, &bd->busy_list);
@@ -375,17 +376,18 @@ static void bsg_add_command(struct bsg_device *bd, request_queue_t *q,
        dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
 
        rq->end_io_data = bc;
-       blk_execute_rq_nowait(q, bd->disk, rq, 1, bsg_rq_end_io);
+       blk_execute_rq_nowait(q, NULL, rq, at_head, bsg_rq_end_io);
 }
 
-static inline struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
+static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 {
        struct bsg_command *bc = NULL;
 
        spin_lock_irq(&bd->lock);
        if (bd->done_cmds) {
-               bc = list_entry_bc(bd->done_list.next);
-               bsg_del_done_cmd(bd, bc);
+               bc = list_first_entry(&bd->done_list, struct bsg_command, list);
+               list_del(&bc->list);
+               bd->done_cmds--;
        }
        spin_unlock_irq(&bd->lock);
 
@@ -395,7 +397,7 @@ static inline struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
 /*
  * Get a finished command from the done list
  */
-static struct bsg_command *__bsg_get_done_cmd(struct bsg_device *bd, int state)
+static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
 {
        struct bsg_command *bc;
        int ret;
@@ -405,9 +407,14 @@ static struct bsg_command *__bsg_get_done_cmd(struct bsg_device *bd, int state)
                if (bc)
                        break;
 
-               ret = bsg_io_schedule(bd, state);
+               if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
+                       bc = ERR_PTR(-EAGAIN);
+                       break;
+               }
+
+               ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
                if (ret) {
-                       bc = ERR_PTR(ret);
+                       bc = ERR_PTR(-ERESTARTSYS);
                        break;
                }
        } while (1);
@@ -417,41 +424,28 @@ static struct bsg_command *__bsg_get_done_cmd(struct bsg_device *bd, int state)
        return bc;
 }
 
-static struct bsg_command *
-bsg_get_done_cmd(struct bsg_device *bd, const struct iovec *iov)
-{
-       return __bsg_get_done_cmd(bd, TASK_INTERRUPTIBLE);
-}
-
-static struct bsg_command *
-bsg_get_done_cmd_nosignals(struct bsg_device *bd)
-{
-       return __bsg_get_done_cmd(bd, TASK_UNINTERRUPTIBLE);
-}
-
 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
-                                   struct bio *bio)
+                                   struct bio *bio, struct bio *bidi_bio)
 {
        int ret = 0;
 
-       dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
+       dprintk("rq %p bio %p 0x%x\n", rq, bio, rq->errors);
        /*
         * fill in all the output members
         */
-       hdr->device_status = status_byte(rq->errors);
+       hdr->device_status = rq->errors & 0xff;
        hdr->transport_status = host_byte(rq->errors);
        hdr->driver_status = driver_byte(rq->errors);
        hdr->info = 0;
        if (hdr->device_status || hdr->transport_status || hdr->driver_status)
                hdr->info |= SG_INFO_CHECK;
-       hdr->din_resid = rq->data_len;
        hdr->response_len = 0;
 
        if (rq->sense_len && hdr->response) {
-               int len = min((unsigned int) hdr->max_response_len,
-                             rq->sense_len);
+               int len = min_t(unsigned int, hdr->max_response_len,
+                                       rq->sense_len);
 
-               ret = copy_to_user((void*)(unsigned long)hdr->response,
+               ret = copy_to_user((void __user *)(unsigned long)hdr->response,
                                   rq->sense, len);
                if (!ret)
                        hdr->response_len = len;
@@ -459,7 +453,28 @@ static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
                        ret = -EFAULT;
        }
 
+       if (rq->next_rq) {
+               hdr->dout_resid = rq->resid_len;
+               hdr->din_resid = rq->next_rq->resid_len;
+               blk_rq_unmap_user(bidi_bio);
+               blk_put_request(rq->next_rq);
+       } else if (rq_data_dir(rq) == READ)
+               hdr->din_resid = rq->resid_len;
+       else
+               hdr->dout_resid = rq->resid_len;
+
+       /*
+        * If the request generated a negative error number, return it
+        * (providing we aren't already returning an error); if it's
+        * just a protocol response (i.e. non negative), that gets
+        * processed above.
+        */
+       if (!ret && rq->errors < 0)
+               ret = rq->errors;
+
        blk_rq_unmap_user(bio);
+       if (rq->cmd != rq->__cmd)
+               kfree(rq->cmd);
        blk_put_request(rq);
 
        return ret;
@@ -472,14 +487,12 @@ static int bsg_complete_all_commands(struct bsg_device *bd)
 
        dprintk("%s: entered\n", bd->name);
 
