1/*
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>
5 *
6 * May be copied or modified under the terms of the GNU General Public
7 * License.  See linux/COPYING for more information.
8 *
9 * Packet writing layer for ATAPI and SCSI CD-RW, DVD+RW, DVD-RW and
10 * DVD-RAM devices.
11 *
12 * Theory of operation:
13 *
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.
22 *
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.
26 *
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.
33 *
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.
38 *
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.
44 *
45 *************************************************************************/
46
47#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
48
49#include <linux/pktcdvd.h>
50#include <linux/module.h>
51#include <linux/types.h>
52#include <linux/kernel.h>
53#include <linux/compat.h>
54#include <linux/kthread.h>
55#include <linux/errno.h>
56#include <linux/spinlock.h>
57#include <linux/file.h>
58#include <linux/proc_fs.h>
59#include <linux/seq_file.h>
60#include <linux/miscdevice.h>
61#include <linux/freezer.h>
62#include <linux/mutex.h>
63#include <linux/slab.h>
64#include <scsi/scsi_cmnd.h>
65#include <scsi/scsi_ioctl.h>
66#include <scsi/scsi.h>
67#include <linux/debugfs.h>
68#include <linux/device.h>
69
70#include <asm/uaccess.h>
71
72#define DRIVER_NAME	"pktcdvd"
73
74#define pkt_err(pd, fmt, ...)						\
75	pr_err("%s: " fmt, pd->name, ##__VA_ARGS__)
76#define pkt_notice(pd, fmt, ...)					\
77	pr_notice("%s: " fmt, pd->name, ##__VA_ARGS__)
78#define pkt_info(pd, fmt, ...)						\
79	pr_info("%s: " fmt, pd->name, ##__VA_ARGS__)
80
81#define pkt_dbg(level, pd, fmt, ...)					\
82do {									\
83	if (level == 2 && PACKET_DEBUG >= 2)				\
84		pr_notice("%s: %s():" fmt,				\
85			  pd->name, __func__, ##__VA_ARGS__);		\
86	else if (level == 1 && PACKET_DEBUG >= 1)			\
87		pr_notice("%s: " fmt, pd->name, ##__VA_ARGS__);		\
88} while (0)
89
90#define MAX_SPEED 0xffff
91
92static DEFINE_MUTEX(pktcdvd_mutex);
93static struct pktcdvd_device *pkt_devs[MAX_WRITERS];
94static struct proc_dir_entry *pkt_proc;
95static int pktdev_major;
96static int write_congestion_on  = PKT_WRITE_CONGESTION_ON;
97static int write_congestion_off = PKT_WRITE_CONGESTION_OFF;
98static struct mutex ctl_mutex;	/* Serialize open/close/setup/teardown */
99static mempool_t *psd_pool;
100
101static struct class	*class_pktcdvd = NULL;    /* /sys/class/pktcdvd */
102static struct dentry	*pkt_debugfs_root = NULL; /* /sys/kernel/debug/pktcdvd */
103
104/* forward declaration */
105static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev);
106static int pkt_remove_dev(dev_t pkt_dev);
107static int pkt_seq_show(struct seq_file *m, void *p);
108
109static sector_t get_zone(sector_t sector, struct pktcdvd_device *pd)
110{
111	return (sector + pd->offset) & ~(sector_t)(pd->settings.size - 1);
112}
113
114/*
115 * create and register a pktcdvd kernel object.
116 */
117static struct pktcdvd_kobj* pkt_kobj_create(struct pktcdvd_device *pd,
118					const char* name,
119					struct kobject* parent,
120					struct kobj_type* ktype)
121{
122	struct pktcdvd_kobj *p;
123	int error;
124
125	p = kzalloc(sizeof(*p), GFP_KERNEL);
126	if (!p)
127		return NULL;
128	p->pd = pd;
129	error = kobject_init_and_add(&p->kobj, ktype, parent, "%s", name);
130	if (error) {
131		kobject_put(&p->kobj);
132		return NULL;
133	}
134	kobject_uevent(&p->kobj, KOBJ_ADD);
135	return p;
136}
137/*
138 * remove a pktcdvd kernel object.
139 */
140static void pkt_kobj_remove(struct pktcdvd_kobj *p)
141{
142	if (p)
143		kobject_put(&p->kobj);
144}
145/*
146 * default release function for pktcdvd kernel objects.
147 */
148static void pkt_kobj_release(struct kobject *kobj)
149{
150	kfree(to_pktcdvdkobj(kobj));
151}
152
153
154/**********************************************************
155 *
156 * sysfs interface for pktcdvd
157 * by (C) 2006  Thomas Maier <balagi@justmail.de>
158 *
159 **********************************************************/
160
161#define DEF_ATTR(_obj,_name,_mode) \
162	static struct attribute _obj = { .name = _name, .mode = _mode }
163
164/**********************************************************
165  /sys/class/pktcdvd/pktcdvd[0-7]/
166                     stat/reset
167                     stat/packets_started
168                     stat/packets_finished
169                     stat/kb_written
170                     stat/kb_read
171                     stat/kb_read_gather
172                     write_queue/size
173                     write_queue/congestion_off
174                     write_queue/congestion_on
175 **********************************************************/
176
177DEF_ATTR(kobj_pkt_attr_st1, "reset", 0200);
178DEF_ATTR(kobj_pkt_attr_st2, "packets_started", 0444);
179DEF_ATTR(kobj_pkt_attr_st3, "packets_finished", 0444);
180DEF_ATTR(kobj_pkt_attr_st4, "kb_written", 0444);
181DEF_ATTR(kobj_pkt_attr_st5, "kb_read", 0444);
182DEF_ATTR(kobj_pkt_attr_st6, "kb_read_gather", 0444);
183
184static struct attribute *kobj_pkt_attrs_stat[] = {
185	&kobj_pkt_attr_st1,
186	&kobj_pkt_attr_st2,
187	&kobj_pkt_attr_st3,
188	&kobj_pkt_attr_st4,
189	&kobj_pkt_attr_st5,
190	&kobj_pkt_attr_st6,
191	NULL
192};
193
194DEF_ATTR(kobj_pkt_attr_wq1, "size", 0444);
195DEF_ATTR(kobj_pkt_attr_wq2, "congestion_off", 0644);
196DEF_ATTR(kobj_pkt_attr_wq3, "congestion_on",  0644);
197
198static struct attribute *kobj_pkt_attrs_wqueue[] = {
199	&kobj_pkt_attr_wq1,
200	&kobj_pkt_attr_wq2,
201	&kobj_pkt_attr_wq3,
202	NULL
203};
204
205static ssize_t kobj_pkt_show(struct kobject *kobj,
206			struct attribute *attr, char *data)
207{
208	struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
209	int n = 0;
210	int v;
211	if (strcmp(attr->name, "packets_started") == 0) {
212		n = sprintf(data, "%lu\n", pd->stats.pkt_started);
213
214	} else if (strcmp(attr->name, "packets_finished") == 0) {
215		n = sprintf(data, "%lu\n", pd->stats.pkt_ended);
216
217	} else if (strcmp(attr->name, "kb_written") == 0) {
218		n = sprintf(data, "%lu\n", pd->stats.secs_w >> 1);
219
220	} else if (strcmp(attr->name, "kb_read") == 0) {
221		n = sprintf(data, "%lu\n", pd->stats.secs_r >> 1);
222
223	} else if (strcmp(attr->name, "kb_read_gather") == 0) {
224		n = sprintf(data, "%lu\n", pd->stats.secs_rg >> 1);
225
226	} else if (strcmp(attr->name, "size") == 0) {
227		spin_lock(&pd->lock);
228		v = pd->bio_queue_size;
229		spin_unlock(&pd->lock);
230		n = sprintf(data, "%d\n", v);
231
232	} else if (strcmp(attr->name, "congestion_off") == 0) {
233		spin_lock(&pd->lock);
234		v = pd->write_congestion_off;
235		spin_unlock(&pd->lock);
236		n = sprintf(data, "%d\n", v);
237
238	} else if (strcmp(attr->name, "congestion_on") == 0) {
239		spin_lock(&pd->lock);
240		v = pd->write_congestion_on;
241		spin_unlock(&pd->lock);
242		n = sprintf(data, "%d\n", v);
243	}
244	return n;
245}
246
247static void init_write_congestion_marks(int* lo, int* hi)
248{
249	if (*hi > 0) {
250		*hi = max(*hi, 500);
251		*hi = min(*hi, 1000000);
252		if (*lo <= 0)
253			*lo = *hi - 100;
254		else {
255			*lo = min(*lo, *hi - 100);
256			*lo = max(*lo, 100);
257		}
258	} else {
259		*hi = -1;
260		*lo = -1;
261	}
262}
263
264static ssize_t kobj_pkt_store(struct kobject *kobj,
265			struct attribute *attr,
266			const char *data, size_t len)
267{
268	struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
269	int val;
270
271	if (strcmp(attr->name, "reset") == 0 && len > 0) {
272		pd->stats.pkt_started = 0;
273		pd->stats.pkt_ended = 0;
274		pd->stats.secs_w = 0;
275		pd->stats.secs_rg = 0;
276		pd->stats.secs_r = 0;
277
278	} else if (strcmp(attr->name, "congestion_off") == 0
279		   && sscanf(data, "%d", &val) == 1) {
280		spin_lock(&pd->lock);
281		pd->write_congestion_off = val;
282		init_write_congestion_marks(&pd->write_congestion_off,
283					&pd->write_congestion_on);
284		spin_unlock(&pd->lock);
285
286	} else if (strcmp(attr->name, "congestion_on") == 0
287		   && sscanf(data, "%d", &val) == 1) {
288		spin_lock(&pd->lock);
289		pd->write_congestion_on = val;
290		init_write_congestion_marks(&pd->write_congestion_off,
291					&pd->write_congestion_on);
292		spin_unlock(&pd->lock);
293	}
294	return len;
295}
296
297static const struct sysfs_ops kobj_pkt_ops = {
298	.show = kobj_pkt_show,
299	.store = kobj_pkt_store
300};
301static struct kobj_type kobj_pkt_type_stat = {
302	.release = pkt_kobj_release,
303	.sysfs_ops = &kobj_pkt_ops,
304	.default_attrs = kobj_pkt_attrs_stat
305};
306static struct kobj_type kobj_pkt_type_wqueue = {
307	.release = pkt_kobj_release,
308	.sysfs_ops = &kobj_pkt_ops,
309	.default_attrs = kobj_pkt_attrs_wqueue
310};
311
312static void pkt_sysfs_dev_new(struct pktcdvd_device *pd)
313{
314	if (class_pktcdvd) {
315		pd->dev = device_create(class_pktcdvd, NULL, MKDEV(0, 0), NULL,
316					"%s", pd->name);
317		if (IS_ERR(pd->dev))
318			pd->dev = NULL;
319	}
320	if (pd->dev) {
321		pd->kobj_stat = pkt_kobj_create(pd, "stat",
322					&pd->dev->kobj,
323					&kobj_pkt_type_stat);
324		pd->kobj_wqueue = pkt_kobj_create(pd, "write_queue",
325					&pd->dev->kobj,
326					&kobj_pkt_type_wqueue);
327	}
328}
329
330static void pkt_sysfs_dev_remove(struct pktcdvd_device *pd)
331{
332	pkt_kobj_remove(pd->kobj_stat);
333	pkt_kobj_remove(pd->kobj_wqueue);
334	if (class_pktcdvd)
335		device_unregister(pd->dev);
336}
337
338
339/********************************************************************
340  /sys/class/pktcdvd/
341                     add            map block device
342                     remove         unmap packet dev
343                     device_map     show mappings
344 *******************************************************************/
345
346static void class_pktcdvd_release(struct class *cls)
347{
348	kfree(cls);
349}
350static ssize_t class_pktcdvd_show_map(struct class *c,
351					struct class_attribute *attr,
352					char *data)
353{
354	int n = 0;
355	int idx;
356	mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
357	for (idx = 0; idx < MAX_WRITERS; idx++) {
358		struct pktcdvd_device *pd = pkt_devs[idx];
359		if (!pd)
360			continue;
361		n += sprintf(data+n, "%s %u:%u %u:%u\n",
362			pd->name,
363			MAJOR(pd->pkt_dev), MINOR(pd->pkt_dev),
364			MAJOR(pd->bdev->bd_dev),
365			MINOR(pd->bdev->bd_dev));
366	}
367	mutex_unlock(&ctl_mutex);
368	return n;
369}
370
371static ssize_t class_pktcdvd_store_add(struct class *c,
372					struct class_attribute *attr,
373					const char *buf,
374					size_t count)
375{
376	unsigned int major, minor;
377
378	if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
379		/* pkt_setup_dev() expects caller to hold reference to self */
380		if (!try_module_get(THIS_MODULE))
381			return -ENODEV;
382
383		pkt_setup_dev(MKDEV(major, minor), NULL);
384
385		module_put(THIS_MODULE);
386
387		return count;
388	}
389
390	return -EINVAL;
391}
392
393static ssize_t class_pktcdvd_store_remove(struct class *c,
394					  struct class_attribute *attr,
395					  const char *buf,
396					size_t count)
397{
398	unsigned int major, minor;
399	if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
400		pkt_remove_dev(MKDEV(major, minor));
401		return count;
402	}
403	return -EINVAL;
404}
405
406static struct class_attribute class_pktcdvd_attrs[] = {
407 __ATTR(add,            0200, NULL, class_pktcdvd_store_add),
408 __ATTR(remove,         0200, NULL, class_pktcdvd_store_remove),
409 __ATTR(device_map,     0444, class_pktcdvd_show_map, NULL),
410 __ATTR_NULL
411};
412
413
414static int pkt_sysfs_init(void)
415{
416	int ret = 0;
417
418	/*
419	 * create control files in sysfs
420	 * /sys/class/pktcdvd/...
