2 * Copyright (C) 2000 Jens Axboe <axboe@suse.de>
3 * Copyright (C) 2001-2004 Peter Osterlund <petero2@telia.com>
4 * Copyright (C) 2006 Thomas Maier <balagi@justmail.de>
6 * May be copied or modified under the terms of the GNU General Public
7 * License. See linux/COPYING for more information.
9 * Packet writing layer for ATAPI and SCSI CD-RW, DVD+RW, DVD-RW and
12 * Theory of operation:
14 * At the lowest level, there is the standard driver for the CD/DVD device,
15 * typically ide-cd.c or sr.c. This driver can handle read and write requests,
16 * but it doesn't know anything about the special restrictions that apply to
17 * packet writing. One restriction is that write requests must be aligned to
18 * packet boundaries on the physical media, and the size of a write request
19 * must be equal to the packet size. Another restriction is that a
20 * GPCMD_FLUSH_CACHE command has to be issued to the drive before a read
21 * command, if the previous command was a write.
23 * The purpose of the packet writing driver is to hide these restrictions from
24 * higher layers, such as file systems, and present a block device that can be
25 * randomly read and written using 2kB-sized blocks.
27 * The lowest layer in the packet writing driver is the packet I/O scheduler.
28 * Its data is defined by the struct packet_iosched and includes two bio
29 * queues with pending read and write requests. These queues are processed
30 * by the pkt_iosched_process_queue() function. The write requests in this
31 * queue are already properly aligned and sized. This layer is responsible for
32 * issuing the flush cache commands and scheduling the I/O in a good order.
34 * The next layer transforms unaligned write requests to aligned writes. This
35 * transformation requires reading missing pieces of data from the underlying
36 * block device, assembling the pieces to full packets and queuing them to the
37 * packet I/O scheduler.
39 * At the top layer there is a custom make_request_fn function that forwards
40 * read requests directly to the iosched queue and puts write requests in the
41 * unaligned write queue. A kernel thread performs the necessary read
42 * gathering to convert the unaligned writes to aligned writes and then feeds
43 * them to the packet I/O scheduler.
45 *************************************************************************/
47 #include <linux/pktcdvd.h>
48 #include <linux/module.h>
49 #include <linux/types.h>
50 #include <linux/kernel.h>
51 #include <linux/kthread.h>
52 #include <linux/errno.h>
53 #include <linux/spinlock.h>
54 #include <linux/file.h>
55 #include <linux/proc_fs.h>
56 #include <linux/seq_file.h>
57 #include <linux/miscdevice.h>
58 #include <linux/freezer.h>
59 #include <linux/mutex.h>
60 #include <scsi/scsi_cmnd.h>
61 #include <scsi/scsi_ioctl.h>
62 #include <scsi/scsi.h>
63 #include <linux/debugfs.h>
64 #include <linux/device.h>
66 #include <asm/uaccess.h>
68 #define DRIVER_NAME "pktcdvd"
71 #define DPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
73 #define DPRINTK(fmt, args...)
77 #define VPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
79 #define VPRINTK(fmt, args...)
82 #define MAX_SPEED 0xffff
84 #define ZONE(sector, pd) (((sector) + (pd)->offset) & ~((pd)->settings.size - 1))
86 static struct pktcdvd_device *pkt_devs[MAX_WRITERS];
87 static struct proc_dir_entry *pkt_proc;
88 static int pktdev_major;
89 static int write_congestion_on = PKT_WRITE_CONGESTION_ON;
90 static int write_congestion_off = PKT_WRITE_CONGESTION_OFF;
91 static struct mutex ctl_mutex; /* Serialize open/close/setup/teardown */
92 static mempool_t *psd_pool;
94 static struct class *class_pktcdvd = NULL; /* /sys/class/pktcdvd */
95 static struct dentry *pkt_debugfs_root = NULL; /* /debug/pktcdvd */
97 /* forward declaration */
98 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev);
99 static int pkt_remove_dev(dev_t pkt_dev);
100 static int pkt_seq_show(struct seq_file *m, void *p);
105 * create and register a pktcdvd kernel object.
107 static struct pktcdvd_kobj* pkt_kobj_create(struct pktcdvd_device *pd,
109 struct kobject* parent,
110 struct kobj_type* ktype)
112 struct pktcdvd_kobj *p;
113 p = kzalloc(sizeof(*p), GFP_KERNEL);
116 kobject_set_name(&p->kobj, "%s", name);
117 p->kobj.parent = parent;
118 p->kobj.ktype = ktype;
120 if (kobject_register(&p->kobj) != 0)
125 * remove a pktcdvd kernel object.
127 static void pkt_kobj_remove(struct pktcdvd_kobj *p)
130 kobject_unregister(&p->kobj);
133 * default release function for pktcdvd kernel objects.
135 static void pkt_kobj_release(struct kobject *kobj)
137 kfree(to_pktcdvdkobj(kobj));
141 /**********************************************************
143 * sysfs interface for pktcdvd
144 * by (C) 2006 Thomas Maier <balagi@justmail.de>
146 **********************************************************/
148 #define DEF_ATTR(_obj,_name,_mode) \
149 static struct attribute _obj = { .name = _name, .mode = _mode }
151 /**********************************************************
152 /sys/class/pktcdvd/pktcdvd[0-7]/
155 stat/packets_finished
160 write_queue/congestion_off
161 write_queue/congestion_on
162 **********************************************************/
164 DEF_ATTR(kobj_pkt_attr_st1, "reset", 0200);
165 DEF_ATTR(kobj_pkt_attr_st2, "packets_started", 0444);
166 DEF_ATTR(kobj_pkt_attr_st3, "packets_finished", 0444);
167 DEF_ATTR(kobj_pkt_attr_st4, "kb_written", 0444);
168 DEF_ATTR(kobj_pkt_attr_st5, "kb_read", 0444);
169 DEF_ATTR(kobj_pkt_attr_st6, "kb_read_gather", 0444);
171 static struct attribute *kobj_pkt_attrs_stat[] = {
181 DEF_ATTR(kobj_pkt_attr_wq1, "size", 0444);
182 DEF_ATTR(kobj_pkt_attr_wq2, "congestion_off", 0644);
183 DEF_ATTR(kobj_pkt_attr_wq3, "congestion_on", 0644);
185 static struct attribute *kobj_pkt_attrs_wqueue[] = {
192 static ssize_t kobj_pkt_show(struct kobject *kobj,
193 struct attribute *attr, char *data)
195 struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
198 if (strcmp(attr->name, "packets_started") == 0) {
199 n = sprintf(data, "%lu\n", pd->stats.pkt_started);
201 } else if (strcmp(attr->name, "packets_finished") == 0) {
202 n = sprintf(data, "%lu\n", pd->stats.pkt_ended);
204 } else if (strcmp(attr->name, "kb_written") == 0) {
205 n = sprintf(data, "%lu\n", pd->stats.secs_w >> 1);
207 } else if (strcmp(attr->name, "kb_read") == 0) {
208 n = sprintf(data, "%lu\n", pd->stats.secs_r >> 1);
210 } else if (strcmp(attr->name, "kb_read_gather") == 0) {
211 n = sprintf(data, "%lu\n", pd->stats.secs_rg >> 1);
213 } else if (strcmp(attr->name, "size") == 0) {
214 spin_lock(&pd->lock);
215 v = pd->bio_queue_size;
216 spin_unlock(&pd->lock);
217 n = sprintf(data, "%d\n", v);
219 } else if (strcmp(attr->name, "congestion_off") == 0) {
220 spin_lock(&pd->lock);
221 v = pd->write_congestion_off;
222 spin_unlock(&pd->lock);
223 n = sprintf(data, "%d\n", v);
225 } else if (strcmp(attr->name, "congestion_on") == 0) {
226 spin_lock(&pd->lock);
227 v = pd->write_congestion_on;
228 spin_unlock(&pd->lock);
229 n = sprintf(data, "%d\n", v);
234 static void init_write_congestion_marks(int* lo, int* hi)
238 *hi = min(*hi, 1000000);
242 *lo = min(*lo, *hi - 100);
251 static ssize_t kobj_pkt_store(struct kobject *kobj,
252 struct attribute *attr,
253 const char *data, size_t len)
255 struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
258 if (strcmp(attr->name, "reset") == 0 && len > 0) {
259 pd->stats.pkt_started = 0;
260 pd->stats.pkt_ended = 0;
261 pd->stats.secs_w = 0;
262 pd->stats.secs_rg = 0;
263 pd->stats.secs_r = 0;
265 } else if (strcmp(attr->name, "congestion_off") == 0
266 && sscanf(data, "%d", &val) == 1) {
267 spin_lock(&pd->lock);
268 pd->write_congestion_off = val;
269 init_write_congestion_marks(&pd->write_congestion_off,
270 &pd->write_congestion_on);
271 spin_unlock(&pd->lock);
273 } else if (strcmp(attr->name, "congestion_on") == 0
274 && sscanf(data, "%d", &val) == 1) {
275 spin_lock(&pd->lock);
276 pd->write_congestion_on = val;
277 init_write_congestion_marks(&pd->write_congestion_off,
278 &pd->write_congestion_on);
279 spin_unlock(&pd->lock);
284 static struct sysfs_ops kobj_pkt_ops = {
285 .show = kobj_pkt_show,
286 .store = kobj_pkt_store
288 static struct kobj_type kobj_pkt_type_stat = {
289 .release = pkt_kobj_release,
290 .sysfs_ops = &kobj_pkt_ops,
291 .default_attrs = kobj_pkt_attrs_stat
293 static struct kobj_type kobj_pkt_type_wqueue = {
294 .release = pkt_kobj_release,
295 .sysfs_ops = &kobj_pkt_ops,
296 .default_attrs = kobj_pkt_attrs_wqueue
299 static void pkt_sysfs_dev_new(struct pktcdvd_device *pd)
302 pd->clsdev = class_device_create(class_pktcdvd,
304 NULL, "%s", pd->name);
305 if (IS_ERR(pd->clsdev))
309 pd->kobj_stat = pkt_kobj_create(pd, "stat",
311 &kobj_pkt_type_stat);
312 pd->kobj_wqueue = pkt_kobj_create(pd, "write_queue",
314 &kobj_pkt_type_wqueue);
318 static void pkt_sysfs_dev_remove(struct pktcdvd_device *pd)
320 pkt_kobj_remove(pd->kobj_stat);
321 pkt_kobj_remove(pd->kobj_wqueue);
323 class_device_destroy(class_pktcdvd, pd->pkt_dev);
327 /********************************************************************
330 remove unmap packet dev
331 device_map show mappings
332 *******************************************************************/
334 static void class_pktcdvd_release(struct class *cls)
338 static ssize_t class_pktcdvd_show_map(struct class *c, char *data)
342 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
343 for (idx = 0; idx < MAX_WRITERS; idx++) {
344 struct pktcdvd_device *pd = pkt_devs[idx];
347 n += sprintf(data+n, "%s %u:%u %u:%u\n",
349 MAJOR(pd->pkt_dev), MINOR(pd->pkt_dev),
350 MAJOR(pd->bdev->bd_dev),
351 MINOR(pd->bdev->bd_dev));
353 mutex_unlock(&ctl_mutex);
357 static ssize_t class_pktcdvd_store_add(struct class *c, const char *buf,
360 unsigned int major, minor;
362 if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
363 /* pkt_setup_dev() expects caller to hold reference to self */
364 if (!try_module_get(THIS_MODULE))
367 pkt_setup_dev(MKDEV(major, minor), NULL);
369 module_put(THIS_MODULE);
377 static ssize_t class_pktcdvd_store_remove(struct class *c, const char *buf,
380 unsigned int major, minor;
381 if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
382 pkt_remove_dev(MKDEV(major, minor));
388 static struct class_attribute class_pktcdvd_attrs[] = {
389 __ATTR(add, 0200, NULL, class_pktcdvd_store_add),
390 __ATTR(remove, 0200, NULL, class_pktcdvd_store_remove),
391 __ATTR(device_map, 0444, class_pktcdvd_show_map, NULL),
396 static int pkt_sysfs_init(void)
401 * create control files in sysfs
402 * /sys/class/pktcdvd/...
