xref: /linux/drivers/block/pktcdvd.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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 #include <linux/pktcdvd.h>
48 #include <linux/module.h>
49 #include <linux/types.h>
50 #include <linux/kernel.h>
51 #include <linux/compat.h>
52 #include <linux/kthread.h>
53 #include <linux/errno.h>
54 #include <linux/spinlock.h>
55 #include <linux/file.h>
56 #include <linux/proc_fs.h>
57 #include <linux/seq_file.h>
58 #include <linux/miscdevice.h>
59 #include <linux/freezer.h>
60 #include <linux/mutex.h>
61 #include <linux/slab.h>
62 #include <scsi/scsi_cmnd.h>
63 #include <scsi/scsi_ioctl.h>
64 #include <scsi/scsi.h>
65 #include <linux/debugfs.h>
66 #include <linux/device.h>
67 
68 #include <asm/uaccess.h>
69 
70 #define DRIVER_NAME	"pktcdvd"
71 
72 #if PACKET_DEBUG
73 #define DPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
74 #else
75 #define DPRINTK(fmt, args...)
76 #endif
77 
78 #if PACKET_DEBUG > 1
79 #define VPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
80 #else
81 #define VPRINTK(fmt, args...)
82 #endif
83 
84 #define MAX_SPEED 0xffff
85 
86 #define ZONE(sector, pd) (((sector) + (pd)->offset) & ~((pd)->settings.size - 1))
87 
88 static DEFINE_MUTEX(pktcdvd_mutex);
89 static struct pktcdvd_device *pkt_devs[MAX_WRITERS];
90 static struct proc_dir_entry *pkt_proc;
91 static int pktdev_major;
92 static int write_congestion_on  = PKT_WRITE_CONGESTION_ON;
93 static int write_congestion_off = PKT_WRITE_CONGESTION_OFF;
94 static struct mutex ctl_mutex;	/* Serialize open/close/setup/teardown */
95 static mempool_t *psd_pool;
96 
97 static struct class	*class_pktcdvd = NULL;    /* /sys/class/pktcdvd */
98 static struct dentry	*pkt_debugfs_root = NULL; /* /sys/kernel/debug/pktcdvd */
99 
100 /* forward declaration */
101 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev);
102 static int pkt_remove_dev(dev_t pkt_dev);
103 static int pkt_seq_show(struct seq_file *m, void *p);
104 
105 
106 
107 /*
108  * create and register a pktcdvd kernel object.
109  */
110 static struct pktcdvd_kobj* pkt_kobj_create(struct pktcdvd_device *pd,
111 					const char* name,
112 					struct kobject* parent,
113 					struct kobj_type* ktype)
114 {
115 	struct pktcdvd_kobj *p;
116 	int error;
117 
118 	p = kzalloc(sizeof(*p), GFP_KERNEL);
119 	if (!p)
120 		return NULL;
121 	p->pd = pd;
122 	error = kobject_init_and_add(&p->kobj, ktype, parent, "%s", name);
123 	if (error) {
124 		kobject_put(&p->kobj);
125 		return NULL;
126 	}
127 	kobject_uevent(&p->kobj, KOBJ_ADD);
128 	return p;
129 }
130 /*
131  * remove a pktcdvd kernel object.
132  */
133 static void pkt_kobj_remove(struct pktcdvd_kobj *p)
134 {
135 	if (p)
136 		kobject_put(&p->kobj);
137 }
138 /*
139  * default release function for pktcdvd kernel objects.
140  */
141 static void pkt_kobj_release(struct kobject *kobj)
142 {
143 	kfree(to_pktcdvdkobj(kobj));
144 }
145 
146 
147 /**********************************************************
148  *
149  * sysfs interface for pktcdvd
150  * by (C) 2006  Thomas Maier <balagi@justmail.de>
151  *
152  **********************************************************/
153 
154 #define DEF_ATTR(_obj,_name,_mode) \
155 	static struct attribute _obj = { .name = _name, .mode = _mode }
156 
157 /**********************************************************
158   /sys/class/pktcdvd/pktcdvd[0-7]/
159                      stat/reset
160                      stat/packets_started
161                      stat/packets_finished
162                      stat/kb_written
163                      stat/kb_read
164                      stat/kb_read_gather
165                      write_queue/size
166                      write_queue/congestion_off
167                      write_queue/congestion_on
168  **********************************************************/
169 
170 DEF_ATTR(kobj_pkt_attr_st1, "reset", 0200);
171 DEF_ATTR(kobj_pkt_attr_st2, "packets_started", 0444);
172 DEF_ATTR(kobj_pkt_attr_st3, "packets_finished", 0444);
173 DEF_ATTR(kobj_pkt_attr_st4, "kb_written", 0444);
174 DEF_ATTR(kobj_pkt_attr_st5, "kb_read", 0444);
175 DEF_ATTR(kobj_pkt_attr_st6, "kb_read_gather", 0444);
176 
177 static struct attribute *kobj_pkt_attrs_stat[] = {
178 	&kobj_pkt_attr_st1,
179 	&kobj_pkt_attr_st2,
180 	&kobj_pkt_attr_st3,
181 	&kobj_pkt_attr_st4,
182 	&kobj_pkt_attr_st5,
183 	&kobj_pkt_attr_st6,
184 	NULL
185 };
186 
187 DEF_ATTR(kobj_pkt_attr_wq1, "size", 0444);
188 DEF_ATTR(kobj_pkt_attr_wq2, "congestion_off", 0644);
189 DEF_ATTR(kobj_pkt_attr_wq3, "congestion_on",  0644);
190 
191 static struct attribute *kobj_pkt_attrs_wqueue[] = {
192 	&kobj_pkt_attr_wq1,
193 	&kobj_pkt_attr_wq2,
194 	&kobj_pkt_attr_wq3,
195 	NULL
196 };
197 
198 static ssize_t kobj_pkt_show(struct kobject *kobj,
199 			struct attribute *attr, char *data)
200 {
201 	struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
202 	int n = 0;
203 	int v;
204 	if (strcmp(attr->name, "packets_started") == 0) {
205 		n = sprintf(data, "%lu\n", pd->stats.pkt_started);
206 
207 	} else if (strcmp(attr->name, "packets_finished") == 0) {
208 		n = sprintf(data, "%lu\n", pd->stats.pkt_ended);
209 
210 	} else if (strcmp(attr->name, "kb_written") == 0) {
211 		n = sprintf(data, "%lu\n", pd->stats.secs_w >> 1);
212 
213 	} else if (strcmp(attr->name, "kb_read") == 0) {
214 		n = sprintf(data, "%lu\n", pd->stats.secs_r >> 1);
215 
216 	} else if (strcmp(attr->name, "kb_read_gather") == 0) {
217 		n = sprintf(data, "%lu\n", pd->stats.secs_rg >> 1);
218 
219 	} else if (strcmp(attr->name, "size") == 0) {
220 		spin_lock(&pd->lock);
221 		v = pd->bio_queue_size;
222 		spin_unlock(&pd->lock);
223 		n = sprintf(data, "%d\n", v);
224 
225 	} else if (strcmp(attr->name, "congestion_off") == 0) {
226 		spin_lock(&pd->lock);
227 		v = pd->write_congestion_off;
228 		spin_unlock(&pd->lock);
229 		n = sprintf(data, "%d\n", v);
230 
231 	} else if (strcmp(attr->name, "congestion_on") == 0) {
232 		spin_lock(&pd->lock);
233 		v = pd->write_congestion_on;
234 		spin_unlock(&pd->lock);
235 		n = sprintf(data, "%d\n", v);
236 	}
237 	return n;
238 }
239 
240 static void init_write_congestion_marks(int* lo, int* hi)
241 {
242 	if (*hi > 0) {
243 		*hi = max(*hi, 500);
244 		*hi = min(*hi, 1000000);
245 		if (*lo <= 0)
246 			*lo = *hi - 100;
247 		else {
248 			*lo = min(*lo, *hi - 100);
249 			*lo = max(*lo, 100);
250 		}
251 	} else {
252 		*hi = -1;
253 		*lo = -1;
254 	}
255 }
256 
257 static ssize_t kobj_pkt_store(struct kobject *kobj,
258 			struct attribute *attr,
259 			const char *data, size_t len)
260 {
261 	struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
262 	int val;
263 
264 	if (strcmp(attr->name, "reset") == 0 && len > 0) {
265 		pd->stats.pkt_started = 0;
266 		pd->stats.pkt_ended = 0;
267 		pd->stats.secs_w = 0;
268 		pd->stats.secs_rg = 0;
269 		pd->stats.secs_r = 0;
270 
271 	} else if (strcmp(attr->name, "congestion_off") == 0
272 		   && sscanf(data, "%d", &val) == 1) {
273 		spin_lock(&pd->lock);
274 		pd->write_congestion_off = val;
275 		init_write_congestion_marks(&pd->write_congestion_off,
276 					&pd->write_congestion_on);
277 		spin_unlock(&pd->lock);
278 
279 	} else if (strcmp(attr->name, "congestion_on") == 0
280 		   && sscanf(data, "%d", &val) == 1) {
281 		spin_lock(&pd->lock);
282 		pd->write_congestion_on = val;
283 		init_write_congestion_marks(&pd->write_congestion_off,
284 					&pd->write_congestion_on);
285 		spin_unlock(&pd->lock);
286 	}
287 	return len;
288 }
289 
290 static const struct sysfs_ops kobj_pkt_ops = {
291 	.show = kobj_pkt_show,
292 	.store = kobj_pkt_store
293 };
294 static struct kobj_type kobj_pkt_type_stat = {
295 	.release = pkt_kobj_release,
296 	.sysfs_ops = &kobj_pkt_ops,
297 	.default_attrs = kobj_pkt_attrs_stat
298 };
299 static struct kobj_type kobj_pkt_type_wqueue = {
300 	.release = pkt_kobj_release,
301 	.sysfs_ops = &kobj_pkt_ops,
302 	.default_attrs = kobj_pkt_attrs_wqueue
303 };
304 
305 static void pkt_sysfs_dev_new(struct pktcdvd_device *pd)
306 {
307 	if (class_pktcdvd) {
308 		pd->dev = device_create(class_pktcdvd, NULL, MKDEV(0, 0), NULL,
309 					"%s", pd->name);
310 		if (IS_ERR(pd->dev))
311 			pd->dev = NULL;
312 	}
313 	if (pd->dev) {
314 		pd->kobj_stat = pkt_kobj_create(pd, "stat",
315 					&pd->dev->kobj,
316 					&kobj_pkt_type_stat);
317 		pd->kobj_wqueue = pkt_kobj_create(pd, "write_queue",
318 					&pd->dev->kobj,
319 					&kobj_pkt_type_wqueue);
320 	}
321 }
322 
323 static void pkt_sysfs_dev_remove(struct pktcdvd_device *pd)
324 {
325 	pkt_kobj_remove(pd->kobj_stat);
326 	pkt_kobj_remove(pd->kobj_wqueue);
327 	if (class_pktcdvd)
328 		device_unregister(pd->dev);
329 }
330 
331 
332 /********************************************************************
333   /sys/class/pktcdvd/
334                      add            map block device
335                      remove         unmap packet dev
336                      device_map     show mappings
337  *******************************************************************/
338 
339 static void class_pktcdvd_release(struct class *cls)
340 {
341 	kfree(cls);
342 }
343 static ssize_t class_pktcdvd_show_map(struct class *c,
344 					struct class_attribute *attr,
345 					char *data)
346 {
347 	int n = 0;
348 	int idx;
349 	mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
350 	for (idx = 0; idx < MAX_WRITERS; idx++) {
351 		struct pktcdvd_device *pd = pkt_devs[idx];
352 		if (!pd)
353 			continue;
354 		n += sprintf(data+n, "%s %u:%u %u:%u\n",
355 			pd->name,
356 			MAJOR(pd->pkt_dev), MINOR(pd->pkt_dev),
357 			MAJOR(pd->bdev->bd_dev),
358 			MINOR(pd->bdev->bd_dev));
359 	}
360 	mutex_unlock(&ctl_mutex);
361 	return n;
362 }
363 
364 static ssize_t class_pktcdvd_store_add(struct class *c,
365 					struct class_attribute *attr,
366 					const char *buf,
367 					size_t count)
368 {
369 	unsigned int major, minor;
370 
371 	if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
372 		/* pkt_setup_dev() expects caller to hold reference to self */
373 		if (!try_module_get(THIS_MODULE))
374 			return -ENODEV;
375 
376 		pkt_setup_dev(MKDEV(major, minor), NULL);
377 
378 		module_put(THIS_MODULE);
379 
380 		return count;
381 	}
382 
383 	return -EINVAL;
384 }
385 
386 static ssize_t class_pktcdvd_store_remove(struct class *c,
387 					  struct class_attribute *attr,
388 					  const char *buf,
389 					size_t count)
390 {
391 	unsigned int major, minor;
392 	if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
393 		pkt_remove_dev(MKDEV(major, minor));
394 		return count;
395 	}
396 	return -EINVAL;
397 }
398 
399 static struct class_attribute class_pktcdvd_attrs[] = {
400  __ATTR(add,            0200, NULL, class_pktcdvd_store_add),
401  __ATTR(remove,         0200, NULL, class_pktcdvd_store_remove),
402  __ATTR(device_map,     0444, class_pktcdvd_show_map, NULL),
403  __ATTR_NULL
404 };
405 
406 
407 static int pkt_sysfs_init(void)
408 {
409 	int ret = 0;
410 
411 	/*
412 	 * create control files in sysfs
413 	 * /sys/class/pktcdvd/...