421	 */
422	class_pktcdvd = kzalloc(sizeof(*class_pktcdvd), GFP_KERNEL);
423	if (!class_pktcdvd)
424		return -ENOMEM;
425	class_pktcdvd->name = DRIVER_NAME;
426	class_pktcdvd->owner = THIS_MODULE;
427	class_pktcdvd->class_release = class_pktcdvd_release;
428	class_pktcdvd->class_attrs = class_pktcdvd_attrs;
429	ret = class_register(class_pktcdvd);
430	if (ret) {
431		kfree(class_pktcdvd);
432		class_pktcdvd = NULL;
433		pr_err("failed to create class pktcdvd\n");
434		return ret;
435	}
436	return 0;
437}
438
439static void pkt_sysfs_cleanup(void)
440{
441	if (class_pktcdvd)
442		class_destroy(class_pktcdvd);
443	class_pktcdvd = NULL;
444}
445
446/********************************************************************
447  entries in debugfs
448
449  /sys/kernel/debug/pktcdvd[0-7]/
450			info
451
452 *******************************************************************/
453
454static int pkt_debugfs_seq_show(struct seq_file *m, void *p)
455{
456	return pkt_seq_show(m, p);
457}
458
459static int pkt_debugfs_fops_open(struct inode *inode, struct file *file)
460{
461	return single_open(file, pkt_debugfs_seq_show, inode->i_private);
462}
463
464static const struct file_operations debug_fops = {
465	.open		= pkt_debugfs_fops_open,
466	.read		= seq_read,
467	.llseek		= seq_lseek,
468	.release	= single_release,
469	.owner		= THIS_MODULE,
470};
471
472static void pkt_debugfs_dev_new(struct pktcdvd_device *pd)
473{
474	if (!pkt_debugfs_root)
475		return;
476	pd->dfs_d_root = debugfs_create_dir(pd->name, pkt_debugfs_root);
477	if (!pd->dfs_d_root)
478		return;
479
480	pd->dfs_f_info = debugfs_create_file("info", S_IRUGO,
481				pd->dfs_d_root, pd, &debug_fops);
482}
483
484static void pkt_debugfs_dev_remove(struct pktcdvd_device *pd)
485{
486	if (!pkt_debugfs_root)
487		return;
488	debugfs_remove(pd->dfs_f_info);
489	debugfs_remove(pd->dfs_d_root);
490	pd->dfs_f_info = NULL;
491	pd->dfs_d_root = NULL;
492}
493
494static void pkt_debugfs_init(void)
495{
496	pkt_debugfs_root = debugfs_create_dir(DRIVER_NAME, NULL);
497}
498
499static void pkt_debugfs_cleanup(void)
500{
501	debugfs_remove(pkt_debugfs_root);
502	pkt_debugfs_root = NULL;
503}
504
505/* ----------------------------------------------------------*/
506
507
508static void pkt_bio_finished(struct pktcdvd_device *pd)
509{
510	BUG_ON(atomic_read(&pd->cdrw.pending_bios) <= 0);
511	if (atomic_dec_and_test(&pd->cdrw.pending_bios)) {
512		pkt_dbg(2, pd, "queue empty\n");
513		atomic_set(&pd->iosched.attention, 1);
514		wake_up(&pd->wqueue);
515	}
516}
517
518/*
519 * Allocate a packet_data struct
520 */
521static struct packet_data *pkt_alloc_packet_data(int frames)
522{
523	int i;
524	struct packet_data *pkt;
525
526	pkt = kzalloc(sizeof(struct packet_data), GFP_KERNEL);
527	if (!pkt)
528		goto no_pkt;
529
530	pkt->frames = frames;
531	pkt->w_bio = bio_kmalloc(GFP_KERNEL, frames);
532	if (!pkt->w_bio)
533		goto no_bio;
534
535	for (i = 0; i < frames / FRAMES_PER_PAGE; i++) {
536		pkt->pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
537		if (!pkt->pages[i])
538			goto no_page;
539	}
540
541	spin_lock_init(&pkt->lock);
542	bio_list_init(&pkt->orig_bios);
543
544	for (i = 0; i < frames; i++) {
545		struct bio *bio = bio_kmalloc(GFP_KERNEL, 1);
546		if (!bio)
547			goto no_rd_bio;
548
549		pkt->r_bios[i] = bio;
550	}
551
552	return pkt;
553
554no_rd_bio:
555	for (i = 0; i < frames; i++) {
556		struct bio *bio = pkt->r_bios[i];
557		if (bio)
558			bio_put(bio);
559	}
560
561no_page:
562	for (i = 0; i < frames / FRAMES_PER_PAGE; i++)
563		if (pkt->pages[i])
564			__free_page(pkt->pages[i]);
565	bio_put(pkt->w_bio);
566no_bio:
567	kfree(pkt);
568no_pkt:
569	return NULL;
570}
571
572/*
573 * Free a packet_data struct
574 */
575static void pkt_free_packet_data(struct packet_data *pkt)
576{
577	int i;
578
579	for (i = 0; i < pkt->frames; i++) {
580		struct bio *bio = pkt->r_bios[i];
581		if (bio)
582			bio_put(bio);
583	}
584	for (i = 0; i < pkt->frames / FRAMES_PER_PAGE; i++)
585		__free_page(pkt->pages[i]);
586	bio_put(pkt->w_bio);
587	kfree(pkt);
588}
589
590static void pkt_shrink_pktlist(struct pktcdvd_device *pd)
591{
592	struct packet_data *pkt, *next;
593
594	BUG_ON(!list_empty(&pd->cdrw.pkt_active_list));
595
596	list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_free_list, list) {
597		pkt_free_packet_data(pkt);
598	}
599	INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
600}
601
602static int pkt_grow_pktlist(struct pktcdvd_device *pd, int nr_packets)
603{
604	struct packet_data *pkt;
605
606	BUG_ON(!list_empty(&pd->cdrw.pkt_free_list));
607
608	while (nr_packets > 0) {
609		pkt = pkt_alloc_packet_data(pd->settings.size >> 2);
610		if (!pkt) {
611			pkt_shrink_pktlist(pd);
612			return 0;
613		}
614		pkt->id = nr_packets;
615		pkt->pd = pd;
616		list_add(&pkt->list, &pd->cdrw.pkt_free_list);
617		nr_packets--;
618	}
619	return 1;
620}
621
622static inline struct pkt_rb_node *pkt_rbtree_next(struct pkt_rb_node *node)
623{
624	struct rb_node *n = rb_next(&node->rb_node);
625	if (!n)
626		return NULL;
627	return rb_entry(n, struct pkt_rb_node, rb_node);
628}
629
630static void pkt_rbtree_erase(struct pktcdvd_device *pd, struct pkt_rb_node *node)
631{
632	rb_erase(&node->rb_node, &pd->bio_queue);
633	mempool_free(node, pd->rb_pool);
634	pd->bio_queue_size--;
635	BUG_ON(pd->bio_queue_size < 0);
636}
637
638/*
639 * Find the first node in the pd->bio_queue rb tree with a starting sector >= s.
640 */
641static struct pkt_rb_node *pkt_rbtree_find(struct pktcdvd_device *pd, sector_t s)
642{
643	struct rb_node *n = pd->bio_queue.rb_node;
644	struct rb_node *next;
645	struct pkt_rb_node *tmp;
646
647	if (!n) {
648		BUG_ON(pd->bio_queue_size > 0);
649		return NULL;
650	}
651
652	for (;;) {
653		tmp = rb_entry(n, struct pkt_rb_node, rb_node);
654		if (s <= tmp->bio->bi_iter.bi_sector)
655			next = n->rb_left;
656		else
657			next = n->rb_right;
658		if (!next)
659			break;
660		n = next;
661	}
662
663	if (s > tmp->bio->bi_iter.bi_sector) {
664		tmp = pkt_rbtree_next(tmp);
665		if (!tmp)
666			return NULL;
667	}
668	BUG_ON(s > tmp->bio->bi_iter.bi_sector);
669	return tmp;
670}
671
672/*
673 * Insert a node into the pd->bio_queue rb tree.
674 */
675static void pkt_rbtree_insert(struct pktcdvd_device *pd, struct pkt_rb_node *node)
676{
677	struct rb_node **p = &pd->bio_queue.rb_node;
678	struct rb_node *parent = NULL;
679	sector_t s = node->bio->bi_iter.bi_sector;
680	struct pkt_rb_node *tmp;
681
682	while (*p) {
683		parent = *p;
684		tmp = rb_entry(parent, struct pkt_rb_node, rb_node);
685		if (s < tmp->bio->bi_iter.bi_sector)
686			p = &(*p)->rb_left;
687		else
688			p = &(*p)->rb_right;
689	}
690	rb_link_node(&node->rb_node, parent, p);
691	rb_insert_color(&node->rb_node, &pd->bio_queue);
692	pd->bio_queue_size++;
693}
694
695/*
696 * Send a packet_command to the underlying block device and
697 * wait for completion.
698 */
699static int pkt_generic_packet(struct pktcdvd_device *pd, struct packet_command *cgc)
700{
701	struct request_queue *q = bdev_get_queue(pd->bdev);
702	struct request *rq;
703	int ret = 0;
704
705	rq = blk_get_request(q, (cgc->data_direction == CGC_DATA_WRITE) ?
706			     WRITE : READ, __GFP_WAIT);
707	if (IS_ERR(rq))
708		return PTR_ERR(rq);
709	blk_rq_set_block_pc(rq);
710
711	if (cgc->buflen) {
712		ret = blk_rq_map_kern(q, rq, cgc->buffer, cgc->buflen,
713				      __GFP_WAIT);
714		if (ret)
715			goto out;
716	}
717
718	rq->cmd_len = COMMAND_SIZE(cgc->cmd[0]);
719	memcpy(rq->cmd, cgc->cmd, CDROM_PACKET_SIZE);
720
721	rq->timeout = 60*HZ;
722	if (cgc->quiet)
723		rq->cmd_flags |= REQ_QUIET;
724
725	blk_execute_rq(rq->q, pd->bdev->bd_disk, rq, 0);
726	if (rq->errors)
727		ret = -EIO;
728out:
729	blk_put_request(rq);
730	return ret;
731}
732
733static const char *sense_key_string(__u8 index)
734{
735	static const char * const info[] = {
736		"No sense", "Recovered error", "Not ready",
737		"Medium error", "Hardware error", "Illegal request",
738		"Unit attention", "Data protect", "Blank check",
739	};
740
741	return index < ARRAY_SIZE(info) ? info[index] : "INVALID";
742}
743
744/*
745 * A generic sense dump / resolve mechanism should be implemented across
746 * all ATAPI + SCSI devices.
747 */
748static void pkt_dump_sense(struct pktcdvd_device *pd,
749			   struct packet_command *cgc)
750{
751	struct request_sense *sense = cgc->sense;
752
753	if (sense)
754		pkt_err(pd, "%*ph - sense %02x.%02x.%02x (%s)\n",
755			CDROM_PACKET_SIZE, cgc->cmd,
756			sense->sense_key, sense->asc, sense->ascq,
757			sense_key_string(sense->sense_key));
758	else
759		pkt_err(pd, "%*ph - no sense\n", CDROM_PACKET_SIZE, cgc->cmd);
760}
761
762/*
763 * flush the drive cache to media
764 */
765static int pkt_flush_cache(struct pktcdvd_device *pd)
766{
767	struct packet_command cgc;
768
769	init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
770	cgc.cmd[0] = GPCMD_FLUSH_CACHE;
771	cgc.quiet = 1;
772
773	/*
774	 * the IMMED bit -- we default to not setting it, although that
775	 * would allow a much faster close, this is safer
776	 */
777#if 0
778	cgc.cmd[1] = 1 << 1;
779#endif
780	return pkt_generic_packet(pd, &cgc);
781}
782
783/*
784 * speed is given as the normal factor, e.g. 4 for 4x
785 */
786static noinline_for_stack int pkt_set_speed(struct pktcdvd_device *pd,
787				unsigned write_speed, unsigned read_speed)
788{
789	struct packet_command cgc;
790	struct request_sense sense;
791	int ret;
792
793	init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
794	cgc.sense = &sense;
795	cgc.cmd[0] = GPCMD_SET_SPEED;
796	cgc.cmd[2] = (read_speed >> 8) & 0xff;
797	cgc.cmd[3] = read_speed & 0xff;
798	cgc.cmd[4] = (write_speed >> 8) & 0xff;
799	cgc.cmd[5] = write_speed & 0xff;
800
801	if ((ret = pkt_generic_packet(pd, &cgc)))
802		pkt_dump_sense(pd, &cgc);
803
804	return ret;
805}
806
807/*
808 * Queue a bio for processing by the low-level CD device. Must be called
809 * from process context.
810 */
811static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio)
812{
813	spin_lock(&pd->iosched.lock);
814	if (bio_data_dir(bio) == READ)
815		bio_list_add(&pd->iosched.read_queue, bio);
816	else
817		bio_list_add(&pd->iosched.write_queue, bio);
818	spin_unlock(&pd->iosched.lock);
819
820	atomic_set(&pd->iosched.attention, 1);
821	wake_up(&pd->wqueue);
822}
823
824/*
825 * Process the queued read/write requests. This function handles special
826 * requirements for CDRW drives:
827 * - A cache flush command must be inserted before a read request if the
828 *   previous request was a write.