404 class_pktcdvd = kzalloc(sizeof(*class_pktcdvd), GFP_KERNEL);
407 class_pktcdvd->name = DRIVER_NAME;
408 class_pktcdvd->owner = THIS_MODULE;
409 class_pktcdvd->class_release = class_pktcdvd_release;
410 class_pktcdvd->class_attrs = class_pktcdvd_attrs;
411 ret = class_register(class_pktcdvd);
413 kfree(class_pktcdvd);
414 class_pktcdvd = NULL;
415 printk(DRIVER_NAME": failed to create class pktcdvd\n");
421 static void pkt_sysfs_cleanup(void)
424 class_destroy(class_pktcdvd);
425 class_pktcdvd = NULL;
428 /********************************************************************
431 /debugfs/pktcdvd[0-7]/
434 *******************************************************************/
436 static int pkt_debugfs_seq_show(struct seq_file *m, void *p)
438 return pkt_seq_show(m, p);
441 static int pkt_debugfs_fops_open(struct inode *inode, struct file *file)
443 return single_open(file, pkt_debugfs_seq_show, inode->i_private);
446 static const struct file_operations debug_fops = {
447 .open = pkt_debugfs_fops_open,
450 .release = single_release,
451 .owner = THIS_MODULE,
454 static void pkt_debugfs_dev_new(struct pktcdvd_device *pd)
456 if (!pkt_debugfs_root)
458 pd->dfs_f_info = NULL;
459 pd->dfs_d_root = debugfs_create_dir(pd->name, pkt_debugfs_root);
460 if (IS_ERR(pd->dfs_d_root)) {
461 pd->dfs_d_root = NULL;
464 pd->dfs_f_info = debugfs_create_file("info", S_IRUGO,
465 pd->dfs_d_root, pd, &debug_fops);
466 if (IS_ERR(pd->dfs_f_info)) {
467 pd->dfs_f_info = NULL;
472 static void pkt_debugfs_dev_remove(struct pktcdvd_device *pd)
474 if (!pkt_debugfs_root)
477 debugfs_remove(pd->dfs_f_info);
478 pd->dfs_f_info = NULL;
480 debugfs_remove(pd->dfs_d_root);
481 pd->dfs_d_root = NULL;
484 static void pkt_debugfs_init(void)
486 pkt_debugfs_root = debugfs_create_dir(DRIVER_NAME, NULL);
487 if (IS_ERR(pkt_debugfs_root)) {
488 pkt_debugfs_root = NULL;
493 static void pkt_debugfs_cleanup(void)
495 if (!pkt_debugfs_root)
497 debugfs_remove(pkt_debugfs_root);
498 pkt_debugfs_root = NULL;
501 /* ----------------------------------------------------------*/
504 static void pkt_bio_finished(struct pktcdvd_device *pd)
506 BUG_ON(atomic_read(&pd->cdrw.pending_bios) <= 0);
507 if (atomic_dec_and_test(&pd->cdrw.pending_bios)) {
508 VPRINTK(DRIVER_NAME": queue empty\n");
509 atomic_set(&pd->iosched.attention, 1);
510 wake_up(&pd->wqueue);
514 static void pkt_bio_destructor(struct bio *bio)
516 kfree(bio->bi_io_vec);
520 static struct bio *pkt_bio_alloc(int nr_iovecs)
522 struct bio_vec *bvl = NULL;
525 bio = kmalloc(sizeof(struct bio), GFP_KERNEL);
530 bvl = kcalloc(nr_iovecs, sizeof(struct bio_vec), GFP_KERNEL);
534 bio->bi_max_vecs = nr_iovecs;
535 bio->bi_io_vec = bvl;
536 bio->bi_destructor = pkt_bio_destructor;
547 * Allocate a packet_data struct
549 static struct packet_data *pkt_alloc_packet_data(int frames)
552 struct packet_data *pkt;
554 pkt = kzalloc(sizeof(struct packet_data), GFP_KERNEL);
558 pkt->frames = frames;
559 pkt->w_bio = pkt_bio_alloc(frames);
563 for (i = 0; i < frames / FRAMES_PER_PAGE; i++) {
564 pkt->pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
569 spin_lock_init(&pkt->lock);
571 for (i = 0; i < frames; i++) {
572 struct bio *bio = pkt_bio_alloc(1);
575 pkt->r_bios[i] = bio;
581 for (i = 0; i < frames; i++) {
582 struct bio *bio = pkt->r_bios[i];
588 for (i = 0; i < frames / FRAMES_PER_PAGE; i++)
590 __free_page(pkt->pages[i]);
599 * Free a packet_data struct
601 static void pkt_free_packet_data(struct packet_data *pkt)
605 for (i = 0; i < pkt->frames; i++) {
606 struct bio *bio = pkt->r_bios[i];
610 for (i = 0; i < pkt->frames / FRAMES_PER_PAGE; i++)
611 __free_page(pkt->pages[i]);
616 static void pkt_shrink_pktlist(struct pktcdvd_device *pd)
618 struct packet_data *pkt, *next;
620 BUG_ON(!list_empty(&pd->cdrw.pkt_active_list));
622 list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_free_list, list) {
623 pkt_free_packet_data(pkt);
625 INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
628 static int pkt_grow_pktlist(struct pktcdvd_device *pd, int nr_packets)
630 struct packet_data *pkt;
632 BUG_ON(!list_empty(&pd->cdrw.pkt_free_list));
634 while (nr_packets > 0) {
635 pkt = pkt_alloc_packet_data(pd->settings.size >> 2);
637 pkt_shrink_pktlist(pd);
640 pkt->id = nr_packets;
642 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
648 static inline struct pkt_rb_node *pkt_rbtree_next(struct pkt_rb_node *node)
650 struct rb_node *n = rb_next(&node->rb_node);
653 return rb_entry(n, struct pkt_rb_node, rb_node);
656 static void pkt_rbtree_erase(struct pktcdvd_device *pd, struct pkt_rb_node *node)
658 rb_erase(&node->rb_node, &pd->bio_queue);
659 mempool_free(node, pd->rb_pool);
660 pd->bio_queue_size--;
661 BUG_ON(pd->bio_queue_size < 0);
665 * Find the first node in the pd->bio_queue rb tree with a starting sector >= s.
667 static struct pkt_rb_node *pkt_rbtree_find(struct pktcdvd_device *pd, sector_t s)
669 struct rb_node *n = pd->bio_queue.rb_node;
670 struct rb_node *next;
671 struct pkt_rb_node *tmp;
674 BUG_ON(pd->bio_queue_size > 0);
679 tmp = rb_entry(n, struct pkt_rb_node, rb_node);
680 if (s <= tmp->bio->bi_sector)
689 if (s > tmp->bio->bi_sector) {
690 tmp = pkt_rbtree_next(tmp);
694 BUG_ON(s > tmp->bio->bi_sector);
699 * Insert a node into the pd->bio_queue rb tree.
701 static void pkt_rbtree_insert(struct pktcdvd_device *pd, struct pkt_rb_node *node)
703 struct rb_node **p = &pd->bio_queue.rb_node;
704 struct rb_node *parent = NULL;
705 sector_t s = node->bio->bi_sector;
706 struct pkt_rb_node *tmp;
710 tmp = rb_entry(parent, struct pkt_rb_node, rb_node);
711 if (s < tmp->bio->bi_sector)
716 rb_link_node(&node->rb_node, parent, p);
717 rb_insert_color(&node->rb_node, &pd->bio_queue);
718 pd->bio_queue_size++;
722 * Add a bio to a single linked list defined by its head and tail pointers.
724 static void pkt_add_list_last(struct bio *bio, struct bio **list_head, struct bio **list_tail)
728 BUG_ON((*list_head) == NULL);
729 (*list_tail)->bi_next = bio;
732 BUG_ON((*list_head) != NULL);
739 * Remove and return the first bio from a single linked list defined by its
740 * head and tail pointers.
742 static inline struct bio *pkt_get_list_first(struct bio **list_head, struct bio **list_tail)
746 if (*list_head == NULL)
750 *list_head = bio->bi_next;
751 if (*list_head == NULL)
759 * Send a packet_command to the underlying block device and
760 * wait for completion.
762 static int pkt_generic_packet(struct pktcdvd_device *pd, struct packet_command *cgc)
764 struct request_queue *q = bdev_get_queue(pd->bdev);
768 rq = blk_get_request(q, (cgc->data_direction == CGC_DATA_WRITE) ?
769 WRITE : READ, __GFP_WAIT);
772 if (blk_rq_map_kern(q, rq, cgc->buffer, cgc->buflen, __GFP_WAIT))
776 rq->cmd_len = COMMAND_SIZE(cgc->cmd[0]);
777 memcpy(rq->cmd, cgc->cmd, CDROM_PACKET_SIZE);
778 if (sizeof(rq->cmd) > CDROM_PACKET_SIZE)
779 memset(rq->cmd + CDROM_PACKET_SIZE, 0, sizeof(rq->cmd) - CDROM_PACKET_SIZE);
782 rq->cmd_type = REQ_TYPE_BLOCK_PC;
783 rq->cmd_flags |= REQ_HARDBARRIER;
785 rq->cmd_flags |= REQ_QUIET;
787 blk_execute_rq(rq->q, pd->bdev->bd_disk, rq, 0);
796 * A generic sense dump / resolve mechanism should be implemented across
797 * all ATAPI + SCSI devices.
799 static void pkt_dump_sense(struct packet_command *cgc)
801 static char *info[9] = { "No sense", "Recovered error", "Not ready",
802 "Medium error", "Hardware error", "Illegal request",
803 "Unit attention", "Data protect", "Blank check" };
805 struct request_sense *sense = cgc->sense;
807 printk(DRIVER_NAME":");
808 for (i = 0; i < CDROM_PACKET_SIZE; i++)
809 printk(" %02x", cgc->cmd[i]);
813 printk("no sense\n");
817 printk("sense %02x.%02x.%02x", sense->sense_key, sense->asc, sense->ascq);
819 if (sense->sense_key > 8) {
820 printk(" (INVALID)\n");
824 printk(" (%s)\n", info[sense->sense_key]);
828 * flush the drive cache to media
830 static int pkt_flush_cache(struct pktcdvd_device *pd)
832 struct packet_command cgc;
834 init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
835 cgc.cmd[0] = GPCMD_FLUSH_CACHE;
839 * the IMMED bit -- we default to not setting it, although that
840 * would allow a much faster close, this is safer
845 return pkt_generic_packet(pd, &cgc);
849 * speed is given as the normal factor, e.g. 4 for 4x
851 static int pkt_set_speed(struct pktcdvd_device *pd, unsigned write_speed, unsigned read_speed)
853 struct packet_command cgc;
854 struct request_sense sense;
857 init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
859 cgc.cmd[0] = GPCMD_SET_SPEED;
860 cgc.cmd[2] = (read_speed >> 8) & 0xff;
861 cgc.cmd[3] = read_speed & 0xff;
862 cgc.cmd[4] = (write_speed >> 8) & 0xff;
863 cgc.cmd[5] = write_speed & 0xff;
865 if ((ret = pkt_generic_packet(pd, &cgc)))
866 pkt_dump_sense(&cgc);
872 * Queue a bio for processing by the low-level CD device. Must be called
873 * from process context.
