2 * bsg.c - block layer implementation of the sg v3 interface
4 * Copyright (C) 2004 Jens Axboe <axboe@suse.de> SUSE Labs
5 * Copyright (C) 2004 Peter M. Jones <pjones@redhat.com>
7 * This file is subject to the terms and conditions of the GNU General Public
8 * License version 2. See the file "COPYING" in the main directory of this
9 * archive for more details.
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/file.h>
15 #include <linux/blkdev.h>
16 #include <linux/poll.h>
17 #include <linux/cdev.h>
18 #include <linux/percpu.h>
19 #include <linux/uio.h>
20 #include <linux/idr.h>
21 #include <linux/bsg.h>
23 #include <scsi/scsi.h>
24 #include <scsi/scsi_ioctl.h>
25 #include <scsi/scsi_cmnd.h>
26 #include <scsi/scsi_device.h>
27 #include <scsi/scsi_driver.h>
30 #define BSG_DESCRIPTION "Block layer SCSI generic (bsg) driver"
31 #define BSG_VERSION "0.4"
34 struct request_queue *queue;
36 struct list_head busy_list;
37 struct list_head done_list;
38 struct hlist_node dev_list;
43 wait_queue_head_t wq_done;
44 wait_queue_head_t wq_free;
45 char name[BUS_ID_SIZE];
55 #define BSG_DEFAULT_CMDS 64
56 #define BSG_MAX_DEVS 32768
61 #define dprintk(fmt, args...) printk(KERN_ERR "%s: " fmt, __FUNCTION__, ##args)
63 #define dprintk(fmt, args...)
66 static DEFINE_MUTEX(bsg_mutex);
67 static DEFINE_IDR(bsg_minor_idr);
69 #define BSG_LIST_ARRAY_SIZE 8
70 static struct hlist_head bsg_device_list[BSG_LIST_ARRAY_SIZE];
72 static struct class *bsg_class;
75 static struct kmem_cache *bsg_cmd_cachep;
78 * our internal command type
81 struct bsg_device *bd;
82 struct list_head list;
88 char sense[SCSI_SENSE_BUFFERSIZE];
91 static void bsg_free_command(struct bsg_command *bc)
93 struct bsg_device *bd = bc->bd;
96 kmem_cache_free(bsg_cmd_cachep, bc);
98 spin_lock_irqsave(&bd->lock, flags);
100 spin_unlock_irqrestore(&bd->lock, flags);
102 wake_up(&bd->wq_free);
105 static struct bsg_command *bsg_alloc_command(struct bsg_device *bd)
107 struct bsg_command *bc = ERR_PTR(-EINVAL);
109 spin_lock_irq(&bd->lock);
111 if (bd->queued_cmds >= bd->max_queue)
115 spin_unlock_irq(&bd->lock);
117 bc = kmem_cache_zalloc(bsg_cmd_cachep, GFP_KERNEL);
119 spin_lock_irq(&bd->lock);
121 bc = ERR_PTR(-ENOMEM);
126 INIT_LIST_HEAD(&bc->list);
127 dprintk("%s: returning free cmd %p\n", bd->name, bc);
130 spin_unlock_irq(&bd->lock);
134 static inline struct hlist_head *bsg_dev_idx_hash(int index)
136 return &bsg_device_list[index & (BSG_LIST_ARRAY_SIZE - 1)];
139 static int bsg_io_schedule(struct bsg_device *bd)
144 spin_lock_irq(&bd->lock);
146 BUG_ON(bd->done_cmds > bd->queued_cmds);
149 * -ENOSPC or -ENODATA? I'm going for -ENODATA, meaning "I have no
150 * work to do", even though we return -ENOSPC after this same test
151 * during bsg_write() -- there, it means our buffer can't have more
152 * bsg_commands added to it, thus has no space left.
