bsg: support for full generic block layer SG v3
Jens Axboe [Mon, 9 Jul 2007 10:38:05 +0000 (12:38 +0200)]
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>

block/Kconfig
block/Makefile
block/bsg.c [new file with mode: 0644]
block/ll_rw_blk.c
block/scsi_ioctl.c
drivers/ide/ide-floppy.c
drivers/ide/ide.c
include/linux/blkdev.h
include/linux/bsg.h [new file with mode: 0644]
include/linux/genhd.h

index 2859351..da12f26 100644 (file)
@@ -51,4 +51,11 @@ config LSF
 
 endif # BLOCK
 
+config BLK_DEV_BSG
+       bool "Block layer SG support"
+       default y
+       ---help---
+       Saying Y here will enable generic SG (SCSI generic) v3
+       support for any block device.
+
 source block/Kconfig.iosched
index 4b84d0d..959feeb 100644 (file)
@@ -4,6 +4,7 @@
 
 obj-$(CONFIG_BLOCK) := elevator.o ll_rw_blk.o ioctl.o genhd.o scsi_ioctl.o
 
+obj-$(CONFIG_BLK_DEV_BSG)      += bsg.o
 obj-$(CONFIG_IOSCHED_NOOP)     += noop-iosched.o
 obj-$(CONFIG_IOSCHED_AS)       += as-iosched.o
 obj-$(CONFIG_IOSCHED_DEADLINE) += deadline-iosched.o
diff --git a/block/bsg.c b/block/bsg.c
new file mode 100644 (file)
index 0000000..724b693
--- /dev/null
@@ -0,0 +1,997 @@
+/*
+ * bsg.c - block layer implementation of the sg v3 interface
+ *
+ * Copyright (C) 2004 Jens Axboe <axboe@suse.de> SUSE Labs
+ * Copyright (C) 2004 Peter M. Jones <pjones@redhat.com>
+ *
+ *  This file is subject to the terms and conditions of the GNU General Public
+ *  License version 2.  See the file "COPYING" in the main directory of this
+ *  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/config.h>
+#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/percpu.h>
+#include <linux/uio.h>
+#include <linux/bsg.h>
+
+#include <scsi/scsi.h>
+#include <scsi/scsi_ioctl.h>
+#include <scsi/scsi_cmnd.h>
+#include <scsi/sg.h>
+
+static char bsg_version[] = "block layer sg (bsg) 0.4";
+
+struct bsg_command;
+
+struct bsg_device {
+       struct gendisk *disk;
+       request_queue_t *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;
+       unsigned long *cmd_bitmap;
+       struct bsg_command *cmd_map;
+       wait_queue_head_t wq_done;
+       wait_queue_head_t wq_free;
+       char name[BDEVNAME_SIZE];
+       int max_queue;
+       unsigned long flags;
+};
+
+enum {
+       BSG_F_BLOCK             = 1,
+       BSG_F_WRITE_PERM        = 2,
+};
+
+/*
+ * command allocation bitmap defines
+ */
+#define BSG_CMDS_PAGE_ORDER    (1)
+#define BSG_CMDS_PER_LONG      (sizeof(unsigned long) * 8)
+#define BSG_CMDS_MASK          (BSG_CMDS_PER_LONG - 1)
+#define BSG_CMDS_BYTES         (PAGE_SIZE * (1 << BSG_CMDS_PAGE_ORDER))
+#define BSG_CMDS               (BSG_CMDS_BYTES / sizeof(struct bsg_command))
+
+#undef BSG_DEBUG
+
+#ifdef BSG_DEBUG
+#define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##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;
+
+#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];
+
+static struct class *bsg_class;
+static LIST_HEAD(bsg_class_list);
+
+/*
+ * our internal command type
+ */
+struct bsg_command {
+       struct bsg_device *bd;
+       struct list_head list;
+       struct request *rq;
+       struct bio *bio;
+       int err;
+       struct sg_io_hdr hdr;
+       struct sg_io_hdr __user *uhdr;
+       char sense[SCSI_SENSE_BUFFERSIZE];
+};
+
+static void bsg_free_command(struct bsg_command *bc)
+{
+       struct bsg_device *bd = bc->bd;
+       unsigned long bitnr = bc - bd->cmd_map;
+       unsigned long flags;
+
+       dprintk("%s: command bit offset %lu\n", bd->name, bitnr);
+
+       spin_lock_irqsave(&bd->lock, flags);
+       bd->queued_cmds--;
+       __clear_bit(bitnr, bd->cmd_bitmap);
+       spin_unlock_irqrestore(&bd->lock, flags);
+
+       wake_up(&bd->wq_free);
+}
+
+static struct bsg_command *__bsg_alloc_command(struct bsg_device *bd)
+{
+       struct bsg_command *bc = NULL;
+       unsigned long *map;
+       int free_nr;
+
+       spin_lock_irq(&bd->lock);
+
+       if (bd->queued_cmds >= bd->max_queue)
+               goto out;
+
+       for (free_nr = 0, map = bd->cmd_bitmap; *map == ~0UL; map++)
+               free_nr += BSG_CMDS_PER_LONG;
+
+       BUG_ON(*map == ~0UL);
+
+       bd->queued_cmds++;
+       free_nr += ffz(*map);
+       __set_bit(free_nr, bd->cmd_bitmap);
+       spin_unlock_irq(&bd->lock);
+
+       bc = bd->cmd_map + free_nr;
+       memset(bc, 0, sizeof(*bc));
+       bc->bd = bd;
+       INIT_LIST_HEAD(&bc->list);
+       dprintk("%s: returning free cmd %p (bit %d)\n", bd->name, bc, free_nr);
+       return bc;
+out:
+       dprintk("%s: failed (depth %d)\n", bd->name, bd->queued_cmds);
+       spin_unlock_irq(&bd->lock);