875 static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio)
877 spin_lock(&pd->iosched.lock);
878 if (bio_data_dir(bio) == READ) {
879 pkt_add_list_last(bio, &pd->iosched.read_queue,
880 &pd->iosched.read_queue_tail);
882 pkt_add_list_last(bio, &pd->iosched.write_queue,
883 &pd->iosched.write_queue_tail);
885 spin_unlock(&pd->iosched.lock);
887 atomic_set(&pd->iosched.attention, 1);
888 wake_up(&pd->wqueue);
892 * Process the queued read/write requests. This function handles special
893 * requirements for CDRW drives:
894 * - A cache flush command must be inserted before a read request if the
895 * previous request was a write.
896 * - Switching between reading and writing is slow, so don't do it more often
898 * - Optimize for throughput at the expense of latency. This means that streaming
899 * writes will never be interrupted by a read, but if the drive has to seek
900 * before the next write, switch to reading instead if there are any pending
902 * - Set the read speed according to current usage pattern. When only reading
903 * from the device, it's best to use the highest possible read speed, but
904 * when switching often between reading and writing, it's better to have the
905 * same read and write speeds.
907 static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
910 if (atomic_read(&pd->iosched.attention) == 0)
912 atomic_set(&pd->iosched.attention, 0);
916 int reads_queued, writes_queued;
918 spin_lock(&pd->iosched.lock);
919 reads_queued = (pd->iosched.read_queue != NULL);
920 writes_queued = (pd->iosched.write_queue != NULL);
921 spin_unlock(&pd->iosched.lock);
923 if (!reads_queued && !writes_queued)
926 if (pd->iosched.writing) {
927 int need_write_seek = 1;
928 spin_lock(&pd->iosched.lock);
929 bio = pd->iosched.write_queue;
930 spin_unlock(&pd->iosched.lock);
931 if (bio && (bio->bi_sector == pd->iosched.last_write))
933 if (need_write_seek && reads_queued) {
934 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
935 VPRINTK(DRIVER_NAME": write, waiting\n");
939 pd->iosched.writing = 0;
942 if (!reads_queued && writes_queued) {
943 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
944 VPRINTK(DRIVER_NAME": read, waiting\n");
947 pd->iosched.writing = 1;
951 spin_lock(&pd->iosched.lock);
952 if (pd->iosched.writing) {
953 bio = pkt_get_list_first(&pd->iosched.write_queue,
954 &pd->iosched.write_queue_tail);
956 bio = pkt_get_list_first(&pd->iosched.read_queue,
957 &pd->iosched.read_queue_tail);
959 spin_unlock(&pd->iosched.lock);
964 if (bio_data_dir(bio) == READ)
965 pd->iosched.successive_reads += bio->bi_size >> 10;
967 pd->iosched.successive_reads = 0;
968 pd->iosched.last_write = bio->bi_sector + bio_sectors(bio);
970 if (pd->iosched.successive_reads >= HI_SPEED_SWITCH) {
971 if (pd->read_speed == pd->write_speed) {
972 pd->read_speed = MAX_SPEED;
973 pkt_set_speed(pd, pd->write_speed, pd->read_speed);
976 if (pd->read_speed != pd->write_speed) {
977 pd->read_speed = pd->write_speed;
978 pkt_set_speed(pd, pd->write_speed, pd->read_speed);
982 atomic_inc(&pd->cdrw.pending_bios);
983 generic_make_request(bio);
988 * Special care is needed if the underlying block device has a small
989 * max_phys_segments value.
991 static int pkt_set_segment_merging(struct pktcdvd_device *pd, struct request_queue *q)
993 if ((pd->settings.size << 9) / CD_FRAMESIZE <= q->max_phys_segments) {
995 * The cdrom device can handle one segment/frame
997 clear_bit(PACKET_MERGE_SEGS, &pd->flags);
999 } else if ((pd->settings.size << 9) / PAGE_SIZE <= q->max_phys_segments) {
1001 * We can handle this case at the expense of some extra memory
1002 * copies during write operations
1004 set_bit(PACKET_MERGE_SEGS, &pd->flags);
1007 printk(DRIVER_NAME": cdrom max_phys_segments too small\n");
1013 * Copy CD_FRAMESIZE bytes from src_bio into a destination page
1015 static void pkt_copy_bio_data(struct bio *src_bio, int seg, int offs, struct page *dst_page, int dst_offs)
1017 unsigned int copy_size = CD_FRAMESIZE;
1019 while (copy_size > 0) {
1020 struct bio_vec *src_bvl = bio_iovec_idx(src_bio, seg);
1021 void *vfrom = kmap_atomic(src_bvl->bv_page, KM_USER0) +
1022 src_bvl->bv_offset + offs;
1023 void *vto = page_address(dst_page) + dst_offs;
1024 int len = min_t(int, copy_size, src_bvl->bv_len - offs);
1027 memcpy(vto, vfrom, len);
1028 kunmap_atomic(vfrom, KM_USER0);
1038 * Copy all data for this packet to pkt->pages[], so that
1039 * a) The number of required segments for the write bio is minimized, which
1040 * is necessary for some scsi controllers.
1041 * b) The data can be used as cache to avoid read requests if we receive a
1042 * new write request for the same zone.
1044 static void pkt_make_local_copy(struct packet_data *pkt, struct bio_vec *bvec)
1048 /* Copy all data to pkt->pages[] */
1051 for (f = 0; f < pkt->frames; f++) {
1052 if (bvec[f].bv_page != pkt->pages[p]) {
1053 void *vfrom = kmap_atomic(bvec[f].bv_page, KM_USER0) + bvec[f].bv_offset;
1054 void *vto = page_address(pkt->pages[p]) + offs;
1055 memcpy(vto, vfrom, CD_FRAMESIZE);
1056 kunmap_atomic(vfrom, KM_USER0);
1057 bvec[f].bv_page = pkt->pages[p];
1058 bvec[f].bv_offset = offs;
1060 BUG_ON(bvec[f].bv_offset != offs);
1062 offs += CD_FRAMESIZE;
1063 if (offs >= PAGE_SIZE) {
1070 static void pkt_end_io_read(struct bio *bio, int err)
1072 struct packet_data *pkt = bio->bi_private;
1073 struct pktcdvd_device *pd = pkt->pd;
1076 VPRINTK("pkt_end_io_read: bio=%p sec0=%llx sec=%llx err=%d\n", bio,
1077 (unsigned long long)pkt->sector, (unsigned long long)bio->bi_sector, err);
1080 atomic_inc(&pkt->io_errors);
1081 if (atomic_dec_and_test(&pkt->io_wait)) {
1082 atomic_inc(&pkt->run_sm);
1083 wake_up(&pd->wqueue);
1085 pkt_bio_finished(pd);
1088 static void pkt_end_io_packet_write(struct bio *bio, int err)
1090 struct packet_data *pkt = bio->bi_private;
1091 struct pktcdvd_device *pd = pkt->pd;
1094 VPRINTK("pkt_end_io_packet_write: id=%d, err=%d\n", pkt->id, err);
1096 pd->stats.pkt_ended++;
1098 pkt_bio_finished(pd);
1099 atomic_dec(&pkt->io_wait);
1100 atomic_inc(&pkt->run_sm);
1101 wake_up(&pd->wqueue);
1105 * Schedule reads for the holes in a packet
1107 static void pkt_gather_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1109 int frames_read = 0;
1112 char written[PACKET_MAX_SIZE];
1114 BUG_ON(!pkt->orig_bios);
1116 atomic_set(&pkt->io_wait, 0);
1117 atomic_set(&pkt->io_errors, 0);
1120 * Figure out which frames we need to read before we can write.
1122 memset(written, 0, sizeof(written));
1123 spin_lock(&pkt->lock);
1124 for (bio = pkt->orig_bios; bio; bio = bio->bi_next) {
1125 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1126 int num_frames = bio->bi_size / CD_FRAMESIZE;
1127 pd->stats.secs_w += num_frames * (CD_FRAMESIZE >> 9);
1128 BUG_ON(first_frame < 0);
1129 BUG_ON(first_frame + num_frames > pkt->frames);
1130 for (f = first_frame; f < first_frame + num_frames; f++)
1133 spin_unlock(&pkt->lock);
1135 if (pkt->cache_valid) {
1136 VPRINTK("pkt_gather_data: zone %llx cached\n",
1137 (unsigned long long)pkt->sector);
1142 * Schedule reads for missing parts of the packet.
1144 for (f = 0; f < pkt->frames; f++) {
1145 struct bio_vec *vec;
1150 bio = pkt->r_bios[f];
1151 vec = bio->bi_io_vec;
1153 bio->bi_max_vecs = 1;
1154 bio->bi_sector = pkt->sector + f * (CD_FRAMESIZE >> 9);
1155 bio->bi_bdev = pd->bdev;
1156 bio->bi_end_io = pkt_end_io_read;
1157 bio->bi_private = pkt;
1158 bio->bi_io_vec = vec;
1159 bio->bi_destructor = pkt_bio_destructor;
1161 p = (f * CD_FRAMESIZE) / PAGE_SIZE;
1162 offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1163 VPRINTK("pkt_gather_data: Adding frame %d, page:%p offs:%d\n",
1164 f, pkt->pages[p], offset);
1165 if (!bio_add_page(bio, pkt->pages[p], CD_FRAMESIZE, offset))
1168 atomic_inc(&pkt->io_wait);
1170 pkt_queue_bio(pd, bio);
1175 VPRINTK("pkt_gather_data: need %d frames for zone %llx\n",
1176 frames_read, (unsigned long long)pkt->sector);
1177 pd->stats.pkt_started++;
1178 pd->stats.secs_rg += frames_read * (CD_FRAMESIZE >> 9);
1182 * Find a packet matching zone, or the least recently used packet if
1183 * there is no match.
1185 static struct packet_data *pkt_get_packet_data(struct pktcdvd_device *pd, int zone)
1187 struct packet_data *pkt;
1189 list_for_each_entry(pkt, &pd->cdrw.pkt_free_list, list) {
1190 if (pkt->sector == zone || pkt->list.next == &pd->cdrw.pkt_free_list) {
1191 list_del_init(&pkt->list);
1192 if (pkt->sector != zone)
1193 pkt->cache_valid = 0;
1201 static void pkt_put_packet_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1203 if (pkt->cache_valid) {
1204 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
1206 list_add_tail(&pkt->list, &pd->cdrw.pkt_free_list);
1211 * recover a failed write, query for relocation if possible
1213 * returns 1 if recovery is possible, or 0 if not
1216 static int pkt_start_recovery(struct packet_data *pkt)
1219 * FIXME. We need help from the file system to implement
1220 * recovery handling.