154 if (bd->done_cmds == bd->queued_cmds) {
159 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
164 prepare_to_wait(&bd->wq_done, &wait, TASK_UNINTERRUPTIBLE);
165 spin_unlock_irq(&bd->lock);
167 finish_wait(&bd->wq_done, &wait);
171 spin_unlock_irq(&bd->lock);
175 static int blk_fill_sgv4_hdr_rq(struct request_queue *q, struct request *rq,
176 struct sg_io_v4 *hdr, int has_write_perm)
178 memset(rq->cmd, 0, BLK_MAX_CDB); /* ATAPI hates garbage after CDB */
180 if (copy_from_user(rq->cmd, (void *)(unsigned long)hdr->request,
184 if (hdr->subprotocol == BSG_SUB_PROTOCOL_SCSI_CMD) {
185 if (blk_verify_command(rq->cmd, has_write_perm))
187 } else if (!capable(CAP_SYS_RAWIO))
191 * fill in request structure
193 rq->cmd_len = hdr->request_len;
194 rq->cmd_type = REQ_TYPE_BLOCK_PC;
196 rq->timeout = (hdr->timeout * HZ) / 1000;
198 rq->timeout = q->sg_timeout;
200 rq->timeout = BLK_DEFAULT_SG_TIMEOUT;
206 * Check if sg_io_v4 from user is allowed and valid
209 bsg_validate_sgv4_hdr(struct request_queue *q, struct sg_io_v4 *hdr, int *rw)
213 if (hdr->guard != 'Q')
215 if (hdr->request_len > BLK_MAX_CDB)
217 if (hdr->dout_xfer_len > (q->max_sectors << 9) ||
218 hdr->din_xfer_len > (q->max_sectors << 9))
221 switch (hdr->protocol) {
222 case BSG_PROTOCOL_SCSI:
223 switch (hdr->subprotocol) {
224 case BSG_SUB_PROTOCOL_SCSI_CMD:
225 case BSG_SUB_PROTOCOL_SCSI_TRANSPORT:
235 *rw = hdr->dout_xfer_len ? WRITE : READ;
240 * map sg_io_v4 to a request.
242 static struct request *
243 bsg_map_hdr(struct bsg_device *bd, struct sg_io_v4 *hdr)
245 struct request_queue *q = bd->queue;
246 struct request *rq, *next_rq = NULL;
248 unsigned int dxfer_len;
251 dprintk("map hdr %llx/%u %llx/%u\n", (unsigned long long) hdr->dout_xferp,
252 hdr->dout_xfer_len, (unsigned long long) hdr->din_xferp,
255 ret = bsg_validate_sgv4_hdr(q, hdr, &rw);
260 * map scatter-gather elements seperately and string them to request
262 rq = blk_get_request(q, rw, GFP_KERNEL);
264 return ERR_PTR(-ENOMEM);
265 ret = blk_fill_sgv4_hdr_rq(q, rq, hdr, test_bit(BSG_F_WRITE_PERM,
270 if (rw == WRITE && hdr->din_xfer_len) {
271 if (!test_bit(QUEUE_FLAG_BIDI, &q->queue_flags)) {
276 next_rq = blk_get_request(q, READ, GFP_KERNEL);
281 rq->next_rq = next_rq;
283 dxferp = (void*)(unsigned long)hdr->din_xferp;
284 ret = blk_rq_map_user(q, next_rq, dxferp, hdr->din_xfer_len);
289 if (hdr->dout_xfer_len) {
290 dxfer_len = hdr->dout_xfer_len;
291 dxferp = (void*)(unsigned long)hdr->dout_xferp;
292 } else if (hdr->din_xfer_len) {
293 dxfer_len = hdr->din_xfer_len;
294 dxferp = (void*)(unsigned long)hdr->din_xferp;
299 ret = blk_rq_map_user(q, rq, dxferp, dxfer_len);
307 blk_rq_unmap_user(next_rq->bio);
308 blk_put_request(next_rq);
314 * async completion call-back from the block layer, when scsi/ide/whatever
315 * calls end_that_request_last() on a request
317 static void bsg_rq_end_io(struct request *rq, int uptodate)
319 struct bsg_command *bc = rq->end_io_data;
320 struct bsg_device *bd = bc->bd;
323 dprintk("%s: finished rq %p bc %p, bio %p stat %d\n",
324 bd->name, rq, bc, bc->bio, uptodate);
326 bc->hdr.duration = jiffies_to_msecs(jiffies - bc->hdr.duration);
328 spin_lock_irqsave(&bd->lock, flags);
329 list_move_tail(&bc->list, &bd->done_list);
331 spin_unlock_irqrestore(&bd->lock, flags);
333 wake_up(&bd->wq_done);
337 * do final setup of a 'bc' and submit the matching 'rq' to the block
340 static void bsg_add_command(struct bsg_device *bd, struct request_queue *q,
341 struct bsg_command *bc, struct request *rq)
343 rq->sense = bc->sense;
347 * add bc command to busy queue and submit rq for io
352 bc->bidi_bio = rq->next_rq->bio;
353 bc->hdr.duration = jiffies;