+       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)
+{
+       bd->done_cmds++;
+       list_add_tail(&bc->list, &bd->done_list);
+       wake_up(&bd->wq_done);
+}
+
+static inline int bsg_io_schedule(struct bsg_device *bd, int state)
+{
+       DEFINE_WAIT(wait);
+       int ret = 0;
+
+       spin_lock_irq(&bd->lock);
+
+       BUG_ON(bd->done_cmds > bd->queued_cmds);
+
+       /*
+        * -ENOSPC or -ENODATA?  I'm going for -ENODATA, meaning "I have no
+        * work to do", even though we return -ENOSPC after this same test
+        * during bsg_write() -- there, it means our buffer can't have more
+        * bsg_commands added to it, thus has no space left.
+        */
+       if (bd->done_cmds == bd->queued_cmds) {
+               ret = -ENODATA;
+               goto unlock;
+       }
+
+       if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
+               ret = -EAGAIN;
+               goto unlock;
+       }
+
+       prepare_to_wait(&bd->wq_done, &wait, state);
+       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)
+{
+       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;
+}
+
+/*
+ * Check if sg_io_hdr from user is allowed and valid
+ */
+static int
+bsg_validate_sghdr(request_queue_t *q, struct sg_io_hdr *hdr, int *rw)
+{
+       if (hdr->interface_id != 'S')
+               return -EINVAL;
+       if (hdr->cmd_len > BLK_MAX_CDB)
+               return -EINVAL;
+       if (hdr->dxfer_len > (q->max_sectors << 9))
+               return -EIO;
+
+       /*
+        * looks sane, if no data then it should be fine from our POV
+        */
+       if (!hdr->dxfer_len)
+               return 0;
+
+       switch (hdr->dxfer_direction) {
+               case SG_DXFER_TO_FROM_DEV:
+               case SG_DXFER_FROM_DEV:
+                       *rw = READ;
+                       break;
+               case SG_DXFER_TO_DEV:
+                       *rw = WRITE;
+                       break;
+               default:
+                       return -EINVAL;
+       }
+
+       return 0;
+}
+
+/*
+ * map sg_io_hdr to a request. for scatter-gather sg_io_hdr, we map
+ * each segment to a bio and string multiple bio's to the request
+ */
+static struct request *
+bsg_map_hdr(struct bsg_device *bd, int rw, struct sg_io_hdr *hdr)
+{
+       request_queue_t *q = bd->queue;
+       struct sg_iovec iov;
+       struct sg_iovec __user *u_iov;
+       struct request *rq;
+       int ret, i = 0;
+
+       dprintk("map hdr %p/%d/%d\n", hdr->dxferp, hdr->dxfer_len,
+                                       hdr->iovec_count);
+
+       ret = bsg_validate_sghdr(q, hdr, &rw);
+       if (ret)
+               return ERR_PTR(ret);
+
+       /*
+        * map scatter-gather elements seperately and string them to request
+        */
+       rq = blk_get_request(q, rw, GFP_KERNEL);
+       ret = blk_fill_sghdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM,
+                               &bd->flags));
+       if (ret) {
+               blk_put_request(rq);
+               return ERR_PTR(ret);
+       }
+
+       if (!hdr->iovec_count) {
+               ret = blk_rq_map_user(q, rq, hdr->dxferp, hdr->dxfer_len);
+               if (ret)
+                       goto out;
+       }
+
+       u_iov = hdr->dxferp;
+       for (ret = 0, i = 0; i < hdr->iovec_count; i++, u_iov++) {
+               if (copy_from_user(&iov, u_iov, sizeof(iov))) {
+                       ret = -EFAULT;
+                       break;
+               }
+
+               if (!iov.iov_len || !iov.iov_base) {
+                       ret = -EINVAL;
+                       break;
+               }
+
+               ret = blk_rq_map_user(q, rq, iov.iov_base, iov.iov_len);
+               if (ret)
+                       break;
+       }
+
+       /*
+        * bugger, cleanup
+        */
+       if (ret) {
+out:
+               dprintk("failed map at %d: %d\n", i, ret);
+               blk_unmap_sghdr_rq(rq, hdr);
+               rq = ERR_PTR(ret);
+       }
+
+       return rq;
+}
+
+/*
+ * async completion call-back from the block layer, when scsi/ide/whatever
+ * calls end_that_request_last() on a request
+ */
+static void bsg_rq_end_io(struct request *rq, int uptodate)
+{
+       struct bsg_command *bc = rq->end_io_data;
+       struct bsg_device *bd = bc->bd;
+       unsigned long flags;
+
+       dprintk("%s: finished rq %p bio %p, bc %p offset %ld stat %d\n",
+                       bd->name, rq, bc, bc->bio, bc - bd->cmd_map, 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);
+       spin_unlock_irqrestore(&bd->lock, flags);
+}
+
+/*
+ * 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,