829 * - Switching between reading and writing is slow, so don't do it more often
830 *   than necessary.
831 * - Optimize for throughput at the expense of latency. This means that streaming
832 *   writes will never be interrupted by a read, but if the drive has to seek
833 *   before the next write, switch to reading instead if there are any pending
834 *   read requests.
835 * - Set the read speed according to current usage pattern. When only reading
836 *   from the device, it's best to use the highest possible read speed, but
837 *   when switching often between reading and writing, it's better to have the
838 *   same read and write speeds.
839 */
840static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
841{
842
843	if (atomic_read(&pd->iosched.attention) == 0)
844		return;
845	atomic_set(&pd->iosched.attention, 0);
846
847	for (;;) {
848		struct bio *bio;
849		int reads_queued, writes_queued;
850
851		spin_lock(&pd->iosched.lock);
852		reads_queued = !bio_list_empty(&pd->iosched.read_queue);
853		writes_queued = !bio_list_empty(&pd->iosched.write_queue);
854		spin_unlock(&pd->iosched.lock);
855
856		if (!reads_queued && !writes_queued)
857			break;
858
859		if (pd->iosched.writing) {
860			int need_write_seek = 1;
861			spin_lock(&pd->iosched.lock);
862			bio = bio_list_peek(&pd->iosched.write_queue);
863			spin_unlock(&pd->iosched.lock);
864			if (bio && (bio->bi_iter.bi_sector ==
865				    pd->iosched.last_write))
866				need_write_seek = 0;
867			if (need_write_seek && reads_queued) {
868				if (atomic_read(&pd->cdrw.pending_bios) > 0) {
869					pkt_dbg(2, pd, "write, waiting\n");
870					break;
871				}
872				pkt_flush_cache(pd);
873				pd->iosched.writing = 0;
874			}
875		} else {
876			if (!reads_queued && writes_queued) {
877				if (atomic_read(&pd->cdrw.pending_bios) > 0) {
878					pkt_dbg(2, pd, "read, waiting\n");
879					break;
880				}
881				pd->iosched.writing = 1;
882			}
883		}
884
885		spin_lock(&pd->iosched.lock);
886		if (pd->iosched.writing)
887			bio = bio_list_pop(&pd->iosched.write_queue);
888		else
889			bio = bio_list_pop(&pd->iosched.read_queue);
890		spin_unlock(&pd->iosched.lock);
891
892		if (!bio)
893			continue;
894
895		if (bio_data_dir(bio) == READ)
896			pd->iosched.successive_reads +=
897				bio->bi_iter.bi_size >> 10;
898		else {
899			pd->iosched.successive_reads = 0;
900			pd->iosched.last_write = bio_end_sector(bio);
901		}
902		if (pd->iosched.successive_reads >= HI_SPEED_SWITCH) {
903			if (pd->read_speed == pd->write_speed) {
904				pd->read_speed = MAX_SPEED;
905				pkt_set_speed(pd, pd->write_speed, pd->read_speed);
906			}
907		} else {
908			if (pd->read_speed != pd->write_speed) {
909				pd->read_speed = pd->write_speed;
910				pkt_set_speed(pd, pd->write_speed, pd->read_speed);
911			}
912		}
913
914		atomic_inc(&pd->cdrw.pending_bios);
915		generic_make_request(bio);
916	}
917}
918
919/*
920 * Special care is needed if the underlying block device has a small
921 * max_phys_segments value.
922 */
923static int pkt_set_segment_merging(struct pktcdvd_device *pd, struct request_queue *q)
924{
925	if ((pd->settings.size << 9) / CD_FRAMESIZE
926	    <= queue_max_segments(q)) {
927		/*
928		 * The cdrom device can handle one segment/frame
929		 */
930		clear_bit(PACKET_MERGE_SEGS, &pd->flags);
931		return 0;
932	} else if ((pd->settings.size << 9) / PAGE_SIZE
933		   <= queue_max_segments(q)) {
934		/*
935		 * We can handle this case at the expense of some extra memory
936		 * copies during write operations
937		 */
938		set_bit(PACKET_MERGE_SEGS, &pd->flags);
939		return 0;
940	} else {
941		pkt_err(pd, "cdrom max_phys_segments too small\n");
942		return -EIO;
943	}
944}
945
946/*
947 * Copy all data for this packet to pkt->pages[], so that
948 * a) The number of required segments for the write bio is minimized, which
949 *    is necessary for some scsi controllers.
950 * b) The data can be used as cache to avoid read requests if we receive a
951 *    new write request for the same zone.
952 */
953static void pkt_make_local_copy(struct packet_data *pkt, struct bio_vec *bvec)
954{
955	int f, p, offs;
956
957	/* Copy all data to pkt->pages[] */
958	p = 0;
959	offs = 0;
960	for (f = 0; f < pkt->frames; f++) {
961		if (bvec[f].bv_page != pkt->pages[p]) {
962			void *vfrom = kmap_atomic(bvec[f].bv_page) + bvec[f].bv_offset;
963			void *vto = page_address(pkt->pages[p]) + offs;
964			memcpy(vto, vfrom, CD_FRAMESIZE);
965			kunmap_atomic(vfrom);
966			bvec[f].bv_page = pkt->pages[p];
967			bvec[f].bv_offset = offs;
968		} else {
969			BUG_ON(bvec[f].bv_offset != offs);
970		}
971		offs += CD_FRAMESIZE;
972		if (offs >= PAGE_SIZE) {
973			offs = 0;
974			p++;
975		}
976	}
977}
978
979static void pkt_end_io_read(struct bio *bio, int err)
980{
981	struct packet_data *pkt = bio->bi_private;
982	struct pktcdvd_device *pd = pkt->pd;
983	BUG_ON(!pd);
984
985	pkt_dbg(2, pd, "bio=%p sec0=%llx sec=%llx err=%d\n",
986		bio, (unsigned long long)pkt->sector,
987		(unsigned long long)bio->bi_iter.bi_sector, err);
988
989	if (err)
990		atomic_inc(&pkt->io_errors);
991	if (atomic_dec_and_test(&pkt->io_wait)) {
992		atomic_inc(&pkt->run_sm);
993		wake_up(&pd->wqueue);
994	}
995	pkt_bio_finished(pd);
996}
997
998static void pkt_end_io_packet_write(struct bio *bio, int err)
999{
1000	struct packet_data *pkt = bio->bi_private;
1001	struct pktcdvd_device *pd = pkt->pd;
1002	BUG_ON(!pd);
1003
1004	pkt_dbg(2, pd, "id=%d, err=%d\n", pkt->id, err);
1005
1006	pd->stats.pkt_ended++;
1007
1008	pkt_bio_finished(pd);
1009	atomic_dec(&pkt->io_wait);
1010	atomic_inc(&pkt->run_sm);
1011	wake_up(&pd->wqueue);
1012}
1013
1014/*
1015 * Schedule reads for the holes in a packet
1016 */
1017static void pkt_gather_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1018{
1019	int frames_read = 0;
1020	struct bio *bio;
1021	int f;
1022	char written[PACKET_MAX_SIZE];
1023
1024	BUG_ON(bio_list_empty(&pkt->orig_bios));
1025
1026	atomic_set(&pkt->io_wait, 0);
1027	atomic_set(&pkt->io_errors, 0);
1028
1029	/*
1030	 * Figure out which frames we need to read before we can write.
1031	 */
1032	memset(written, 0, sizeof(written));
1033	spin_lock(&pkt->lock);
1034	bio_list_for_each(bio, &pkt->orig_bios) {
1035		int first_frame = (bio->bi_iter.bi_sector - pkt->sector) /
1036			(CD_FRAMESIZE >> 9);
1037		int num_frames = bio->bi_iter.bi_size / CD_FRAMESIZE;
1038		pd->stats.secs_w += num_frames * (CD_FRAMESIZE >> 9);
1039		BUG_ON(first_frame < 0);
1040		BUG_ON(first_frame + num_frames > pkt->frames);
1041		for (f = first_frame; f < first_frame + num_frames; f++)
1042			written[f] = 1;
1043	}
1044	spin_unlock(&pkt->lock);
1045
1046	if (pkt->cache_valid) {
1047		pkt_dbg(2, pd, "zone %llx cached\n",
1048			(unsigned long long)pkt->sector);
1049		goto out_account;
1050	}
1051
1052	/*
1053	 * Schedule reads for missing parts of the packet.
1054	 */
1055	for (f = 0; f < pkt->frames; f++) {
1056		int p, offset;
1057
1058		if (written[f])
1059			continue;
1060
1061		bio = pkt->r_bios[f];
1062		bio_reset(bio);
1063		bio->bi_iter.bi_sector = pkt->sector + f * (CD_FRAMESIZE >> 9);
1064		bio->bi_bdev = pd->bdev;
1065		bio->bi_end_io = pkt_end_io_read;
1066		bio->bi_private = pkt;
1067
1068		p = (f * CD_FRAMESIZE) / PAGE_SIZE;
1069		offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1070		pkt_dbg(2, pd, "Adding frame %d, page:%p offs:%d\n",
1071			f, pkt->pages[p], offset);
1072		if (!bio_add_page(bio, pkt->pages[p], CD_FRAMESIZE, offset))
1073			BUG();
1074
1075		atomic_inc(&pkt->io_wait);
1076		bio->bi_rw = READ;
1077		pkt_queue_bio(pd, bio);
1078		frames_read++;
1079	}
1080
1081out_account:
1082	pkt_dbg(2, pd, "need %d frames for zone %llx\n",
1083		frames_read, (unsigned long long)pkt->sector);
1084	pd->stats.pkt_started++;
1085	pd->stats.secs_rg += frames_read * (CD_FRAMESIZE >> 9);
1086}
1087
1088/*
1089 * Find a packet matching zone, or the least recently used packet if
1090 * there is no match.
1091 */
1092static struct packet_data *pkt_get_packet_data(struct pktcdvd_device *pd, int zone)
1093{
1094	struct packet_data *pkt;
1095
1096	list_for_each_entry(pkt, &pd->cdrw.pkt_free_list, list) {
1097		if (pkt->sector == zone || pkt->list.next == &pd->cdrw.pkt_free_list) {
1098			list_del_init(&pkt->list);
1099			if (pkt->sector != zone)
1100				pkt->cache_valid = 0;
1101			return pkt;
1102		}
1103	}
1104	BUG();
1105	return NULL;
1106}
1107
1108static void pkt_put_packet_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1109{
1110	if (pkt->cache_valid) {
1111		list_add(&pkt->list, &pd->cdrw.pkt_free_list);
1112	} else {
1113		list_add_tail(&pkt->list, &pd->cdrw.pkt_free_list);
1114	}
1115}
1116
1117/*
1118 * recover a failed write, query for relocation if possible
1119 *
1120 * returns 1 if recovery is possible, or 0 if not
1121 *
1122 */
1123static int pkt_start_recovery(struct packet_data *pkt)
1124{
1125	/*
1126	 * FIXME. We need help from the file system to implement
1127	 * recovery handling.
1128	 */
1129	return 0;
1130#if 0
1131	struct request *rq = pkt->rq;
1132	struct pktcdvd_device *pd = rq->rq_disk->private_data;
1133	struct block_device *pkt_bdev;
1134	struct super_block *sb = NULL;
1135	unsigned long old_block, new_block;
1136	sector_t new_sector;
1137
1138	pkt_bdev = bdget(kdev_t_to_nr(pd->pkt_dev));
1139	if (pkt_bdev) {
1140		sb = get_super(pkt_bdev);
1141		bdput(pkt_bdev);
1142	}
1143
1144	if (!sb)
1145		return 0;
1146
1147	if (!sb->s_op->relocate_blocks)
1148		goto out;
1149
1150	old_block = pkt->sector / (CD_FRAMESIZE >> 9);
1151	if (sb->s_op->relocate_blocks(sb, old_block, &new_block))
1152		goto out;
1153
1154	new_sector = new_block * (CD_FRAMESIZE >> 9);
1155	pkt->sector = new_sector;
1156
1157	bio_reset(pkt->bio);
1158	pkt->bio->bi_bdev = pd->bdev;
1159	pkt->bio->bi_rw = REQ_WRITE;
1160	pkt->bio->bi_iter.bi_sector = new_sector;
1161	pkt->bio->bi_iter.bi_size = pkt->frames * CD_FRAMESIZE;
1162	pkt->bio->bi_vcnt = pkt->frames;
1163
1164	pkt->bio->bi_end_io = pkt_end_io_packet_write;
1165	pkt->bio->bi_private = pkt;
1166
1167	drop_super(sb);
1168	return 1;
1169
1170out:
1171	drop_super(sb);
1172	return 0;
1173#endif
1174}
1175
1176static inline void pkt_set_state(struct packet_data *pkt, enum packet_data_state state)
1177{
1178#if PACKET_DEBUG > 1
1179	static const char *state_name[] = {
1180		"IDLE", "WAITING", "READ_WAIT", "WRITE_WAIT", "RECOVERY", "FINISHED"
1181	};
1182	enum packet_data_state old_state = pkt->state;
1183	pkt_dbg(2, pd, "pkt %2d : s=%6llx %s -> %s\n",
1184		pkt->id, (unsigned long long)pkt->sector,
1185		state_name[old_state], state_name[state]);
1186#endif
1187	pkt->state = state;
1188}
1189
1190/*
1191 * Scan the work queue to see if we can start a new packet.
1192 * returns non-zero if any work was done.