354 spin_lock_irq(&bd->lock);
355 list_add_tail(&bc->list, &bd->busy_list);
356 spin_unlock_irq(&bd->lock);
358 dprintk("%s: queueing rq %p, bc %p\n", bd->name, rq, bc);
360 rq->end_io_data = bc;
361 blk_execute_rq_nowait(q, NULL, rq, 1, bsg_rq_end_io);
364 static struct bsg_command *bsg_next_done_cmd(struct bsg_device *bd)
366 struct bsg_command *bc = NULL;
368 spin_lock_irq(&bd->lock);
370 bc = list_entry(bd->done_list.next, struct bsg_command, list);
374 spin_unlock_irq(&bd->lock);
380 * Get a finished command from the done list
382 static struct bsg_command *bsg_get_done_cmd(struct bsg_device *bd)
384 struct bsg_command *bc;
388 bc = bsg_next_done_cmd(bd);
392 if (!test_bit(BSG_F_BLOCK, &bd->flags)) {
393 bc = ERR_PTR(-EAGAIN);
397 ret = wait_event_interruptible(bd->wq_done, bd->done_cmds);
399 bc = ERR_PTR(-ERESTARTSYS);
404 dprintk("%s: returning done %p\n", bd->name, bc);
409 static int blk_complete_sgv4_hdr_rq(struct request *rq, struct sg_io_v4 *hdr,
410 struct bio *bio, struct bio *bidi_bio)
414 dprintk("rq %p bio %p %u\n", rq, bio, rq->errors);
416 * fill in all the output members
418 hdr->device_status = status_byte(rq->errors);
419 hdr->transport_status = host_byte(rq->errors);
420 hdr->driver_status = driver_byte(rq->errors);
422 if (hdr->device_status || hdr->transport_status || hdr->driver_status)
423 hdr->info |= SG_INFO_CHECK;
424 hdr->din_resid = rq->data_len;
425 hdr->response_len = 0;
427 if (rq->sense_len && hdr->response) {
428 int len = min_t(unsigned int, hdr->max_response_len,
431 ret = copy_to_user((void*)(unsigned long)hdr->response,
434 hdr->response_len = len;
440 blk_rq_unmap_user(bidi_bio);
441 blk_put_request(rq->next_rq);
444 blk_rq_unmap_user(bio);
450 static int bsg_complete_all_commands(struct bsg_device *bd)
452 struct bsg_command *bc;
455 dprintk("%s: entered\n", bd->name);
457 set_bit(BSG_F_BLOCK, &bd->flags);
460 * wait for all commands to complete
464 ret = bsg_io_schedule(bd);
466 * look for -ENODATA specifically -- we'll sometimes get
467 * -ERESTARTSYS when we've taken a signal, but we can't
468 * return until we're done freeing the queue, so ignore
469 * it. The signal will get handled when we're done freeing
472 } while (ret != -ENODATA);
475 * discard done commands
479 spin_lock_irq(&bd->lock);
480 if (!bd->queued_cmds) {
481 spin_unlock_irq(&bd->lock);
484 spin_unlock_irq(&bd->lock);
486 bc = bsg_get_done_cmd(bd);
490 tret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
495 bsg_free_command(bc);
502 __bsg_read(char __user *buf, size_t count, struct bsg_device *bd,
503 const struct iovec *iov, ssize_t *bytes_read)
505 struct bsg_command *bc;
506 int nr_commands, ret;
508 if (count % sizeof(struct sg_io_v4))
512 nr_commands = count / sizeof(struct sg_io_v4);
513 while (nr_commands) {
514 bc = bsg_get_done_cmd(bd);
521 * this is the only case where we need to copy data back
522 * after completing the request. so do that here,
523 * bsg_complete_work() cannot do that for us
525 ret = blk_complete_sgv4_hdr_rq(bc->rq, &bc->hdr, bc->bio,
528 if (copy_to_user(buf, &bc->hdr, sizeof(bc->hdr)))
531 bsg_free_command(bc);
536 buf += sizeof(struct sg_io_v4);
537 *bytes_read += sizeof(struct sg_io_v4);
544 static inline void bsg_set_block(struct bsg_device *bd, struct file *file)
546 if (file->f_flags & O_NONBLOCK)
547 clear_bit(BSG_F_BLOCK, &bd->flags);
549 set_bit(BSG_F_BLOCK, &bd->flags);
552 static inline void bsg_set_write_perm(struct bsg_device *bd, struct file *file)
554 if (file->f_mode & FMODE_WRITE)
555 set_bit(BSG_F_WRITE_PERM, &bd->flags);
557 clear_bit(BSG_F_WRITE_PERM, &bd->flags);
561 * Check if the error is a "real" error that we should return.