414 	 */
415 	class_pktcdvd = kzalloc(sizeof(*class_pktcdvd), GFP_KERNEL);
416 	if (!class_pktcdvd)
417 		return -ENOMEM;
418 	class_pktcdvd->name = DRIVER_NAME;
419 	class_pktcdvd->owner = THIS_MODULE;
420 	class_pktcdvd->class_release = class_pktcdvd_release;
421 	class_pktcdvd->class_attrs = class_pktcdvd_attrs;
422 	ret = class_register(class_pktcdvd);
423 	if (ret) {
424 		kfree(class_pktcdvd);
425 		class_pktcdvd = NULL;
426 		printk(DRIVER_NAME": failed to create class pktcdvd\n");
427 		return ret;
428 	}
429 	return 0;
430 }
431 
432 static void pkt_sysfs_cleanup(void)
433 {
434 	if (class_pktcdvd)
435 		class_destroy(class_pktcdvd);
436 	class_pktcdvd = NULL;
437 }
438 
439 /********************************************************************
440   entries in debugfs
441 
442   /sys/kernel/debug/pktcdvd[0-7]/
443 			info
444 
445  *******************************************************************/
446 
447 static int pkt_debugfs_seq_show(struct seq_file *m, void *p)
448 {
449 	return pkt_seq_show(m, p);
450 }
451 
452 static int pkt_debugfs_fops_open(struct inode *inode, struct file *file)
453 {
454 	return single_open(file, pkt_debugfs_seq_show, inode->i_private);
455 }
456 
457 static const struct file_operations debug_fops = {
458 	.open		= pkt_debugfs_fops_open,
459 	.read		= seq_read,
460 	.llseek		= seq_lseek,
461 	.release	= single_release,
462 	.owner		= THIS_MODULE,
463 };
464 
465 static void pkt_debugfs_dev_new(struct pktcdvd_device *pd)
466 {
467 	if (!pkt_debugfs_root)
468 		return;
469 	pd->dfs_f_info = NULL;
470 	pd->dfs_d_root = debugfs_create_dir(pd->name, pkt_debugfs_root);
471 	if (IS_ERR(pd->dfs_d_root)) {
472 		pd->dfs_d_root = NULL;
473 		return;
474 	}
475 	pd->dfs_f_info = debugfs_create_file("info", S_IRUGO,
476 				pd->dfs_d_root, pd, &debug_fops);
477 	if (IS_ERR(pd->dfs_f_info)) {
478 		pd->dfs_f_info = NULL;
479 		return;
480 	}
481 }
482 
483 static void pkt_debugfs_dev_remove(struct pktcdvd_device *pd)
484 {
485 	if (!pkt_debugfs_root)
486 		return;
487 	if (pd->dfs_f_info)
488 		debugfs_remove(pd->dfs_f_info);
489 	pd->dfs_f_info = NULL;
490 	if (pd->dfs_d_root)
491 		debugfs_remove(pd->dfs_d_root);
492 	pd->dfs_d_root = NULL;
493 }
494 
495 static void pkt_debugfs_init(void)
496 {
497 	pkt_debugfs_root = debugfs_create_dir(DRIVER_NAME, NULL);
498 	if (IS_ERR(pkt_debugfs_root)) {
499 		pkt_debugfs_root = NULL;
500 		return;
501 	}
502 }
503 
504 static void pkt_debugfs_cleanup(void)
505 {
506 	if (!pkt_debugfs_root)
507 		return;
508 	debugfs_remove(pkt_debugfs_root);
509 	pkt_debugfs_root = NULL;
510 }
511 
512 /* ----------------------------------------------------------*/
513 
514 
515 static void pkt_bio_finished(struct pktcdvd_device *pd)
516 {
517 	BUG_ON(atomic_read(&pd->cdrw.pending_bios) <= 0);
518 	if (atomic_dec_and_test(&pd->cdrw.pending_bios)) {
519 		VPRINTK(DRIVER_NAME": queue empty\n");
520 		atomic_set(&pd->iosched.attention, 1);
521 		wake_up(&pd->wqueue);
522 	}
523 }
524 
525 /*
526  * Allocate a packet_data struct
527  */
528 static struct packet_data *pkt_alloc_packet_data(int frames)
529 {
530 	int i;
531 	struct packet_data *pkt;
532 
533 	pkt = kzalloc(sizeof(struct packet_data), GFP_KERNEL);
534 	if (!pkt)
535 		goto no_pkt;
536 
537 	pkt->frames = frames;
538 	pkt->w_bio = bio_kmalloc(GFP_KERNEL, frames);
539 	if (!pkt->w_bio)
540 		goto no_bio;
541 
542 	for (i = 0; i < frames / FRAMES_PER_PAGE; i++) {
543 		pkt->pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
544 		if (!pkt->pages[i])
545 			goto no_page;
546 	}
547 
548 	spin_lock_init(&pkt->lock);
549 	bio_list_init(&pkt->orig_bios);
550 
551 	for (i = 0; i < frames; i++) {
552 		struct bio *bio = bio_kmalloc(GFP_KERNEL, 1);
553 		if (!bio)
554 			goto no_rd_bio;
555 
556 		pkt->r_bios[i] = bio;
557 	}
558 
559 	return pkt;
560 
561 no_rd_bio:
562 	for (i = 0; i < frames; i++) {
563 		struct bio *bio = pkt->r_bios[i];
564 		if (bio)
565 			bio_put(bio);
566 	}
567 
568 no_page:
569 	for (i = 0; i < frames / FRAMES_PER_PAGE; i++)
570 		if (pkt->pages[i])
571 			__free_page(pkt->pages[i]);
572 	bio_put(pkt->w_bio);
573 no_bio:
574 	kfree(pkt);
575 no_pkt:
576 	return NULL;
577 }
578 
579 /*
580  * Free a packet_data struct
581  */
582 static void pkt_free_packet_data(struct packet_data *pkt)
583 {
584 	int i;
585 
586 	for (i = 0; i < pkt->frames; i++) {
587 		struct bio *bio = pkt->r_bios[i];
588 		if (bio)
589 			bio_put(bio);
590 	}
591 	for (i = 0; i < pkt->frames / FRAMES_PER_PAGE; i++)
592 		__free_page(pkt->pages[i]);
593 	bio_put(pkt->w_bio);
594 	kfree(pkt);
595 }
596 
597 static void pkt_shrink_pktlist(struct pktcdvd_device *pd)
598 {
599 	struct packet_data *pkt, *next;
600 
601 	BUG_ON(!list_empty(&pd->cdrw.pkt_active_list));
602 
603 	list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_free_list, list) {
604 		pkt_free_packet_data(pkt);
605 	}
606 	INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
607 }
608 
609 static int pkt_grow_pktlist(struct pktcdvd_device *pd, int nr_packets)
610 {
611 	struct packet_data *pkt;
612 
613 	BUG_ON(!list_empty(&pd->cdrw.pkt_free_list));
614 
615 	while (nr_packets > 0) {
616 		pkt = pkt_alloc_packet_data(pd->settings.size >> 2);
617 		if (!pkt) {
618 			pkt_shrink_pktlist(pd);
619 			return 0;
620 		}
621 		pkt->id = nr_packets;
622 		pkt->pd = pd;
623 		list_add(&pkt->list, &pd->cdrw.pkt_free_list);
624 		nr_packets--;
625 	}
626 	return 1;
627 }
628 
629 static inline struct pkt_rb_node *pkt_rbtree_next(struct pkt_rb_node *node)
630 {
631 	struct rb_node *n = rb_next(&node->rb_node);
632 	if (!n)
633 		return NULL;
634 	return rb_entry(n, struct pkt_rb_node, rb_node);
635 }
636 
637 static void pkt_rbtree_erase(struct pktcdvd_device *pd, struct pkt_rb_node *node)
638 {
639 	rb_erase(&node->rb_node, &pd->bio_queue);
640 	mempool_free(node, pd->rb_pool);
641 	pd->bio_queue_size--;
642 	BUG_ON(pd->bio_queue_size < 0);
643 }
644 
645 /*
646  * Find the first node in the pd->bio_queue rb tree with a starting sector >= s.
647  */
648 static struct pkt_rb_node *pkt_rbtree_find(struct pktcdvd_device *pd, sector_t s)
649 {
650 	struct rb_node *n = pd->bio_queue.rb_node;
651 	struct rb_node *next;
652 	struct pkt_rb_node *tmp;
653 
654 	if (!n) {
655 		BUG_ON(pd->bio_queue_size > 0);
656 		return NULL;
657 	}
658 
659 	for (;;) {
660 		tmp = rb_entry(n, struct pkt_rb_node, rb_node);
661 		if (s <= tmp->bio->bi_sector)
662 			next = n->rb_left;
663 		else
664 			next = n->rb_right;
665 		if (!next)
666 			break;
667 		n = next;
668 	}
669 
670 	if (s > tmp->bio->bi_sector) {
671 		tmp = pkt_rbtree_next(tmp);
672 		if (!tmp)
673 			return NULL;
674 	}
675 	BUG_ON(s > tmp->bio->bi_sector);
676 	return tmp;
677 }
678 
679 /*
680  * Insert a node into the pd->bio_queue rb tree.
681  */
682 static void pkt_rbtree_insert(struct pktcdvd_device *pd, struct pkt_rb_node *node)
683 {
684 	struct rb_node **p = &pd->bio_queue.rb_node;
685 	struct rb_node *parent = NULL;
686 	sector_t s = node->bio->bi_sector;
687 	struct pkt_rb_node *tmp;
688 
689 	while (*p) {
690 		parent = *p;
691 		tmp = rb_entry(parent, struct pkt_rb_node, rb_node);
692 		if (s < tmp->bio->bi_sector)
693 			p = &(*p)->rb_left;
694 		else
695 			p = &(*p)->rb_right;
696 	}
697 	rb_link_node(&node->rb_node, parent, p);
698 	rb_insert_color(&node->rb_node, &pd->bio_queue);
699 	pd->bio_queue_size++;
700 }
701 
702 /*
703  * Send a packet_command to the underlying block device and
704  * wait for completion.
705  */
706 static int pkt_generic_packet(struct pktcdvd_device *pd, struct packet_command *cgc)
707 {
708 	struct request_queue *q = bdev_get_queue(pd->bdev);
709 	struct request *rq;
710 	int ret = 0;
711 
712 	rq = blk_get_request(q, (cgc->data_direction == CGC_DATA_WRITE) ?
713 			     WRITE : READ, __GFP_WAIT);
714 
715 	if (cgc->buflen) {
716 		if (blk_rq_map_kern(q, rq, cgc->buffer, cgc->buflen, __GFP_WAIT))
717 			goto out;
718 	}
719 
720 	rq->cmd_len = COMMAND_SIZE(cgc->cmd[0]);
721 	memcpy(rq->cmd, cgc->cmd, CDROM_PACKET_SIZE);
722 
723 	rq->timeout = 60*HZ;
724 	rq->cmd_type = REQ_TYPE_BLOCK_PC;
725 	if (cgc->quiet)
726 		rq->cmd_flags |= REQ_QUIET;
727 
728 	blk_execute_rq(rq->q, pd->bdev->bd_disk, rq, 0);
729 	if (rq->errors)
730 		ret = -EIO;
731 out:
732 	blk_put_request(rq);
733 	return ret;
734 }
735 
736 /*
737  * A generic sense dump / resolve mechanism should be implemented across
738  * all ATAPI + SCSI devices.