1193 */
1194static int pkt_handle_queue(struct pktcdvd_device *pd)
1195{
1196	struct packet_data *pkt, *p;
1197	struct bio *bio = NULL;
1198	sector_t zone = 0; /* Suppress gcc warning */
1199	struct pkt_rb_node *node, *first_node;
1200	struct rb_node *n;
1201	int wakeup;
1202
1203	atomic_set(&pd->scan_queue, 0);
1204
1205	if (list_empty(&pd->cdrw.pkt_free_list)) {
1206		pkt_dbg(2, pd, "no pkt\n");
1207		return 0;
1208	}
1209
1210	/*
1211	 * Try to find a zone we are not already working on.
1212	 */
1213	spin_lock(&pd->lock);
1214	first_node = pkt_rbtree_find(pd, pd->current_sector);
1215	if (!first_node) {
1216		n = rb_first(&pd->bio_queue);
1217		if (n)
1218			first_node = rb_entry(n, struct pkt_rb_node, rb_node);
1219	}
1220	node = first_node;
1221	while (node) {
1222		bio = node->bio;
1223		zone = get_zone(bio->bi_iter.bi_sector, pd);
1224		list_for_each_entry(p, &pd->cdrw.pkt_active_list, list) {
1225			if (p->sector == zone) {
1226				bio = NULL;
1227				goto try_next_bio;
1228			}
1229		}
1230		break;
1231try_next_bio:
1232		node = pkt_rbtree_next(node);
1233		if (!node) {
1234			n = rb_first(&pd->bio_queue);
1235			if (n)
1236				node = rb_entry(n, struct pkt_rb_node, rb_node);
1237		}
1238		if (node == first_node)
1239			node = NULL;
1240	}
1241	spin_unlock(&pd->lock);
1242	if (!bio) {
1243		pkt_dbg(2, pd, "no bio\n");
1244		return 0;
1245	}
1246
1247	pkt = pkt_get_packet_data(pd, zone);
1248
1249	pd->current_sector = zone + pd->settings.size;
1250	pkt->sector = zone;
1251	BUG_ON(pkt->frames != pd->settings.size >> 2);
1252	pkt->write_size = 0;
1253
1254	/*
1255	 * Scan work queue for bios in the same zone and link them
1256	 * to this packet.
1257	 */
1258	spin_lock(&pd->lock);
1259	pkt_dbg(2, pd, "looking for zone %llx\n", (unsigned long long)zone);
1260	while ((node = pkt_rbtree_find(pd, zone)) != NULL) {
1261		bio = node->bio;
1262		pkt_dbg(2, pd, "found zone=%llx\n", (unsigned long long)
1263			get_zone(bio->bi_iter.bi_sector, pd));
1264		if (get_zone(bio->bi_iter.bi_sector, pd) != zone)
1265			break;
1266		pkt_rbtree_erase(pd, node);
1267		spin_lock(&pkt->lock);
1268		bio_list_add(&pkt->orig_bios, bio);
1269		pkt->write_size += bio->bi_iter.bi_size / CD_FRAMESIZE;
1270		spin_unlock(&pkt->lock);
1271	}
1272	/* check write congestion marks, and if bio_queue_size is
1273	   below, wake up any waiters */
1274	wakeup = (pd->write_congestion_on > 0
1275	 		&& pd->bio_queue_size <= pd->write_congestion_off);
1276	spin_unlock(&pd->lock);
1277	if (wakeup) {
1278		clear_bdi_congested(&pd->disk->queue->backing_dev_info,
1279					BLK_RW_ASYNC);
1280	}
1281
1282	pkt->sleep_time = max(PACKET_WAIT_TIME, 1);
1283	pkt_set_state(pkt, PACKET_WAITING_STATE);
1284	atomic_set(&pkt->run_sm, 1);
1285
1286	spin_lock(&pd->cdrw.active_list_lock);
1287	list_add(&pkt->list, &pd->cdrw.pkt_active_list);
1288	spin_unlock(&pd->cdrw.active_list_lock);
1289
1290	return 1;
1291}
1292
1293/*
1294 * Assemble a bio to write one packet and queue the bio for processing
1295 * by the underlying block device.
1296 */
1297static void pkt_start_write(struct pktcdvd_device *pd, struct packet_data *pkt)
1298{
1299	int f;
1300	struct bio_vec *bvec = pkt->w_bio->bi_io_vec;
1301
1302	bio_reset(pkt->w_bio);
1303	pkt->w_bio->bi_iter.bi_sector = pkt->sector;
1304	pkt->w_bio->bi_bdev = pd->bdev;
1305	pkt->w_bio->bi_end_io = pkt_end_io_packet_write;
1306	pkt->w_bio->bi_private = pkt;
1307
1308	/* XXX: locking? */
1309	for (f = 0; f < pkt->frames; f++) {
1310		bvec[f].bv_page = pkt->pages[(f * CD_FRAMESIZE) / PAGE_SIZE];
1311		bvec[f].bv_offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1312		if (!bio_add_page(pkt->w_bio, bvec[f].bv_page, CD_FRAMESIZE, bvec[f].bv_offset))
1313			BUG();
1314	}
1315	pkt_dbg(2, pd, "vcnt=%d\n", pkt->w_bio->bi_vcnt);
1316
1317	/*
1318	 * Fill-in bvec with data from orig_bios.
1319	 */
1320	spin_lock(&pkt->lock);
1321	bio_copy_data(pkt->w_bio, pkt->orig_bios.head);
1322
1323	pkt_set_state(pkt, PACKET_WRITE_WAIT_STATE);
1324	spin_unlock(&pkt->lock);
1325
1326	pkt_dbg(2, pd, "Writing %d frames for zone %llx\n",
1327		pkt->write_size, (unsigned long long)pkt->sector);
1328
1329	if (test_bit(PACKET_MERGE_SEGS, &pd->flags) || (pkt->write_size < pkt->frames)) {
1330		pkt_make_local_copy(pkt, bvec);
1331		pkt->cache_valid = 1;
1332	} else {
1333		pkt->cache_valid = 0;
1334	}
1335
1336	/* Start the write request */
1337	atomic_set(&pkt->io_wait, 1);
1338	pkt->w_bio->bi_rw = WRITE;
1339	pkt_queue_bio(pd, pkt->w_bio);
1340}
1341
1342static void pkt_finish_packet(struct packet_data *pkt, int uptodate)
1343{
1344	struct bio *bio;
1345
1346	if (!uptodate)
1347		pkt->cache_valid = 0;
1348
1349	/* Finish all bios corresponding to this packet */
1350	while ((bio = bio_list_pop(&pkt->orig_bios)))
1351		bio_endio(bio, uptodate ? 0 : -EIO);
1352}
1353
1354static void pkt_run_state_machine(struct pktcdvd_device *pd, struct packet_data *pkt)
1355{
1356	int uptodate;
1357
1358	pkt_dbg(2, pd, "pkt %d\n", pkt->id);
1359
1360	for (;;) {
1361		switch (pkt->state) {
1362		case PACKET_WAITING_STATE:
1363			if ((pkt->write_size < pkt->frames) && (pkt->sleep_time > 0))
1364				return;
1365
1366			pkt->sleep_time = 0;
1367			pkt_gather_data(pd, pkt);
1368			pkt_set_state(pkt, PACKET_READ_WAIT_STATE);
1369			break;
1370
1371		case PACKET_READ_WAIT_STATE:
1372			if (atomic_read(&pkt->io_wait) > 0)
1373				return;
1374
1375			if (atomic_read(&pkt->io_errors) > 0) {
1376				pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1377			} else {
1378				pkt_start_write(pd, pkt);
1379			}
1380			break;
1381
1382		case PACKET_WRITE_WAIT_STATE:
1383			if (atomic_read(&pkt->io_wait) > 0)
1384				return;
1385
1386			if (test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags)) {
1387				pkt_set_state(pkt, PACKET_FINISHED_STATE);
1388			} else {
1389				pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1390			}
1391			break;
1392
1393		case PACKET_RECOVERY_STATE:
1394			if (pkt_start_recovery(pkt)) {
1395				pkt_start_write(pd, pkt);
1396			} else {
1397				pkt_dbg(2, pd, "No recovery possible\n");
1398				pkt_set_state(pkt, PACKET_FINISHED_STATE);
1399			}
1400			break;
1401
1402		case PACKET_FINISHED_STATE:
1403			uptodate = test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags);
1404			pkt_finish_packet(pkt, uptodate);
1405			return;
1406
1407		default:
1408			BUG();
1409			break;
1410		}
1411	}
1412}
1413
1414static void pkt_handle_packets(struct pktcdvd_device *pd)
1415{
1416	struct packet_data *pkt, *next;
1417
1418	/*
1419	 * Run state machine for active packets
1420	 */
1421	list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1422		if (atomic_read(&pkt->run_sm) > 0) {
1423			atomic_set(&pkt->run_sm, 0);
1424			pkt_run_state_machine(pd, pkt);
1425		}
1426	}
1427
1428	/*
1429	 * Move no longer active packets to the free list
1430	 */
1431	spin_lock(&pd->cdrw.active_list_lock);
1432	list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_active_list, list) {
1433		if (pkt->state == PACKET_FINISHED_STATE) {
1434			list_del(&pkt->list);
1435			pkt_put_packet_data(pd, pkt);
1436			pkt_set_state(pkt, PACKET_IDLE_STATE);
1437			atomic_set(&pd->scan_queue, 1);
1438		}
1439	}
1440	spin_unlock(&pd->cdrw.active_list_lock);
1441}
1442
1443static void pkt_count_states(struct pktcdvd_device *pd, int *states)
1444{
1445	struct packet_data *pkt;
1446	int i;
1447
1448	for (i = 0; i < PACKET_NUM_STATES; i++)
1449		states[i] = 0;
1450
1451	spin_lock(&pd->cdrw.active_list_lock);
1452	list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1453		states[pkt->state]++;
1454	}
1455	spin_unlock(&pd->cdrw.active_list_lock);
1456}
1457
1458/*
1459 * kcdrwd is woken up when writes have been queued for one of our
1460 * registered devices
1461 */
1462static int kcdrwd(void *foobar)
1463{
1464	struct pktcdvd_device *pd = foobar;
1465	struct packet_data *pkt;
1466	long min_sleep_time, residue;
1467
1468	set_user_nice(current, MIN_NICE);
1469	set_freezable();
1470
1471	for (;;) {
1472		DECLARE_WAITQUEUE(wait, current);
1473
1474		/*
1475		 * Wait until there is something to do
1476		 */
1477		add_wait_queue(&pd->wqueue, &wait);
1478		for (;;) {
1479			set_current_state(TASK_INTERRUPTIBLE);
1480
1481			/* Check if we need to run pkt_handle_queue */
1482			if (atomic_read(&pd->scan_queue) > 0)
1483				goto work_to_do;
1484
1485			/* Check if we need to run the state machine for some packet */
1486			list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1487				if (atomic_read(&pkt->run_sm) > 0)
1488					goto work_to_do;
1489			}
1490
1491			/* Check if we need to process the iosched queues */
1492			if (atomic_read(&pd->iosched.attention) != 0)
1493				goto work_to_do;
1494
1495			/* Otherwise, go to sleep */
1496			if (PACKET_DEBUG > 1) {
1497				int states[PACKET_NUM_STATES];
1498				pkt_count_states(pd, states);
1499				pkt_dbg(2, pd, "i:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
1500					states[0], states[1], states[2],
1501					states[3], states[4], states[5]);
1502			}
1503
1504			min_sleep_time = MAX_SCHEDULE_TIMEOUT;
1505			list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1506				if (pkt->sleep_time && pkt->sleep_time < min_sleep_time)
1507					min_sleep_time = pkt->sleep_time;
1508			}
1509
1510			pkt_dbg(2, pd, "sleeping\n");
1511			residue = schedule_timeout(min_sleep_time);
1512			pkt_dbg(2, pd, "wake up\n");
1513
1514			/* make swsusp happy with our thread */
1515			try_to_freeze();
1516
1517			list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1518				if (!pkt->sleep_time)
1519					continue;
1520				pkt->sleep_time -= min_sleep_time - residue;
1521				if (pkt->sleep_time <= 0) {
1522					pkt->sleep_time = 0;
1523					atomic_inc(&pkt->run_sm);
1524				}
1525			}
1526
1527			if (kthread_should_stop())
1528				break;
1529		}
1530work_to_do:
1531		set_current_state(TASK_RUNNING);
1532		remove_wait_queue(&pd->wqueue, &wait);
1533
1534		if (kthread_should_stop())
1535			break;
1536
1537		/*
1538		 * if pkt_handle_queue returns true, we can queue
1539		 * another request.