-       set_bit(BSG_F_BLOCK, &bd->flags);
-
        /*
         * wait for all commands to complete
         */
        ret = 0;
        do {
-               ret = bsg_io_schedule(bd, TASK_UNINTERRUPTIBLE);
+               ret = bsg_io_schedule(bd);
                /*
                 * look for -ENODATA specifically -- we'll sometimes get
                 * -ERESTARTSYS when we've taken a signal, but we can't
@@ -494,20 +507,19 @@ static int bsg_complete_all_commands(struct bsg_device *bd)
         */
        ret = 0;
        do {
-               bc = bsg_get_done_cmd_nosignals(bd);
-
-               /*
-                * we _must_ complete before restarting, because
-                * bsg_release can't handle this failing.
-                */
-               if (PTR_ERR(bc) == -ERESTARTSYS)
-                       continue;
-               if (IS_ERR(bc)) {
-                       ret = PTR_ERR(bc);
+               spin_lock_irq(&bd->lock);
+               if (!bd->queued_cmds) {
+                       spin_unlock_irq(&bd->lock);
                        break;
                }
+               spin_unlock_irq(&bd->lock);
 
-               tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio);
+               bc = bsg_get_done_cmd(bd);
+               if (IS_ERR(bc))
+                       break;
+
+               tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
+                                               bc->bidi_bio);
                if (!ret)
                        ret = tret;
 
@@ -517,11 +529,9 @@ static int bsg_complete_all_commands(struct bsg_device *bd)
        return ret;
 }
 
-typedef struct bsg_command *(*bsg_command_callback)(struct bsg_device *bd, const struct iovec *iov);
-
-static ssize_t
-__bsg_read(char __user *buf, size_t count, bsg_command_callback get_bc,
-          struct bsg_device *bd, const struct iovec *iov, ssize_t *bytes_read)
+static int
+__bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
+          const struct iovec *iov, ssize_t *bytes_read)
 {
        struct bsg_command *bc;
        int nr_commands, ret;
@@ -532,7 +542,7 @@ __bsg_read(char __user *buf, size_t count, bsg_command_callback get_bc,
        ret = 0;
        nr_commands = count / sizeof(struct sg_io_v4);
        while (nr_commands) {
-               bc = get_bc(bd, iov);
+               bc = bsg_get_done_cmd(bd);
                if (IS_ERR(bc)) {
                        ret = PTR_ERR(bc);
                        break;
@@ -543,9 +553,10 @@ __bsg_read(char __user *buf, size_t count, bsg_command_callback get_bc,
                 * after completing the request. so do that here,
                 * bsg_complete_work() cannot do that for us
                 */
-               ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio);
+               ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
+                                              bc->bidi_bio);
 
-               if (copy_to_user(buf, (char *) &bc->hdr, sizeof(bc->hdr)))
+               if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
                        ret = -EFAULT;
 
                bsg_free_command(bc);
@@ -569,14 +580,9 @@ static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
                set_bit(BSG_F_BLOCK, &bd->flags);
 }
 
-static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file)
-{
-       if (file->f_mode & FMODE_WRITE)
-               set_bit(BSG_F_WRITE_PERM, &bd->flags);
-       else
-               clear_bit(BSG_F_WRITE_PERM, &bd->flags);
-}
-
+/*
+ * Check if the error is a "real" error that we should return.
+ */
 static inline int err_block_err(int ret)
 {
        if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
@@ -595,19 +601,20 @@ bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
        dprintk("%s: read %Zd bytes\n", bd->name, count);
 
        bsg_set_block(bd, file);
+
        bytes_read = 0;
-       ret = __bsg_read(buf, count, bsg_get_done_cmd,
-                       bd, NULL, &bytes_read);
+       ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
        *ppos = bytes_read;
 
-       if (!bytes_read || (bytes_read && err_block_err(ret)))
+       if (!bytes_read || err_block_err(ret))
                bytes_read = ret;
 
        return bytes_read;
 }
 
-static ssize_t __bsg_write(struct bsg_device *bd, const char __user *buf,
-                          size_t count, ssize_t *bytes_read)
+static int __bsg_write(struct bsg_device *bd, const char __user *buf,
+                      size_t count, ssize_t *bytes_written,
+                      fmode_t has_write_perm)
 {
        struct bsg_command *bc;
        struct request *rq;
@@ -621,18 +628,15 @@ static ssize_t __bsg_write(struct bsg_device *bd, const char __user *buf,
        bc = NULL;
        ret = 0;
        while (nr_commands) {
-               request_queue_t *q = bd->queue;
+               struct request_queue *q = bd->queue;
 