739  */
740 static void pkt_dump_sense(struct packet_command *cgc)
741 {
742 	static char *info[9] = { "No sense", "Recovered error", "Not ready",
743 				 "Medium error", "Hardware error", "Illegal request",
744 				 "Unit attention", "Data protect", "Blank check" };
745 	int i;
746 	struct request_sense *sense = cgc->sense;
747 
748 	printk(DRIVER_NAME":");
749 	for (i = 0; i < CDROM_PACKET_SIZE; i++)
750 		printk(" %02x", cgc->cmd[i]);
751 	printk(" - ");
752 
753 	if (sense == NULL) {
754 		printk("no sense\n");
755 		return;
756 	}
757 
758 	printk("sense %02x.%02x.%02x", sense->sense_key, sense->asc, sense->ascq);
759 
760 	if (sense->sense_key > 8) {
761 		printk(" (INVALID)\n");
762 		return;
763 	}
764 
765 	printk(" (%s)\n", info[sense->sense_key]);
766 }
767 
768 /*
769  * flush the drive cache to media
770  */
771 static int pkt_flush_cache(struct pktcdvd_device *pd)
772 {
773 	struct packet_command cgc;
774 
775 	init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
776 	cgc.cmd[0] = GPCMD_FLUSH_CACHE;
777 	cgc.quiet = 1;
778 
779 	/*
780 	 * the IMMED bit -- we default to not setting it, although that
781 	 * would allow a much faster close, this is safer
782 	 */
783 #if 0
784 	cgc.cmd[1] = 1 << 1;
785 #endif
786 	return pkt_generic_packet(pd, &cgc);
787 }
788 
789 /*
790  * speed is given as the normal factor, e.g. 4 for 4x
791  */
792 static noinline_for_stack int pkt_set_speed(struct pktcdvd_device *pd,
793 				unsigned write_speed, unsigned read_speed)
794 {
795 	struct packet_command cgc;
796 	struct request_sense sense;
797 	int ret;
798 
799 	init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
800 	cgc.sense = &sense;
801 	cgc.cmd[0] = GPCMD_SET_SPEED;
802 	cgc.cmd[2] = (read_speed >> 8) & 0xff;
803 	cgc.cmd[3] = read_speed & 0xff;
804 	cgc.cmd[4] = (write_speed >> 8) & 0xff;
805 	cgc.cmd[5] = write_speed & 0xff;
806 
807 	if ((ret = pkt_generic_packet(pd, &cgc)))
808 		pkt_dump_sense(&cgc);
809 
810 	return ret;
811 }
812 
813 /*
814  * Queue a bio for processing by the low-level CD device. Must be called
815  * from process context.
816  */
817 static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio)
818 {
819 	spin_lock(&pd->iosched.lock);
820 	if (bio_data_dir(bio) == READ)
821 		bio_list_add(&pd->iosched.read_queue, bio);
822 	else
823 		bio_list_add(&pd->iosched.write_queue, bio);
824 	spin_unlock(&pd->iosched.lock);
825 
826 	atomic_set(&pd->iosched.attention, 1);
827 	wake_up(&pd->wqueue);
828 }
829 
830 /*
831  * Process the queued read/write requests. This function handles special
832  * requirements for CDRW drives:
833  * - A cache flush command must be inserted before a read request if the
834  *   previous request was a write.
835  * - Switching between reading and writing is slow, so don't do it more often
836  *   than necessary.
837  * - Optimize for throughput at the expense of latency. This means that streaming
838  *   writes will never be interrupted by a read, but if the drive has to seek
839  *   before the next write, switch to reading instead if there are any pending
840  *   read requests.
841  * - Set the read speed according to current usage pattern. When only reading
842  *   from the device, it's best to use the highest possible read speed, but
843  *   when switching often between reading and writing, it's better to have the
844  *   same read and write speeds.
845  */
846 static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
847 {
848 
849 	if (atomic_read(&pd->iosched.attention) == 0)
850 		return;
851 	atomic_set(&pd->iosched.attention, 0);
852 
853 	for (;;) {
854 		struct bio *bio;
855 		int reads_queued, writes_queued;
856 
857 		spin_lock(&pd->iosched.lock);
858 		reads_queued = !bio_list_empty(&pd->iosched.read_queue);
859 		writes_queued = !bio_list_empty(&pd->iosched.write_queue);
860 		spin_unlock(&pd->iosched.lock);
861 
862 		if (!reads_queued && !writes_queued)
863 			break;
864 
865 		if (pd->iosched.writing) {
866 			int need_write_seek = 1;
867 			spin_lock(&pd->iosched.lock);
868 			bio = bio_list_peek(&pd->iosched.write_queue);
869 			spin_unlock(&pd->iosched.lock);
870 			if (bio && (bio->bi_sector == pd->iosched.last_write))
871 				need_write_seek = 0;
872 			if (need_write_seek && reads_queued) {
873 				if (atomic_read(&pd->cdrw.pending_bios) > 0) {
874 					VPRINTK(DRIVER_NAME": write, waiting\n");
875 					break;
876 				}
877 				pkt_flush_cache(pd);
878 				pd->iosched.writing = 0;
879 			}
880 		} else {
881 			if (!reads_queued && writes_queued) {
882 				if (atomic_read(&pd->cdrw.pending_bios) > 0) {
883 					VPRINTK(DRIVER_NAME": read, waiting\n");
884 					break;
885 				}
886 				pd->iosched.writing = 1;
887 			}
888 		}
889 
890 		spin_lock(&pd->iosched.lock);
891 		if (pd->iosched.writing)
892 			bio = bio_list_pop(&pd->iosched.write_queue);
893 		else
894 			bio = bio_list_pop(&pd->iosched.read_queue);
895 		spin_unlock(&pd->iosched.lock);
896 
897 		if (!bio)
898 			continue;
899 
900 		if (bio_data_dir(bio) == READ)
901 			pd->iosched.successive_reads += bio->bi_size >> 10;
902 		else {
903 			pd->iosched.successive_reads = 0;
904 			pd->iosched.last_write = bio->bi_sector + bio_sectors(bio);
905 		}
906 		if (pd->iosched.successive_reads >= HI_SPEED_SWITCH) {
907 			if (pd->read_speed == pd->write_speed) {
908 				pd->read_speed = MAX_SPEED;
909 				pkt_set_speed(pd, pd->write_speed, pd->read_speed);
910 			}
911 		} else {
912 			if (pd->read_speed != pd->write_speed) {
913 				pd->read_speed = pd->write_speed;
914 				pkt_set_speed(pd, pd->write_speed, pd->read_speed);
915 			}
916 		}
917 
918 		atomic_inc(&pd->cdrw.pending_bios);
919 		generic_make_request(bio);
920 	}
921 }
922 
923 /*
924  * Special care is needed if the underlying block device has a small
925  * max_phys_segments value.
926  */
927 static int pkt_set_segment_merging(struct pktcdvd_device *pd, struct request_queue *q)
928 {
929 	if ((pd->settings.size << 9) / CD_FRAMESIZE
930 	    <= queue_max_segments(q)) {
931 		/*
932 		 * The cdrom device can handle one segment/frame
933 		 */
934 		clear_bit(PACKET_MERGE_SEGS, &pd->flags);
935 		return 0;
936 	} else if ((pd->settings.size << 9) / PAGE_SIZE
937 		   <= queue_max_segments(q)) {
938 		/*
939 		 * We can handle this case at the expense of some extra memory
940 		 * copies during write operations
941 		 */
942 		set_bit(PACKET_MERGE_SEGS, &pd->flags);
943 		return 0;
944 	} else {
945 		printk(DRIVER_NAME": cdrom max_phys_segments too small\n");
946 		return -EIO;
947 	}
948 }
949 
950 /*
951  * Copy CD_FRAMESIZE bytes from src_bio into a destination page
952  */
953 static void pkt_copy_bio_data(struct bio *src_bio, int seg, int offs, struct page *dst_page, int dst_offs)
954 {
955 	unsigned int copy_size = CD_FRAMESIZE;
956 
957 	while (copy_size > 0) {
958 		struct bio_vec *src_bvl = bio_iovec_idx(src_bio, seg);
959 		void *vfrom = kmap_atomic(src_bvl->bv_page) +
960 			src_bvl->bv_offset + offs;
961 		void *vto = page_address(dst_page) + dst_offs;
962 		int len = min_t(int, copy_size, src_bvl->bv_len - offs);
963 
964 		BUG_ON(len < 0);
965 		memcpy(vto, vfrom, len);
966 		kunmap_atomic(vfrom);
967 
968 		seg++;
969 		offs = 0;
970 		dst_offs += len;
971 		copy_size -= len;
972 	}
973 }
974 
975 /*
976  * Copy all data for this packet to pkt->pages[], so that
977  * a) The number of required segments for the write bio is minimized, which
978  *    is necessary for some scsi controllers.
979  * b) The data can be used as cache to avoid read requests if we receive a
980  *    new write request for the same zone.
981  */
982 static void pkt_make_local_copy(struct packet_data *pkt, struct bio_vec *bvec)
983 {
984 	int f, p, offs;
985 
986 	/* Copy all data to pkt->pages[] */
987 	p = 0;
988 	offs = 0;
989 	for (f = 0; f < pkt->frames; f++) {
990 		if (bvec[f].bv_page != pkt->pages[p]) {
991 			void *vfrom = kmap_atomic(bvec[f].bv_page) + bvec[f].bv_offset;
992 			void *vto = page_address(pkt->pages[p]) + offs;
993 			memcpy(vto, vfrom, CD_FRAMESIZE);
994 			kunmap_atomic(vfrom);
995 			bvec[f].bv_page = pkt->pages[p];
996 			bvec[f].bv_offset = offs;
997 		} else {
998 			BUG_ON(bvec[f].bv_offset != offs);
999 		}
1000 		offs += CD_FRAMESIZE;
1001 		if (offs >= PAGE_SIZE) {
1002 			offs = 0;
1003 			p++;
1004 		}
1005 	}
1006 }
1007 
1008 static void pkt_end_io_read(struct bio *bio, int err)
1009 {
1010 	struct packet_data *pkt = bio->bi_private;
1011 	struct pktcdvd_device *pd = pkt->pd;
1012 	BUG_ON(!pd);
1013 
1014 	VPRINTK("pkt_end_io_read: bio=%p sec0=%llx sec=%llx err=%d\n", bio,
1015 		(unsigned long long)pkt->sector, (unsigned long long)bio->bi_sector, err);
1016 
1017 	if (err)
1018 		atomic_inc(&pkt->io_errors);
1019 	if (atomic_dec_and_test(&pkt->io_wait)) {
1020 		atomic_inc(&pkt->run_sm);
1021 		wake_up(&pd->wqueue);
1022 	}
1023 	pkt_bio_finished(pd);
1024 }
1025 
1026 static void pkt_end_io_packet_write(struct bio *bio, int err)
1027 {
1028 	struct packet_data *pkt = bio->bi_private;
1029 	struct pktcdvd_device *pd = pkt->pd;
1030 	BUG_ON(!pd);
1031 
1032 	VPRINTK("pkt_end_io_packet_write: id=%d, err=%d\n", pkt->id, err);
1033 
1034 	pd->stats.pkt_ended++;
1035 
1036 	pkt_bio_finished(pd);
1037 	atomic_dec(&pkt->io_wait);
1038 	atomic_inc(&pkt->run_sm);
1039 	wake_up(&pd->wqueue);
1040 }
1041 
1042 /*
1043  * Schedule reads for the holes in a packet
1044  */
1045 static void pkt_gather_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1046 {
1047 	int frames_read = 0;
1048 	struct bio *bio;
1049 	int f;
1050 	char written[PACKET_MAX_SIZE];
1051 
1052 	BUG_ON(bio_list_empty(&pkt->orig_bios));
1053 
1054 	atomic_set(&pkt->io_wait, 0);
1055 	atomic_set(&pkt->io_errors, 0);
1056 
1057 	/*
1058 	 * Figure out which frames we need to read before we can write.
1059 	 */
1060 	memset(written, 0, sizeof(written));
1061 	spin_lock(&pkt->lock);
1062 	bio_list_for_each(bio, &pkt->orig_bios) {
1063 		int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1064 		int num_frames = bio->bi_size / CD_FRAMESIZE;
1065 		pd->stats.secs_w += num_frames * (CD_FRAMESIZE >> 9);
1066 		BUG_ON(first_frame < 0);
1067 		BUG_ON(first_frame + num_frames > pkt->frames);
1068 		for (f = first_frame; f < first_frame + num_frames; f++)
1069 			written[f] = 1;
1070 	}
1071 	spin_unlock(&pkt->lock);
1072 
1073 	if (pkt->cache_valid) {
1074 		VPRINTK("pkt_gather_data: zone %llx cached\n",
1075 			(unsigned long long)pkt->sector);
1076 		goto out_account;
1077 	}
1078 
1079 	/*
1080 	 * Schedule reads for missing parts of the packet.