1540		 */
1541		while (pkt_handle_queue(pd))
1542			;
1543
1544		/*
1545		 * Handle packet state machine
1546		 */
1547		pkt_handle_packets(pd);
1548
1549		/*
1550		 * Handle iosched queues
1551		 */
1552		pkt_iosched_process_queue(pd);
1553	}
1554
1555	return 0;
1556}
1557
1558static void pkt_print_settings(struct pktcdvd_device *pd)
1559{
1560	pkt_info(pd, "%s packets, %u blocks, Mode-%c disc\n",
1561		 pd->settings.fp ? "Fixed" : "Variable",
1562		 pd->settings.size >> 2,
1563		 pd->settings.block_mode == 8 ? '1' : '2');
1564}
1565
1566static int pkt_mode_sense(struct pktcdvd_device *pd, struct packet_command *cgc, int page_code, int page_control)
1567{
1568	memset(cgc->cmd, 0, sizeof(cgc->cmd));
1569
1570	cgc->cmd[0] = GPCMD_MODE_SENSE_10;
1571	cgc->cmd[2] = page_code | (page_control << 6);
1572	cgc->cmd[7] = cgc->buflen >> 8;
1573	cgc->cmd[8] = cgc->buflen & 0xff;
1574	cgc->data_direction = CGC_DATA_READ;
1575	return pkt_generic_packet(pd, cgc);
1576}
1577
1578static int pkt_mode_select(struct pktcdvd_device *pd, struct packet_command *cgc)
1579{
1580	memset(cgc->cmd, 0, sizeof(cgc->cmd));
1581	memset(cgc->buffer, 0, 2);
1582	cgc->cmd[0] = GPCMD_MODE_SELECT_10;
1583	cgc->cmd[1] = 0x10;		/* PF */
1584	cgc->cmd[7] = cgc->buflen >> 8;
1585	cgc->cmd[8] = cgc->buflen & 0xff;
1586	cgc->data_direction = CGC_DATA_WRITE;
1587	return pkt_generic_packet(pd, cgc);
1588}
1589
1590static int pkt_get_disc_info(struct pktcdvd_device *pd, disc_information *di)
1591{
1592	struct packet_command cgc;
1593	int ret;
1594
1595	/* set up command and get the disc info */
1596	init_cdrom_command(&cgc, di, sizeof(*di), CGC_DATA_READ);
1597	cgc.cmd[0] = GPCMD_READ_DISC_INFO;
1598	cgc.cmd[8] = cgc.buflen = 2;
1599	cgc.quiet = 1;
1600
1601	if ((ret = pkt_generic_packet(pd, &cgc)))
1602		return ret;
1603
1604	/* not all drives have the same disc_info length, so requeue
1605	 * packet with the length the drive tells us it can supply
1606	 */
1607	cgc.buflen = be16_to_cpu(di->disc_information_length) +
1608		     sizeof(di->disc_information_length);
1609
1610	if (cgc.buflen > sizeof(disc_information))
1611		cgc.buflen = sizeof(disc_information);
1612
1613	cgc.cmd[8] = cgc.buflen;
1614	return pkt_generic_packet(pd, &cgc);
1615}
1616
1617static int pkt_get_track_info(struct pktcdvd_device *pd, __u16 track, __u8 type, track_information *ti)
1618{
1619	struct packet_command cgc;
1620	int ret;
1621
1622	init_cdrom_command(&cgc, ti, 8, CGC_DATA_READ);
1623	cgc.cmd[0] = GPCMD_READ_TRACK_RZONE_INFO;
1624	cgc.cmd[1] = type & 3;
1625	cgc.cmd[4] = (track & 0xff00) >> 8;
1626	cgc.cmd[5] = track & 0xff;
1627	cgc.cmd[8] = 8;
1628	cgc.quiet = 1;
1629
1630	if ((ret = pkt_generic_packet(pd, &cgc)))
1631		return ret;
1632
1633	cgc.buflen = be16_to_cpu(ti->track_information_length) +
1634		     sizeof(ti->track_information_length);
1635
1636	if (cgc.buflen > sizeof(track_information))
1637		cgc.buflen = sizeof(track_information);
1638
1639	cgc.cmd[8] = cgc.buflen;
1640	return pkt_generic_packet(pd, &cgc);
1641}
1642
1643static noinline_for_stack int pkt_get_last_written(struct pktcdvd_device *pd,
1644						long *last_written)
1645{
1646	disc_information di;
1647	track_information ti;
1648	__u32 last_track;
1649	int ret = -1;
1650
1651	if ((ret = pkt_get_disc_info(pd, &di)))
1652		return ret;
1653
1654	last_track = (di.last_track_msb << 8) | di.last_track_lsb;
1655	if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1656		return ret;
1657
1658	/* if this track is blank, try the previous. */
1659	if (ti.blank) {
1660		last_track--;
1661		if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1662			return ret;
1663	}
1664
1665	/* if last recorded field is valid, return it. */
1666	if (ti.lra_v) {
1667		*last_written = be32_to_cpu(ti.last_rec_address);
1668	} else {
1669		/* make it up instead */
1670		*last_written = be32_to_cpu(ti.track_start) +
1671				be32_to_cpu(ti.track_size);
1672		if (ti.free_blocks)
1673			*last_written -= (be32_to_cpu(ti.free_blocks) + 7);
1674	}
1675	return 0;
1676}
1677
1678/*
1679 * write mode select package based on pd->settings
1680 */
1681static noinline_for_stack int pkt_set_write_settings(struct pktcdvd_device *pd)
1682{
1683	struct packet_command cgc;
1684	struct request_sense sense;
1685	write_param_page *wp;
1686	char buffer[128];
1687	int ret, size;
1688
1689	/* doesn't apply to DVD+RW or DVD-RAM */
1690	if ((pd->mmc3_profile == 0x1a) || (pd->mmc3_profile == 0x12))
1691		return 0;
1692
1693	memset(buffer, 0, sizeof(buffer));
1694	init_cdrom_command(&cgc, buffer, sizeof(*wp), CGC_DATA_READ);
1695	cgc.sense = &sense;
1696	if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1697		pkt_dump_sense(pd, &cgc);
1698		return ret;
1699	}
1700
1701	size = 2 + ((buffer[0] << 8) | (buffer[1] & 0xff));
1702	pd->mode_offset = (buffer[6] << 8) | (buffer[7] & 0xff);
1703	if (size > sizeof(buffer))
1704		size = sizeof(buffer);
1705
1706	/*
1707	 * now get it all
1708	 */
1709	init_cdrom_command(&cgc, buffer, size, CGC_DATA_READ);
1710	cgc.sense = &sense;
1711	if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1712		pkt_dump_sense(pd, &cgc);
1713		return ret;
1714	}
1715
1716	/*
1717	 * write page is offset header + block descriptor length
1718	 */
1719	wp = (write_param_page *) &buffer[sizeof(struct mode_page_header) + pd->mode_offset];
1720
1721	wp->fp = pd->settings.fp;
1722	wp->track_mode = pd->settings.track_mode;
1723	wp->write_type = pd->settings.write_type;
1724	wp->data_block_type = pd->settings.block_mode;
1725
1726	wp->multi_session = 0;
1727
1728#ifdef PACKET_USE_LS
1729	wp->link_size = 7;
1730	wp->ls_v = 1;
1731#endif
1732
1733	if (wp->data_block_type == PACKET_BLOCK_MODE1) {
1734		wp->session_format = 0;
1735		wp->subhdr2 = 0x20;
1736	} else if (wp->data_block_type == PACKET_BLOCK_MODE2) {
1737		wp->session_format = 0x20;
1738		wp->subhdr2 = 8;
1739#if 0
1740		wp->mcn[0] = 0x80;
1741		memcpy(&wp->mcn[1], PACKET_MCN, sizeof(wp->mcn) - 1);
1742#endif
1743	} else {
1744		/*
1745		 * paranoia
1746		 */
1747		pkt_err(pd, "write mode wrong %d\n", wp->data_block_type);
1748		return 1;
1749	}
1750	wp->packet_size = cpu_to_be32(pd->settings.size >> 2);
1751
1752	cgc.buflen = cgc.cmd[8] = size;
1753	if ((ret = pkt_mode_select(pd, &cgc))) {
1754		pkt_dump_sense(pd, &cgc);
1755		return ret;
1756	}
1757
1758	pkt_print_settings(pd);
1759	return 0;
1760}
1761
1762/*
1763 * 1 -- we can write to this track, 0 -- we can't
1764 */
1765static int pkt_writable_track(struct pktcdvd_device *pd, track_information *ti)
1766{
1767	switch (pd->mmc3_profile) {
1768		case 0x1a: /* DVD+RW */
1769		case 0x12: /* DVD-RAM */
1770			/* The track is always writable on DVD+RW/DVD-RAM */
1771			return 1;
1772		default:
1773			break;
1774	}
1775
1776	if (!ti->packet || !ti->fp)
1777		return 0;
1778
1779	/*
1780	 * "good" settings as per Mt Fuji.
1781	 */
1782	if (ti->rt == 0 && ti->blank == 0)
1783		return 1;
1784
1785	if (ti->rt == 0 && ti->blank == 1)
1786		return 1;
1787
1788	if (ti->rt == 1 && ti->blank == 0)
1789		return 1;
1790
1791	pkt_err(pd, "bad state %d-%d-%d\n", ti->rt, ti->blank, ti->packet);
1792	return 0;
1793}
1794
1795/*
1796 * 1 -- we can write to this disc, 0 -- we can't
1797 */
1798static int pkt_writable_disc(struct pktcdvd_device *pd, disc_information *di)
1799{
1800	switch (pd->mmc3_profile) {
1801		case 0x0a: /* CD-RW */
1802		case 0xffff: /* MMC3 not supported */
1803			break;
1804		case 0x1a: /* DVD+RW */
1805		case 0x13: /* DVD-RW */
1806		case 0x12: /* DVD-RAM */
1807			return 1;
1808		default:
1809			pkt_dbg(2, pd, "Wrong disc profile (%x)\n",
1810				pd->mmc3_profile);
1811			return 0;
1812	}
1813
1814	/*
1815	 * for disc type 0xff we should probably reserve a new track.
1816	 * but i'm not sure, should we leave this to user apps? probably.
1817	 */
1818	if (di->disc_type == 0xff) {
1819		pkt_notice(pd, "unknown disc - no track?\n");
1820		return 0;
1821	}
1822
1823	if (di->disc_type != 0x20 && di->disc_type != 0) {
1824		pkt_err(pd, "wrong disc type (%x)\n", di->disc_type);
1825		return 0;
1826	}
1827
1828	if (di->erasable == 0) {
1829		pkt_notice(pd, "disc not erasable\n");
1830		return 0;
1831	}
1832
1833	if (di->border_status == PACKET_SESSION_RESERVED) {
1834		pkt_err(pd, "can't write to last track (reserved)\n");
1835		return 0;
1836	}
1837
1838	return 1;
1839}
1840
1841static noinline_for_stack int pkt_probe_settings(struct pktcdvd_device *pd)
1842{
1843	struct packet_command cgc;
1844	unsigned char buf[12];
1845	disc_information di;
1846	track_information ti;
1847	int ret, track;
1848
1849	init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
1850	cgc.cmd[0] = GPCMD_GET_CONFIGURATION;
1851	cgc.cmd[8] = 8;
1852	ret = pkt_generic_packet(pd, &cgc);
1853	pd->mmc3_profile = ret ? 0xffff : buf[6] << 8 | buf[7];
1854
1855	memset(&di, 0, sizeof(disc_information));
1856	memset(&ti, 0, sizeof(track_information));
1857
1858	if ((ret = pkt_get_disc_info(pd, &di))) {
1859		pkt_err(pd, "failed get_disc\n");
1860		return ret;
1861	}
1862
1863	if (!pkt_writable_disc(pd, &di))
1864		return -EROFS;
1865
1866	pd->type = di.erasable ? PACKET_CDRW : PACKET_CDR;
1867
1868	track = 1; /* (di.last_track_msb << 8) | di.last_track_lsb; */
1869	if ((ret = pkt_get_track_info(pd, track, 1, &ti))) {
1870		pkt_err(pd, "failed get_track\n");
1871		return ret;
1872	}
1873
1874	if (!pkt_writable_track(pd, &ti)) {
1875		pkt_err(pd, "can't write to this track\n");
1876		return -EROFS;
1877	}
1878
1879	/*
1880	 * we keep packet size in 512 byte units, makes it easier to
1881	 * deal with request calculations.
1882	 */
1883	pd->settings.size = be32_to_cpu(ti.fixed_packet_size) << 2;
1884	if (pd->settings.size == 0) {
1885		pkt_notice(pd, "detected zero packet size!\n");
1886		return -ENXIO;
1887	}
1888	if (pd->settings.size > PACKET_MAX_SECTORS) {
1889		pkt_err(pd, "packet size is too big\n");
1890		return -EROFS;
1891	}
1892	pd->settings.fp = ti.fp;
1893	pd->offset = (be32_to_cpu(ti.track_start) << 2) & (pd->settings.size - 1);
1894
1895	if (ti.nwa_v) {
1896		pd->nwa = be32_to_cpu(ti.next_writable);
1897		set_bit(PACKET_NWA_VALID, &pd->flags);
1898	}
1899
1900	/*
1901	 * in theory we could use lra on -RW media as well and just zero
1902	 * blocks that haven't been written yet, but in practice that
1903	 * is just a no-go. we'll use that for -R, naturally.