+                           struct bsg_command *bc, struct request *rq)
+{
+       rq->sense = bc->sense;
+       rq->sense_len = 0;
+
+       /*
+        * add bc command to busy queue and submit rq for io
+        */
+       bc->rq = rq;
+       bc->bio = rq->bio;
+       bc->hdr.duration = jiffies;
+       spin_lock_irq(&bd->lock);
+       list_add_tail(&bc->list, &bd->busy_list);
+       spin_unlock_irq(&bd->lock);
+
+       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);
+}
+
+static inline 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);
+       }
+       spin_unlock_irq(&bd->lock);
+
+       return bc;
+}
+
+/*
+ * Get a finished command from the done list
+ */
+static struct bsg_command *__bsg_get_done_cmd(struct bsg_device *bd, int state)
+{
+       struct bsg_command *bc;
+       int ret;
+
+       do {
+               bc = bsg_next_done_cmd(bd);
+               if (bc)
+                       break;
+
+               ret = bsg_io_schedule(bd, state);
+               if (ret) {
+                       bc = ERR_PTR(ret);
+                       break;
+               }
+       } while (1);
+
+       dprintk("%s: returning done %p\n", bd->name, bc);
+
+       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 bsg_complete_all_commands(struct bsg_device *bd)
+{
+       struct bsg_command *bc;
+       int ret, tret;
+
+       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);
+               /*
+                * look for -ENODATA specifically -- we'll sometimes get
+                * -ERESTARTSYS when we've taken a signal, but we can't
+                * return until we're done freeing the queue, so ignore
+                * it.  The signal will get handled when we're done freeing
+                * the bsg_device.
+                */
+       } while (ret != -ENODATA);
+
+       /*
+        * discard done commands
+        */
+       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);
+                       break;
+               }
+
+               tret = blk_complete_sghdr_rq(bc->rq, &bc->hdr, bc->bio);
+               if (!ret)
+                       ret = tret;
+
+               bsg_free_command(bc);
+       } while (1);
+
+       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)
+{
+       struct bsg_command *bc;
+       int nr_commands, ret;
+
+       if (count % sizeof(struct sg_io_hdr))
+               return -EINVAL;
+
+       ret = 0;
+       nr_commands = count / sizeof(struct sg_io_hdr);
+       while (nr_commands) {
+               bc = get_bc(bd, iov);
+               if (IS_ERR(bc)) {
+                       ret = PTR_ERR(bc);
+                       break;
+               }
+
+               /*
+                * this is the only case where we need to copy data back
+                * after completing the request. so do that here,
+                * bsg_complete_work() cannot do that for us
+                */
+               ret = blk_complete_sghdr_rq(bc->rq, &bc->hdr, bc->bio);
+
+               if (copy_to_user(buf, (char *) &bc->hdr, sizeof(bc->hdr)))
+                       ret = -EFAULT;
+
+               bsg_free_command(bc);
+
+               if (ret)
+                       break;
+
+               buf += sizeof(struct sg_io_hdr);
+               *bytes_read += sizeof(struct sg_io_hdr);
+               nr_commands--;
+       }
+
+       return ret;
+}
+
+static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
+{
+       if (file->f_flags & O_NONBLOCK)
+               clear_bit(BSG_F_BLOCK, &bd->flags);
+       else
+               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);
+}
+
+static inline int err_block_err(int ret)
+{
+       if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
+               return 1;
+
+       return 0;
+}
+
+static ssize_t
+bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
+{
+       struct bsg_device *bd = file->private_data;
+       int ret;
+       ssize_t bytes_read;
+
+       dprintk("%s: read %lu 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);
+       *ppos = bytes_read;
+
+       if (!bytes_read || (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)
+{
+       struct bsg_command *bc;
+       struct request *rq;
+       int ret, nr_commands;
+
+       if (count % sizeof(struct sg_io_hdr))
+               return -EINVAL;
+
+       nr_commands = count / sizeof(struct sg_io_hdr);
+       rq = NULL;
+       bc = NULL;
+       ret = 0;
+       while (nr_commands) {
+               request_queue_t *q = bd->queue;
+               int rw = READ;
+
+               bc = bsg_get_command(bd);
+               if (!bc)