563 static inline int err_block_err(int ret)
565 if (ret && ret != -ENOSPC && ret != -ENODATA && ret != -EAGAIN)
572 bsg_read(struct file *file, char __user *buf, size_t count, loff_t *ppos)
574 struct bsg_device *bd = file->private_data;
578 dprintk("%s: read %Zd bytes\n", bd->name, count);
580 bsg_set_block(bd, file);
582 ret = __bsg_read(buf, count, bd, NULL, &bytes_read);
585 if (!bytes_read || (bytes_read && err_block_err(ret)))
591 static int __bsg_write(struct bsg_device *bd, const char __user *buf,
592 size_t count, ssize_t *bytes_written)
594 struct bsg_command *bc;
596 int ret, nr_commands;
598 if (count % sizeof(struct sg_io_v4))
601 nr_commands = count / sizeof(struct sg_io_v4);
605 while (nr_commands) {
606 struct request_queue *q = bd->queue;
608 bc = bsg_alloc_command(bd);
615 if (copy_from_user(&bc->hdr, buf, sizeof(bc->hdr))) {
621 * get a request, fill in the blanks, and add to request queue
623 rq = bsg_map_hdr(bd, &bc->hdr);
630 bsg_add_command(bd, q, bc, rq);
634 buf += sizeof(struct sg_io_v4);
635 *bytes_written += sizeof(struct sg_io_v4);
639 bsg_free_command(bc);
645 bsg_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
647 struct bsg_device *bd = file->private_data;
648 ssize_t bytes_written;
651 dprintk("%s: write %Zd bytes\n", bd->name, count);
653 bsg_set_block(bd, file);
654 bsg_set_write_perm(bd, file);
657 ret = __bsg_write(bd, buf, count, &bytes_written);
658 *ppos = bytes_written;
661 * return bytes written on non-fatal errors
663 if (!bytes_written || (bytes_written && err_block_err(ret)))
666 dprintk("%s: returning %Zd\n", bd->name, bytes_written);
667 return bytes_written;
670 static struct bsg_device *bsg_alloc_device(void)
672 struct bsg_device *bd;
674 bd = kzalloc(sizeof(struct bsg_device), GFP_KERNEL);
678 spin_lock_init(&bd->lock);
680 bd->max_queue = BSG_DEFAULT_CMDS;
682 INIT_LIST_HEAD(&bd->busy_list);
683 INIT_LIST_HEAD(&bd->done_list);
684 INIT_HLIST_NODE(&bd->dev_list);
686 init_waitqueue_head(&bd->wq_free);
687 init_waitqueue_head(&bd->wq_done);
691 static int bsg_put_device(struct bsg_device *bd)
695 mutex_lock(&bsg_mutex);
697 if (!atomic_dec_and_test(&bd->ref_count))
700 dprintk("%s: tearing down\n", bd->name);
703 * close can always block
705 set_bit(BSG_F_BLOCK, &bd->flags);
708 * correct error detection baddies here again. it's the responsibility
709 * of the app to properly reap commands before close() if it wants
710 * fool-proof error detection
712 ret = bsg_complete_all_commands(bd);
714 blk_put_queue(bd->queue);
715 hlist_del(&bd->dev_list);
718 mutex_unlock(&bsg_mutex);
722 static struct bsg_device *bsg_add_device(struct inode *inode,
723 struct request_queue *rq,
726 struct bsg_device *bd;
728 unsigned char buf[32];
731 bd = bsg_alloc_device();
733 return ERR_PTR(-ENOMEM);
736 kobject_get(&rq->kobj);
737 bsg_set_block(bd, file);