1081 	 */
1082 	for (f = 0; f < pkt->frames; f++) {
1083 		int p, offset;
1084 
1085 		if (written[f])
1086 			continue;
1087 
1088 		bio = pkt->r_bios[f];
1089 		bio_reset(bio);
1090 		bio->bi_sector = pkt->sector + f * (CD_FRAMESIZE >> 9);
1091 		bio->bi_bdev = pd->bdev;
1092 		bio->bi_end_io = pkt_end_io_read;
1093 		bio->bi_private = pkt;
1094 
1095 		p = (f * CD_FRAMESIZE) / PAGE_SIZE;
1096 		offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1097 		VPRINTK("pkt_gather_data: Adding frame %d, page:%p offs:%d\n",
1098 			f, pkt->pages[p], offset);
1099 		if (!bio_add_page(bio, pkt->pages[p], CD_FRAMESIZE, offset))
1100 			BUG();
1101 
1102 		atomic_inc(&pkt->io_wait);
1103 		bio->bi_rw = READ;
1104 		pkt_queue_bio(pd, bio);
1105 		frames_read++;
1106 	}
1107 
1108 out_account:
1109 	VPRINTK("pkt_gather_data: need %d frames for zone %llx\n",
1110 		frames_read, (unsigned long long)pkt->sector);
1111 	pd->stats.pkt_started++;
1112 	pd->stats.secs_rg += frames_read * (CD_FRAMESIZE >> 9);
1113 }
1114 
1115 /*
1116  * Find a packet matching zone, or the least recently used packet if
1117  * there is no match.
1118  */
1119 static struct packet_data *pkt_get_packet_data(struct pktcdvd_device *pd, int zone)
1120 {
1121 	struct packet_data *pkt;
1122 
1123 	list_for_each_entry(pkt, &pd->cdrw.pkt_free_list, list) {
1124 		if (pkt->sector == zone || pkt->list.next == &pd->cdrw.pkt_free_list) {
1125 			list_del_init(&pkt->list);
1126 			if (pkt->sector != zone)
1127 				pkt->cache_valid = 0;
1128 			return pkt;
1129 		}
1130 	}
1131 	BUG();
1132 	return NULL;
1133 }
1134 
1135 static void pkt_put_packet_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1136 {
1137 	if (pkt->cache_valid) {
1138 		list_add(&pkt->list, &pd->cdrw.pkt_free_list);
1139 	} else {
1140 		list_add_tail(&pkt->list, &pd->cdrw.pkt_free_list);
1141 	}
1142 }
1143 
1144 /*
1145  * recover a failed write, query for relocation if possible
1146  *
1147  * returns 1 if recovery is possible, or 0 if not
1148  *
1149  */
1150 static int pkt_start_recovery(struct packet_data *pkt)
1151 {
1152 	/*
1153 	 * FIXME. We need help from the file system to implement
1154 	 * recovery handling.
1155 	 */
1156 	return 0;
1157 #if 0
1158 	struct request *rq = pkt->rq;
1159 	struct pktcdvd_device *pd = rq->rq_disk->private_data;
1160 	struct block_device *pkt_bdev;
1161 	struct super_block *sb = NULL;
1162 	unsigned long old_block, new_block;
1163 	sector_t new_sector;
1164 
1165 	pkt_bdev = bdget(kdev_t_to_nr(pd->pkt_dev));
1166 	if (pkt_bdev) {
1167 		sb = get_super(pkt_bdev);
1168 		bdput(pkt_bdev);
1169 	}
1170 
1171 	if (!sb)
1172 		return 0;
1173 
1174 	if (!sb->s_op->relocate_blocks)
1175 		goto out;
1176 
1177 	old_block = pkt->sector / (CD_FRAMESIZE >> 9);
1178 	if (sb->s_op->relocate_blocks(sb, old_block, &new_block))
1179 		goto out;
1180 
1181 	new_sector = new_block * (CD_FRAMESIZE >> 9);
1182 	pkt->sector = new_sector;
1183 
1184 	pkt->bio->bi_sector = new_sector;
1185 	pkt->bio->bi_next = NULL;
1186 	pkt->bio->bi_flags = 1 << BIO_UPTODATE;
1187 	pkt->bio->bi_idx = 0;
1188 
1189 	BUG_ON(pkt->bio->bi_rw != REQ_WRITE);
1190 	BUG_ON(pkt->bio->bi_vcnt != pkt->frames);
1191 	BUG_ON(pkt->bio->bi_size != pkt->frames * CD_FRAMESIZE);
1192 	BUG_ON(pkt->bio->bi_end_io != pkt_end_io_packet_write);
1193 	BUG_ON(pkt->bio->bi_private != pkt);
1194 
1195 	drop_super(sb);
1196 	return 1;
1197 
1198 out:
1199 	drop_super(sb);
1200 	return 0;
1201 #endif
1202 }
1203 
1204 static inline void pkt_set_state(struct packet_data *pkt, enum packet_data_state state)
1205 {
1206 #if PACKET_DEBUG > 1
1207 	static const char *state_name[] = {
1208 		"IDLE", "WAITING", "READ_WAIT", "WRITE_WAIT", "RECOVERY", "FINISHED"
1209 	};
1210 	enum packet_data_state old_state = pkt->state;
1211 	VPRINTK("pkt %2d : s=%6llx %s -> %s\n", pkt->id, (unsigned long long)pkt->sector,
1212 		state_name[old_state], state_name[state]);
1213 #endif
1214 	pkt->state = state;
1215 }
1216 
1217 /*
1218  * Scan the work queue to see if we can start a new packet.
1219  * returns non-zero if any work was done.
1220  */
1221 static int pkt_handle_queue(struct pktcdvd_device *pd)
1222 {
1223 	struct packet_data *pkt, *p;
1224 	struct bio *bio = NULL;
1225 	sector_t zone = 0; /* Suppress gcc warning */
1226 	struct pkt_rb_node *node, *first_node;
1227 	struct rb_node *n;
1228 	int wakeup;
1229 
1230 	VPRINTK("handle_queue\n");
1231 
1232 	atomic_set(&pd->scan_queue, 0);
1233 
1234 	if (list_empty(&pd->cdrw.pkt_free_list)) {
1235 		VPRINTK("handle_queue: no pkt\n");
1236 		return 0;
1237 	}
1238 
1239 	/*
1240 	 * Try to find a zone we are not already working on.
1241 	 */
1242 	spin_lock(&pd->lock);
1243 	first_node = pkt_rbtree_find(pd, pd->current_sector);
1244 	if (!first_node) {
1245 		n = rb_first(&pd->bio_queue);
1246 		if (n)
1247 			first_node = rb_entry(n, struct pkt_rb_node, rb_node);
1248 	}
1249 	node = first_node;
1250 	while (node) {
1251 		bio = node->bio;
1252 		zone = ZONE(bio->bi_sector, pd);
1253 		list_for_each_entry(p, &pd->cdrw.pkt_active_list, list) {
1254 			if (p->sector == zone) {
1255 				bio = NULL;
1256 				goto try_next_bio;
1257 			}
1258 		}
1259 		break;
1260 try_next_bio:
1261 		node = pkt_rbtree_next(node);
1262 		if (!node) {
1263 			n = rb_first(&pd->bio_queue);
1264 			if (n)
1265 				node = rb_entry(n, struct pkt_rb_node, rb_node);
1266 		}
1267 		if (node == first_node)
1268 			node = NULL;
1269 	}
1270 	spin_unlock(&pd->lock);
1271 	if (!bio) {
1272 		VPRINTK("handle_queue: no bio\n");
1273 		return 0;
1274 	}
1275 
1276 	pkt = pkt_get_packet_data(pd, zone);
1277 
1278 	pd->current_sector = zone + pd->settings.size;
1279 	pkt->sector = zone;
1280 	BUG_ON(pkt->frames != pd->settings.size >> 2);
1281 	pkt->write_size = 0;
1282 
1283 	/*
1284 	 * Scan work queue for bios in the same zone and link them
1285 	 * to this packet.
1286 	 */
1287 	spin_lock(&pd->lock);
1288 	VPRINTK("pkt_handle_queue: looking for zone %llx\n", (unsigned long long)zone);
1289 	while ((node = pkt_rbtree_find(pd, zone)) != NULL) {
1290 		bio = node->bio;
1291 		VPRINTK("pkt_handle_queue: found zone=%llx\n",
1292 			(unsigned long long)ZONE(bio->bi_sector, pd));
1293 		if (ZONE(bio->bi_sector, pd) != zone)
1294 			break;
1295 		pkt_rbtree_erase(pd, node);
1296 		spin_lock(&pkt->lock);
1297 		bio_list_add(&pkt->orig_bios, bio);
1298 		pkt->write_size += bio->bi_size / CD_FRAMESIZE;
1299 		spin_unlock(&pkt->lock);
1300 	}
1301 	/* check write congestion marks, and if bio_queue_size is
1302 	   below, wake up any waiters */
1303 	wakeup = (pd->write_congestion_on > 0
1304 	 		&& pd->bio_queue_size <= pd->write_congestion_off);
1305 	spin_unlock(&pd->lock);
1306 	if (wakeup) {
1307 		clear_bdi_congested(&pd->disk->queue->backing_dev_info,
1308 					BLK_RW_ASYNC);
1309 	}
1310 
1311 	pkt->sleep_time = max(PACKET_WAIT_TIME, 1);
1312 	pkt_set_state(pkt, PACKET_WAITING_STATE);
1313 	atomic_set(&pkt->run_sm, 1);
1314 
1315 	spin_lock(&pd->cdrw.active_list_lock);
1316 	list_add(&pkt->list, &pd->cdrw.pkt_active_list);
1317 	spin_unlock(&pd->cdrw.active_list_lock);
1318 
1319 	return 1;
1320 }
1321 
1322 /*
1323  * Assemble a bio to write one packet and queue the bio for processing
1324  * by the underlying block device.
1325  */
1326 static void pkt_start_write(struct pktcdvd_device *pd, struct packet_data *pkt)
1327 {
1328 	struct bio *bio;
1329 	int f;
1330 	int frames_write;
1331 	struct bio_vec *bvec = pkt->w_bio->bi_io_vec;
1332 
1333 	for (f = 0; f < pkt->frames; f++) {
1334 		bvec[f].bv_page = pkt->pages[(f * CD_FRAMESIZE) / PAGE_SIZE];
1335 		bvec[f].bv_offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1336 	}
1337 
1338 	/*
1339 	 * Fill-in bvec with data from orig_bios.