+                       break;
+               if (IS_ERR(bc)) {
+                       ret = PTR_ERR(bc);
+                       bc = NULL;
+                       break;
+               }
+
+               bc->uhdr = (struct sg_io_hdr __user *) buf;
+               if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
+                       ret = -EFAULT;
+                       break;
+               }
+
+               /*
+                * get a request, fill in the blanks, and add to request queue
+                */
+               rq = bsg_map_hdr(bd, rw, &bc->hdr);
+               if (IS_ERR(rq)) {
+                       ret = PTR_ERR(rq);
+                       rq = NULL;
+                       break;
+               }
+
+               bsg_add_command(bd, q, bc, rq);
+               bc = NULL;
+               rq = NULL;
+               nr_commands--;
+               buf += sizeof(struct sg_io_hdr);
+               *bytes_read += sizeof(struct sg_io_hdr);
+       }
+
+       if (rq)
+               blk_unmap_sghdr_rq(rq, &bc->hdr);
+       if (bc)
+               bsg_free_command(bc);
+
+       return ret;
+}
+
+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;
+       int ret;
+
+       dprintk("%s: write %lu 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;
+
+       /*
+        * return bytes written on non-fatal errors
+        */
+       if (!bytes_read || (bytes_read && err_block_err(ret)))
+               bytes_read = ret;
+
+       dprintk("%s: returning %lu\n", bd->name, bytes_read);
+       return bytes_read;
+}
+
+static void bsg_free_device(struct bsg_device *bd)
+{
+       if (bd->cmd_map)
+               free_pages((unsigned long) bd->cmd_map, BSG_CMDS_PAGE_ORDER);
+
+       kfree(bd->cmd_bitmap);
+       kfree(bd);
+}
+
+static struct bsg_device *bsg_alloc_device(void)
+{
+       struct bsg_command *cmd_map;
+       unsigned long *cmd_bitmap;
+       struct bsg_device *bd;
+       int bits;
+
+       bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
+       if (unlikely(!bd))
+               return NULL;
+
+       spin_lock_init(&bd->lock);
+
+       bd->max_queue = BSG_CMDS;
+
+       bits = (BSG_CMDS / BSG_CMDS_PER_LONG) + 1;
+       cmd_bitmap = kzalloc(bits * sizeof(unsigned long), GFP_KERNEL);
+       if (!cmd_bitmap)
+               goto out_free_bd;
+       bd->cmd_bitmap = cmd_bitmap;
+
+       cmd_map = (void *) __get_free_pages(GFP_KERNEL | __GFP_ZERO,
+                                               BSG_CMDS_PAGE_ORDER);
+       if (!cmd_map)
+               goto out_free_bitmap;
+       bd->cmd_map = cmd_map;
+
+       INIT_LIST_HEAD(&bd->busy_list);
+       INIT_LIST_HEAD(&bd->done_list);
+       INIT_HLIST_NODE(&bd->dev_list);
+
+       init_waitqueue_head(&bd->wq_free);
+       init_waitqueue_head(&bd->wq_done);
+       return bd;
+
+out_free_bitmap:
+       kfree(cmd_bitmap);
+out_free_bd:
+       kfree(bd);
+       return NULL;
+}
+
+static int bsg_put_device(struct bsg_device *bd)
+{
+       int ret = 0;
+
+       mutex_lock(&bsg_mutex);
+
+       if (!atomic_dec_and_test(&bd->ref_count))
+               goto out;
+
+       dprintk("%s: tearing down\n", bd->name);
+
+       /*
+        * close can always block
+        */
+       set_bit(BSG_F_BLOCK, &bd->flags);
+
+       /*
+        * correct error detection baddies here again. it's the responsibility
+        * of the app to properly reap commands before close() if it wants
+        * fool-proof error detection
+        */
+       ret = bsg_complete_all_commands(bd);
+
+       blk_put_queue(bd->queue);
+       hlist_del(&bd->dev_list);
+       bsg_free_device(bd);
+out:
+       mutex_unlock(&bsg_mutex);
+       return ret;
+}
+
+static struct bsg_device *bsg_add_device(struct inode *inode,
+                                        struct gendisk *disk,
+                                        struct file *file)
+{
+       struct bsg_device *bd = NULL;
+#ifdef BSG_DEBUG
+       unsigned char buf[32];
+#endif
+
+       bd = bsg_alloc_device();
+       if (!bd)
+               return ERR_PTR(-ENOMEM);
+
+       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)]);
+
+       strncpy(bd->name, disk->disk_name, sizeof(bd->name) - 1);
+       dprintk("bound to <%s>, max queue %d\n",
+               format_dev_t(buf, i->i_rdev), bd->max_queue);
+
+       mutex_unlock(&bsg_mutex);
+       return bd;
+}
+
+static struct bsg_device *__bsg_get_device(int minor)
+{
+       struct hlist_head *list = &bsg_device_list[bsg_list_idx(minor)];
+       struct bsg_device *bd = NULL;
+       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) {
+                       atomic_inc(&bd->ref_count);