1904	 */
1905	if (ti.lra_v) {
1906		pd->lra = be32_to_cpu(ti.last_rec_address);
1907		set_bit(PACKET_LRA_VALID, &pd->flags);
1908	} else {
1909		pd->lra = 0xffffffff;
1910		set_bit(PACKET_LRA_VALID, &pd->flags);
1911	}
1912
1913	/*
1914	 * fine for now
1915	 */
1916	pd->settings.link_loss = 7;
1917	pd->settings.write_type = 0;	/* packet */
1918	pd->settings.track_mode = ti.track_mode;
1919
1920	/*
1921	 * mode1 or mode2 disc
1922	 */
1923	switch (ti.data_mode) {
1924		case PACKET_MODE1:
1925			pd->settings.block_mode = PACKET_BLOCK_MODE1;
1926			break;
1927		case PACKET_MODE2:
1928			pd->settings.block_mode = PACKET_BLOCK_MODE2;
1929			break;
1930		default:
1931			pkt_err(pd, "unknown data mode\n");
1932			return -EROFS;
1933	}
1934	return 0;
1935}
1936
1937/*
1938 * enable/disable write caching on drive
1939 */
1940static noinline_for_stack int pkt_write_caching(struct pktcdvd_device *pd,
1941						int set)
1942{
1943	struct packet_command cgc;
1944	struct request_sense sense;
1945	unsigned char buf[64];
1946	int ret;
1947
1948	init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
1949	cgc.sense = &sense;
1950	cgc.buflen = pd->mode_offset + 12;
1951
1952	/*
1953	 * caching mode page might not be there, so quiet this command
1954	 */
1955	cgc.quiet = 1;
1956
1957	if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WCACHING_PAGE, 0)))
1958		return ret;
1959
1960	buf[pd->mode_offset + 10] |= (!!set << 2);
1961
1962	cgc.buflen = cgc.cmd[8] = 2 + ((buf[0] << 8) | (buf[1] & 0xff));
1963	ret = pkt_mode_select(pd, &cgc);
1964	if (ret) {
1965		pkt_err(pd, "write caching control failed\n");
1966		pkt_dump_sense(pd, &cgc);
1967	} else if (!ret && set)
1968		pkt_notice(pd, "enabled write caching\n");
1969	return ret;
1970}
1971
1972static int pkt_lock_door(struct pktcdvd_device *pd, int lockflag)
1973{
1974	struct packet_command cgc;
1975
1976	init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
1977	cgc.cmd[0] = GPCMD_PREVENT_ALLOW_MEDIUM_REMOVAL;
1978	cgc.cmd[4] = lockflag ? 1 : 0;
1979	return pkt_generic_packet(pd, &cgc);
1980}
1981
1982/*
1983 * Returns drive maximum write speed
1984 */
1985static noinline_for_stack int pkt_get_max_speed(struct pktcdvd_device *pd,
1986						unsigned *write_speed)
1987{
1988	struct packet_command cgc;
1989	struct request_sense sense;
1990	unsigned char buf[256+18];
1991	unsigned char *cap_buf;
1992	int ret, offset;
1993
1994	cap_buf = &buf[sizeof(struct mode_page_header) + pd->mode_offset];
1995	init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_UNKNOWN);
1996	cgc.sense = &sense;
1997
1998	ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
1999	if (ret) {
2000		cgc.buflen = pd->mode_offset + cap_buf[1] + 2 +
2001			     sizeof(struct mode_page_header);
2002		ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2003		if (ret) {
2004			pkt_dump_sense(pd, &cgc);
2005			return ret;
2006		}
2007	}
2008
2009	offset = 20;			    /* Obsoleted field, used by older drives */
2010	if (cap_buf[1] >= 28)
2011		offset = 28;		    /* Current write speed selected */
2012	if (cap_buf[1] >= 30) {
2013		/* If the drive reports at least one "Logical Unit Write
2014		 * Speed Performance Descriptor Block", use the information
2015		 * in the first block. (contains the highest speed)
2016		 */
2017		int num_spdb = (cap_buf[30] << 8) + cap_buf[31];
2018		if (num_spdb > 0)
2019			offset = 34;
2020	}
2021
2022	*write_speed = (cap_buf[offset] << 8) | cap_buf[offset + 1];
2023	return 0;
2024}
2025
2026/* These tables from cdrecord - I don't have orange book */
2027/* standard speed CD-RW (1-4x) */
2028static char clv_to_speed[16] = {
2029	/* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2030	   0, 2, 4, 6, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2031};
2032/* high speed CD-RW (-10x) */
2033static char hs_clv_to_speed[16] = {
2034	/* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2035	   0, 2, 4, 6, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2036};
2037/* ultra high speed CD-RW */
2038static char us_clv_to_speed[16] = {
2039	/* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2040	   0, 2, 4, 8, 0, 0,16, 0,24,32,40,48, 0, 0, 0, 0
2041};
2042
2043/*
2044 * reads the maximum media speed from ATIP
2045 */
2046static noinline_for_stack int pkt_media_speed(struct pktcdvd_device *pd,
2047						unsigned *speed)
2048{
2049	struct packet_command cgc;
2050	struct request_sense sense;
2051	unsigned char buf[64];
2052	unsigned int size, st, sp;
2053	int ret;
2054
2055	init_cdrom_command(&cgc, buf, 2, CGC_DATA_READ);
2056	cgc.sense = &sense;
2057	cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2058	cgc.cmd[1] = 2;
2059	cgc.cmd[2] = 4; /* READ ATIP */
2060	cgc.cmd[8] = 2;
2061	ret = pkt_generic_packet(pd, &cgc);
2062	if (ret) {
2063		pkt_dump_sense(pd, &cgc);
2064		return ret;
2065	}
2066	size = ((unsigned int) buf[0]<<8) + buf[1] + 2;
2067	if (size > sizeof(buf))
2068		size = sizeof(buf);
2069
2070	init_cdrom_command(&cgc, buf, size, CGC_DATA_READ);
2071	cgc.sense = &sense;
2072	cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2073	cgc.cmd[1] = 2;
2074	cgc.cmd[2] = 4;
2075	cgc.cmd[8] = size;
2076	ret = pkt_generic_packet(pd, &cgc);
2077	if (ret) {
2078		pkt_dump_sense(pd, &cgc);
2079		return ret;
2080	}
2081
2082	if (!(buf[6] & 0x40)) {
2083		pkt_notice(pd, "disc type is not CD-RW\n");
2084		return 1;
2085	}
2086	if (!(buf[6] & 0x4)) {
2087		pkt_notice(pd, "A1 values on media are not valid, maybe not CDRW?\n");
2088		return 1;
2089	}
2090
2091	st = (buf[6] >> 3) & 0x7; /* disc sub-type */
2092
2093	sp = buf[16] & 0xf; /* max speed from ATIP A1 field */
2094
2095	/* Info from cdrecord */
2096	switch (st) {
2097		case 0: /* standard speed */
2098			*speed = clv_to_speed[sp];
2099			break;
2100		case 1: /* high speed */
2101			*speed = hs_clv_to_speed[sp];
2102			break;
2103		case 2: /* ultra high speed */
2104			*speed = us_clv_to_speed[sp];
2105			break;
2106		default:
2107			pkt_notice(pd, "unknown disc sub-type %d\n", st);
2108			return 1;
2109	}
2110	if (*speed) {
2111		pkt_info(pd, "maximum media speed: %d\n", *speed);
2112		return 0;
2113	} else {
2114		pkt_notice(pd, "unknown speed %d for sub-type %d\n", sp, st);
2115		return 1;
2116	}
2117}
2118
2119static noinline_for_stack int pkt_perform_opc(struct pktcdvd_device *pd)
2120{
2121	struct packet_command cgc;
2122	struct request_sense sense;
2123	int ret;
2124
2125	pkt_dbg(2, pd, "Performing OPC\n");
2126
2127	init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2128	cgc.sense = &sense;
2129	cgc.timeout = 60*HZ;
2130	cgc.cmd[0] = GPCMD_SEND_OPC;
2131	cgc.cmd[1] = 1;
2132	if ((ret = pkt_generic_packet(pd, &cgc)))
2133		pkt_dump_sense(pd, &cgc);
2134	return ret;
2135}
2136
2137static int pkt_open_write(struct pktcdvd_device *pd)
2138{
2139	int ret;
2140	unsigned int write_speed, media_write_speed, read_speed;
2141
2142	if ((ret = pkt_probe_settings(pd))) {
2143		pkt_dbg(2, pd, "failed probe\n");
2144		return ret;
2145	}
2146
2147	if ((ret = pkt_set_write_settings(pd))) {
2148		pkt_dbg(1, pd, "failed saving write settings\n");
2149		return -EIO;
2150	}
2151
2152	pkt_write_caching(pd, USE_WCACHING);
2153
2154	if ((ret = pkt_get_max_speed(pd, &write_speed)))
2155		write_speed = 16 * 177;
2156	switch (pd->mmc3_profile) {
2157		case 0x13: /* DVD-RW */
2158		case 0x1a: /* DVD+RW */
2159		case 0x12: /* DVD-RAM */
2160			pkt_dbg(1, pd, "write speed %ukB/s\n", write_speed);
2161			break;
2162		default:
2163			if ((ret = pkt_media_speed(pd, &media_write_speed)))
2164				media_write_speed = 16;
2165			write_speed = min(write_speed, media_write_speed * 177);
2166			pkt_dbg(1, pd, "write speed %ux\n", write_speed / 176);
2167			break;
2168	}
2169	read_speed = write_speed;
2170
2171	if ((ret = pkt_set_speed(pd, write_speed, read_speed))) {
2172		pkt_dbg(1, pd, "couldn't set write speed\n");
2173		return -EIO;
2174	}
2175	pd->write_speed = write_speed;
2176	pd->read_speed = read_speed;
2177
2178	if ((ret = pkt_perform_opc(pd))) {
2179		pkt_dbg(1, pd, "Optimum Power Calibration failed\n");
2180	}
2181
2182	return 0;
2183}
2184
2185/*
2186 * called at open time.
2187 */
2188static int pkt_open_dev(struct pktcdvd_device *pd, fmode_t write)
2189{
2190	int ret;
2191	long lba;
2192	struct request_queue *q;
2193
2194	/*
2195	 * We need to re-open the cdrom device without O_NONBLOCK to be able
2196	 * to read/write from/to it. It is already opened in O_NONBLOCK mode
2197	 * so bdget() can't fail.
2198	 */
2199	bdget(pd->bdev->bd_dev);
2200	if ((ret = blkdev_get(pd->bdev, FMODE_READ | FMODE_EXCL, pd)))
2201		goto out;
2202
2203	if ((ret = pkt_get_last_written(pd, &lba))) {
2204		pkt_err(pd, "pkt_get_last_written failed\n");
2205		goto out_putdev;
2206	}
2207
2208	set_capacity(pd->disk, lba << 2);
2209	set_capacity(pd->bdev->bd_disk, lba << 2);
2210	bd_set_size(pd->bdev, (loff_t)lba << 11);
2211
2212	q = bdev_get_queue(pd->bdev);
2213	if (write) {
2214		if ((ret = pkt_open_write(pd)))
2215			goto out_putdev;
2216		/*
2217		 * Some CDRW drives can not handle writes larger than one packet,
2218		 * even if the size is a multiple of the packet size.
2219		 */
2220		spin_lock_irq(q->queue_lock);
2221		blk_queue_max_hw_sectors(q, pd->settings.size);
2222		spin_unlock_irq(q->queue_lock);
2223		set_bit(PACKET_WRITABLE, &pd->flags);
2224	} else {
2225		pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2226		clear_bit(PACKET_WRITABLE, &pd->flags);
2227	}
2228
2229	if ((ret = pkt_set_segment_merging(pd, q)))
2230		goto out_putdev;
2231
2232	if (write) {
2233		if (!pkt_grow_pktlist(pd, CONFIG_CDROM_PKTCDVD_BUFFERS)) {
2234			pkt_err(pd, "not enough memory for buffers\n");
2235			ret = -ENOMEM;
2236			goto out_putdev;
2237		}
2238		pkt_info(pd, "%lukB available on disc\n", lba << 1);
2239	}
2240
2241	return 0;
2242
2243out_putdev:
2244	blkdev_put(pd->bdev, FMODE_READ | FMODE_EXCL);
2245out:
2246	return ret;
2247}
2248
2249/*
2250 * called when the device is closed. makes sure that the device flushes
2251 * the internal cache before we close.
2252 */
2253static void pkt_release_dev(struct pktcdvd_device *pd, int flush)
2254{
2255	if (flush && pkt_flush_cache(pd))
2256		pkt_dbg(1, pd, "not flushing cache\n");
2257
2258	pkt_lock_door(pd, 0);
2259
2260	pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2261	blkdev_put(pd->bdev, FMODE_READ | FMODE_EXCL);
2262
2263	pkt_shrink_pktlist(pd);
2264}
2265
2266static struct pktcdvd_device *pkt_find_dev_from_minor(unsigned int dev_minor)
2267{
2268	if (dev_minor >= MAX_WRITERS)
2269		return NULL;
2270	return pkt_devs[dev_minor];
2271}
2272
2273static int pkt_open(struct block_device *bdev, fmode_t mode)
2274{
2275	struct pktcdvd_device *pd = NULL;
2276	int ret;
2277
2278	mutex_lock(&pktcdvd_mutex);
2279	mutex_lock(&ctl_mutex);
2280	pd = pkt_find_dev_from_minor(MINOR(bdev->bd_dev));
2281	if (!pd) {
2282		ret = -ENODEV;
2283		goto out;
2284	}
2285	BUG_ON(pd->refcnt < 0);
2286
2287	pd->refcnt++;
2288	if (pd->refcnt > 1) {
2289		if ((mode & FMODE_WRITE) &&
2290		    !test_bit(PACKET_WRITABLE, &pd->flags)) {
2291			ret = -EBUSY;
2292			goto out_dec;
2293		}
2294	} else {
2295		ret = pkt_open_dev(pd, mode & FMODE_WRITE);
2296		if (ret)
2297			goto out_dec;
2298		/*
2299		 * needed here as well, since ext2 (among others) may change
2300		 * the blocksize at mount time
2301		 */
2302		set_blocksize(bdev, CD_FRAMESIZE);
2303	}
2304
2305	mutex_unlock(&ctl_mutex);
2306	mutex_unlock(&pktcdvd_mutex);
2307	return 0;
2308
2309out_dec:
2310	pd->refcnt--;
2311out:
2312	mutex_unlock(&ctl_mutex);
2313	mutex_unlock(&pktcdvd_mutex);
2314	return ret;
2315}
2316
2317static void pkt_close(struct gendisk *disk, fmode_t mode)
2318{
2319	struct pktcdvd_device *pd = disk->private_data;
2320
2321	mutex_lock(&pktcdvd_mutex);
2322	mutex_lock(&ctl_mutex);
2323	pd->refcnt--;
2324	BUG_ON(pd->refcnt < 0);
2325	if (pd->refcnt == 0) {
2326		int flush = test_bit(PACKET_WRITABLE, &pd->flags);
2327		pkt_release_dev(pd, flush);
2328	}
2329	mutex_unlock(&ctl_mutex);
2330	mutex_unlock(&pktcdvd_mutex);
2331}
2332
2333
2334static void pkt_end_io_read_cloned(struct bio *bio, int err)
2335{
2336	struct packet_stacked_data *psd = bio->bi_private;
2337	struct pktcdvd_device *pd = psd->pd;
2338
2339	bio_put(bio);
2340	bio_endio(psd->bio, err);
2341	mempool_free(psd, psd_pool);
2342	pkt_bio_finished(pd);
2343}
2344
2345static void pkt_make_request_read(struct pktcdvd_device *pd, struct bio *bio)
2346{
2347	struct bio *cloned_bio = bio_clone(bio, GFP_NOIO);
2348	struct packet_stacked_data *psd = mempool_alloc(psd_pool, GFP_NOIO);
2349
2350	psd->pd = pd;
2351	psd->bio = bio;
2352	cloned_bio->bi_bdev = pd->bdev;
2353	cloned_bio->bi_private = psd;
2354	cloned_bio->bi_end_io = pkt_end_io_read_cloned;
2355	pd->stats.secs_r += bio_sectors(bio);
2356	pkt_queue_bio(pd, cloned_bio);
2357}
2358
2359static void pkt_make_request_write(struct request_queue *q, struct bio *bio)
2360{
2361	struct pktcdvd_device *pd = q->queuedata;
2362	sector_t zone;
2363	struct packet_data *pkt;
2364	int was_empty, blocked_bio;
2365	struct pkt_rb_node *node;
2366
2367	zone = get_zone(bio->bi_iter.bi_sector, pd);
2368
2369	/*
2370	 * If we find a matching packet in state WAITING or READ_WAIT, we can
2371	 * just append this bio to that packet.