1340 	 */
1341 	frames_write = 0;
1342 	spin_lock(&pkt->lock);
1343 	bio_list_for_each(bio, &pkt->orig_bios) {
1344 		int segment = bio->bi_idx;
1345 		int src_offs = 0;
1346 		int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1347 		int num_frames = bio->bi_size / CD_FRAMESIZE;
1348 		BUG_ON(first_frame < 0);
1349 		BUG_ON(first_frame + num_frames > pkt->frames);
1350 		for (f = first_frame; f < first_frame + num_frames; f++) {
1351 			struct bio_vec *src_bvl = bio_iovec_idx(bio, segment);
1352 
1353 			while (src_offs >= src_bvl->bv_len) {
1354 				src_offs -= src_bvl->bv_len;
1355 				segment++;
1356 				BUG_ON(segment >= bio->bi_vcnt);
1357 				src_bvl = bio_iovec_idx(bio, segment);
1358 			}
1359 
1360 			if (src_bvl->bv_len - src_offs >= CD_FRAMESIZE) {
1361 				bvec[f].bv_page = src_bvl->bv_page;
1362 				bvec[f].bv_offset = src_bvl->bv_offset + src_offs;
1363 			} else {
1364 				pkt_copy_bio_data(bio, segment, src_offs,
1365 						  bvec[f].bv_page, bvec[f].bv_offset);
1366 			}
1367 			src_offs += CD_FRAMESIZE;
1368 			frames_write++;
1369 		}
1370 	}
1371 	pkt_set_state(pkt, PACKET_WRITE_WAIT_STATE);
1372 	spin_unlock(&pkt->lock);
1373 
1374 	VPRINTK("pkt_start_write: Writing %d frames for zone %llx\n",
1375 		frames_write, (unsigned long long)pkt->sector);
1376 	BUG_ON(frames_write != pkt->write_size);
1377 
1378 	if (test_bit(PACKET_MERGE_SEGS, &pd->flags) || (pkt->write_size < pkt->frames)) {
1379 		pkt_make_local_copy(pkt, bvec);
1380 		pkt->cache_valid = 1;
1381 	} else {
1382 		pkt->cache_valid = 0;
1383 	}
1384 
1385 	/* Start the write request */
1386 	bio_reset(pkt->w_bio);
1387 	pkt->w_bio->bi_sector = pkt->sector;
1388 	pkt->w_bio->bi_bdev = pd->bdev;
1389 	pkt->w_bio->bi_end_io = pkt_end_io_packet_write;
1390 	pkt->w_bio->bi_private = pkt;
1391 	for (f = 0; f < pkt->frames; f++)
1392 		if (!bio_add_page(pkt->w_bio, bvec[f].bv_page, CD_FRAMESIZE, bvec[f].bv_offset))
1393 			BUG();
1394 	VPRINTK(DRIVER_NAME": vcnt=%d\n", pkt->w_bio->bi_vcnt);
1395 
1396 	atomic_set(&pkt->io_wait, 1);
1397 	pkt->w_bio->bi_rw = WRITE;
1398 	pkt_queue_bio(pd, pkt->w_bio);
1399 }
1400 
1401 static void pkt_finish_packet(struct packet_data *pkt, int uptodate)
1402 {
1403 	struct bio *bio;
1404 
1405 	if (!uptodate)
1406 		pkt->cache_valid = 0;
1407 
1408 	/* Finish all bios corresponding to this packet */
1409 	while ((bio = bio_list_pop(&pkt->orig_bios)))
1410 		bio_endio(bio, uptodate ? 0 : -EIO);
1411 }
1412 
1413 static void pkt_run_state_machine(struct pktcdvd_device *pd, struct packet_data *pkt)
1414 {
1415 	int uptodate;
1416 
1417 	VPRINTK("run_state_machine: pkt %d\n", pkt->id);
1418 
1419 	for (;;) {
1420 		switch (pkt->state) {
1421 		case PACKET_WAITING_STATE:
1422 			if ((pkt->write_size < pkt->frames) && (pkt->sleep_time > 0))
1423 				return;
1424 
1425 			pkt->sleep_time = 0;
1426 			pkt_gather_data(pd, pkt);
1427 			pkt_set_state(pkt, PACKET_READ_WAIT_STATE);
1428 			break;
1429 
1430 		case PACKET_READ_WAIT_STATE:
1431 			if (atomic_read(&pkt->io_wait) > 0)
1432 				return;
1433 
1434 			if (atomic_read(&pkt->io_errors) > 0) {
1435 				pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1436 			} else {
1437 				pkt_start_write(pd, pkt);
1438 			}
1439 			break;
1440 
1441 		case PACKET_WRITE_WAIT_STATE:
1442 			if (atomic_read(&pkt->io_wait) > 0)
1443 				return;
1444 
1445 			if (test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags)) {
1446 				pkt_set_state(pkt, PACKET_FINISHED_STATE);
1447 			} else {
1448 				pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1449 			}
1450 			break;
1451 
1452 		case PACKET_RECOVERY_STATE:
1453 			if (pkt_start_recovery(pkt)) {
1454 				pkt_start_write(pd, pkt);
1455 			} else {
1456 				VPRINTK("No recovery possible\n");
1457 				pkt_set_state(pkt, PACKET_FINISHED_STATE);
1458 			}
1459 			break;
1460 
1461 		case PACKET_FINISHED_STATE:
1462 			uptodate = test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags);
1463 			pkt_finish_packet(pkt, uptodate);
1464 			return;
1465 
1466 		default:
1467 			BUG();
1468 			break;
1469 		}
1470 	}
1471 }
1472 
1473 static void pkt_handle_packets(struct pktcdvd_device *pd)
1474 {
1475 	struct packet_data *pkt, *next;
1476 
1477 	VPRINTK("pkt_handle_packets\n");
1478 
1479 	/*
1480 	 * Run state machine for active packets
1481 	 */
1482 	list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1483 		if (atomic_read(&pkt->run_sm) > 0) {
1484 			atomic_set(&pkt->run_sm, 0);
1485 			pkt_run_state_machine(pd, pkt);
1486 		}
1487 	}
1488 
1489 	/*
1490 	 * Move no longer active packets to the free list
1491 	 */
1492 	spin_lock(&pd->cdrw.active_list_lock);
1493 	list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_active_list, list) {
1494 		if (pkt->state == PACKET_FINISHED_STATE) {
1495 			list_del(&pkt->list);
1496 			pkt_put_packet_data(pd, pkt);
1497 			pkt_set_state(pkt, PACKET_IDLE_STATE);
1498 			atomic_set(&pd->scan_queue, 1);
1499 		}
1500 	}
1501 	spin_unlock(&pd->cdrw.active_list_lock);
1502 }
1503 
1504 static void pkt_count_states(struct pktcdvd_device *pd, int *states)
1505 {
1506 	struct packet_data *pkt;
1507 	int i;
1508 
1509 	for (i = 0; i < PACKET_NUM_STATES; i++)
1510 		states[i] = 0;
1511 
1512 	spin_lock(&pd->cdrw.active_list_lock);
1513 	list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1514 		states[pkt->state]++;
1515 	}
1516 	spin_unlock(&pd->cdrw.active_list_lock);
1517 }
1518 
1519 /*
1520  * kcdrwd is woken up when writes have been queued for one of our
1521  * registered devices
1522  */
1523 static int kcdrwd(void *foobar)
1524 {
1525 	struct pktcdvd_device *pd = foobar;
1526 	struct packet_data *pkt;
1527 	long min_sleep_time, residue;
1528 
1529 	set_user_nice(current, -20);
1530 	set_freezable();
1531 
1532 	for (;;) {
1533 		DECLARE_WAITQUEUE(wait, current);
1534 
1535 		/*
1536 		 * Wait until there is something to do
1537 		 */
1538 		add_wait_queue(&pd->wqueue, &wait);
1539 		for (;;) {
1540 			set_current_state(TASK_INTERRUPTIBLE);
1541 
1542 			/* Check if we need to run pkt_handle_queue */
1543 			if (atomic_read(&pd->scan_queue) > 0)
1544 				goto work_to_do;
1545 
1546 			/* Check if we need to run the state machine for some packet */
1547 			list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1548 				if (atomic_read(&pkt->run_sm) > 0)
1549 					goto work_to_do;
1550 			}
1551 
1552 			/* Check if we need to process the iosched queues */
1553 			if (atomic_read(&pd->iosched.attention) != 0)
1554 				goto work_to_do;
1555 
1556 			/* Otherwise, go to sleep */
1557 			if (PACKET_DEBUG > 1) {
1558 				int states[PACKET_NUM_STATES];
1559 				pkt_count_states(pd, states);
1560 				VPRINTK("kcdrwd: i:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
1561 					states[0], states[1], states[2], states[3],
1562 					states[4], states[5]);
1563 			}
1564 
1565 			min_sleep_time = MAX_SCHEDULE_TIMEOUT;
1566 			list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1567 				if (pkt->sleep_time && pkt->sleep_time < min_sleep_time)
1568 					min_sleep_time = pkt->sleep_time;
1569 			}
1570 
1571 			VPRINTK("kcdrwd: sleeping\n");
1572 			residue = schedule_timeout(min_sleep_time);
1573 			VPRINTK("kcdrwd: wake up\n");
1574 
1575 			/* make swsusp happy with our thread */
1576 			try_to_freeze();
1577 
1578 			list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1579 				if (!pkt->sleep_time)
1580 					continue;
1581 				pkt->sleep_time -= min_sleep_time - residue;
1582 				if (pkt->sleep_time <= 0) {
1583 					pkt->sleep_time = 0;
1584 					atomic_inc(&pkt->run_sm);
1585 				}
1586 			}
1587 
1588 			if (kthread_should_stop())
1589 				break;
1590 		}
1591 work_to_do:
1592 		set_current_state(TASK_RUNNING);
1593 		remove_wait_queue(&pd->wqueue, &wait);
1594 
1595 		if (kthread_should_stop())
1596 			break;
1597 
1598 		/*
1599 		 * if pkt_handle_queue returns true, we can queue
1600 		 * another request.
1601 		 */
1602 		while (pkt_handle_queue(pd))
1603 			;
1604 
1605 		/*
1606 		 * Handle packet state machine
1607 		 */
1608 		pkt_handle_packets(pd);
1609 
1610 		/*
1611 		 * Handle iosched queues
1612 		 */
1613 		pkt_iosched_process_queue(pd);
1614 	}
1615 
1616 	return 0;
1617 }
1618 
1619 static void pkt_print_settings(struct pktcdvd_device *pd)
1620 {
1621 	printk(DRIVER_NAME": %s packets, ", pd->settings.fp ? "Fixed" : "Variable");
1622 	printk("%u blocks, ", pd->settings.size >> 2);
1623 	printk("Mode-%c disc\n", pd->settings.block_mode == 8 ? '1' : '2');
1624 }
1625 
1626 static int pkt_mode_sense(struct pktcdvd_device *pd, struct packet_command *cgc, int page_code, int page_control)
1627 {
1628 	memset(cgc->cmd, 0, sizeof(cgc->cmd));
1629 
1630 	cgc->cmd[0] = GPCMD_MODE_SENSE_10;
1631 	cgc->cmd[2] = page_code | (page_control << 6);
1632 	cgc->cmd[7] = cgc->buflen >> 8;
1633 	cgc->cmd[8] = cgc->buflen & 0xff;
1634 	cgc->data_direction = CGC_DATA_READ;
1635 	return pkt_generic_packet(pd, cgc);
1636 }
1637 
1638 static int pkt_mode_select(struct pktcdvd_device *pd, struct packet_command *cgc)
1639 {
1640 	memset(cgc->cmd, 0, sizeof(cgc->cmd));
1641 	memset(cgc->buffer, 0, 2);
1642 	cgc->cmd[0] = GPCMD_MODE_SELECT_10;
1643 	cgc->cmd[1] = 0x10;		/* PF */
1644 	cgc->cmd[7] = cgc->buflen >> 8;
1645 	cgc->cmd[8] = cgc->buflen & 0xff;
1646 	cgc->data_direction = CGC_DATA_WRITE;
1647 	return pkt_generic_packet(pd, cgc);
1648 }
1649 
1650 static int pkt_get_disc_info(struct pktcdvd_device *pd, disc_information *di)
1651 {
1652 	struct packet_command cgc;
1653 	int ret;
1654 
1655 	/* set up command and get the disc info */
1656 	init_cdrom_command(&cgc, di, sizeof(*di), CGC_DATA_READ);
1657 	cgc.cmd[0] = GPCMD_READ_DISC_INFO;
1658 	cgc.cmd[8] = cgc.buflen = 2;
1659 	cgc.quiet = 1;
1660 
1661 	if ((ret = pkt_generic_packet(pd, &cgc)))
1662 		return ret;
1663 
1664 	/* not all drives have the same disc_info length, so requeue
1665 	 * packet with the length the drive tells us it can supply
1666 	 */
1667 	cgc.buflen = be16_to_cpu(di->disc_information_length) +
1668 		     sizeof(di->disc_information_length);
1669 
1670 	if (cgc.buflen > sizeof(disc_information))
1671 		cgc.buflen = sizeof(disc_information);
1672 
1673 	cgc.cmd[8] = cgc.buflen;
1674 	return pkt_generic_packet(pd, &cgc);
1675 }
1676 
1677 static int pkt_get_track_info(struct pktcdvd_device *pd, __u16 track, __u8 type, track_information *ti)
1678 {
1679 	struct packet_command cgc;
1680 	int ret;
1681 
1682 	init_cdrom_command(&cgc, ti, 8, CGC_DATA_READ);
1683 	cgc.cmd[0] = GPCMD_READ_TRACK_RZONE_INFO;
1684 	cgc.cmd[1] = type & 3;
1685 	cgc.cmd[4] = (track & 0xff00) >> 8;
1686 	cgc.cmd[5] = track & 0xff;
1687 	cgc.cmd[8] = 8;
1688 	cgc.quiet = 1;
1689 
1690 	if ((ret = pkt_generic_packet(pd, &cgc)))
1691 		return ret;