+                       break;
+               }
+
+               bd = NULL;
+       }
+
+       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;
+
+       /*
+        * 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;
+               }
+       }
+       mutex_unlock(&bsg_mutex);
+
+       if (!bcd)
+               return ERR_PTR(-ENODEV);
+
+       return bsg_add_device(inode, bcd->disk, file);
+}
+
+static int bsg_open(struct inode *inode, struct file *file)
+{
+       struct bsg_device *bd = bsg_get_device(inode, file);
+
+       if (IS_ERR(bd))
+               return PTR_ERR(bd);
+
+       file->private_data = bd;
+       return 0;
+}
+
+static int bsg_release(struct inode *inode, struct file *file)
+{
+       struct bsg_device *bd = file->private_data;
+
+       file->private_data = NULL;
+       return bsg_put_device(bd);
+}
+
+static unsigned int bsg_poll(struct file *file, poll_table *wait)
+{
+       struct bsg_device *bd = file->private_data;
+       unsigned int mask = 0;
+
+       poll_wait(file, &bd->wq_done, wait);
+       poll_wait(file, &bd->wq_free, wait);
+
+       spin_lock_irq(&bd->lock);
+       if (!list_empty(&bd->done_list))
+               mask |= POLLIN | POLLRDNORM;
+       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)
+{
+       struct bsg_device *bd = file->private_data;
+       int __user *uarg = (int __user *) arg;
+
+       if (!bd)
+               return -ENXIO;
+
+       switch (cmd) {
+               /*
+                * our own ioctls
+                */
+       case SG_GET_COMMAND_Q:
+               return put_user(bd->max_queue, uarg);
+               case SG_SET_COMMAND_Q: {
+               int queue;
+
+               if (get_user(queue, uarg))
+                       return -EFAULT;
+               if (queue > BSG_CMDS || queue < 1)
+                       return -EINVAL;
+
+               bd->max_queue = queue;
+               return 0;
+       }
+
+       /*
+        * SCSI/sg ioctls
+        */
+       case SG_GET_VERSION_NUM:
+       case SCSI_IOCTL_GET_IDLUN:
+       case SCSI_IOCTL_GET_BUS_NUMBER:
+       case SG_SET_TIMEOUT:
+       case SG_GET_TIMEOUT:
+       case SG_GET_RESERVED_SIZE:
+       case SG_SET_RESERVED_SIZE:
+       case SG_EMULATED_HOST:
+       case SG_IO:
+       case SCSI_IOCTL_SEND_COMMAND: {
+               void __user *uarg = (void __user *) arg;
+               return scsi_cmd_ioctl(file, bd->disk, cmd, uarg);
+       }
+       /*
+        * block device ioctls
+        */
+       default:
+#if 0
+               return ioctl_by_bdev(bd->bdev, cmd, arg);
+#else
+               return -ENOTTY;
+#endif
+       }
+}
+
+static struct file_operations bsg_fops = {
+       .read           =       bsg_read,
+       .write          =       bsg_write,
+       .poll           =       bsg_poll,
+       .open           =       bsg_open,
+       .release        =       bsg_release,
+       .ioctl          =       bsg_ioctl,
+       .owner          =       THIS_MODULE,
+};
+
+void bsg_unregister_disk(struct gendisk *disk)
+{
+       struct bsg_class_device *bcd = &disk->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));
+       bcd->class_dev = NULL;
+       list_del_init(&bcd->list);
+       mutex_unlock(&bsg_mutex);
+}
+
+int bsg_register_disk(struct gendisk *disk)
+{
+       request_queue_t *q = disk->queue;
+       struct bsg_class_device *bcd;
+       dev_t dev;
+
+       /*
+        * we need a proper transport to send commands, not a stacked device
+        */
+       if (!q->request_fn)
+               return 0;
+
+       bcd = &disk->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);
+       list_add_tail(&bcd->list, &bsg_class_list);
+       sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
+       mutex_unlock(&bsg_mutex);
+       return 0;
+}
+
+static int __init bsg_init(void)
+{
+       int ret, i;
+
+       for (i = 0; i < BSG_LIST_SIZE; i++)
+               INIT_HLIST_HEAD(&bsg_device_list[i]);
+
+       bsg_class = class_create(THIS_MODULE, "bsg");
+       if (IS_ERR(bsg_class))
+               return PTR_ERR(bsg_class);
+
+       ret = register_chrdev(BSG_MAJOR, "bsg", &bsg_fops);
+       if (ret) {
+               class_destroy(bsg_class);
+               return ret;
+       }
+
+       printk(KERN_INFO "%s loaded\n", bsg_version);
+       return 0;
+}
+
+MODULE_AUTHOR("Jens Axboe");
+MODULE_DESCRIPTION("Block layer SGSI generic (sg) driver");
+MODULE_LICENSE("GPL");
+
+subsys_initcall(bsg_init);
index ef42bb2..3795e07 100644 (file)
@@ -4091,6 +4091,13 @@ int blk_register_queue(struct gendisk *disk)
                return ret;
        }
 