1224 struct request *rq = pkt->rq;
1225 struct pktcdvd_device *pd = rq->rq_disk->private_data;
1226 struct block_device *pkt_bdev;
1227 struct super_block *sb = NULL;
1228 unsigned long old_block, new_block;
1229 sector_t new_sector;
1231 pkt_bdev = bdget(kdev_t_to_nr(pd->pkt_dev));
1233 sb = get_super(pkt_bdev);
1240 if (!sb->s_op || !sb->s_op->relocate_blocks)
1243 old_block = pkt->sector / (CD_FRAMESIZE >> 9);
1244 if (sb->s_op->relocate_blocks(sb, old_block, &new_block))
1247 new_sector = new_block * (CD_FRAMESIZE >> 9);
1248 pkt->sector = new_sector;
1250 pkt->bio->bi_sector = new_sector;
1251 pkt->bio->bi_next = NULL;
1252 pkt->bio->bi_flags = 1 << BIO_UPTODATE;
1253 pkt->bio->bi_idx = 0;
1255 BUG_ON(pkt->bio->bi_rw != (1 << BIO_RW));
1256 BUG_ON(pkt->bio->bi_vcnt != pkt->frames);
1257 BUG_ON(pkt->bio->bi_size != pkt->frames * CD_FRAMESIZE);
1258 BUG_ON(pkt->bio->bi_end_io != pkt_end_io_packet_write);
1259 BUG_ON(pkt->bio->bi_private != pkt);
1270 static inline void pkt_set_state(struct packet_data *pkt, enum packet_data_state state)
1272 #if PACKET_DEBUG > 1
1273 static const char *state_name[] = {
1274 "IDLE", "WAITING", "READ_WAIT", "WRITE_WAIT", "RECOVERY", "FINISHED"
1276 enum packet_data_state old_state = pkt->state;
1277 VPRINTK("pkt %2d : s=%6llx %s -> %s\n", pkt->id, (unsigned long long)pkt->sector,
1278 state_name[old_state], state_name[state]);
1284 * Scan the work queue to see if we can start a new packet.
1285 * returns non-zero if any work was done.
1287 static int pkt_handle_queue(struct pktcdvd_device *pd)
1289 struct packet_data *pkt, *p;
1290 struct bio *bio = NULL;
1291 sector_t zone = 0; /* Suppress gcc warning */
1292 struct pkt_rb_node *node, *first_node;
1296 VPRINTK("handle_queue\n");
1298 atomic_set(&pd->scan_queue, 0);
1300 if (list_empty(&pd->cdrw.pkt_free_list)) {
1301 VPRINTK("handle_queue: no pkt\n");
1306 * Try to find a zone we are not already working on.
1308 spin_lock(&pd->lock);
1309 first_node = pkt_rbtree_find(pd, pd->current_sector);
1311 n = rb_first(&pd->bio_queue);
1313 first_node = rb_entry(n, struct pkt_rb_node, rb_node);
1318 zone = ZONE(bio->bi_sector, pd);
1319 list_for_each_entry(p, &pd->cdrw.pkt_active_list, list) {
1320 if (p->sector == zone) {
1327 node = pkt_rbtree_next(node);
1329 n = rb_first(&pd->bio_queue);
1331 node = rb_entry(n, struct pkt_rb_node, rb_node);
1333 if (node == first_node)
1336 spin_unlock(&pd->lock);
1338 VPRINTK("handle_queue: no bio\n");
1342 pkt = pkt_get_packet_data(pd, zone);
1344 pd->current_sector = zone + pd->settings.size;
1346 BUG_ON(pkt->frames != pd->settings.size >> 2);
1347 pkt->write_size = 0;
1350 * Scan work queue for bios in the same zone and link them
1353 spin_lock(&pd->lock);
1354 VPRINTK("pkt_handle_queue: looking for zone %llx\n", (unsigned long long)zone);
1355 while ((node = pkt_rbtree_find(pd, zone)) != NULL) {
1357 VPRINTK("pkt_handle_queue: found zone=%llx\n",
1358 (unsigned long long)ZONE(bio->bi_sector, pd));
1359 if (ZONE(bio->bi_sector, pd) != zone)
1361 pkt_rbtree_erase(pd, node);
1362 spin_lock(&pkt->lock);
1363 pkt_add_list_last(bio, &pkt->orig_bios, &pkt->orig_bios_tail);
1364 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
1365 spin_unlock(&pkt->lock);
1367 /* check write congestion marks, and if bio_queue_size is
1368 below, wake up any waiters */
1369 wakeup = (pd->write_congestion_on > 0
1370 && pd->bio_queue_size <= pd->write_congestion_off);
1371 spin_unlock(&pd->lock);
1373 clear_bdi_congested(&pd->disk->queue->backing_dev_info, WRITE);
1375 pkt->sleep_time = max(PACKET_WAIT_TIME, 1);
1376 pkt_set_state(pkt, PACKET_WAITING_STATE);
1377 atomic_set(&pkt->run_sm, 1);
1379 spin_lock(&pd->cdrw.active_list_lock);
1380 list_add(&pkt->list, &pd->cdrw.pkt_active_list);
1381 spin_unlock(&pd->cdrw.active_list_lock);
1387 * Assemble a bio to write one packet and queue the bio for processing
1388 * by the underlying block device.
1390 static void pkt_start_write(struct pktcdvd_device *pd, struct packet_data *pkt)
1395 struct bio_vec *bvec = pkt->w_bio->bi_io_vec;
1397 for (f = 0; f < pkt->frames; f++) {
1398 bvec[f].bv_page = pkt->pages[(f * CD_FRAMESIZE) / PAGE_SIZE];
1399 bvec[f].bv_offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1403 * Fill-in bvec with data from orig_bios.
1406 spin_lock(&pkt->lock);
1407 for (bio = pkt->orig_bios; bio; bio = bio->bi_next) {
1408 int segment = bio->bi_idx;
1410 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1411 int num_frames = bio->bi_size / CD_FRAMESIZE;
1412 BUG_ON(first_frame < 0);
1413 BUG_ON(first_frame + num_frames > pkt->frames);
1414 for (f = first_frame; f < first_frame + num_frames; f++) {
1415 struct bio_vec *src_bvl = bio_iovec_idx(bio, segment);
1417 while (src_offs >= src_bvl->bv_len) {
1418 src_offs -= src_bvl->bv_len;
1420 BUG_ON(segment >= bio->bi_vcnt);
1421 src_bvl = bio_iovec_idx(bio, segment);
1424 if (src_bvl->bv_len - src_offs >= CD_FRAMESIZE) {
1425 bvec[f].bv_page = src_bvl->bv_page;
1426 bvec[f].bv_offset = src_bvl->bv_offset + src_offs;
1428 pkt_copy_bio_data(bio, segment, src_offs,
1429 bvec[f].bv_page, bvec[f].bv_offset);
1431 src_offs += CD_FRAMESIZE;
1435 pkt_set_state(pkt, PACKET_WRITE_WAIT_STATE);
1436 spin_unlock(&pkt->lock);
1438 VPRINTK("pkt_start_write: Writing %d frames for zone %llx\n",
1439 frames_write, (unsigned long long)pkt->sector);
1440 BUG_ON(frames_write != pkt->write_size);
1442 if (test_bit(PACKET_MERGE_SEGS, &pd->flags) || (pkt->write_size < pkt->frames)) {
1443 pkt_make_local_copy(pkt, bvec);
1444 pkt->cache_valid = 1;
1446 pkt->cache_valid = 0;
1449 /* Start the write request */
1450 bio_init(pkt->w_bio);
1451 pkt->w_bio->bi_max_vecs = PACKET_MAX_SIZE;
1452 pkt->w_bio->bi_sector = pkt->sector;
1453 pkt->w_bio->bi_bdev = pd->bdev;
1454 pkt->w_bio->bi_end_io = pkt_end_io_packet_write;
1455 pkt->w_bio->bi_private = pkt;
1456 pkt->w_bio->bi_io_vec = bvec;
1457 pkt->w_bio->bi_destructor = pkt_bio_destructor;
1458 for (f = 0; f < pkt->frames; f++)
1459 if (!bio_add_page(pkt->w_bio, bvec[f].bv_page, CD_FRAMESIZE, bvec[f].bv_offset))
1461 VPRINTK(DRIVER_NAME": vcnt=%d\n", pkt->w_bio->bi_vcnt);
1463 atomic_set(&pkt->io_wait, 1);
1464 pkt->w_bio->bi_rw = WRITE;
1465 pkt_queue_bio(pd, pkt->w_bio);
1468 static void pkt_finish_packet(struct packet_data *pkt, int uptodate)
1470 struct bio *bio, *next;
1473 pkt->cache_valid = 0;
1475 /* Finish all bios corresponding to this packet */
1476 bio = pkt->orig_bios;
1478 next = bio->bi_next;
1479 bio->bi_next = NULL;
1480 bio_endio(bio, uptodate ? 0 : -EIO);
1483 pkt->orig_bios = pkt->orig_bios_tail = NULL;
1486 static void pkt_run_state_machine(struct pktcdvd_device *pd, struct packet_data *pkt)
1490 VPRINTK("run_state_machine: pkt %d\n", pkt->id);
1493 switch (pkt->state) {
1494 case PACKET_WAITING_STATE:
1495 if ((pkt->write_size < pkt->frames) && (pkt->sleep_time > 0))
1498 pkt->sleep_time = 0;
1499 pkt_gather_data(pd, pkt);
1500 pkt_set_state(pkt, PACKET_READ_WAIT_STATE);
1503 case PACKET_READ_WAIT_STATE:
1504 if (atomic_read(&pkt->io_wait) > 0)
1507 if (atomic_read(&pkt->io_errors) > 0) {
1508 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1510 pkt_start_write(pd, pkt);
1514 case PACKET_WRITE_WAIT_STATE:
1515 if (atomic_read(&pkt->io_wait) > 0)
1518 if (test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags)) {
1519 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1521 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1525 case PACKET_RECOVERY_STATE:
1526 if (pkt_start_recovery(pkt)) {
1527 pkt_start_write(pd, pkt);
1529 VPRINTK("No recovery possible\n");
1530 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1534 case PACKET_FINISHED_STATE:
1535 uptodate = test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags);
1536 pkt_finish_packet(pkt, uptodate);
1546 static void pkt_handle_packets(struct pktcdvd_device *pd)
1548 struct packet_data *pkt, *next;
1550 VPRINTK("pkt_handle_packets\n");
1553 * Run state machine for active packets
1555 list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1556 if (atomic_read(&pkt->run_sm) > 0) {
1557 atomic_set(&pkt->run_sm, 0);
1558 pkt_run_state_machine(pd, pkt);
1563 * Move no longer active packets to the free list
1565 spin_lock(&pd->cdrw.active_list_lock);
1566 list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_active_list, list) {
1567 if (pkt->state == PACKET_FINISHED_STATE) {
1568 list_del(&pkt->list);
1569 pkt_put_packet_data(pd, pkt);
1570 pkt_set_state(pkt, PACKET_IDLE_STATE);
1571 atomic_set(&pd->scan_queue, 1);