1692 
1693 	cgc.buflen = be16_to_cpu(ti->track_information_length) +
1694 		     sizeof(ti->track_information_length);
1695 
1696 	if (cgc.buflen > sizeof(track_information))
1697 		cgc.buflen = sizeof(track_information);
1698 
1699 	cgc.cmd[8] = cgc.buflen;
1700 	return pkt_generic_packet(pd, &cgc);
1701 }
1702 
1703 static noinline_for_stack int pkt_get_last_written(struct pktcdvd_device *pd,
1704 						long *last_written)
1705 {
1706 	disc_information di;
1707 	track_information ti;
1708 	__u32 last_track;
1709 	int ret = -1;
1710 
1711 	if ((ret = pkt_get_disc_info(pd, &di)))
1712 		return ret;
1713 
1714 	last_track = (di.last_track_msb << 8) | di.last_track_lsb;
1715 	if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1716 		return ret;
1717 
1718 	/* if this track is blank, try the previous. */
1719 	if (ti.blank) {
1720 		last_track--;
1721 		if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1722 			return ret;
1723 	}
1724 
1725 	/* if last recorded field is valid, return it. */
1726 	if (ti.lra_v) {
1727 		*last_written = be32_to_cpu(ti.last_rec_address);
1728 	} else {
1729 		/* make it up instead */
1730 		*last_written = be32_to_cpu(ti.track_start) +
1731 				be32_to_cpu(ti.track_size);
1732 		if (ti.free_blocks)
1733 			*last_written -= (be32_to_cpu(ti.free_blocks) + 7);
1734 	}
1735 	return 0;
1736 }
1737 
1738 /*
1739  * write mode select package based on pd->settings
1740  */
1741 static noinline_for_stack int pkt_set_write_settings(struct pktcdvd_device *pd)
1742 {
1743 	struct packet_command cgc;
1744 	struct request_sense sense;
1745 	write_param_page *wp;
1746 	char buffer[128];
1747 	int ret, size;
1748 
1749 	/* doesn't apply to DVD+RW or DVD-RAM */
1750 	if ((pd->mmc3_profile == 0x1a) || (pd->mmc3_profile == 0x12))
1751 		return 0;
1752 
1753 	memset(buffer, 0, sizeof(buffer));
1754 	init_cdrom_command(&cgc, buffer, sizeof(*wp), CGC_DATA_READ);
1755 	cgc.sense = &sense;
1756 	if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1757 		pkt_dump_sense(&cgc);
1758 		return ret;
1759 	}
1760 
1761 	size = 2 + ((buffer[0] << 8) | (buffer[1] & 0xff));
1762 	pd->mode_offset = (buffer[6] << 8) | (buffer[7] & 0xff);
1763 	if (size > sizeof(buffer))
1764 		size = sizeof(buffer);
1765 
1766 	/*
1767 	 * now get it all
1768 	 */
1769 	init_cdrom_command(&cgc, buffer, size, CGC_DATA_READ);
1770 	cgc.sense = &sense;
1771 	if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1772 		pkt_dump_sense(&cgc);
1773 		return ret;
1774 	}
1775 
1776 	/*
1777 	 * write page is offset header + block descriptor length
1778 	 */
1779 	wp = (write_param_page *) &buffer[sizeof(struct mode_page_header) + pd->mode_offset];
1780 
1781 	wp->fp = pd->settings.fp;
1782 	wp->track_mode = pd->settings.track_mode;
1783 	wp->write_type = pd->settings.write_type;
1784 	wp->data_block_type = pd->settings.block_mode;
1785 
1786 	wp->multi_session = 0;
1787 
1788 #ifdef PACKET_USE_LS
1789 	wp->link_size = 7;
1790 	wp->ls_v = 1;
1791 #endif
1792 
1793 	if (wp->data_block_type == PACKET_BLOCK_MODE1) {
1794 		wp->session_format = 0;
1795 		wp->subhdr2 = 0x20;
1796 	} else if (wp->data_block_type == PACKET_BLOCK_MODE2) {
1797 		wp->session_format = 0x20;
1798 		wp->subhdr2 = 8;
1799 #if 0
1800 		wp->mcn[0] = 0x80;
1801 		memcpy(&wp->mcn[1], PACKET_MCN, sizeof(wp->mcn) - 1);
1802 #endif
1803 	} else {
1804 		/*
1805 		 * paranoia
1806 		 */
1807 		printk(DRIVER_NAME": write mode wrong %d\n", wp->data_block_type);
1808 		return 1;
1809 	}
1810 	wp->packet_size = cpu_to_be32(pd->settings.size >> 2);
1811 
1812 	cgc.buflen = cgc.cmd[8] = size;
1813 	if ((ret = pkt_mode_select(pd, &cgc))) {
1814 		pkt_dump_sense(&cgc);
1815 		return ret;
1816 	}
1817 
1818 	pkt_print_settings(pd);
1819 	return 0;
1820 }
1821 
1822 /*
1823  * 1 -- we can write to this track, 0 -- we can't
1824  */
1825 static int pkt_writable_track(struct pktcdvd_device *pd, track_information *ti)
1826 {
1827 	switch (pd->mmc3_profile) {
1828 		case 0x1a: /* DVD+RW */
1829 		case 0x12: /* DVD-RAM */
1830 			/* The track is always writable on DVD+RW/DVD-RAM */
1831 			return 1;
1832 		default:
1833 			break;
1834 	}
1835 
1836 	if (!ti->packet || !ti->fp)
1837 		return 0;
1838 
1839 	/*
1840 	 * "good" settings as per Mt Fuji.
1841 	 */
1842 	if (ti->rt == 0 && ti->blank == 0)
1843 		return 1;
1844 
1845 	if (ti->rt == 0 && ti->blank == 1)
1846 		return 1;
1847 
1848 	if (ti->rt == 1 && ti->blank == 0)
1849 		return 1;
1850 
1851 	printk(DRIVER_NAME": bad state %d-%d-%d\n", ti->rt, ti->blank, ti->packet);
1852 	return 0;
1853 }
1854 
1855 /*
1856  * 1 -- we can write to this disc, 0 -- we can't
1857  */
1858 static int pkt_writable_disc(struct pktcdvd_device *pd, disc_information *di)
1859 {
1860 	switch (pd->mmc3_profile) {
1861 		case 0x0a: /* CD-RW */
1862 		case 0xffff: /* MMC3 not supported */
1863 			break;
1864 		case 0x1a: /* DVD+RW */
1865 		case 0x13: /* DVD-RW */
1866 		case 0x12: /* DVD-RAM */
1867 			return 1;
1868 		default:
1869 			VPRINTK(DRIVER_NAME": Wrong disc profile (%x)\n", pd->mmc3_profile);
1870 			return 0;
1871 	}
1872 
1873 	/*
1874 	 * for disc type 0xff we should probably reserve a new track.
1875 	 * but i'm not sure, should we leave this to user apps? probably.
1876 	 */
1877 	if (di->disc_type == 0xff) {
1878 		printk(DRIVER_NAME": Unknown disc. No track?\n");
1879 		return 0;
1880 	}
1881 
1882 	if (di->disc_type != 0x20 && di->disc_type != 0) {
1883 		printk(DRIVER_NAME": Wrong disc type (%x)\n", di->disc_type);
1884 		return 0;
1885 	}
1886 
1887 	if (di->erasable == 0) {
1888 		printk(DRIVER_NAME": Disc not erasable\n");
1889 		return 0;
1890 	}
1891 
1892 	if (di->border_status == PACKET_SESSION_RESERVED) {
1893 		printk(DRIVER_NAME": Can't write to last track (reserved)\n");
1894 		return 0;
1895 	}
1896 
1897 	return 1;
1898 }
1899 
1900 static noinline_for_stack int pkt_probe_settings(struct pktcdvd_device *pd)
1901 {
1902 	struct packet_command cgc;
1903 	unsigned char buf[12];
1904 	disc_information di;
1905 	track_information ti;
1906 	int ret, track;
1907 
1908 	init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
1909 	cgc.cmd[0] = GPCMD_GET_CONFIGURATION;
1910 	cgc.cmd[8] = 8;
1911 	ret = pkt_generic_packet(pd, &cgc);
1912 	pd->mmc3_profile = ret ? 0xffff : buf[6] << 8 | buf[7];
1913 
1914 	memset(&di, 0, sizeof(disc_information));
1915 	memset(&ti, 0, sizeof(track_information));
1916 
1917 	if ((ret = pkt_get_disc_info(pd, &di))) {
1918 		printk("failed get_disc\n");
1919 		return ret;
1920 	}
1921 
1922 	if (!pkt_writable_disc(pd, &di))
1923 		return -EROFS;
1924 
1925 	pd->type = di.erasable ? PACKET_CDRW : PACKET_CDR;
1926 
1927 	track = 1; /* (di.last_track_msb << 8) | di.last_track_lsb; */
1928 	if ((ret = pkt_get_track_info(pd, track, 1, &ti))) {
1929 		printk(DRIVER_NAME": failed get_track\n");
1930 		return ret;
1931 	}
1932 
1933 	if (!pkt_writable_track(pd, &ti)) {
1934 		printk(DRIVER_NAME": can't write to this track\n");
1935 		return -EROFS;
1936 	}
1937 
1938 	/*
1939 	 * we keep packet size in 512 byte units, makes it easier to
1940 	 * deal with request calculations.
1941 	 */
1942 	pd->settings.size = be32_to_cpu(ti.fixed_packet_size) << 2;
1943 	if (pd->settings.size == 0) {
1944 		printk(DRIVER_NAME": detected zero packet size!\n");
1945 		return -ENXIO;
1946 	}
1947 	if (pd->settings.size > PACKET_MAX_SECTORS) {
1948 		printk(DRIVER_NAME": packet size is too big\n");
1949 		return -EROFS;
1950 	}
1951 	pd->settings.fp = ti.fp;
1952 	pd->offset = (be32_to_cpu(ti.track_start) << 2) & (pd->settings.size - 1);
1953 
1954 	if (ti.nwa_v) {
1955 		pd->nwa = be32_to_cpu(ti.next_writable);
1956 		set_bit(PACKET_NWA_VALID, &pd->flags);
1957 	}
1958 
1959 	/*
1960 	 * in theory we could use lra on -RW media as well and just zero
1961 	 * blocks that haven't been written yet, but in practice that
1962 	 * is just a no-go. we'll use that for -R, naturally.
1963 	 */
1964 	if (ti.lra_v) {
1965 		pd->lra = be32_to_cpu(ti.last_rec_address);
1966 		set_bit(PACKET_LRA_VALID, &pd->flags);
1967 	} else {
1968 		pd->lra = 0xffffffff;
1969 		set_bit(PACKET_LRA_VALID, &pd->flags);
1970 	}
1971 
1972 	/*
1973 	 * fine for now
1974 	 */
1975 	pd->settings.link_loss = 7;
1976 	pd->settings.write_type = 0;	/* packet */
1977 	pd->settings.track_mode = ti.track_mode;
1978 
1979 	/*
1980 	 * mode1 or mode2 disc
1981 	 */
1982 	switch (ti.data_mode) {
1983 		case PACKET_MODE1:
1984 			pd->settings.block_mode = PACKET_BLOCK_MODE1;
1985 			break;
1986 		case PACKET_MODE2:
1987 			pd->settings.block_mode = PACKET_BLOCK_MODE2;
1988 			break;
1989 		default:
1990 			printk(DRIVER_NAME": unknown data mode\n");
1991 			return -EROFS;
1992 	}
1993 	return 0;
1994 }
1995 
1996 /*
1997  * enable/disable write caching on drive
1998  */
1999 static noinline_for_stack int pkt_write_caching(struct pktcdvd_device *pd,
2000 						int set)
2001 {
2002 	struct packet_command cgc;
2003 	struct request_sense sense;
2004 	unsigned char buf[64];
2005 	int ret;
2006 
2007 	init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
2008 	cgc.sense = &sense;
2009 	cgc.buflen = pd->mode_offset + 12;
2010 
2011 	/*
2012 	 * caching mode page might not be there, so quiet this command
2013 	 */
2014 	cgc.quiet = 1;
2015 
2016 	if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WCACHING_PAGE, 0)))
2017 		return ret;
2018 
2019 	buf[pd->mode_offset + 10] |= (!!set << 2);
2020 
2021 	cgc.buflen = cgc.cmd[8] = 2 + ((buf[0] << 8) | (buf[1] & 0xff));
2022 	ret = pkt_mode_select(pd, &cgc);
2023 	if (ret) {
2024 		printk(DRIVER_NAME": write caching control failed\n");
2025 		pkt_dump_sense(&cgc);
2026 	} else if (!ret && set)
2027 		printk(DRIVER_NAME": enabled write caching on %s\n", pd->name);
2028 	return ret;
2029 }
2030 
2031 static int pkt_lock_door(struct pktcdvd_device *pd, int lockflag)
2032 {
2033 	struct packet_command cgc;
2034 
2035 	init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2036 	cgc.cmd[0] = GPCMD_PREVENT_ALLOW_MEDIUM_REMOVAL;
2037 	cgc.cmd[4] = lockflag ? 1 : 0;
2038 	return pkt_generic_packet(pd, &cgc);
2039 }
2040 
2041 /*
2042  * Returns drive maximum write speed
2043  */
2044 static noinline_for_stack int pkt_get_max_speed(struct pktcdvd_device *pd,
2045 						unsigned *write_speed)
2046 {
2047 	struct packet_command cgc;
2048 	struct request_sense sense;
2049 	unsigned char buf[256+18];
2050 	unsigned char *cap_buf;
2051 	int ret, offset;
2052 
2053 	cap_buf = &buf[sizeof(struct mode_page_header) + pd->mode_offset];
2054 	init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_UNKNOWN);
2055 	cgc.sense = &sense;
2056 
2057 	ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2058 	if (ret) {
2059 		cgc.buflen = pd->mode_offset + cap_buf[1] + 2 +