+       ret = bsg_register_disk(disk);
+       if (ret) {
+               elv_unregister_queue(q);
+               kobject_unregister(&q->kobj);
+               return ret;
+       }
+
        return 0;
 }
 
@@ -4099,6 +4106,7 @@ void blk_unregister_queue(struct gendisk *disk)
        request_queue_t *q = disk->queue;
 
        if (q && q->request_fn) {
+               bsg_unregister_disk(disk);
                elv_unregister_queue(q);
 
                kobject_uevent(&q->kobj, KOBJ_REMOVE);
index e83f1db..88fd008 100644 (file)
@@ -41,8 +41,6 @@ const unsigned char scsi_command_size[8] =
 
 EXPORT_SYMBOL(scsi_command_size);
 
-#define BLK_DEFAULT_TIMEOUT    (60 * HZ)
-
 #include <scsi/sg.h>
 
 static int sg_get_version(int __user *p)
@@ -114,7 +112,7 @@ static int sg_emulated_host(request_queue_t *q, int __user *p)
 #define safe_for_read(cmd)     [cmd] = CMD_READ_SAFE
 #define safe_for_write(cmd)    [cmd] = CMD_WRITE_SAFE
 
-static int verify_command(struct file *file, unsigned char *cmd)
+static int verify_command(unsigned char *cmd, int has_write_perm)
 {
        static unsigned char cmd_type[256] = {
 
@@ -193,18 +191,11 @@ static int verify_command(struct file *file, unsigned char *cmd)
                safe_for_write(GPCMD_SET_STREAMING),
        };
        unsigned char type = cmd_type[cmd[0]];
-       int has_write_perm = 0;
 
        /* Anybody who can open the device can do a read-safe command */
        if (type & CMD_READ_SAFE)
                return 0;
 
-       /*
-        * file can be NULL from ioctl_by_bdev()...
-        */
-       if (file)
-               has_write_perm = file->f_mode & FMODE_WRITE;
-
        /* Write-safe commands just require a writable open.. */
        if ((type & CMD_WRITE_SAFE) && has_write_perm)
                return 0;
@@ -222,24 +213,104 @@ static int verify_command(struct file *file, unsigned char *cmd)
        return -EPERM;
 }
 
+int blk_fill_sghdr_rq(request_queue_t *q, struct request *rq,
+                     struct sg_io_hdr *hdr, int has_write_perm)
+{
+       memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
+
+       if (copy_from_user(rq->cmd, hdr->cmdp, hdr->cmd_len))
+               return -EFAULT;
+       if (verify_command(rq->cmd, has_write_perm))
+               return -EPERM;
+
+       /*
+        * fill in request structure
+        */
+       rq->cmd_len = hdr->cmd_len;
+       rq->cmd_type = REQ_TYPE_BLOCK_PC;
+
+       rq->timeout = (hdr->timeout * HZ) / 1000;
+       if (!rq->timeout)
+               rq->timeout = q->sg_timeout;
+       if (!rq->timeout)
+               rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(blk_fill_sghdr_rq);
+
+/*
+ * unmap a request that was previously mapped to this sg_io_hdr. handles
+ * both sg and non-sg sg_io_hdr.
+ */
+int blk_unmap_sghdr_rq(struct request *rq, struct sg_io_hdr *hdr)
+{
+       struct bio *bio = rq->bio;
+
+       /*
+        * also releases request
+        */
+       if (!hdr->iovec_count)
+               return blk_rq_unmap_user(bio, hdr->dxfer_len);
+
+       rq_for_each_bio(bio, rq)
+               bio_unmap_user(bio);
+
+       blk_put_request(rq);
+       return 0;
+}
+EXPORT_SYMBOL_GPL(blk_unmap_sghdr_rq);
+
+int blk_complete_sghdr_rq(struct request *rq, struct sg_io_hdr *hdr,
+                         struct bio *bio)
+{
+       int r, ret = 0;
+
+       /*
+        * fill in all the output members
+        */
+       hdr->status = rq->errors & 0xff;
+       hdr->masked_status = status_byte(rq->errors);
+       hdr->msg_status = msg_byte(rq->errors);
+       hdr->host_status = host_byte(rq->errors);
+       hdr->driver_status = driver_byte(rq->errors);
+       hdr->info = 0;
+       if (hdr->masked_status || hdr->host_status || hdr->driver_status)
+               hdr->info |= SG_INFO_CHECK;
+       hdr->resid = rq->data_len;
+       hdr->sb_len_wr = 0;
+
+       if (rq->sense_len && hdr->sbp) {
+               int len = min((unsigned int) hdr->mx_sb_len, rq->sense_len);
+
+               if (!copy_to_user(hdr->sbp, rq->sense, len))
+                       hdr->sb_len_wr = len;
+               else
+                       ret = -EFAULT;
+       }
+
+       rq->bio = bio;
+       r = blk_unmap_sghdr_rq(rq, hdr);
+       if (ret)
+               r = ret;
+
+       return r;
+}
+EXPORT_SYMBOL_GPL(blk_complete_sghdr_rq);
+
 static int sg_io(struct file *file, request_queue_t *q,
                struct gendisk *bd_disk, struct sg_io_hdr *hdr)
 {
-       unsigned long start_time, timeout;
-       int writing = 0, ret = 0;
+       unsigned long start_time;
+       int writing = 0, ret = 0, has_write_perm = 0;
        struct request *rq;
        char sense[SCSI_SENSE_BUFFERSIZE];
-       unsigned char cmd[BLK_MAX_CDB];
        struct bio *bio;
 
        if (hdr->interface_id != 'S')
                return -EINVAL;
        if (hdr->cmd_len > BLK_MAX_CDB)
                return -EINVAL;
-       if (copy_from_user(cmd, hdr->cmdp, hdr->cmd_len))
-               return -EFAULT;
-       if (verify_command(file, cmd))
-               return -EPERM;
 
        if (hdr->dxfer_len > (q->max_hw_sectors << 9))
                return -EIO;
@@ -260,25 +331,14 @@ static int sg_io(struct file *file, request_queue_t *q,
        if (!rq)
                return -ENOMEM;
 