739 atomic_set(&bd->ref_count, 1);
740 bd->minor = iminor(inode);
741 mutex_lock(&bsg_mutex);
742 hlist_add_head(&bd->dev_list, bsg_dev_idx_hash(bd->minor));
744 strncpy(bd->name, rq->bsg_dev.class_dev->class_id, sizeof(bd->name) - 1);
745 dprintk("bound to <%s>, max queue %d\n",
746 format_dev_t(buf, inode->i_rdev), bd->max_queue);
748 mutex_unlock(&bsg_mutex);
752 static struct bsg_device *__bsg_get_device(int minor)
754 struct bsg_device *bd = NULL;
755 struct hlist_node *entry;
757 mutex_lock(&bsg_mutex);
759 hlist_for_each(entry, bsg_dev_idx_hash(minor)) {
760 bd = hlist_entry(entry, struct bsg_device, dev_list);
761 if (bd->minor == minor) {
762 atomic_inc(&bd->ref_count);
769 mutex_unlock(&bsg_mutex);
773 static struct bsg_device *bsg_get_device(struct inode *inode, struct file *file)
775 struct bsg_device *bd;
776 struct bsg_class_device *bcd;
778 bd = __bsg_get_device(iminor(inode));
783 * find the class device
785 mutex_lock(&bsg_mutex);
786 bcd = idr_find(&bsg_minor_idr, iminor(inode));
787 mutex_unlock(&bsg_mutex);
790 return ERR_PTR(-ENODEV);
792 return bsg_add_device(inode, bcd->queue, file);
795 static int bsg_open(struct inode *inode, struct file *file)
797 struct bsg_device *bd = bsg_get_device(inode, file);
802 file->private_data = bd;
806 static int bsg_release(struct inode *inode, struct file *file)
808 struct bsg_device *bd = file->private_data;
810 file->private_data = NULL;
811 return bsg_put_device(bd);
814 static unsigned int bsg_poll(struct file *file, poll_table *wait)
816 struct bsg_device *bd = file->private_data;
817 unsigned int mask = 0;
819 poll_wait(file, &bd->wq_done, wait);
820 poll_wait(file, &bd->wq_free, wait);
822 spin_lock_irq(&bd->lock);
823 if (!list_empty(&bd->done_list))
824 mask |= POLLIN | POLLRDNORM;
825 if (bd->queued_cmds >= bd->max_queue)
827 spin_unlock_irq(&bd->lock);
832 static long bsg_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
834 struct bsg_device *bd = file->private_data;
835 int __user *uarg = (int __user *) arg;
841 case SG_GET_COMMAND_Q:
842 return put_user(bd->max_queue, uarg);
843 case SG_SET_COMMAND_Q: {
846 if (get_user(queue, uarg))
851 spin_lock_irq(&bd->lock);
852 bd->max_queue = queue;
853 spin_unlock_irq(&bd->lock);
860 case SG_GET_VERSION_NUM:
861 case SCSI_IOCTL_GET_IDLUN:
862 case SCSI_IOCTL_GET_BUS_NUMBER:
865 case SG_GET_RESERVED_SIZE:
866 case SG_SET_RESERVED_SIZE:
867 case SG_EMULATED_HOST:
868 case SCSI_IOCTL_SEND_COMMAND: {
869 void __user *uarg = (void __user *) arg;
870 return scsi_cmd_ioctl(file, bd->queue, NULL, cmd, uarg);
874 struct bio *bio, *bidi_bio = NULL;
877 if (copy_from_user(&hdr, uarg, sizeof(hdr)))
880 rq = bsg_map_hdr(bd, &hdr);
886 bidi_bio = rq->next_rq->bio;
887 blk_execute_rq(bd->queue, NULL, rq, 0);
888 blk_complete_sgv4_hdr_rq(rq, &hdr, bio, bidi_bio);
890 if (copy_to_user(uarg, &hdr, sizeof(hdr)))