-               bc = bsg_get_command(bd);
-               if (!bc)
-                       break;
+               bc = bsg_alloc_command(bd);
                if (IS_ERR(bc)) {
                        ret = PTR_ERR(bc);
                        bc = NULL;
                        break;
                }
 
-               bc->uhdr = (struct sg_io_v4 __user *) buf;
                if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
                        ret = -EFAULT;
                        break;
@@ -641,7 +645,7 @@ static ssize_t __bsg_write(struct bsg_device *bd, const char __user *buf,
                /*
                 * get a request, fill in the blanks, and add to request queue
                 */
-               rq = bsg_map_hdr(bd, &bc->hdr);
+               rq = bsg_map_hdr(bd, &bc->hdr, has_write_perm, bc->sense);
                if (IS_ERR(rq)) {
                        ret = PTR_ERR(rq);
                        rq = NULL;
@@ -653,7 +657,7 @@ static ssize_t __bsg_write(struct bsg_device *bd, const char __user *buf,
                rq = NULL;
                nr_commands--;
                buf += sizeof(struct sg_io_v4);
-               *bytes_read += sizeof(struct sg_io_v4);
+               *bytes_written += sizeof(struct sg_io_v4);
        }
 
        if (bc)
@@ -666,26 +670,27 @@ static ssize_t
 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
 {
        struct bsg_device *bd = file->private_data;
-       ssize_t bytes_read;
+       ssize_t bytes_written;
        int ret;
 
        dprintk("%s: write %Zd bytes\n", bd->name, count);
 
        bsg_set_block(bd, file);
-       bsg_set_write_perm(bd, file);
 
-       bytes_read = 0;
-       ret = __bsg_write(bd, buf, count, &bytes_read);
-       *ppos = bytes_read;
+       bytes_written = 0;
+       ret = __bsg_write(bd, buf, count, &bytes_written,
+                         file->f_mode & FMODE_WRITE);
+
+       *ppos = bytes_written;
 
        /*
         * return bytes written on non-fatal errors
         */
-       if (!bytes_read || (bytes_read && err_block_err(ret)))
-               bytes_read = ret;
+       if (!bytes_written || err_block_err(ret))
+               bytes_written = ret;
 
-       dprintk("%s: returning %Zd\n", bd->name, bytes_read);
-       return bytes_read;
+       dprintk("%s: returning %Zd\n", bd->name, bytes_written);
+       return bytes_written;
 }
 
 static struct bsg_device *bsg_alloc_device(void)
@@ -709,14 +714,33 @@ static struct bsg_device *bsg_alloc_device(void)
        return bd;
 }
 
+static void bsg_kref_release_function(struct kref *kref)
+{
+       struct bsg_class_device *bcd =
+               container_of(kref, struct bsg_class_device, ref);
+       struct device *parent = bcd->parent;
+
+       if (bcd->release)
+               bcd->release(bcd->parent);
+
+       put_device(parent);
+}
+
 static int bsg_put_device(struct bsg_device *bd)
 {
-       int ret = 0;
+       int ret = 0, do_free;
+       struct request_queue *q = bd->queue;
 
        mutex_lock(&bsg_mutex);
 
-       if (!atomic_dec_and_test(&bd->ref_count))
+       do_free = atomic_dec_and_test(&bd->ref_count);
+       if (!do_free) {
+               mutex_unlock(&bsg_mutex);
                goto out;
+       }
+
+       hlist_del(&bd->dev_list);
+       mutex_unlock(&bsg_mutex);
 
        dprintk("%s: tearing down\n", bd->name);
 
@@ -732,38 +756,42 @@ static int bsg_put_device(struct bsg_device *bd)
         */
        ret = bsg_complete_all_commands(bd);
 