-       /*
-        * fill in request structure
-        */
-       rq->cmd_len = hdr->cmd_len;
-       memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
-       memcpy(rq->cmd, cmd, hdr->cmd_len);
-
-       memset(sense, 0, sizeof(sense));
-       rq->sense = sense;
-       rq->sense_len = 0;
-
-       rq->cmd_type = REQ_TYPE_BLOCK_PC;
+       if (file)
+               has_write_perm = file->f_mode & FMODE_WRITE;
 
-       timeout = msecs_to_jiffies(hdr->timeout);
-       rq->timeout = (timeout < INT_MAX) ? timeout : INT_MAX;
-       if (!rq->timeout)
-               rq->timeout = q->sg_timeout;
-       if (!rq->timeout)
-               rq->timeout = BLK_DEFAULT_TIMEOUT;
+       if (blk_fill_sghdr_rq(q, rq, hdr, has_write_perm)) {
+               blk_rq_unmap_user(bio, hdr->dxfer_len);
+               blk_put_request(rq);
+               return -EFAULT;
+       }
 
        if (hdr->iovec_count) {
                const int size = sizeof(struct sg_iovec) * hdr->iovec_count;
@@ -306,6 +366,9 @@ static int sg_io(struct file *file, request_queue_t *q,
                goto out;
 
        bio = rq->bio;
+       memset(sense, 0, sizeof(sense));
+       rq->sense = sense;
+       rq->sense_len = 0;
        rq->retries = 0;
 
        start_time = jiffies;
@@ -316,31 +379,9 @@ static int sg_io(struct file *file, request_queue_t *q,
         */
        blk_execute_rq(q, bd_disk, rq, 0);
 
-       /* write to all output members */
-       hdr->status = 0xff & rq->errors;
-       hdr->masked_status = status_byte(rq->errors);
-       hdr->msg_status = msg_byte(rq->errors);
-       hdr->host_status = host_byte(rq->errors);
-       hdr->driver_status = driver_byte(rq->errors);
-       hdr->info = 0;
-       if (hdr->masked_status || hdr->host_status || hdr->driver_status)
-               hdr->info |= SG_INFO_CHECK;
-       hdr->resid = rq->data_len;
        hdr->duration = ((jiffies - start_time) * 1000) / HZ;
-       hdr->sb_len_wr = 0;
-
-       if (rq->sense_len && hdr->sbp) {
-               int len = min((unsigned int) hdr->mx_sb_len, rq->sense_len);
-
-               if (!copy_to_user(hdr->sbp, rq->sense, len))
-                       hdr->sb_len_wr = len;
-       }
-
-       if (blk_rq_unmap_user(bio))
-               ret = -EFAULT;
 
-       /* may not have succeeded, but output values written to control
-        * structure (struct sg_io_hdr).  */
+       return blk_complete_sghdr_rq(rq, hdr, bio);
 out:
        blk_put_request(rq);
        return ret;
@@ -427,7 +468,7 @@ int sg_scsi_ioctl(struct file *file, struct request_queue *q,
        if (in_len && copy_from_user(buffer, sic->data + cmdlen, in_len))
                goto error;
 
-       err = verify_command(file, rq->cmd);
+       err = verify_command(rq->cmd, file->f_mode & FMODE_WRITE);
        if (err)
                goto error;
 
@@ -454,7 +495,7 @@ int sg_scsi_ioctl(struct file *file, struct request_queue *q,
                rq->retries = 1;
                break;
        default:
-               rq->timeout = BLK_DEFAULT_TIMEOUT;
+               rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
                break;
        }
 
@@ -501,7 +542,7 @@ static int __blk_send_generic(request_queue_t *q, struct gendisk *bd_disk, int c
        rq->cmd_type = REQ_TYPE_BLOCK_PC;
        rq->data = NULL;
        rq->data_len = 0;
-       rq->timeout = BLK_DEFAULT_TIMEOUT;
+       rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
        memset(rq->cmd, 0, sizeof(rq->cmd));
        rq->cmd[0] = cmd;
        rq->cmd[4] = data;
index f429be8..a21f585 100644 (file)
@@ -1258,19 +1258,25 @@ static void idefloppy_create_rw_cmd (idefloppy_floppy_t *floppy, idefloppy_pc_t
        set_bit(PC_DMA_RECOMMENDED, &pc->flags);
 }
 
-static int
+static void
 idefloppy_blockpc_cmd(idefloppy_floppy_t *floppy, idefloppy_pc_t *pc, struct request *rq)
 {
-       /*
-        * just support eject for now, it would not be hard to make the
-        * REQ_BLOCK_PC support fully-featured
-        */
-       if (rq->cmd[0] != IDEFLOPPY_START_STOP_CMD)
-               return 1;
-
        idefloppy_init_pc(pc);
+       pc->callback = &idefloppy_rw_callback;
        memcpy(pc->c, rq->cmd, sizeof(pc->c));
-       return 0;
+       pc->rq = rq;
+       pc->b_count = rq->data_len;
+       if (rq->data_len && rq_data_dir(rq) == WRITE)
+               set_bit(PC_WRITING, &pc->flags);
+       pc->buffer = rq->data;
+       if (rq->bio)
+               set_bit(PC_DMA_RECOMMENDED, &pc->flags);
+               
+       /*
+        * possibly problematic, doesn't look like ide-floppy correctly
+        * handled scattered requests if dma fails...
+        */
+       pc->request_transfer = pc->buffer_size = rq->data_len;
 }
 