2372	 */
2373	spin_lock(&pd->cdrw.active_list_lock);
2374	blocked_bio = 0;
2375	list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
2376		if (pkt->sector == zone) {
2377			spin_lock(&pkt->lock);
2378			if ((pkt->state == PACKET_WAITING_STATE) ||
2379			    (pkt->state == PACKET_READ_WAIT_STATE)) {
2380				bio_list_add(&pkt->orig_bios, bio);
2381				pkt->write_size +=
2382					bio->bi_iter.bi_size / CD_FRAMESIZE;
2383				if ((pkt->write_size >= pkt->frames) &&
2384				    (pkt->state == PACKET_WAITING_STATE)) {
2385					atomic_inc(&pkt->run_sm);
2386					wake_up(&pd->wqueue);
2387				}
2388				spin_unlock(&pkt->lock);
2389				spin_unlock(&pd->cdrw.active_list_lock);
2390				return;
2391			} else {
2392				blocked_bio = 1;
2393			}
2394			spin_unlock(&pkt->lock);
2395		}
2396	}
2397	spin_unlock(&pd->cdrw.active_list_lock);
2398
2399 	/*
2400	 * Test if there is enough room left in the bio work queue
2401	 * (queue size >= congestion on mark).
2402	 * If not, wait till the work queue size is below the congestion off mark.
2403	 */
2404	spin_lock(&pd->lock);
2405	if (pd->write_congestion_on > 0
2406	    && pd->bio_queue_size >= pd->write_congestion_on) {
2407		set_bdi_congested(&q->backing_dev_info, BLK_RW_ASYNC);
2408		do {
2409			spin_unlock(&pd->lock);
2410			congestion_wait(BLK_RW_ASYNC, HZ);
2411			spin_lock(&pd->lock);
2412		} while(pd->bio_queue_size > pd->write_congestion_off);
2413	}
2414	spin_unlock(&pd->lock);
2415
2416	/*
2417	 * No matching packet found. Store the bio in the work queue.
2418	 */
2419	node = mempool_alloc(pd->rb_pool, GFP_NOIO);
2420	node->bio = bio;
2421	spin_lock(&pd->lock);
2422	BUG_ON(pd->bio_queue_size < 0);
2423	was_empty = (pd->bio_queue_size == 0);
2424	pkt_rbtree_insert(pd, node);
2425	spin_unlock(&pd->lock);
2426
2427	/*
2428	 * Wake up the worker thread.
2429	 */
2430	atomic_set(&pd->scan_queue, 1);
2431	if (was_empty) {
2432		/* This wake_up is required for correct operation */
2433		wake_up(&pd->wqueue);
2434	} else if (!list_empty(&pd->cdrw.pkt_free_list) && !blocked_bio) {
2435		/*
2436		 * This wake up is not required for correct operation,
2437		 * but improves performance in some cases.
2438		 */
2439		wake_up(&pd->wqueue);
2440	}
2441}
2442
2443static void pkt_make_request(struct request_queue *q, struct bio *bio)
2444{
2445	struct pktcdvd_device *pd;
2446	char b[BDEVNAME_SIZE];
2447	struct bio *split;
2448
2449	pd = q->queuedata;
2450	if (!pd) {
2451		pr_err("%s incorrect request queue\n",
2452		       bdevname(bio->bi_bdev, b));
2453		goto end_io;
2454	}
2455
2456	pkt_dbg(2, pd, "start = %6llx stop = %6llx\n",
2457		(unsigned long long)bio->bi_iter.bi_sector,
2458		(unsigned long long)bio_end_sector(bio));
2459
2460	/*
2461	 * Clone READ bios so we can have our own bi_end_io callback.
2462	 */
2463	if (bio_data_dir(bio) == READ) {
2464		pkt_make_request_read(pd, bio);
2465		return;
2466	}
2467
2468	if (!test_bit(PACKET_WRITABLE, &pd->flags)) {
2469		pkt_notice(pd, "WRITE for ro device (%llu)\n",
2470			   (unsigned long long)bio->bi_iter.bi_sector);
2471		goto end_io;
2472	}
2473
2474	if (!bio->bi_iter.bi_size || (bio->bi_iter.bi_size % CD_FRAMESIZE)) {
2475		pkt_err(pd, "wrong bio size\n");
2476		goto end_io;
2477	}
2478
2479	blk_queue_bounce(q, &bio);
2480
2481	do {
2482		sector_t zone = get_zone(bio->bi_iter.bi_sector, pd);
2483		sector_t last_zone = get_zone(bio_end_sector(bio) - 1, pd);
2484
2485		if (last_zone != zone) {
2486			BUG_ON(last_zone != zone + pd->settings.size);
2487
2488			split = bio_split(bio, last_zone -
2489					  bio->bi_iter.bi_sector,
2490					  GFP_NOIO, fs_bio_set);
2491			bio_chain(split, bio);
2492		} else {
2493			split = bio;
2494		}
2495
2496		pkt_make_request_write(q, split);
2497	} while (split != bio);
2498
2499	return;
2500end_io:
2501	bio_io_error(bio);
2502}
2503
2504
2505
2506static int pkt_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2507			  struct bio_vec *bvec)
2508{
2509	struct pktcdvd_device *pd = q->queuedata;
2510	sector_t zone = get_zone(bmd->bi_sector, pd);
2511	int used = ((bmd->bi_sector - zone) << 9) + bmd->bi_size;
2512	int remaining = (pd->settings.size << 9) - used;
2513	int remaining2;
2514
2515	/*
2516	 * A bio <= PAGE_SIZE must be allowed. If it crosses a packet
2517	 * boundary, pkt_make_request() will split the bio.
2518	 */
2519	remaining2 = PAGE_SIZE - bmd->bi_size;
2520	remaining = max(remaining, remaining2);
2521
2522	BUG_ON(remaining < 0);
2523	return remaining;
2524}
2525
2526static void pkt_init_queue(struct pktcdvd_device *pd)
2527{
2528	struct request_queue *q = pd->disk->queue;
2529
2530	blk_queue_make_request(q, pkt_make_request);
2531	blk_queue_logical_block_size(q, CD_FRAMESIZE);
2532	blk_queue_max_hw_sectors(q, PACKET_MAX_SECTORS);
2533	blk_queue_merge_bvec(q, pkt_merge_bvec);
2534	q->queuedata = pd;
2535}
2536
2537static int pkt_seq_show(struct seq_file *m, void *p)
2538{
2539	struct pktcdvd_device *pd = m->private;
2540	char *msg;
2541	char bdev_buf[BDEVNAME_SIZE];
2542	int states[PACKET_NUM_STATES];
2543
2544	seq_printf(m, "Writer %s mapped to %s:\n", pd->name,
2545		   bdevname(pd->bdev, bdev_buf));
2546
2547	seq_printf(m, "\nSettings:\n");
2548	seq_printf(m, "\tpacket size:\t\t%dkB\n", pd->settings.size / 2);
2549
2550	if (pd->settings.write_type == 0)
2551		msg = "Packet";
2552	else
2553		msg = "Unknown";
2554	seq_printf(m, "\twrite type:\t\t%s\n", msg);
2555
2556	seq_printf(m, "\tpacket type:\t\t%s\n", pd->settings.fp ? "Fixed" : "Variable");
2557	seq_printf(m, "\tlink loss:\t\t%d\n", pd->settings.link_loss);
2558
2559	seq_printf(m, "\ttrack mode:\t\t%d\n", pd->settings.track_mode);
2560
2561	if (pd->settings.block_mode == PACKET_BLOCK_MODE1)
2562		msg = "Mode 1";
2563	else if (pd->settings.block_mode == PACKET_BLOCK_MODE2)
2564		msg = "Mode 2";
2565	else
2566		msg = "Unknown";
2567	seq_printf(m, "\tblock mode:\t\t%s\n", msg);
2568
2569	seq_printf(m, "\nStatistics:\n");
2570	seq_printf(m, "\tpackets started:\t%lu\n", pd->stats.pkt_started);
2571	seq_printf(m, "\tpackets ended:\t\t%lu\n", pd->stats.pkt_ended);
2572	seq_printf(m, "\twritten:\t\t%lukB\n", pd->stats.secs_w >> 1);
2573	seq_printf(m, "\tread gather:\t\t%lukB\n", pd->stats.secs_rg >> 1);
2574	seq_printf(m, "\tread:\t\t\t%lukB\n", pd->stats.secs_r >> 1);
2575
2576	seq_printf(m, "\nMisc:\n");
2577	seq_printf(m, "\treference count:\t%d\n", pd->refcnt);
2578	seq_printf(m, "\tflags:\t\t\t0x%lx\n", pd->flags);
2579	seq_printf(m, "\tread speed:\t\t%ukB/s\n", pd->read_speed);
2580	seq_printf(m, "\twrite speed:\t\t%ukB/s\n", pd->write_speed);
2581	seq_printf(m, "\tstart offset:\t\t%lu\n", pd->offset);
2582	seq_printf(m, "\tmode page offset:\t%u\n", pd->mode_offset);
2583
2584	seq_printf(m, "\nQueue state:\n");
2585	seq_printf(m, "\tbios queued:\t\t%d\n", pd->bio_queue_size);
2586	seq_printf(m, "\tbios pending:\t\t%d\n", atomic_read(&pd->cdrw.pending_bios));
2587	seq_printf(m, "\tcurrent sector:\t\t0x%llx\n", (unsigned long long)pd->current_sector);
2588
2589	pkt_count_states(pd, states);
2590	seq_printf(m, "\tstate:\t\t\ti:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
2591		   states[0], states[1], states[2], states[3], states[4], states[5]);
2592
2593	seq_printf(m, "\twrite congestion marks:\toff=%d on=%d\n",
2594			pd->write_congestion_off,
2595			pd->write_congestion_on);
2596	return 0;
2597}
2598
2599static int pkt_seq_open(struct inode *inode, struct file *file)
2600{
2601	return single_open(file, pkt_seq_show, PDE_DATA(inode));
2602}
2603
2604static const struct file_operations pkt_proc_fops = {
2605	.open	= pkt_seq_open,
2606	.read	= seq_read,
2607	.llseek	= seq_lseek,
2608	.release = single_release
2609};
2610
2611static int pkt_new_dev(struct pktcdvd_device *pd, dev_t dev)
2612{
2613	int i;
2614	int ret = 0;
2615	char b[BDEVNAME_SIZE];
2616	struct block_device *bdev;
2617
2618	if (pd->pkt_dev == dev) {
2619		pkt_err(pd, "recursive setup not allowed\n");
2620		return -EBUSY;
2621	}
2622	for (i = 0; i < MAX_WRITERS; i++) {
2623		struct pktcdvd_device *pd2 = pkt_devs[i];
2624		if (!pd2)
2625			continue;
2626		if (pd2->bdev->bd_dev == dev) {
2627			pkt_err(pd, "%s already setup\n",
2628				bdevname(pd2->bdev, b));
2629			return -EBUSY;
2630		}
2631		if (pd2->pkt_dev == dev) {
2632			pkt_err(pd, "can't chain pktcdvd devices\n");
2633			return -EBUSY;
2634		}
2635	}
2636
2637	bdev = bdget(dev);
2638	if (!bdev)
2639		return -ENOMEM;
2640	ret = blkdev_get(bdev, FMODE_READ | FMODE_NDELAY, NULL);
2641	if (ret)
2642		return ret;
2643
2644	/* This is safe, since we have a reference from open(). */
2645	__module_get(THIS_MODULE);
2646
2647	pd->bdev = bdev;
2648	set_blocksize(bdev, CD_FRAMESIZE);
2649
2650	pkt_init_queue(pd);
2651
2652	atomic_set(&pd->cdrw.pending_bios, 0);
2653	pd->cdrw.thread = kthread_run(kcdrwd, pd, "%s", pd->name);
2654	if (IS_ERR(pd->cdrw.thread)) {
2655		pkt_err(pd, "can't start kernel thread\n");
2656		ret = -ENOMEM;
2657		goto out_mem;
2658	}
2659
2660	proc_create_data(pd->name, 0, pkt_proc, &pkt_proc_fops, pd);
2661	pkt_dbg(1, pd, "writer mapped to %s\n", bdevname(bdev, b));
2662	return 0;
2663
2664out_mem:
2665	blkdev_put(bdev, FMODE_READ | FMODE_NDELAY);
2666	/* This is safe: open() is still holding a reference. */
2667	module_put(THIS_MODULE);
2668	return ret;
2669}
2670
2671static int pkt_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, unsigned long arg)
2672{
2673	struct pktcdvd_device *pd = bdev->bd_disk->private_data;
2674	int ret;
2675
2676	pkt_dbg(2, pd, "cmd %x, dev %d:%d\n",
2677		cmd, MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev));
2678
2679	mutex_lock(&pktcdvd_mutex);
2680	switch (cmd) {
2681	case CDROMEJECT:
2682		/*
2683		 * The door gets locked when the device is opened, so we
2684		 * have to unlock it or else the eject command fails.