-       blk_put_queue(bd->queue);
-       hlist_del(&bd->dev_list);
        kfree(bd);
 out:
-       mutex_unlock(&bsg_mutex);
+       kref_put(&q->bsg_dev.ref, bsg_kref_release_function);
+       if (do_free)
+               blk_put_queue(q);
        return ret;
 }
 
 static struct bsg_device *bsg_add_device(struct inode *inode,
-                                        struct gendisk *disk,
+                                        struct request_queue *rq,
                                         struct file *file)
 {
-       struct bsg_device *bd = NULL;
+       struct bsg_device *bd;
+       int ret;
 #ifdef BSG_DEBUG
        unsigned char buf[32];
 #endif
+       ret = blk_get_queue(rq);
+       if (ret)
+               return ERR_PTR(-ENXIO);
 
        bd = bsg_alloc_device();
-       if (!bd)
+       if (!bd) {
+               blk_put_queue(rq);
                return ERR_PTR(-ENOMEM);
+       }
+
+       bd->queue = rq;
 
-       bd->disk = disk;
-       bd->queue = disk->queue;
-       kobject_get(&disk->queue->kobj);
        bsg_set_block(bd, file);
 
        atomic_set(&bd->ref_count, 1);
-       bd->minor = iminor(inode);
        mutex_lock(&bsg_mutex);
-       hlist_add_head(&bd->dev_list,&bsg_device_list[bsg_list_idx(bd->minor)]);
+       hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(iminor(inode)));
 
-       strncpy(bd->name, disk->disk_name, sizeof(bd->name) - 1);
+       strncpy(bd->name, dev_name(rq->bsg_dev.class_dev), sizeof(bd->name) - 1);
        dprintk("bound to <%s>, max queue %d\n",
                format_dev_t(buf, inode->i_rdev), bd->max_queue);
 
@@ -771,58 +799,58 @@ static struct bsg_device *bsg_add_device(struct inode *inode,
        return bd;
 }
 
-static struct bsg_device *__bsg_get_device(int minor)
+static struct bsg_device *__bsg_get_device(int minor, struct request_queue *q)
 {
-       struct hlist_head *list = &bsg_device_list[bsg_list_idx(minor)];
-       struct bsg_device *bd = NULL;
+       struct bsg_device *bd;
        struct hlist_node *entry;
 
        mutex_lock(&bsg_mutex);
 
-       hlist_for_each(entry, list) {
-               bd = hlist_entry(entry, struct bsg_device, dev_list);
-               if (bd->minor == minor) {
+       hlist_for_each_entry(bd, entry, bsg_dev_idx_hash(minor), dev_list) {
+               if (bd->queue == q) {
                        atomic_inc(&bd->ref_count);
-                       break;
+                       goto found;
                }
-
-               bd = NULL;
        }
-
+       bd = NULL;
+found:
        mutex_unlock(&bsg_mutex);
        return bd;
 }
 
 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
 {
-       struct bsg_device *bd = __bsg_get_device(iminor(inode));
-       struct bsg_class_device *bcd, *__bcd;
-
-       if (bd)
-               return bd;
+       struct bsg_device *bd;
+       struct bsg_class_device *bcd;
 
        /*
         * find the class device
         */
-       bcd = NULL;
        mutex_lock(&bsg_mutex);
-       list_for_each_entry(__bcd, &bsg_class_list, list) {
-               if (__bcd->minor == iminor(inode)) {
-                       bcd = __bcd;
-                       break;
-               }
-       }
+       bcd = idr_find(&bsg_minor_idr, iminor(inode));
+       if (bcd)
+               kref_get(&bcd->ref);
        mutex_unlock(&bsg_mutex);
 
        if (!bcd)
                return ERR_PTR(-ENODEV);
 
-       return bsg_add_device(inode, bcd->disk, file);
+       bd = __bsg_get_device(iminor(inode), bcd->queue);
+       if (bd)
+               return bd;
+
+       bd = bsg_add_device(inode, bcd->queue, file);
+       if (IS_ERR(bd))
+               kref_put(&bcd->ref, bsg_kref_release_function);
+
+       return bd;
 }
 
 static int bsg_open(struct inode *inode, struct file *file)
 {
-       struct bsg_device *bd = bsg_get_device(inode, file);
+       struct bsg_device *bd;
+
+       bd = bsg_get_device(inode, file);
 
        if (IS_ERR(bd))
                return PTR_ERR(bd);
@@ -850,22 +878,18 @@ static unsigned int bsg_poll(struct file *file, poll_table *wait)
        spin_lock_irq(&bd->lock);
        if (!list_empty(&bd->done_list))
                mask |= POLLIN | POLLRDNORM;
-       if (bd->queued_cmds >= bd->max_queue)
+       if (bd->queued_cmds < bd->max_queue)
                mask |= POLLOUT;
        spin_unlock_irq(&bd->lock);
 