896 * block device ioctls
900 return ioctl_by_bdev(bd->bdev, cmd, arg);
907 static struct file_operations bsg_fops = {
912 .release = bsg_release,
913 .unlocked_ioctl = bsg_ioctl,
914 .owner = THIS_MODULE,
917 void bsg_unregister_queue(struct request_queue *q)
919 struct bsg_class_device *bcd = &q->bsg_dev;
924 mutex_lock(&bsg_mutex);
925 idr_remove(&bsg_minor_idr, bcd->minor);
926 sysfs_remove_link(&q->kobj, "bsg");
927 class_device_unregister(bcd->class_dev);
928 put_device(bcd->dev);
929 bcd->class_dev = NULL;
931 mutex_unlock(&bsg_mutex);
933 EXPORT_SYMBOL_GPL(bsg_unregister_queue);
935 int bsg_register_queue(struct request_queue *q, struct device *gdev,
938 struct bsg_class_device *bcd;
941 struct class_device *class_dev = NULL;
947 devname = gdev->bus_id;
950 * we need a proper transport to send commands, not a stacked device
956 memset(bcd, 0, sizeof(*bcd));
958 mutex_lock(&bsg_mutex);
960 ret = idr_pre_get(&bsg_minor_idr, GFP_KERNEL);
966 ret = idr_get_new(&bsg_minor_idr, bcd, &minor);
970 if (minor >= BSG_MAX_DEVS) {
971 printk(KERN_ERR "bsg: too many bsg devices\n");
978 bcd->dev = get_device(gdev);
979 dev = MKDEV(bsg_major, bcd->minor);
980 class_dev = class_device_create(bsg_class, NULL, dev, gdev, "%s",
982 if (IS_ERR(class_dev)) {
983 ret = PTR_ERR(class_dev);
986 bcd->class_dev = class_dev;
989 ret = sysfs_create_link(&q->kobj, &bcd->class_dev->kobj, "bsg");
991 goto unregister_class_dev;
994 mutex_unlock(&bsg_mutex);
997 unregister_class_dev:
998 class_device_unregister(class_dev);
1002 idr_remove(&bsg_minor_idr, minor);
1004 mutex_unlock(&bsg_mutex);
1007 EXPORT_SYMBOL_GPL(bsg_register_queue);
1009 static struct cdev bsg_cdev = {
1010 .kobj = {.name = "bsg", },
1011 .owner = THIS_MODULE,
1014 static int __init bsg_init(void)
1019 bsg_cmd_cachep = kmem_cache_create("bsg_cmd",
1020 sizeof(struct bsg_command), 0, 0, NULL);
1021 if (!bsg_cmd_cachep) {
1022 printk(KERN_ERR "bsg: failed creating slab cache\n");
1026 for (i = 0; i < BSG_LIST_ARRAY_SIZE; i++)
1027 INIT_HLIST_HEAD(&bsg_device_list[i]);
1029 bsg_class = class_create(THIS_MODULE, "bsg");
1030 if (IS_ERR(bsg_class)) {
1031 ret = PTR_ERR(bsg_class);
1032 goto destroy_kmemcache;
1035 ret = alloc_chrdev_region(&devid, 0, BSG_MAX_DEVS, "bsg");
1037 goto destroy_bsg_class;
1039 bsg_major = MAJOR(devid);
1041 cdev_init(&bsg_cdev, &bsg_fops);
1042 ret = cdev_add(&bsg_cdev, MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1044 goto unregister_chrdev;
1046 printk(KERN_INFO BSG_DESCRIPTION " version " BSG_VERSION
1047 " loaded (major %d)\n", bsg_major);
1050 unregister_chrdev_region(MKDEV(bsg_major, 0), BSG_MAX_DEVS);
1052 class_destroy(bsg_class);
1054 kmem_cache_destroy(bsg_cmd_cachep);
1058 MODULE_AUTHOR("Jens Axboe");
1059 MODULE_DESCRIPTION(BSG_DESCRIPTION);
1060 MODULE_LICENSE("GPL");
1062 device_initcall(bsg_init);