 /*
@@ -1317,10 +1323,7 @@ static ide_startstop_t idefloppy_do_request (ide_drive_t *drive, struct request
                pc = (idefloppy_pc_t *) rq->buffer;
        } else if (blk_pc_request(rq)) {
                pc = idefloppy_next_pc_storage(drive);
-               if (idefloppy_blockpc_cmd(floppy, pc, rq)) {
-                       idefloppy_do_end_request(drive, 0, 0);
-                       return ide_stopped;
-               }
+               idefloppy_blockpc_cmd(floppy, pc, rq);
        } else {
                blk_dump_rq_flags(rq,
                        "ide-floppy: unsupported command in queue");
index c948a5c..9ae60a7 100644 (file)
@@ -1049,9 +1049,13 @@ int generic_ide_ioctl(ide_drive_t *drive, struct file *file, struct block_device
        unsigned long flags;
        ide_driver_t *drv;
        void __user *p = (void __user *)arg;
-       int err = 0, (*setfunc)(ide_drive_t *, int);
+       int err, (*setfunc)(ide_drive_t *, int);
        u8 *val;
 
+       err = scsi_cmd_ioctl(file, bdev->bd_disk, cmd, p);
+       if (err != -ENOTTY)
+               return err;
+
        switch (cmd) {
        case HDIO_GET_32BIT:        val = &drive->io_32bit;      goto read_val;
        case HDIO_GET_KEEPSETTINGS: val = &drive->keep_settings; goto read_val;
@@ -1171,10 +1175,6 @@ int generic_ide_ioctl(ide_drive_t *drive, struct file *file, struct block_device
                        return 0;
                }
 
-               case CDROMEJECT:
-               case CDROMCLOSETRAY:
-                       return scsi_cmd_ioctl(file, bdev->bd_disk, cmd, p);
-
                case HDIO_GET_BUSSTATE:
                        if (!capable(CAP_SYS_ADMIN))
                                return -EACCES;
index fae138b..53002d4 100644 (file)
@@ -41,6 +41,8 @@ struct elevator_queue;
 typedef struct elevator_queue elevator_t;
 struct request_pm_state;
 struct blk_trace;
+struct request;
+struct sg_io_hdr;
 
 #define BLKDEV_MIN_RQ  4
 #define BLKDEV_MAX_RQ  128     /* Default maximum */
@@ -607,6 +609,11 @@ extern unsigned long blk_max_low_pfn, blk_max_pfn;
 #define BLK_BOUNCE_ANY         ((u64)blk_max_pfn << PAGE_SHIFT)
 #define BLK_BOUNCE_ISA         (ISA_DMA_THRESHOLD)
 
+/*
+ * default timeout for SG_IO if none specified
+ */
+#define BLK_DEFAULT_SG_TIMEOUT (60 * HZ)
+
 #ifdef CONFIG_MMU
 extern int init_emergency_isa_pool(void);
 extern void blk_queue_bounce(request_queue_t *q, struct bio **bio);
@@ -680,6 +687,11 @@ extern int blk_execute_rq(request_queue_t *, struct gendisk *,
                          struct request *, int);
 extern void blk_execute_rq_nowait(request_queue_t *, struct gendisk *,
                                  struct request *, int, rq_end_io_fn *);
+extern int blk_fill_sghdr_rq(request_queue_t *, struct request *,
+                     struct sg_io_hdr *, int);
+extern int blk_unmap_sghdr_rq(struct request *, struct sg_io_hdr *);
+extern int blk_complete_sghdr_rq(struct request *, struct sg_io_hdr *,
+                         struct bio *);
 
 static inline request_queue_t *bdev_get_queue(struct block_device *bdev)
 {
diff --git a/include/linux/bsg.h b/include/linux/bsg.h
new file mode 100644 (file)
index 0000000..dc0d728
--- /dev/null
@@ -0,0 +1,21 @@
+#ifndef BSG_H
+#define BSG_H
+
+#if defined(CONFIG_BLK_DEV_BSG)
+struct bsg_class_device {
+       struct class_device *class_dev;
+       struct device *dev;
+       int minor;
+       struct gendisk *disk;
+       struct list_head list;
+};
+
+extern int bsg_register_disk(struct gendisk *);
+extern void bsg_unregister_disk(struct gendisk *);
+#else
+struct bsg_class_device { };
+#define bsg_register_disk(disk)                (0)
+#define bsg_unregister_disk(disk)      do { } while (0)
+#endif
+
+#endif
index 9756fc1..8c43d70 100644 (file)
@@ -67,6 +67,7 @@ struct partition {
 #include <linux/string.h>
 #include <linux/fs.h>
 #include <linux/workqueue.h>
+#include <linux/bsg.h>
 
 struct partition {
        unsigned char boot_ind;         /* 0x80 - active */
@@ -91,6 +92,7 @@ struct hd_struct {
 #ifdef CONFIG_FAIL_MAKE_REQUEST
        int make_it_fail;
 #endif
+       struct bsg_class_device bsg_dev;
 };
 
 #define GENHD_FL_REMOVABLE                     1