        return mask;
 }
 
-static int
-bsg_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
-         unsigned long arg)
+static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
 {
        struct bsg_device *bd = file->private_data;
        int __user *uarg = (int __user *) arg;
-
-       if (!bd)
-               return -ENXIO;
+       int ret;
 
        switch (cmd) {
                /*
@@ -900,28 +924,34 @@ bsg_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
        case SG_EMULATED_HOST:
        case SCSI_IOCTL_SEND_COMMAND: {
                void __user *uarg = (void __user *) arg;
-               return scsi_cmd_ioctl(file, bd->disk, cmd, uarg);
+               return scsi_cmd_ioctl(bd->queue, NULL, file->f_mode, cmd, uarg);
        }
        case SG_IO: {
                struct request *rq;
-               struct bio *bio;
+               struct bio *bio, *bidi_bio = NULL;
                struct sg_io_v4 hdr;
+               int at_head;
+               u8 sense[SCSI_SENSE_BUFFERSIZE];
 
                if (copy_from_user(&hdr, uarg, sizeof(hdr)))
                        return -EFAULT;
 
-               rq = bsg_map_hdr(bd, &hdr);
+               rq = bsg_map_hdr(bd, &hdr, file->f_mode & FMODE_WRITE, sense);
                if (IS_ERR(rq))
                        return PTR_ERR(rq);
 
                bio = rq->bio;
-               blk_execute_rq(bd->queue, bd->disk, rq, 0);
-               blk_complete_sgv4_hdr_rq(rq, &hdr, bio);
+               if (rq->next_rq)
+                       bidi_bio = rq->next_rq->bio;
+
+               at_head = (0 == (hdr.flags & BSG_FLAG_Q_AT_TAIL));
+               blk_execute_rq(bd->queue, NULL, rq, at_head);
+               ret = blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
 
                if (copy_to_user(uarg, &hdr, sizeof(hdr)))
                        return -EFAULT;
 
-               return 0;
+               return ret;
        }
        /*
         * block device ioctls
@@ -935,36 +965,47 @@ bsg_ioctl(struct inode *inode, struct file *file, unsigned int cmd,
        }
 }
 
-static struct file_operations bsg_fops = {
+static const struct file_operations bsg_fops = {
        .read           =       bsg_read,
        .write          =       bsg_write,
        .poll           =       bsg_poll,
        .open           =       bsg_open,
        .release        =       bsg_release,
-       .ioctl          =       bsg_ioctl,
+       .unlocked_ioctl =       bsg_ioctl,
        .owner          =       THIS_MODULE,
+       .llseek         =       default_llseek,
 };
 
-void bsg_unregister_disk(struct gendisk *disk)
+void bsg_unregister_queue(struct request_queue *q)
 {
-       struct bsg_class_device *bcd = &disk->bsg_dev;
+       struct bsg_class_device *bcd = &q->bsg_dev;
 
        if (!bcd->class_dev)
                return;
 
        mutex_lock(&bsg_mutex);
-       sysfs_remove_link(&bcd->disk->queue->kobj, "bsg");
-       class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor));
+       idr_remove(&bsg_minor_idr, bcd->minor);
+       sysfs_remove_link(&q->kobj, "bsg");
+       device_unregister(bcd->class_dev);
        bcd->class_dev = NULL;
-       list_del_init(&bcd->list);
+       kref_put(&bcd->ref, bsg_kref_release_function);
        mutex_unlock(&bsg_mutex);
 }
+EXPORT_SYMBOL_GPL(bsg_unregister_queue);
 
-int bsg_register_disk(struct gendisk *disk)
+int bsg_register_queue(struct request_queue *q, struct device *parent,
+                      const char *name, void (*release)(struct device *))
 {
-       request_queue_t *q = disk->queue;
        struct bsg_class_device *bcd;
        dev_t dev;
+       int ret, minor;
+       struct device *class_dev = NULL;
+       const char *devname;
+
+       if (name)
+               devname = name;
+       else
+               devname = dev_name(parent);
 
        /*
         * we need a proper transport to send commands, not a stacked device
@@ -972,64 +1013,115 @@ int bsg_register_disk(struct gendisk *disk)
        if (!q->request_fn)
                return 0;
 