1574 spin_unlock(&pd->cdrw.active_list_lock);
1577 static void pkt_count_states(struct pktcdvd_device *pd, int *states)
1579 struct packet_data *pkt;
1582 for (i = 0; i < PACKET_NUM_STATES; i++)
1585 spin_lock(&pd->cdrw.active_list_lock);
1586 list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1587 states[pkt->state]++;
1589 spin_unlock(&pd->cdrw.active_list_lock);
1593 * kcdrwd is woken up when writes have been queued for one of our
1594 * registered devices
1596 static int kcdrwd(void *foobar)
1598 struct pktcdvd_device *pd = foobar;
1599 struct packet_data *pkt;
1600 long min_sleep_time, residue;
1602 set_user_nice(current, -20);
1606 DECLARE_WAITQUEUE(wait, current);
1609 * Wait until there is something to do
1611 add_wait_queue(&pd->wqueue, &wait);
1613 set_current_state(TASK_INTERRUPTIBLE);
1615 /* Check if we need to run pkt_handle_queue */
1616 if (atomic_read(&pd->scan_queue) > 0)
1619 /* Check if we need to run the state machine for some packet */
1620 list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1621 if (atomic_read(&pkt->run_sm) > 0)
1625 /* Check if we need to process the iosched queues */
1626 if (atomic_read(&pd->iosched.attention) != 0)
1629 /* Otherwise, go to sleep */
1630 if (PACKET_DEBUG > 1) {
1631 int states[PACKET_NUM_STATES];
1632 pkt_count_states(pd, states);
1633 VPRINTK("kcdrwd: i:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
1634 states[0], states[1], states[2], states[3],
1635 states[4], states[5]);
1638 min_sleep_time = MAX_SCHEDULE_TIMEOUT;
1639 list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1640 if (pkt->sleep_time && pkt->sleep_time < min_sleep_time)
1641 min_sleep_time = pkt->sleep_time;
1644 generic_unplug_device(bdev_get_queue(pd->bdev));
1646 VPRINTK("kcdrwd: sleeping\n");
1647 residue = schedule_timeout(min_sleep_time);
1648 VPRINTK("kcdrwd: wake up\n");
1650 /* make swsusp happy with our thread */
1653 list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1654 if (!pkt->sleep_time)
1656 pkt->sleep_time -= min_sleep_time - residue;
1657 if (pkt->sleep_time <= 0) {
1658 pkt->sleep_time = 0;
1659 atomic_inc(&pkt->run_sm);
1663 if (kthread_should_stop())
1667 set_current_state(TASK_RUNNING);
1668 remove_wait_queue(&pd->wqueue, &wait);
1670 if (kthread_should_stop())
1674 * if pkt_handle_queue returns true, we can queue
1677 while (pkt_handle_queue(pd))
1681 * Handle packet state machine
1683 pkt_handle_packets(pd);
1686 * Handle iosched queues
1688 pkt_iosched_process_queue(pd);
1694 static void pkt_print_settings(struct pktcdvd_device *pd)
1696 printk(DRIVER_NAME": %s packets, ", pd->settings.fp ? "Fixed" : "Variable");
1697 printk("%u blocks, ", pd->settings.size >> 2);
1698 printk("Mode-%c disc\n", pd->settings.block_mode == 8 ? '1' : '2');
1701 static int pkt_mode_sense(struct pktcdvd_device *pd, struct packet_command *cgc, int page_code, int page_control)
1703 memset(cgc->cmd, 0, sizeof(cgc->cmd));
1705 cgc->cmd[0] = GPCMD_MODE_SENSE_10;
1706 cgc->cmd[2] = page_code | (page_control << 6);
1707 cgc->cmd[7] = cgc->buflen >> 8;
1708 cgc->cmd[8] = cgc->buflen & 0xff;
1709 cgc->data_direction = CGC_DATA_READ;
1710 return pkt_generic_packet(pd, cgc);
1713 static int pkt_mode_select(struct pktcdvd_device *pd, struct packet_command *cgc)
1715 memset(cgc->cmd, 0, sizeof(cgc->cmd));
1716 memset(cgc->buffer, 0, 2);
1717 cgc->cmd[0] = GPCMD_MODE_SELECT_10;
1718 cgc->cmd[1] = 0x10; /* PF */
1719 cgc->cmd[7] = cgc->buflen >> 8;
1720 cgc->cmd[8] = cgc->buflen & 0xff;
1721 cgc->data_direction = CGC_DATA_WRITE;
1722 return pkt_generic_packet(pd, cgc);
1725 static int pkt_get_disc_info(struct pktcdvd_device *pd, disc_information *di)
1727 struct packet_command cgc;
1730 /* set up command and get the disc info */
1731 init_cdrom_command(&cgc, di, sizeof(*di), CGC_DATA_READ);
1732 cgc.cmd[0] = GPCMD_READ_DISC_INFO;
1733 cgc.cmd[8] = cgc.buflen = 2;
1736 if ((ret = pkt_generic_packet(pd, &cgc)))
1739 /* not all drives have the same disc_info length, so requeue
1740 * packet with the length the drive tells us it can supply
1742 cgc.buflen = be16_to_cpu(di->disc_information_length) +
1743 sizeof(di->disc_information_length);
1745 if (cgc.buflen > sizeof(disc_information))
1746 cgc.buflen = sizeof(disc_information);
1748 cgc.cmd[8] = cgc.buflen;
1749 return pkt_generic_packet(pd, &cgc);
1752 static int pkt_get_track_info(struct pktcdvd_device *pd, __u16 track, __u8 type, track_information *ti)
1754 struct packet_command cgc;
1757 init_cdrom_command(&cgc, ti, 8, CGC_DATA_READ);
1758 cgc.cmd[0] = GPCMD_READ_TRACK_RZONE_INFO;
1759 cgc.cmd[1] = type & 3;
1760 cgc.cmd[4] = (track & 0xff00) >> 8;
1761 cgc.cmd[5] = track & 0xff;
1765 if ((ret = pkt_generic_packet(pd, &cgc)))
1768 cgc.buflen = be16_to_cpu(ti->track_information_length) +
1769 sizeof(ti->track_information_length);
1771 if (cgc.buflen > sizeof(track_information))
1772 cgc.buflen = sizeof(track_information);
1774 cgc.cmd[8] = cgc.buflen;
1775 return pkt_generic_packet(pd, &cgc);
1778 static int pkt_get_last_written(struct pktcdvd_device *pd, long *last_written)
1780 disc_information di;
1781 track_information ti;
1785 if ((ret = pkt_get_disc_info(pd, &di)))
1788 last_track = (di.last_track_msb << 8) | di.last_track_lsb;
1789 if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1792 /* if this track is blank, try the previous. */
1795 if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1799 /* if last recorded field is valid, return it. */
1801 *last_written = be32_to_cpu(ti.last_rec_address);
1803 /* make it up instead */
1804 *last_written = be32_to_cpu(ti.track_start) +
1805 be32_to_cpu(ti.track_size);
1807 *last_written -= (be32_to_cpu(ti.free_blocks) + 7);
1813 * write mode select package based on pd->settings
1815 static int pkt_set_write_settings(struct pktcdvd_device *pd)
1817 struct packet_command cgc;
1818 struct request_sense sense;
1819 write_param_page *wp;
1823 /* doesn't apply to DVD+RW or DVD-RAM */
1824 if ((pd->mmc3_profile == 0x1a) || (pd->mmc3_profile == 0x12))
1827 memset(buffer, 0, sizeof(buffer));
1828 init_cdrom_command(&cgc, buffer, sizeof(*wp), CGC_DATA_READ);
1830 if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1831 pkt_dump_sense(&cgc);
1835 size = 2 + ((buffer[0] << 8) | (buffer[1] & 0xff));
1836 pd->mode_offset = (buffer[6] << 8) | (buffer[7] & 0xff);
1837 if (size > sizeof(buffer))
1838 size = sizeof(buffer);
1843 init_cdrom_command(&cgc, buffer, size, CGC_DATA_READ);
1845 if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1846 pkt_dump_sense(&cgc);
1851 * write page is offset header + block descriptor length
1853 wp = (write_param_page *) &buffer[sizeof(struct mode_page_header) + pd->mode_offset];
1855 wp->fp = pd->settings.fp;
1856 wp->track_mode = pd->settings.track_mode;
1857 wp->write_type = pd->settings.write_type;
1858 wp->data_block_type = pd->settings.block_mode;
1860 wp->multi_session = 0;
1862 #ifdef PACKET_USE_LS
1867 if (wp->data_block_type == PACKET_BLOCK_MODE1) {
1868 wp->session_format = 0;
1870 } else if (wp->data_block_type == PACKET_BLOCK_MODE2) {
1871 wp->session_format = 0x20;
1875 memcpy(&wp->mcn[1], PACKET_MCN, sizeof(wp->mcn) - 1);
1881 printk(DRIVER_NAME": write mode wrong %d\n", wp->data_block_type);
1884 wp->packet_size = cpu_to_be32(pd->settings.size >> 2);
1886 cgc.buflen = cgc.cmd[8] = size;
1887 if ((ret = pkt_mode_select(pd, &cgc))) {
1888 pkt_dump_sense(&cgc);
1892 pkt_print_settings(pd);
1897 * 1 -- we can write to this track, 0 -- we can't
1899 static int pkt_writable_track(struct pktcdvd_device *pd, track_information *ti)
1901 switch (pd->mmc3_profile) {
1902 case 0x1a: /* DVD+RW */
1903 case 0x12: /* DVD-RAM */
1904 /* The track is always writable on DVD+RW/DVD-RAM */
1910 if (!ti->packet || !ti->fp)
1914 * "good" settings as per Mt Fuji.
1916 if (ti->rt == 0 && ti->blank == 0)
1919 if (ti->rt == 0 && ti->blank == 1)
1922 if (ti->rt == 1 && ti->blank == 0)
1925 printk(DRIVER_NAME": bad state %d-%d-%d\n", ti->rt, ti->blank, ti->packet);
1930 * 1 -- we can write to this disc, 0 -- we can't
1932 static int pkt_writable_disc(struct pktcdvd_device *pd, disc_information *di)
1934 switch (pd->mmc3_profile) {
1935 case 0x0a: /* CD-RW */
1936 case 0xffff: /* MMC3 not supported */
1938 case 0x1a: /* DVD+RW */
1939 case 0x13: /* DVD-RW */
1940 case 0x12: /* DVD-RAM */
1943 VPRINTK(DRIVER_NAME": Wrong disc profile (%x)\n", pd->mmc3_profile);
1948 * for disc type 0xff we should probably reserve a new track.
1949 * but i'm not sure, should we leave this to user apps? probably.