2685		 */
2686		if (pd->refcnt == 1)
2687			pkt_lock_door(pd, 0);
2688		/* fallthru */
2689	/*
2690	 * forward selected CDROM ioctls to CD-ROM, for UDF
2691	 */
2692	case CDROMMULTISESSION:
2693	case CDROMREADTOCENTRY:
2694	case CDROM_LAST_WRITTEN:
2695	case CDROM_SEND_PACKET:
2696	case SCSI_IOCTL_SEND_COMMAND:
2697		ret = __blkdev_driver_ioctl(pd->bdev, mode, cmd, arg);
2698		break;
2699
2700	default:
2701		pkt_dbg(2, pd, "Unknown ioctl (%x)\n", cmd);
2702		ret = -ENOTTY;
2703	}
2704	mutex_unlock(&pktcdvd_mutex);
2705
2706	return ret;
2707}
2708
2709static unsigned int pkt_check_events(struct gendisk *disk,
2710				     unsigned int clearing)
2711{
2712	struct pktcdvd_device *pd = disk->private_data;
2713	struct gendisk *attached_disk;
2714
2715	if (!pd)
2716		return 0;
2717	if (!pd->bdev)
2718		return 0;
2719	attached_disk = pd->bdev->bd_disk;
2720	if (!attached_disk || !attached_disk->fops->check_events)
2721		return 0;
2722	return attached_disk->fops->check_events(attached_disk, clearing);
2723}
2724
2725static const struct block_device_operations pktcdvd_ops = {
2726	.owner =		THIS_MODULE,
2727	.open =			pkt_open,
2728	.release =		pkt_close,
2729	.ioctl =		pkt_ioctl,
2730	.check_events =		pkt_check_events,
2731};
2732
2733static char *pktcdvd_devnode(struct gendisk *gd, umode_t *mode)
2734{
2735	return kasprintf(GFP_KERNEL, "pktcdvd/%s", gd->disk_name);
2736}
2737
2738/*
2739 * Set up mapping from pktcdvd device to CD-ROM device.
2740 */
2741static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev)
2742{
2743	int idx;
2744	int ret = -ENOMEM;
2745	struct pktcdvd_device *pd;
2746	struct gendisk *disk;
2747
2748	mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2749
2750	for (idx = 0; idx < MAX_WRITERS; idx++)
2751		if (!pkt_devs[idx])
2752			break;
2753	if (idx == MAX_WRITERS) {
2754		pr_err("max %d writers supported\n", MAX_WRITERS);
2755		ret = -EBUSY;
2756		goto out_mutex;
2757	}
2758
2759	pd = kzalloc(sizeof(struct pktcdvd_device), GFP_KERNEL);
2760	if (!pd)
2761		goto out_mutex;
2762
2763	pd->rb_pool = mempool_create_kmalloc_pool(PKT_RB_POOL_SIZE,
2764						  sizeof(struct pkt_rb_node));
2765	if (!pd->rb_pool)
2766		goto out_mem;
2767
2768	INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
2769	INIT_LIST_HEAD(&pd->cdrw.pkt_active_list);
2770	spin_lock_init(&pd->cdrw.active_list_lock);
2771
2772	spin_lock_init(&pd->lock);
2773	spin_lock_init(&pd->iosched.lock);
2774	bio_list_init(&pd->iosched.read_queue);
2775	bio_list_init(&pd->iosched.write_queue);
2776	sprintf(pd->name, DRIVER_NAME"%d", idx);
2777	init_waitqueue_head(&pd->wqueue);
2778	pd->bio_queue = RB_ROOT;
2779
2780	pd->write_congestion_on  = write_congestion_on;
2781	pd->write_congestion_off = write_congestion_off;
2782
2783	disk = alloc_disk(1);
2784	if (!disk)
2785		goto out_mem;
2786	pd->disk = disk;
2787	disk->major = pktdev_major;
2788	disk->first_minor = idx;
2789	disk->fops = &pktcdvd_ops;
2790	disk->flags = GENHD_FL_REMOVABLE;
2791	strcpy(disk->disk_name, pd->name);
2792	disk->devnode = pktcdvd_devnode;
2793	disk->private_data = pd;
2794	disk->queue = blk_alloc_queue(GFP_KERNEL);
2795	if (!disk->queue)
2796		goto out_mem2;
2797
2798	pd->pkt_dev = MKDEV(pktdev_major, idx);
2799	ret = pkt_new_dev(pd, dev);
2800	if (ret)
2801		goto out_new_dev;
2802
2803	/* inherit events of the host device */
2804	disk->events = pd->bdev->bd_disk->events;
2805	disk->async_events = pd->bdev->bd_disk->async_events;
2806
2807	add_disk(disk);
2808
2809	pkt_sysfs_dev_new(pd);
2810	pkt_debugfs_dev_new(pd);
2811
2812	pkt_devs[idx] = pd;
2813	if (pkt_dev)
2814		*pkt_dev = pd->pkt_dev;
2815
2816	mutex_unlock(&ctl_mutex);
2817	return 0;
2818
2819out_new_dev:
2820	blk_cleanup_queue(disk->queue);
2821out_mem2:
2822	put_disk(disk);
2823out_mem:
2824	if (pd->rb_pool)
2825		mempool_destroy(pd->rb_pool);
2826	kfree(pd);
2827out_mutex:
2828	mutex_unlock(&ctl_mutex);
2829	pr_err("setup of pktcdvd device failed\n");
2830	return ret;
2831}
2832
2833/*
2834 * Tear down mapping from pktcdvd device to CD-ROM device.
2835 */
2836static int pkt_remove_dev(dev_t pkt_dev)
2837{
2838	struct pktcdvd_device *pd;
2839	int idx;
2840	int ret = 0;
2841
2842	mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2843
2844	for (idx = 0; idx < MAX_WRITERS; idx++) {
2845		pd = pkt_devs[idx];
2846		if (pd && (pd->pkt_dev == pkt_dev))
2847			break;
2848	}
2849	if (idx == MAX_WRITERS) {
2850		pr_debug("dev not setup\n");
2851		ret = -ENXIO;
2852		goto out;
2853	}
2854
2855	if (pd->refcnt > 0) {
2856		ret = -EBUSY;
2857		goto out;
2858	}
2859	if (!IS_ERR(pd->cdrw.thread))
2860		kthread_stop(pd->cdrw.thread);
2861
2862	pkt_devs[idx] = NULL;
2863
2864	pkt_debugfs_dev_remove(pd);
2865	pkt_sysfs_dev_remove(pd);
2866
2867	blkdev_put(pd->bdev, FMODE_READ | FMODE_NDELAY);
2868
2869	remove_proc_entry(pd->name, pkt_proc);
2870	pkt_dbg(1, pd, "writer unmapped\n");
2871
2872	del_gendisk(pd->disk);
2873	blk_cleanup_queue(pd->disk->queue);
2874	put_disk(pd->disk);
2875
2876	mempool_destroy(pd->rb_pool);
2877	kfree(pd);
2878
2879	/* This is safe: open() is still holding a reference. */
2880	module_put(THIS_MODULE);
2881
2882out:
2883	mutex_unlock(&ctl_mutex);
2884	return ret;
2885}
2886
2887static void pkt_get_status(struct pkt_ctrl_command *ctrl_cmd)
2888{
2889	struct pktcdvd_device *pd;
2890
2891	mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2892
2893	pd = pkt_find_dev_from_minor(ctrl_cmd->dev_index);
2894	if (pd) {
2895		ctrl_cmd->dev = new_encode_dev(pd->bdev->bd_dev);
2896		ctrl_cmd->pkt_dev = new_encode_dev(pd->pkt_dev);
2897	} else {
2898		ctrl_cmd->dev = 0;
2899		ctrl_cmd->pkt_dev = 0;
2900	}
2901	ctrl_cmd->num_devices = MAX_WRITERS;
2902
2903	mutex_unlock(&ctl_mutex);
2904}
2905
2906static long pkt_ctl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2907{
2908	void __user *argp = (void __user *)arg;
2909	struct pkt_ctrl_command ctrl_cmd;
2910	int ret = 0;
2911	dev_t pkt_dev = 0;
2912
2913	if (cmd != PACKET_CTRL_CMD)
2914		return -ENOTTY;
2915
2916	if (copy_from_user(&ctrl_cmd, argp, sizeof(struct pkt_ctrl_command)))
2917		return -EFAULT;
2918
2919	switch (ctrl_cmd.command) {
2920	case PKT_CTRL_CMD_SETUP:
2921		if (!capable(CAP_SYS_ADMIN))
2922			return -EPERM;
2923		ret = pkt_setup_dev(new_decode_dev(ctrl_cmd.dev), &pkt_dev);
2924		ctrl_cmd.pkt_dev = new_encode_dev(pkt_dev);
2925		break;
2926	case PKT_CTRL_CMD_TEARDOWN:
2927		if (!capable(CAP_SYS_ADMIN))
2928			return -EPERM;
2929		ret = pkt_remove_dev(new_decode_dev(ctrl_cmd.pkt_dev));
2930		break;
2931	case PKT_CTRL_CMD_STATUS:
2932		pkt_get_status(&ctrl_cmd);
2933		break;
2934	default:
2935		return -ENOTTY;
2936	}
2937
2938	if (copy_to_user(argp, &ctrl_cmd, sizeof(struct pkt_ctrl_command)))
2939		return -EFAULT;
2940	return ret;
2941}
2942
2943#ifdef CONFIG_COMPAT
2944static long pkt_ctl_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2945{
2946	return pkt_ctl_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
2947}
2948#endif
2949
2950static const struct file_operations pkt_ctl_fops = {
2951	.open		= nonseekable_open,
2952	.unlocked_ioctl	= pkt_ctl_ioctl,
2953#ifdef CONFIG_COMPAT
2954	.compat_ioctl	= pkt_ctl_compat_ioctl,
2955#endif
2956	.owner		= THIS_MODULE,
2957	.llseek		= no_llseek,
2958};
2959
2960static struct miscdevice pkt_misc = {
2961	.minor 		= MISC_DYNAMIC_MINOR,
2962	.name  		= DRIVER_NAME,
2963	.nodename	= "pktcdvd/control",
2964	.fops  		= &pkt_ctl_fops
2965};
2966
2967static int __init pkt_init(void)
2968{
2969	int ret;
2970
2971	mutex_init(&ctl_mutex);
2972
2973	psd_pool = mempool_create_kmalloc_pool(PSD_POOL_SIZE,
2974					sizeof(struct packet_stacked_data));
2975	if (!psd_pool)
2976		return -ENOMEM;
2977
2978	ret = register_blkdev(pktdev_major, DRIVER_NAME);
2979	if (ret < 0) {
2980		pr_err("unable to register block device\n");
2981		goto out2;
2982	}
2983	if (!pktdev_major)
2984		pktdev_major = ret;
2985
2986	ret = pkt_sysfs_init();
2987	if (ret)
2988		goto out;
2989
2990	pkt_debugfs_init();
2991
2992	ret = misc_register(&pkt_misc);
2993	if (ret) {
2994		pr_err("unable to register misc device\n");
2995		goto out_misc;
2996	}
2997
2998	pkt_proc = proc_mkdir("driver/"DRIVER_NAME, NULL);
2999
3000	return 0;
3001
3002out_misc:
3003	pkt_debugfs_cleanup();
3004	pkt_sysfs_cleanup();
3005out:
3006	unregister_blkdev(pktdev_major, DRIVER_NAME);
3007out2:
3008	mempool_destroy(psd_pool);
3009	return ret;
3010}
3011
3012static void __exit pkt_exit(void)
3013{
3014	remove_proc_entry("driver/"DRIVER_NAME, NULL);
3015	misc_deregister(&pkt_misc);
3016
3017	pkt_debugfs_cleanup();
3018	pkt_sysfs_cleanup();
3019
3020	unregister_blkdev(pktdev_major, DRIVER_NAME);
3021	mempool_destroy(psd_pool);
3022}
3023
3024MODULE_DESCRIPTION("Packet writing layer for CD/DVD drives");
3025MODULE_AUTHOR("Jens Axboe <axboe@suse.de>");
3026MODULE_LICENSE("GPL");
3027
3028module_init(pkt_init);
3029module_exit(pkt_exit);
3030