-       bcd = &disk->bsg_dev;
+       bcd = &q->bsg_dev;
        memset(bcd, 0, sizeof(*bcd));
-       INIT_LIST_HEAD(&bcd->list);
 
        mutex_lock(&bsg_mutex);
-       dev = MKDEV(BSG_MAJOR, bsg_device_nr);
-       bcd->minor = bsg_device_nr;
-       bsg_device_nr++;
-       bcd->disk = disk;
-       bcd->class_dev = class_device_create(bsg_class, NULL, dev, bcd->dev, "%s", disk->disk_name);
-       if (!bcd->class_dev)
-               goto err;
-       list_add_tail(&bcd->list, &bsg_class_list);
-       if (sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg"))
-               goto err;
+
+       ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
+       if (!ret) {
+               ret = -ENOMEM;
+               goto unlock;
+       }
+
+       ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
+       if (ret < 0)
+               goto unlock;
+
+       if (minor >= BSG_MAX_DEVS) {
+               printk(KERN_ERR "bsg: too many bsg devices\n");
+               ret = -EINVAL;
+               goto remove_idr;
+       }
+
+       bcd->minor = minor;
+       bcd->queue = q;
+       bcd->parent = get_device(parent);
+       bcd->release = release;
+       kref_init(&bcd->ref);
+       dev = MKDEV(bsg_major, bcd->minor);
+       class_dev = device_create(bsg_class, parent, dev, NULL, "%s", devname);
+       if (IS_ERR(class_dev)) {
+               ret = PTR_ERR(class_dev);
+               goto put_dev;
+       }
+       bcd->class_dev = class_dev;
+
+       if (q->kobj.sd) {
+               ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
+               if (ret)
+                       goto unregister_class_dev;
+       }
+
        mutex_unlock(&bsg_mutex);
        return 0;
-err:
-       bsg_device_nr--;
-       if (bcd->class_dev)
-               class_device_destroy(bsg_class, MKDEV(BSG_MAJOR, bcd->minor));
+
+unregister_class_dev:
+       device_unregister(class_dev);
+put_dev:
+       put_device(parent);
+remove_idr:
+       idr_remove(&bsg_minor_idr, minor);
+unlock:
        mutex_unlock(&bsg_mutex);
-       return -ENOMEM;
+       return ret;
+}
+EXPORT_SYMBOL_GPL(bsg_register_queue);
+
+static struct cdev bsg_cdev;
+
+static char *bsg_devnode(struct device *dev, mode_t *mode)
+{
+       return kasprintf(GFP_KERNEL, "bsg/%s", dev_name(dev));
 }
 
 static int __init bsg_init(void)
 {
        int ret, i;
+       dev_t devid;
 
        bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
-                               sizeof(struct bsg_command), 0, 0, NULL, NULL);
+                               sizeof(struct bsg_command), 0, 0, NULL);
        if (!bsg_cmd_cachep) {
                printk(KERN_ERR "bsg: failed creating slab cache\n");
                return -ENOMEM;
        }
 
-       for (i = 0; i < BSG_LIST_SIZE; i++)
+       for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
                INIT_HLIST_HEAD(&bsg_device_list[i]);
 
        bsg_class = class_create(THIS_MODULE, "bsg");
        if (IS_ERR(bsg_class)) {
-               kmem_cache_destroy(bsg_cmd_cachep);
-               return PTR_ERR(bsg_class);
+               ret = PTR_ERR(bsg_class);
+               goto destroy_kmemcache;
        }
+       bsg_class->devnode = bsg_devnode;
 
-       ret = register_chrdev(BSG_MAJOR, "bsg", &bsg_fops);
-       if (ret) {
-               kmem_cache_destroy(bsg_cmd_cachep);
-               class_destroy(bsg_class);
-               return ret;
-       }
+       ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
+       if (ret)
+               goto destroy_bsg_class;
+
+       bsg_major = MAJOR(devid);
 
-       printk(KERN_INFO "%s loaded\n", bsg_version);
+       cdev_init(&bsg_cdev, &bsg_fops);
+       ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
+       if (ret)
+               goto unregister_chrdev;
+
+       printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
+              " loaded (major %d)\n", bsg_major);
        return 0;
+unregister_chrdev:
+       unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
+destroy_bsg_class:
+       class_destroy(bsg_class);
+destroy_kmemcache:
+       kmem_cache_destroy(bsg_cmd_cachep);
+       return ret;
 }
 
 MODULE_AUTHOR("Jens Axboe");
-MODULE_DESCRIPTION("Block layer SGSI generic (sg) driver");
+MODULE_DESCRIPTION(BSG_DESCRIPTION);
 MODULE_LICENSE("GPL");
 
-subsys_initcall(bsg_init);
+device_initcall(bsg_init);