1951 if (di->disc_type == 0xff) {
1952 printk(DRIVER_NAME": Unknown disc. No track?\n");
1956 if (di->disc_type != 0x20 && di->disc_type != 0) {
1957 printk(DRIVER_NAME": Wrong disc type (%x)\n", di->disc_type);
1961 if (di->erasable == 0) {
1962 printk(DRIVER_NAME": Disc not erasable\n");
1966 if (di->border_status == PACKET_SESSION_RESERVED) {
1967 printk(DRIVER_NAME": Can't write to last track (reserved)\n");
1974 static int pkt_probe_settings(struct pktcdvd_device *pd)
1976 struct packet_command cgc;
1977 unsigned char buf[12];
1978 disc_information di;
1979 track_information ti;
1982 init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
1983 cgc.cmd[0] = GPCMD_GET_CONFIGURATION;
1985 ret = pkt_generic_packet(pd, &cgc);
1986 pd->mmc3_profile = ret ? 0xffff : buf[6] << 8 | buf[7];
1988 memset(&di, 0, sizeof(disc_information));
1989 memset(&ti, 0, sizeof(track_information));
1991 if ((ret = pkt_get_disc_info(pd, &di))) {
1992 printk("failed get_disc\n");
1996 if (!pkt_writable_disc(pd, &di))
1999 pd->type = di.erasable ? PACKET_CDRW : PACKET_CDR;
2001 track = 1; /* (di.last_track_msb << 8) | di.last_track_lsb; */
2002 if ((ret = pkt_get_track_info(pd, track, 1, &ti))) {
2003 printk(DRIVER_NAME": failed get_track\n");
2007 if (!pkt_writable_track(pd, &ti)) {
2008 printk(DRIVER_NAME": can't write to this track\n");
2013 * we keep packet size in 512 byte units, makes it easier to
2014 * deal with request calculations.
2016 pd->settings.size = be32_to_cpu(ti.fixed_packet_size) << 2;
2017 if (pd->settings.size == 0) {
2018 printk(DRIVER_NAME": detected zero packet size!\n");
2021 if (pd->settings.size > PACKET_MAX_SECTORS) {
2022 printk(DRIVER_NAME": packet size is too big\n");
2025 pd->settings.fp = ti.fp;
2026 pd->offset = (be32_to_cpu(ti.track_start) << 2) & (pd->settings.size - 1);
2029 pd->nwa = be32_to_cpu(ti.next_writable);
2030 set_bit(PACKET_NWA_VALID, &pd->flags);
2034 * in theory we could use lra on -RW media as well and just zero
2035 * blocks that haven't been written yet, but in practice that
2036 * is just a no-go. we'll use that for -R, naturally.
2039 pd->lra = be32_to_cpu(ti.last_rec_address);
2040 set_bit(PACKET_LRA_VALID, &pd->flags);
2042 pd->lra = 0xffffffff;
2043 set_bit(PACKET_LRA_VALID, &pd->flags);
2049 pd->settings.link_loss = 7;
2050 pd->settings.write_type = 0; /* packet */
2051 pd->settings.track_mode = ti.track_mode;
2054 * mode1 or mode2 disc
2056 switch (ti.data_mode) {
2058 pd->settings.block_mode = PACKET_BLOCK_MODE1;
2061 pd->settings.block_mode = PACKET_BLOCK_MODE2;
2064 printk(DRIVER_NAME": unknown data mode\n");
2071 * enable/disable write caching on drive
2073 static int pkt_write_caching(struct pktcdvd_device *pd, int set)
2075 struct packet_command cgc;
2076 struct request_sense sense;
2077 unsigned char buf[64];
2080 memset(buf, 0, sizeof(buf));
2081 init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
2083 cgc.buflen = pd->mode_offset + 12;
2086 * caching mode page might not be there, so quiet this command
2090 if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WCACHING_PAGE, 0)))
2093 buf[pd->mode_offset + 10] |= (!!set << 2);
2095 cgc.buflen = cgc.cmd[8] = 2 + ((buf[0] << 8) | (buf[1] & 0xff));
2096 ret = pkt_mode_select(pd, &cgc);
2098 printk(DRIVER_NAME": write caching control failed\n");
2099 pkt_dump_sense(&cgc);
2100 } else if (!ret && set)
2101 printk(DRIVER_NAME": enabled write caching on %s\n", pd->name);
2105 static int pkt_lock_door(struct pktcdvd_device *pd, int lockflag)
2107 struct packet_command cgc;
2109 init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2110 cgc.cmd[0] = GPCMD_PREVENT_ALLOW_MEDIUM_REMOVAL;
2111 cgc.cmd[4] = lockflag ? 1 : 0;
2112 return pkt_generic_packet(pd, &cgc);
2116 * Returns drive maximum write speed
2118 static int pkt_get_max_speed(struct pktcdvd_device *pd, unsigned *write_speed)
2120 struct packet_command cgc;
2121 struct request_sense sense;
2122 unsigned char buf[256+18];
2123 unsigned char *cap_buf;
2126 memset(buf, 0, sizeof(buf));
2127 cap_buf = &buf[sizeof(struct mode_page_header) + pd->mode_offset];
2128 init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_UNKNOWN);
2131 ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2133 cgc.buflen = pd->mode_offset + cap_buf[1] + 2 +
2134 sizeof(struct mode_page_header);
2135 ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2137 pkt_dump_sense(&cgc);
2142 offset = 20; /* Obsoleted field, used by older drives */
2143 if (cap_buf[1] >= 28)
2144 offset = 28; /* Current write speed selected */
2145 if (cap_buf[1] >= 30) {
2146 /* If the drive reports at least one "Logical Unit Write
2147 * Speed Performance Descriptor Block", use the information
2148 * in the first block. (contains the highest speed)
2150 int num_spdb = (cap_buf[30] << 8) + cap_buf[31];
2155 *write_speed = (cap_buf[offset] << 8) | cap_buf[offset + 1];
2159 /* These tables from cdrecord - I don't have orange book */
2160 /* standard speed CD-RW (1-4x) */
2161 static char clv_to_speed[16] = {
2162 /* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 */
2163 0, 2, 4, 6, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2165 /* high speed CD-RW (-10x) */
2166 static char hs_clv_to_speed[16] = {
2167 /* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 */
2168 0, 2, 4, 6, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2170 /* ultra high speed CD-RW */
2171 static char us_clv_to_speed[16] = {
2172 /* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 */
2173 0, 2, 4, 8, 0, 0,16, 0,24,32,40,48, 0, 0, 0, 0
2177 * reads the maximum media speed from ATIP
2179 static int pkt_media_speed(struct pktcdvd_device *pd, unsigned *speed)
2181 struct packet_command cgc;
2182 struct request_sense sense;
2183 unsigned char buf[64];
2184 unsigned int size, st, sp;
2187 init_cdrom_command(&cgc, buf, 2, CGC_DATA_READ);
2189 cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2191 cgc.cmd[2] = 4; /* READ ATIP */
2193 ret = pkt_generic_packet(pd, &cgc);
2195 pkt_dump_sense(&cgc);
2198 size = ((unsigned int) buf[0]<<8) + buf[1] + 2;
2199 if (size > sizeof(buf))
2202 init_cdrom_command(&cgc, buf, size, CGC_DATA_READ);
2204 cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2208 ret = pkt_generic_packet(pd, &cgc);
2210 pkt_dump_sense(&cgc);
2214 if (!buf[6] & 0x40) {
2215 printk(DRIVER_NAME": Disc type is not CD-RW\n");
2218 if (!buf[6] & 0x4) {
2219 printk(DRIVER_NAME": A1 values on media are not valid, maybe not CDRW?\n");
2223 st = (buf[6] >> 3) & 0x7; /* disc sub-type */
2225 sp = buf[16] & 0xf; /* max speed from ATIP A1 field */
2227 /* Info from cdrecord */
2229 case 0: /* standard speed */
2230 *speed = clv_to_speed[sp];
2232 case 1: /* high speed */
2233 *speed = hs_clv_to_speed[sp];
2235 case 2: /* ultra high speed */
2236 *speed = us_clv_to_speed[sp];
2239 printk(DRIVER_NAME": Unknown disc sub-type %d\n",st);
2243 printk(DRIVER_NAME": Max. media speed: %d\n",*speed);
2246 printk(DRIVER_NAME": Unknown speed %d for sub-type %d\n",sp,st);
2251 static int pkt_perform_opc(struct pktcdvd_device *pd)
2253 struct packet_command cgc;
2254 struct request_sense sense;
2257 VPRINTK(DRIVER_NAME": Performing OPC\n");
2259 init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2261 cgc.timeout = 60*HZ;
2262 cgc.cmd[0] = GPCMD_SEND_OPC;
2264 if ((ret = pkt_generic_packet(pd, &cgc)))
2265 pkt_dump_sense(&cgc);
2269 static int pkt_open_write(struct pktcdvd_device *pd)
2272 unsigned int write_speed, media_write_speed, read_speed;
2274 if ((ret = pkt_probe_settings(pd))) {
2275 VPRINTK(DRIVER_NAME": %s failed probe\n", pd->name);
2279 if ((ret = pkt_set_write_settings(pd))) {
2280 DPRINTK(DRIVER_NAME": %s failed saving write settings\n", pd->name);
2284 pkt_write_caching(pd, USE_WCACHING);
2286 if ((ret = pkt_get_max_speed(pd, &write_speed)))
2287 write_speed = 16 * 177;
2288 switch (pd->mmc3_profile) {
2289 case 0x13: /* DVD-RW */
2290 case 0x1a: /* DVD+RW */
2291 case 0x12: /* DVD-RAM */
2292 DPRINTK(DRIVER_NAME": write speed %ukB/s\n", write_speed);
2295 if ((ret = pkt_media_speed(pd, &media_write_speed)))
2296 media_write_speed = 16;
2297 write_speed = min(write_speed, media_write_speed * 177);
2298 DPRINTK(DRIVER_NAME": write speed %ux\n", write_speed / 176);
2301 read_speed = write_speed;
2303 if ((ret = pkt_set_speed(pd, write_speed, read_speed))) {
2304 DPRINTK(DRIVER_NAME": %s couldn't set write speed\n", pd->name);
2307 pd->write_speed = write_speed;
2308 pd->read_speed = read_speed;
2310 if ((ret = pkt_perform_opc(pd))) {
2311 DPRINTK(DRIVER_NAME": %s Optimum Power Calibration failed\n", pd->name);
2318 * called at open time.
2320 static int pkt_open_dev(struct pktcdvd_device *pd, int write)
2324 struct request_queue *q;
2327 * We need to re-open the cdrom device without O_NONBLOCK to be able
2328 * to read/write from/to it. It is already opened in O_NONBLOCK mode
2329 * so bdget() can't fail.
2331 bdget(pd->bdev->bd_dev);
2332 if ((ret = blkdev_get(pd->bdev, FMODE_READ, O_RDONLY)))
2335 if ((ret = bd_claim(pd->bdev, pd)))
2338 if ((ret = pkt_get_last_written(pd, &lba))) {
2339 printk(DRIVER_NAME": pkt_get_last_written failed\n");
2343 set_capacity(pd->disk, lba << 2);
2344 set_capacity(pd->bdev->bd_disk, lba << 2);
2345 bd_set_size(pd->bdev, (loff_t)lba << 11);
2347 q = bdev_get_queue(pd->bdev);
2349 if ((ret = pkt_open_write(pd)))
2352 * Some CDRW drives can not handle writes larger than one packet,
2353 * even if the size is a multiple of the packet size.