2060 			     sizeof(struct mode_page_header);
2061 		ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2062 		if (ret) {
2063 			pkt_dump_sense(&cgc);
2064 			return ret;
2065 		}
2066 	}
2067 
2068 	offset = 20;			    /* Obsoleted field, used by older drives */
2069 	if (cap_buf[1] >= 28)
2070 		offset = 28;		    /* Current write speed selected */
2071 	if (cap_buf[1] >= 30) {
2072 		/* If the drive reports at least one "Logical Unit Write
2073 		 * Speed Performance Descriptor Block", use the information
2074 		 * in the first block. (contains the highest speed)
2075 		 */
2076 		int num_spdb = (cap_buf[30] << 8) + cap_buf[31];
2077 		if (num_spdb > 0)
2078 			offset = 34;
2079 	}
2080 
2081 	*write_speed = (cap_buf[offset] << 8) | cap_buf[offset + 1];
2082 	return 0;
2083 }
2084 
2085 /* These tables from cdrecord - I don't have orange book */
2086 /* standard speed CD-RW (1-4x) */
2087 static char clv_to_speed[16] = {
2088 	/* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2089 	   0, 2, 4, 6, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2090 };
2091 /* high speed CD-RW (-10x) */
2092 static char hs_clv_to_speed[16] = {
2093 	/* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2094 	   0, 2, 4, 6, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2095 };
2096 /* ultra high speed CD-RW */
2097 static char us_clv_to_speed[16] = {
2098 	/* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2099 	   0, 2, 4, 8, 0, 0,16, 0,24,32,40,48, 0, 0, 0, 0
2100 };
2101 
2102 /*
2103  * reads the maximum media speed from ATIP
2104  */
2105 static noinline_for_stack int pkt_media_speed(struct pktcdvd_device *pd,
2106 						unsigned *speed)
2107 {
2108 	struct packet_command cgc;
2109 	struct request_sense sense;
2110 	unsigned char buf[64];
2111 	unsigned int size, st, sp;
2112 	int ret;
2113 
2114 	init_cdrom_command(&cgc, buf, 2, CGC_DATA_READ);
2115 	cgc.sense = &sense;
2116 	cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2117 	cgc.cmd[1] = 2;
2118 	cgc.cmd[2] = 4; /* READ ATIP */
2119 	cgc.cmd[8] = 2;
2120 	ret = pkt_generic_packet(pd, &cgc);
2121 	if (ret) {
2122 		pkt_dump_sense(&cgc);
2123 		return ret;
2124 	}
2125 	size = ((unsigned int) buf[0]<<8) + buf[1] + 2;
2126 	if (size > sizeof(buf))
2127 		size = sizeof(buf);
2128 
2129 	init_cdrom_command(&cgc, buf, size, CGC_DATA_READ);
2130 	cgc.sense = &sense;
2131 	cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2132 	cgc.cmd[1] = 2;
2133 	cgc.cmd[2] = 4;
2134 	cgc.cmd[8] = size;
2135 	ret = pkt_generic_packet(pd, &cgc);
2136 	if (ret) {
2137 		pkt_dump_sense(&cgc);
2138 		return ret;
2139 	}
2140 
2141 	if (!(buf[6] & 0x40)) {
2142 		printk(DRIVER_NAME": Disc type is not CD-RW\n");
2143 		return 1;
2144 	}
2145 	if (!(buf[6] & 0x4)) {
2146 		printk(DRIVER_NAME": A1 values on media are not valid, maybe not CDRW?\n");
2147 		return 1;
2148 	}
2149 
2150 	st = (buf[6] >> 3) & 0x7; /* disc sub-type */
2151 
2152 	sp = buf[16] & 0xf; /* max speed from ATIP A1 field */
2153 
2154 	/* Info from cdrecord */
2155 	switch (st) {
2156 		case 0: /* standard speed */
2157 			*speed = clv_to_speed[sp];
2158 			break;
2159 		case 1: /* high speed */
2160 			*speed = hs_clv_to_speed[sp];
2161 			break;
2162 		case 2: /* ultra high speed */
2163 			*speed = us_clv_to_speed[sp];
2164 			break;
2165 		default:
2166 			printk(DRIVER_NAME": Unknown disc sub-type %d\n",st);
2167 			return 1;
2168 	}
2169 	if (*speed) {
2170 		printk(DRIVER_NAME": Max. media speed: %d\n",*speed);
2171 		return 0;
2172 	} else {
2173 		printk(DRIVER_NAME": Unknown speed %d for sub-type %d\n",sp,st);
2174 		return 1;
2175 	}
2176 }
2177 
2178 static noinline_for_stack int pkt_perform_opc(struct pktcdvd_device *pd)
2179 {
2180 	struct packet_command cgc;
2181 	struct request_sense sense;
2182 	int ret;
2183 
2184 	VPRINTK(DRIVER_NAME": Performing OPC\n");
2185 
2186 	init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2187 	cgc.sense = &sense;
2188 	cgc.timeout = 60*HZ;
2189 	cgc.cmd[0] = GPCMD_SEND_OPC;
2190 	cgc.cmd[1] = 1;
2191 	if ((ret = pkt_generic_packet(pd, &cgc)))
2192 		pkt_dump_sense(&cgc);
2193 	return ret;
2194 }
2195 
2196 static int pkt_open_write(struct pktcdvd_device *pd)
2197 {
2198 	int ret;
2199 	unsigned int write_speed, media_write_speed, read_speed;
2200 
2201 	if ((ret = pkt_probe_settings(pd))) {
2202 		VPRINTK(DRIVER_NAME": %s failed probe\n", pd->name);
2203 		return ret;
2204 	}
2205 
2206 	if ((ret = pkt_set_write_settings(pd))) {
2207 		DPRINTK(DRIVER_NAME": %s failed saving write settings\n", pd->name);
2208 		return -EIO;
2209 	}
2210 
2211 	pkt_write_caching(pd, USE_WCACHING);
2212 
2213 	if ((ret = pkt_get_max_speed(pd, &write_speed)))
2214 		write_speed = 16 * 177;
2215 	switch (pd->mmc3_profile) {
2216 		case 0x13: /* DVD-RW */
2217 		case 0x1a: /* DVD+RW */
2218 		case 0x12: /* DVD-RAM */
2219 			DPRINTK(DRIVER_NAME": write speed %ukB/s\n", write_speed);
2220 			break;
2221 		default:
2222 			if ((ret = pkt_media_speed(pd, &media_write_speed)))
2223 				media_write_speed = 16;
2224 			write_speed = min(write_speed, media_write_speed * 177);
2225 			DPRINTK(DRIVER_NAME": write speed %ux\n", write_speed / 176);
2226 			break;
2227 	}
2228 	read_speed = write_speed;
2229 
2230 	if ((ret = pkt_set_speed(pd, write_speed, read_speed))) {
2231 		DPRINTK(DRIVER_NAME": %s couldn't set write speed\n", pd->name);
2232 		return -EIO;
2233 	}
2234 	pd->write_speed = write_speed;
2235 	pd->read_speed = read_speed;
2236 
2237 	if ((ret = pkt_perform_opc(pd))) {
2238 		DPRINTK(DRIVER_NAME": %s Optimum Power Calibration failed\n", pd->name);
2239 	}
2240 
2241 	return 0;
2242 }
2243 
2244 /*
2245  * called at open time.
2246  */
2247 static int pkt_open_dev(struct pktcdvd_device *pd, fmode_t write)
2248 {
2249 	int ret;
2250 	long lba;
2251 	struct request_queue *q;
2252 
2253 	/*
2254 	 * We need to re-open the cdrom device without O_NONBLOCK to be able
2255 	 * to read/write from/to it. It is already opened in O_NONBLOCK mode
2256 	 * so bdget() can't fail.
2257 	 */
2258 	bdget(pd->bdev->bd_dev);
2259 	if ((ret = blkdev_get(pd->bdev, FMODE_READ | FMODE_EXCL, pd)))
2260 		goto out;
2261 
2262 	if ((ret = pkt_get_last_written(pd, &lba))) {
2263 		printk(DRIVER_NAME": pkt_get_last_written failed\n");
2264 		goto out_putdev;
2265 	}
2266 
2267 	set_capacity(pd->disk, lba << 2);
2268 	set_capacity(pd->bdev->bd_disk, lba << 2);
2269 	bd_set_size(pd->bdev, (loff_t)lba << 11);
2270 
2271 	q = bdev_get_queue(pd->bdev);
2272 	if (write) {
2273 		if ((ret = pkt_open_write(pd)))
2274 			goto out_putdev;
2275 		/*
2276 		 * Some CDRW drives can not handle writes larger than one packet,
2277 		 * even if the size is a multiple of the packet size.
2278 		 */
2279 		spin_lock_irq(q->queue_lock);
2280 		blk_queue_max_hw_sectors(q, pd->settings.size);
2281 		spin_unlock_irq(q->queue_lock);
2282 		set_bit(PACKET_WRITABLE, &pd->flags);
2283 	} else {
2284 		pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2285 		clear_bit(PACKET_WRITABLE, &pd->flags);
2286 	}
2287 
2288 	if ((ret = pkt_set_segment_merging(pd, q)))
2289 		goto out_putdev;
2290 
2291 	if (write) {
2292 		if (!pkt_grow_pktlist(pd, CONFIG_CDROM_PKTCDVD_BUFFERS)) {
2293 			printk(DRIVER_NAME": not enough memory for buffers\n");
2294 			ret = -ENOMEM;
2295 			goto out_putdev;
2296 		}
2297 		printk(DRIVER_NAME": %lukB available on disc\n", lba << 1);
2298 	}
2299 
2300 	return 0;
2301 
2302 out_putdev:
2303 	blkdev_put(pd->bdev, FMODE_READ | FMODE_EXCL);
2304 out:
2305 	return ret;
2306 }
2307 
2308 /*
2309  * called when the device is closed. makes sure that the device flushes
2310  * the internal cache before we close.
2311  */
2312 static void pkt_release_dev(struct pktcdvd_device *pd, int flush)
2313 {
2314 	if (flush && pkt_flush_cache(pd))
2315 		DPRINTK(DRIVER_NAME": %s not flushing cache\n", pd->name);
2316 
2317 	pkt_lock_door(pd, 0);
2318 
2319 	pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2320 	blkdev_put(pd->bdev, FMODE_READ | FMODE_EXCL);
2321 
2322 	pkt_shrink_pktlist(pd);
2323 }
2324 
2325 static struct pktcdvd_device *pkt_find_dev_from_minor(unsigned int dev_minor)
2326 {
2327 	if (dev_minor >= MAX_WRITERS)
2328 		return NULL;
2329 	return pkt_devs[dev_minor];
2330 }
2331 
2332 static int pkt_open(struct block_device *bdev, fmode_t mode)
2333 {
2334 	struct pktcdvd_device *pd = NULL;
2335 	int ret;
2336 
2337 	VPRINTK(DRIVER_NAME": entering open\n");
2338 
2339 	mutex_lock(&pktcdvd_mutex);
2340 	mutex_lock(&ctl_mutex);
2341 	pd = pkt_find_dev_from_minor(MINOR(bdev->bd_dev));
2342 	if (!pd) {
2343 		ret = -ENODEV;
2344 		goto out;
2345 	}
2346 	BUG_ON(pd->refcnt < 0);
2347 
2348 	pd->refcnt++;
2349 	if (pd->refcnt > 1) {
2350 		if ((mode & FMODE_WRITE) &&
2351 		    !test_bit(PACKET_WRITABLE, &pd->flags)) {
2352 			ret = -EBUSY;
2353 			goto out_dec;
2354 		}
2355 	} else {
2356 		ret = pkt_open_dev(pd, mode & FMODE_WRITE);
2357 		if (ret)
2358 			goto out_dec;
2359 		/*
2360 		 * needed here as well, since ext2 (among others) may change
2361 		 * the blocksize at mount time
2362 		 */
2363 		set_blocksize(bdev, CD_FRAMESIZE);
2364 	}
2365 
2366 	mutex_unlock(&ctl_mutex);
2367 	mutex_unlock(&pktcdvd_mutex);
2368 	return 0;
2369 
2370 out_dec:
2371 	pd->refcnt--;
2372 out:
2373 	VPRINTK(DRIVER_NAME": failed open (%d)\n", ret);
2374 	mutex_unlock(&ctl_mutex);
2375 	mutex_unlock(&pktcdvd_mutex);
2376 	return ret;
2377 }
2378 
2379 static int pkt_close(struct gendisk *disk, fmode_t mode)
2380 {
2381 	struct pktcdvd_device *pd = disk->private_data;
2382 	int ret = 0;
2383 
2384 	mutex_lock(&pktcdvd_mutex);
2385 	mutex_lock(&ctl_mutex);
2386 	pd->refcnt--;
2387 	BUG_ON(pd->refcnt < 0);
2388 	if (pd->refcnt == 0) {
2389 		int flush = test_bit(PACKET_WRITABLE, &pd->flags);
2390 		pkt_release_dev(pd, flush);
2391 	}
2392 	mutex_unlock(&ctl_mutex);
2393 	mutex_unlock(&pktcdvd_mutex);
2394 	return ret;
2395 }
2396 
2397 
2398 static void pkt_end_io_read_cloned(struct bio *bio, int err)
2399 {
2400 	struct packet_stacked_data *psd = bio->bi_private;
2401 	struct pktcdvd_device *pd = psd->pd;
2402 
2403 	bio_put(bio);
2404 	bio_endio(psd->bio, err);
2405 	mempool_free(psd, psd_pool);
2406 	pkt_bio_finished(pd);
2407 }
2408 
2409 static void pkt_make_request(struct request_queue *q, struct bio *bio)
2410 {
2411 	struct pktcdvd_device *pd;
2412 	char b[BDEVNAME_SIZE];
2413 	sector_t zone;
2414 	struct packet_data *pkt;
2415 	int was_empty, blocked_bio;
2416 	struct pkt_rb_node *node;
2417 
2418 	pd = q->queuedata;
2419 	if (!pd) {
2420 		printk(DRIVER_NAME": %s incorrect request queue\n", bdevname(bio->bi_bdev, b));
2421 		goto end_io;
2422 	}
2423 
2424 	/*
2425 	 * Clone READ bios so we can have our own bi_end_io callback.