2355 spin_lock_irq(q->queue_lock);
2356 blk_queue_max_sectors(q, pd->settings.size);
2357 spin_unlock_irq(q->queue_lock);
2358 set_bit(PACKET_WRITABLE, &pd->flags);
2360 pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2361 clear_bit(PACKET_WRITABLE, &pd->flags);
2364 if ((ret = pkt_set_segment_merging(pd, q)))
2368 if (!pkt_grow_pktlist(pd, CONFIG_CDROM_PKTCDVD_BUFFERS)) {
2369 printk(DRIVER_NAME": not enough memory for buffers\n");
2373 printk(DRIVER_NAME": %lukB available on disc\n", lba << 1);
2379 bd_release(pd->bdev);
2381 blkdev_put(pd->bdev);
2387 * called when the device is closed. makes sure that the device flushes
2388 * the internal cache before we close.
2390 static void pkt_release_dev(struct pktcdvd_device *pd, int flush)
2392 if (flush && pkt_flush_cache(pd))
2393 DPRINTK(DRIVER_NAME": %s not flushing cache\n", pd->name);
2395 pkt_lock_door(pd, 0);
2397 pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2398 bd_release(pd->bdev);
2399 blkdev_put(pd->bdev);
2401 pkt_shrink_pktlist(pd);
2404 static struct pktcdvd_device *pkt_find_dev_from_minor(int dev_minor)
2406 if (dev_minor >= MAX_WRITERS)
2408 return pkt_devs[dev_minor];
2411 static int pkt_open(struct inode *inode, struct file *file)
2413 struct pktcdvd_device *pd = NULL;
2416 VPRINTK(DRIVER_NAME": entering open\n");
2418 mutex_lock(&ctl_mutex);
2419 pd = pkt_find_dev_from_minor(iminor(inode));
2424 BUG_ON(pd->refcnt < 0);
2427 if (pd->refcnt > 1) {
2428 if ((file->f_mode & FMODE_WRITE) &&
2429 !test_bit(PACKET_WRITABLE, &pd->flags)) {
2434 ret = pkt_open_dev(pd, file->f_mode & FMODE_WRITE);
2438 * needed here as well, since ext2 (among others) may change
2439 * the blocksize at mount time
2441 set_blocksize(inode->i_bdev, CD_FRAMESIZE);
2444 mutex_unlock(&ctl_mutex);
2450 VPRINTK(DRIVER_NAME": failed open (%d)\n", ret);
2451 mutex_unlock(&ctl_mutex);
2455 static int pkt_close(struct inode *inode, struct file *file)
2457 struct pktcdvd_device *pd = inode->i_bdev->bd_disk->private_data;
2460 mutex_lock(&ctl_mutex);
2462 BUG_ON(pd->refcnt < 0);
2463 if (pd->refcnt == 0) {
2464 int flush = test_bit(PACKET_WRITABLE, &pd->flags);
2465 pkt_release_dev(pd, flush);
2467 mutex_unlock(&ctl_mutex);
2472 static void pkt_end_io_read_cloned(struct bio *bio, int err)
2474 struct packet_stacked_data *psd = bio->bi_private;
2475 struct pktcdvd_device *pd = psd->pd;
2478 bio_endio(psd->bio, err);
2479 mempool_free(psd, psd_pool);
2480 pkt_bio_finished(pd);
2483 static int pkt_make_request(struct request_queue *q, struct bio *bio)
2485 struct pktcdvd_device *pd;
2486 char b[BDEVNAME_SIZE];
2488 struct packet_data *pkt;
2489 int was_empty, blocked_bio;
2490 struct pkt_rb_node *node;
2494 printk(DRIVER_NAME": %s incorrect request queue\n", bdevname(bio->bi_bdev, b));
2499 * Clone READ bios so we can have our own bi_end_io callback.
2501 if (bio_data_dir(bio) == READ) {
2502 struct bio *cloned_bio = bio_clone(bio, GFP_NOIO);
2503 struct packet_stacked_data *psd = mempool_alloc(psd_pool, GFP_NOIO);
2507 cloned_bio->bi_bdev = pd->bdev;
2508 cloned_bio->bi_private = psd;
2509 cloned_bio->bi_end_io = pkt_end_io_read_cloned;
2510 pd->stats.secs_r += bio->bi_size >> 9;
2511 pkt_queue_bio(pd, cloned_bio);
2515 if (!test_bit(PACKET_WRITABLE, &pd->flags)) {
2516 printk(DRIVER_NAME": WRITE for ro device %s (%llu)\n",
2517 pd->name, (unsigned long long)bio->bi_sector);
2521 if (!bio->bi_size || (bio->bi_size % CD_FRAMESIZE)) {
2522 printk(DRIVER_NAME": wrong bio size\n");
2526 blk_queue_bounce(q, &bio);
2528 zone = ZONE(bio->bi_sector, pd);
2529 VPRINTK("pkt_make_request: start = %6llx stop = %6llx\n",
2530 (unsigned long long)bio->bi_sector,
2531 (unsigned long long)(bio->bi_sector + bio_sectors(bio)));
2533 /* Check if we have to split the bio */
2535 struct bio_pair *bp;
2539 last_zone = ZONE(bio->bi_sector + bio_sectors(bio) - 1, pd);
2540 if (last_zone != zone) {
2541 BUG_ON(last_zone != zone + pd->settings.size);
2542 first_sectors = last_zone - bio->bi_sector;
2543 bp = bio_split(bio, bio_split_pool, first_sectors);
2545 pkt_make_request(q, &bp->bio1);
2546 pkt_make_request(q, &bp->bio2);
2547 bio_pair_release(bp);
2553 * If we find a matching packet in state WAITING or READ_WAIT, we can
2554 * just append this bio to that packet.
2556 spin_lock(&pd->cdrw.active_list_lock);
2558 list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
2559 if (pkt->sector == zone) {
2560 spin_lock(&pkt->lock);
2561 if ((pkt->state == PACKET_WAITING_STATE) ||
2562 (pkt->state == PACKET_READ_WAIT_STATE)) {
2563 pkt_add_list_last(bio, &pkt->orig_bios,
2564 &pkt->orig_bios_tail);
2565 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
2566 if ((pkt->write_size >= pkt->frames) &&
2567 (pkt->state == PACKET_WAITING_STATE)) {
2568 atomic_inc(&pkt->run_sm);
2569 wake_up(&pd->wqueue);
2571 spin_unlock(&pkt->lock);
2572 spin_unlock(&pd->cdrw.active_list_lock);
2577 spin_unlock(&pkt->lock);
2580 spin_unlock(&pd->cdrw.active_list_lock);
2583 * Test if there is enough room left in the bio work queue
2584 * (queue size >= congestion on mark).
2585 * If not, wait till the work queue size is below the congestion off mark.
2587 spin_lock(&pd->lock);
2588 if (pd->write_congestion_on > 0
2589 && pd->bio_queue_size >= pd->write_congestion_on) {
2590 set_bdi_congested(&q->backing_dev_info, WRITE);
2592 spin_unlock(&pd->lock);
2593 congestion_wait(WRITE, HZ);
2594 spin_lock(&pd->lock);
2595 } while(pd->bio_queue_size > pd->write_congestion_off);
2597 spin_unlock(&pd->lock);
2600 * No matching packet found. Store the bio in the work queue.
2602 node = mempool_alloc(pd->rb_pool, GFP_NOIO);
2604 spin_lock(&pd->lock);
2605 BUG_ON(pd->bio_queue_size < 0);
2606 was_empty = (pd->bio_queue_size == 0);
2607 pkt_rbtree_insert(pd, node);
2608 spin_unlock(&pd->lock);
2611 * Wake up the worker thread.
2613 atomic_set(&pd->scan_queue, 1);
2615 /* This wake_up is required for correct operation */
2616 wake_up(&pd->wqueue);
2617 } else if (!list_empty(&pd->cdrw.pkt_free_list) && !blocked_bio) {
2619 * This wake up is not required for correct operation,
2620 * but improves performance in some cases.
2622 wake_up(&pd->wqueue);
2632 static int pkt_merge_bvec(struct request_queue *q, struct bio *bio, struct bio_vec *bvec)
2634 struct pktcdvd_device *pd = q->queuedata;
2635 sector_t zone = ZONE(bio->bi_sector, pd);
2636 int used = ((bio->bi_sector - zone) << 9) + bio->bi_size;
2637 int remaining = (pd->settings.size << 9) - used;
2641 * A bio <= PAGE_SIZE must be allowed. If it crosses a packet
2642 * boundary, pkt_make_request() will split the bio.
2644 remaining2 = PAGE_SIZE - bio->bi_size;
2645 remaining = max(remaining, remaining2);
2647 BUG_ON(remaining < 0);
2651 static void pkt_init_queue(struct pktcdvd_device *pd)
2653 struct request_queue *q = pd->disk->queue;
2655 blk_queue_make_request(q, pkt_make_request);
2656 blk_queue_hardsect_size(q, CD_FRAMESIZE);
2657 blk_queue_max_sectors(q, PACKET_MAX_SECTORS);
2658 blk_queue_merge_bvec(q, pkt_merge_bvec);
2662 static int pkt_seq_show(struct seq_file *m, void *p)
2664 struct pktcdvd_device *pd = m->private;
2666 char bdev_buf[BDEVNAME_SIZE];
2667 int states[PACKET_NUM_STATES];
2669 seq_printf(m, "Writer %s mapped to %s:\n", pd->name,
2670 bdevname(pd->bdev, bdev_buf));
2672 seq_printf(m, "\nSettings:\n");
2673 seq_printf(m, "\tpacket size:\t\t%dkB\n", pd->settings.size / 2);
2675 if (pd->settings.write_type == 0)
2679 seq_printf(m, "\twrite type:\t\t%s\n", msg);
2681 seq_printf(m, "\tpacket type:\t\t%s\n", pd->settings.fp ? "Fixed" : "Variable");
2682 seq_printf(m, "\tlink loss:\t\t%d\n", pd->settings.link_loss);
2684 seq_printf(m, "\ttrack mode:\t\t%d\n", pd->settings.track_mode);
2686 if (pd->settings.block_mode == PACKET_BLOCK_MODE1)
2688 else if (pd->settings.block_mode == PACKET_BLOCK_MODE2)
2692 seq_printf(m, "\tblock mode:\t\t%s\n", msg);
2694 seq_printf(m, "\nStatistics:\n");
2695 seq_printf(m, "\tpackets started:\t%lu\n", pd->stats.pkt_started);
2696 seq_printf(m, "\tpackets ended:\t\t%lu\n", pd->stats.pkt_ended);
2697 seq_printf(m, "\twritten:\t\t%lukB\n", pd->stats.secs_w >> 1);
2698 seq_printf(m, "\tread gather:\t\t%lukB\n", pd->stats.secs_rg >> 1);
2699 seq_printf(m, "\tread:\t\t\t%lukB\n", pd->stats.secs_r >> 1);
2701 seq_printf(m, "\nMisc:\n");
2702 seq_printf(m, "\treference count:\t%d\n", pd->refcnt);
2703 seq_printf(m, "\tflags:\t\t\t0x%lx\n", pd->flags);
2704 seq_printf(m, "\tread speed:\t\t%ukB/s\n", pd->read_speed);
2705 seq_printf(m, "\twrite speed:\t\t%ukB/s\n", pd->write_speed);
2706 seq_printf(m, "\tstart offset:\t\t%lu\n", pd->offset);
2707 seq_printf(m, "\tmode page offset:\t%u\n", pd->mode_offset);
2709 seq_printf(m, "\nQueue state:\n");
2710 seq_printf(m, "\tbios queued:\t\t%d\n", pd->bio_queue_size);
2711 seq_printf(m, "\tbios pending:\t\t%d\n", atomic_read(&pd->cdrw.pending_bios));
2712 seq_printf(m, "\tcurrent sector:\t\t0x%llx\n", (unsigned long long)pd->current_sector);
2714 pkt_count_states(pd, states);
2715 seq_printf(m, "\tstate:\t\t\ti:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
2716 states[0], states[1], states[2], states[3], states[4], states[5]);
2718 seq_printf(m, "\twrite congestion marks:\toff=%d on=%d\n",
2719 pd->write_congestion_off,
2720 pd->write_congestion_on);
2724 static int pkt_seq_open(struct inode *inode, struct file *file)
2726 return single_open(file, pkt_seq_show, PDE(inode)->data);
2729 static const struct file_operations pkt_proc_fops = {
2730 .open = pkt_seq_open,
2732 .llseek = seq_lseek,
2733 .release = single_release
2736 static int pkt_new_dev(struct pktcdvd_device *pd, dev_t dev)
2740 char b[BDEVNAME_SIZE];
2741 struct proc_dir_entry *proc;
2742 struct block_device *bdev;
2744 if (pd->pkt_dev == dev) {
2745 printk(DRIVER_NAME": Recursive setup not allowed\n");
2748 for (i = 0; i < MAX_WRITERS; i++) {
2749 struct pktcdvd_device *pd2 = pkt_devs[i];
2752 if (pd2->bdev->bd_dev == dev) {
2753 printk(DRIVER_NAME": %s already setup\n", bdevname(pd2->bdev, b));
2756 if (pd2->pkt_dev == dev) {
2757 printk(DRIVER_NAME": Can't chain pktcdvd devices\n");
2765 ret = blkdev_get(bdev, FMODE_READ, O_RDONLY | O_NONBLOCK);
2769 /* This is safe, since we have a reference from open(). */
2770 __module_get(THIS_MODULE);
2773 set_blocksize(bdev, CD_FRAMESIZE);
2777 atomic_set(&pd->cdrw.pending_bios, 0);
2778 pd->cdrw.thread = kthread_run(kcdrwd, pd, "%s", pd->name);
2779 if (IS_ERR(pd->cdrw.thread)) {
2780 printk(DRIVER_NAME": can't start kernel thread\n");
2785 proc = create_proc_entry(pd->name, 0, pkt_proc);
2788 proc->proc_fops = &pkt_proc_fops;
2790 DPRINTK(DRIVER_NAME": writer %s mapped to %s\n", pd->name, bdevname(bdev, b));
2795 /* This is safe: open() is still holding a reference. */
2796 module_put(THIS_MODULE);
2800 static int pkt_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg)
2802 struct pktcdvd_device *pd = inode->i_bdev->bd_disk->private_data;
2804 VPRINTK("pkt_ioctl: cmd %x, dev %d:%d\n", cmd, imajor(inode), iminor(inode));
2808 * forward selected CDROM ioctls to CD-ROM, for UDF
2810 case CDROMMULTISESSION:
2811 case CDROMREADTOCENTRY:
2812 case CDROM_LAST_WRITTEN:
2813 case CDROM_SEND_PACKET:
2814 case SCSI_IOCTL_SEND_COMMAND:
2815 return blkdev_ioctl(pd->bdev->bd_inode, file, cmd, arg);
2819 * The door gets locked when the device is opened, so we
2820 * have to unlock it or else the eject command fails.
2822 if (pd->refcnt == 1)
2823 pkt_lock_door(pd, 0);
2824 return blkdev_ioctl(pd->bdev->bd_inode, file, cmd, arg);
2827 VPRINTK(DRIVER_NAME": Unknown ioctl for %s (%x)\n", pd->name, cmd);
2834 static int pkt_media_changed(struct gendisk *disk)
2836 struct pktcdvd_device *pd = disk->private_data;
2837 struct gendisk *attached_disk;
2843 attached_disk = pd->bdev->bd_disk;
2846 return attached_disk->fops->media_changed(attached_disk);
2849 static struct block_device_operations pktcdvd_ops = {
2850 .owner = THIS_MODULE,
2852 .release = pkt_close,
2854 .media_changed = pkt_media_changed,
2858 * Set up mapping from pktcdvd device to CD-ROM device.
2860 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev)
2864 struct pktcdvd_device *pd;
2865 struct gendisk *disk;
2867 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2869 for (idx = 0; idx < MAX_WRITERS; idx++)
2872 if (idx == MAX_WRITERS) {
2873 printk(DRIVER_NAME": max %d writers supported\n", MAX_WRITERS);
2878 pd = kzalloc(sizeof(struct pktcdvd_device), GFP_KERNEL);
2882 pd->rb_pool = mempool_create_kmalloc_pool(PKT_RB_POOL_SIZE,
2883 sizeof(struct pkt_rb_node));
2887 INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
2888 INIT_LIST_HEAD(&pd->cdrw.pkt_active_list);
2889 spin_lock_init(&pd->cdrw.active_list_lock);
2891 spin_lock_init(&pd->lock);
2892 spin_lock_init(&pd->iosched.lock);
2893 sprintf(pd->name, DRIVER_NAME"%d", idx);
2894 init_waitqueue_head(&pd->wqueue);
2895 pd->bio_queue = RB_ROOT;
2897 pd->write_congestion_on = write_congestion_on;
2898 pd->write_congestion_off = write_congestion_off;
2900 disk = alloc_disk(1);
2904 disk->major = pktdev_major;
2905 disk->first_minor = idx;
2906 disk->fops = &pktcdvd_ops;
2907 disk->flags = GENHD_FL_REMOVABLE;
2908 strcpy(disk->disk_name, pd->name);
2909 disk->private_data = pd;
2910 disk->queue = blk_alloc_queue(GFP_KERNEL);
2914 pd->pkt_dev = MKDEV(disk->major, disk->first_minor);
2915 ret = pkt_new_dev(pd, dev);
2921 pkt_sysfs_dev_new(pd);
2922 pkt_debugfs_dev_new(pd);
2926 *pkt_dev = pd->pkt_dev;
2928 mutex_unlock(&ctl_mutex);
2932 blk_cleanup_queue(disk->queue);
2937 mempool_destroy(pd->rb_pool);
2940 mutex_unlock(&ctl_mutex);
2941 printk(DRIVER_NAME": setup of pktcdvd device failed\n");
2946 * Tear down mapping from pktcdvd device to CD-ROM device.
2948 static int pkt_remove_dev(dev_t pkt_dev)
2950 struct pktcdvd_device *pd;
2954 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2956 for (idx = 0; idx < MAX_WRITERS; idx++) {
2958 if (pd && (pd->pkt_dev == pkt_dev))
2961 if (idx == MAX_WRITERS) {
2962 DPRINTK(DRIVER_NAME": dev not setup\n");
2967 if (pd->refcnt > 0) {
2971 if (!IS_ERR(pd->cdrw.thread))
2972 kthread_stop(pd->cdrw.thread);
2974 pkt_devs[idx] = NULL;
2976 pkt_debugfs_dev_remove(pd);
2977 pkt_sysfs_dev_remove(pd);
2979 blkdev_put(pd->bdev);
2981 remove_proc_entry(pd->name, pkt_proc);
2982 DPRINTK(DRIVER_NAME": writer %s unmapped\n", pd->name);
2984 del_gendisk(pd->disk);
2985 blk_cleanup_queue(pd->disk->queue);
2988 mempool_destroy(pd->rb_pool);
2991 /* This is safe: open() is still holding a reference. */
2992 module_put(THIS_MODULE);
2995 mutex_unlock(&ctl_mutex);
2999 static void pkt_get_status(struct pkt_ctrl_command *ctrl_cmd)
3001 struct pktcdvd_device *pd;
3003 mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3005 pd = pkt_find_dev_from_minor(ctrl_cmd->dev_index);
3007 ctrl_cmd->dev = new_encode_dev(pd->bdev->bd_dev);
3008 ctrl_cmd->pkt_dev = new_encode_dev(pd->pkt_dev);
3011 ctrl_cmd->pkt_dev = 0;
3013 ctrl_cmd->num_devices = MAX_WRITERS;
3015 mutex_unlock(&ctl_mutex);
3018 static int pkt_ctl_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg)
3020 void __user *argp = (void __user *)arg;
3021 struct pkt_ctrl_command ctrl_cmd;
3025 if (cmd != PACKET_CTRL_CMD)
3028 if (copy_from_user(&ctrl_cmd, argp, sizeof(struct pkt_ctrl_command)))
3031 switch (ctrl_cmd.command) {
3032 case PKT_CTRL_CMD_SETUP:
3033 if (!capable(CAP_SYS_ADMIN))
3035 ret = pkt_setup_dev(new_decode_dev(ctrl_cmd.dev), &pkt_dev);
3036 ctrl_cmd.pkt_dev = new_encode_dev(pkt_dev);
3038 case PKT_CTRL_CMD_TEARDOWN:
3039 if (!capable(CAP_SYS_ADMIN))
3041 ret = pkt_remove_dev(new_decode_dev(ctrl_cmd.pkt_dev));
3043 case PKT_CTRL_CMD_STATUS:
3044 pkt_get_status(&ctrl_cmd);
3050 if (copy_to_user(argp, &ctrl_cmd, sizeof(struct pkt_ctrl_command)))
3056 static const struct file_operations pkt_ctl_fops = {
3057 .ioctl = pkt_ctl_ioctl,
3058 .owner = THIS_MODULE,
3061 static struct miscdevice pkt_misc = {
3062 .minor = MISC_DYNAMIC_MINOR,
3063 .name = DRIVER_NAME,
3064 .fops = &pkt_ctl_fops
3067 static int __init pkt_init(void)
3071 mutex_init(&ctl_mutex);
3073 psd_pool = mempool_create_kmalloc_pool(PSD_POOL_SIZE,
3074 sizeof(struct packet_stacked_data));
3078 ret = register_blkdev(pktdev_major, DRIVER_NAME);
3080 printk(DRIVER_NAME": Unable to register block device\n");
3086 ret = pkt_sysfs_init();
3092 ret = misc_register(&pkt_misc);
3094 printk(DRIVER_NAME": Unable to register misc device\n");
3098 pkt_proc = proc_mkdir(DRIVER_NAME, proc_root_driver);
3103 pkt_debugfs_cleanup();
3104 pkt_sysfs_cleanup();
3106 unregister_blkdev(pktdev_major, DRIVER_NAME);
3108 mempool_destroy(psd_pool);
3112 static void __exit pkt_exit(void)
3114 remove_proc_entry(DRIVER_NAME, proc_root_driver);
3115 misc_deregister(&pkt_misc);
3117 pkt_debugfs_cleanup();
3118 pkt_sysfs_cleanup();
3120 unregister_blkdev(pktdev_major, DRIVER_NAME);
3121 mempool_destroy(psd_pool);
3124 MODULE_DESCRIPTION("Packet writing layer for CD/DVD drives");
3125 MODULE_AUTHOR("Jens Axboe <axboe@suse.de>");
3126 MODULE_LICENSE("GPL");
3128 module_init(pkt_init);
3129 module_exit(pkt_exit);