2426 	 */
2427 	if (bio_data_dir(bio) == READ) {
2428 		struct bio *cloned_bio = bio_clone(bio, GFP_NOIO);
2429 		struct packet_stacked_data *psd = mempool_alloc(psd_pool, GFP_NOIO);
2430 
2431 		psd->pd = pd;
2432 		psd->bio = bio;
2433 		cloned_bio->bi_bdev = pd->bdev;
2434 		cloned_bio->bi_private = psd;
2435 		cloned_bio->bi_end_io = pkt_end_io_read_cloned;
2436 		pd->stats.secs_r += bio->bi_size >> 9;
2437 		pkt_queue_bio(pd, cloned_bio);
2438 		return;
2439 	}
2440 
2441 	if (!test_bit(PACKET_WRITABLE, &pd->flags)) {
2442 		printk(DRIVER_NAME": WRITE for ro device %s (%llu)\n",
2443 			pd->name, (unsigned long long)bio->bi_sector);
2444 		goto end_io;
2445 	}
2446 
2447 	if (!bio->bi_size || (bio->bi_size % CD_FRAMESIZE)) {
2448 		printk(DRIVER_NAME": wrong bio size\n");
2449 		goto end_io;
2450 	}
2451 
2452 	blk_queue_bounce(q, &bio);
2453 
2454 	zone = ZONE(bio->bi_sector, pd);
2455 	VPRINTK("pkt_make_request: start = %6llx stop = %6llx\n",
2456 		(unsigned long long)bio->bi_sector,
2457 		(unsigned long long)(bio->bi_sector + bio_sectors(bio)));
2458 
2459 	/* Check if we have to split the bio */
2460 	{
2461 		struct bio_pair *bp;
2462 		sector_t last_zone;
2463 		int first_sectors;
2464 
2465 		last_zone = ZONE(bio->bi_sector + bio_sectors(bio) - 1, pd);
2466 		if (last_zone != zone) {
2467 			BUG_ON(last_zone != zone + pd->settings.size);
2468 			first_sectors = last_zone - bio->bi_sector;
2469 			bp = bio_split(bio, first_sectors);
2470 			BUG_ON(!bp);
2471 			pkt_make_request(q, &bp->bio1);
2472 			pkt_make_request(q, &bp->bio2);
2473 			bio_pair_release(bp);
2474 			return;
2475 		}
2476 	}
2477 
2478 	/*
2479 	 * If we find a matching packet in state WAITING or READ_WAIT, we can
2480 	 * just append this bio to that packet.
2481 	 */
2482 	spin_lock(&pd->cdrw.active_list_lock);
2483 	blocked_bio = 0;
2484 	list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
2485 		if (pkt->sector == zone) {
2486 			spin_lock(&pkt->lock);
2487 			if ((pkt->state == PACKET_WAITING_STATE) ||
2488 			    (pkt->state == PACKET_READ_WAIT_STATE)) {
2489 				bio_list_add(&pkt->orig_bios, bio);
2490 				pkt->write_size += bio->bi_size / CD_FRAMESIZE;
2491 				if ((pkt->write_size >= pkt->frames) &&
2492 				    (pkt->state == PACKET_WAITING_STATE)) {
2493 					atomic_inc(&pkt->run_sm);
2494 					wake_up(&pd->wqueue);
2495 				}
2496 				spin_unlock(&pkt->lock);
2497 				spin_unlock(&pd->cdrw.active_list_lock);
2498 				return;
2499 			} else {
2500 				blocked_bio = 1;
2501 			}
2502 			spin_unlock(&pkt->lock);
2503 		}
2504 	}
2505 	spin_unlock(&pd->cdrw.active_list_lock);
2506 
2507  	/*
2508 	 * Test if there is enough room left in the bio work queue
2509 	 * (queue size >= congestion on mark).
2510 	 * If not, wait till the work queue size is below the congestion off mark.
2511 	 */
2512 	spin_lock(&pd->lock);
2513 	if (pd->write_congestion_on > 0
2514 	    && pd->bio_queue_size >= pd->write_congestion_on) {
2515 		set_bdi_congested(&q->backing_dev_info, BLK_RW_ASYNC);
2516 		do {
2517 			spin_unlock(&pd->lock);
2518 			congestion_wait(BLK_RW_ASYNC, HZ);
2519 			spin_lock(&pd->lock);
2520 		} while(pd->bio_queue_size > pd->write_congestion_off);
2521 	}
2522 	spin_unlock(&pd->lock);
2523 
2524 	/*
2525 	 * No matching packet found. Store the bio in the work queue.
2526 	 */
2527 	node = mempool_alloc(pd->rb_pool, GFP_NOIO);
2528 	node->bio = bio;
2529 	spin_lock(&pd->lock);
2530 	BUG_ON(pd->bio_queue_size < 0);
2531 	was_empty = (pd->bio_queue_size == 0);
2532 	pkt_rbtree_insert(pd, node);
2533 	spin_unlock(&pd->lock);
2534 
2535 	/*
2536 	 * Wake up the worker thread.
2537 	 */
2538 	atomic_set(&pd->scan_queue, 1);
2539 	if (was_empty) {
2540 		/* This wake_up is required for correct operation */
2541 		wake_up(&pd->wqueue);
2542 	} else if (!list_empty(&pd->cdrw.pkt_free_list) && !blocked_bio) {
2543 		/*
2544 		 * This wake up is not required for correct operation,
2545 		 * but improves performance in some cases.
2546 		 */
2547 		wake_up(&pd->wqueue);
2548 	}
2549 	return;
2550 end_io:
2551 	bio_io_error(bio);
2552 }
2553 
2554 
2555 
2556 static int pkt_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2557 			  struct bio_vec *bvec)
2558 {
2559 	struct pktcdvd_device *pd = q->queuedata;
2560 	sector_t zone = ZONE(bmd->bi_sector, pd);
2561 	int used = ((bmd->bi_sector - zone) << 9) + bmd->bi_size;
2562 	int remaining = (pd->settings.size << 9) - used;
2563 	int remaining2;
2564 
2565 	/*
2566 	 * A bio <= PAGE_SIZE must be allowed. If it crosses a packet
2567 	 * boundary, pkt_make_request() will split the bio.
2568 	 */
2569 	remaining2 = PAGE_SIZE - bmd->bi_size;
2570 	remaining = max(remaining, remaining2);
2571 
2572 	BUG_ON(remaining < 0);
2573 	return remaining;
2574 }
2575 
2576 static void pkt_init_queue(struct pktcdvd_device *pd)
2577 {
2578 	struct request_queue *q = pd->disk->queue;
2579 
2580 	blk_queue_make_request(q, pkt_make_request);
2581 	blk_queue_logical_block_size(q, CD_FRAMESIZE);
2582 	blk_queue_max_hw_sectors(q, PACKET_MAX_SECTORS);
2583 	blk_queue_merge_bvec(q, pkt_merge_bvec);
2584 	q->queuedata = pd;
2585 }
2586 
2587 static int pkt_seq_show(struct seq_file *m, void *p)
2588 {
2589 	struct pktcdvd_device *pd = m->private;
2590 	char *msg;
2591 	char bdev_buf[BDEVNAME_SIZE];
2592 	int states[PACKET_NUM_STATES];
2593 
2594 	seq_printf(m, "Writer %s mapped to %s:\n", pd->name,
2595 		   bdevname(pd->bdev, bdev_buf));
2596 
2597 	seq_printf(m, "\nSettings:\n");
2598 	seq_printf(m, "\tpacket size:\t\t%dkB\n", pd->settings.size / 2);
2599 
2600 	if (pd->settings.write_type == 0)
2601 		msg = "Packet";
2602 	else
2603 		msg = "Unknown";
2604 	seq_printf(m, "\twrite type:\t\t%s\n", msg);
2605 
2606 	seq_printf(m, "\tpacket type:\t\t%s\n", pd->settings.fp ? "Fixed" : "Variable");
2607 	seq_printf(m, "\tlink loss:\t\t%d\n", pd->settings.link_loss);
2608 
2609 	seq_printf(m, "\ttrack mode:\t\t%d\n", pd->settings.track_mode);
2610 
2611 	if (pd->settings.block_mode == PACKET_BLOCK_MODE1)
2612 		msg = "Mode 1";
2613 	else if (pd->settings.block_mode == PACKET_BLOCK_MODE2)
2614 		msg = "Mode 2";
2615 	else
2616 		msg = "Unknown";
2617 	seq_printf(m, "\tblock mode:\t\t%s\n", msg);
2618 
2619 	seq_printf(m, "\nStatistics:\n");
2620 	seq_printf(m, "\tpackets started:\t%lu\n", pd->stats.pkt_started);
2621 	seq_printf(m, "\tpackets ended:\t\t%lu\n", pd->stats.pkt_ended);
2622 	seq_printf(m, "\twritten:\t\t%lukB\n", pd->stats.secs_w >> 1);
2623 	seq_printf(m, "\tread gather:\t\t%lukB\n", pd->stats.secs_rg >> 1);
2624 	seq_printf(m, "\tread:\t\t\t%lukB\n", pd->stats.secs_r >> 1);
2625 
2626 	seq_printf(m, "\nMisc:\n");
2627 	seq_printf(m, "\treference count:\t%d\n", pd->refcnt);
2628 	seq_printf(m, "\tflags:\t\t\t0x%lx\n", pd->flags);
2629 	seq_printf(m, "\tread speed:\t\t%ukB/s\n", pd->read_speed);
2630 	seq_printf(m, "\twrite speed:\t\t%ukB/s\n", pd->write_speed);
2631 	seq_printf(m, "\tstart offset:\t\t%lu\n", pd->offset);
2632 	seq_printf(m, "\tmode page offset:\t%u\n", pd->mode_offset);
2633 
2634 	seq_printf(m, "\nQueue state:\n");
2635 	seq_printf(m, "\tbios queued:\t\t%d\n", pd->bio_queue_size);
2636 	seq_printf(m, "\tbios pending:\t\t%d\n", atomic_read(&pd->cdrw.pending_bios));
2637 	seq_printf(m, "\tcurrent sector:\t\t0x%llx\n", (unsigned long long)pd->current_sector);
2638 
2639 	pkt_count_states(pd, states);
2640 	seq_printf(m, "\tstate:\t\t\ti:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
2641 		   states[0], states[1], states[2], states[3], states[4], states[5]);
2642 
2643 	seq_printf(m, "\twrite congestion marks:\toff=%d on=%d\n",
2644 			pd->write_congestion_off,
2645 			pd->write_congestion_on);
2646 	return 0;
2647 }
2648 
2649 static int pkt_seq_open(struct inode *inode, struct file *file)
2650 {
2651 	return single_open(file, pkt_seq_show, PDE(inode)->data);
2652 }
2653 
2654 static const struct file_operations pkt_proc_fops = {
2655 	.open	= pkt_seq_open,
2656 	.read	= seq_read,
2657 	.llseek	= seq_lseek,
2658 	.release = single_release
2659 };
2660 
2661 static int pkt_new_dev(struct pktcdvd_device *pd, dev_t dev)
2662 {
2663 	int i;
2664 	int ret = 0;
2665 	char b[BDEVNAME_SIZE];
2666 	struct block_device *bdev;
2667 
2668 	if (pd->pkt_dev == dev) {
2669 		printk(DRIVER_NAME": Recursive setup not allowed\n");
2670 		return -EBUSY;
2671 	}
2672 	for (i = 0; i < MAX_WRITERS; i++) {
2673 		struct pktcdvd_device *pd2 = pkt_devs[i];
2674 		if (!pd2)
2675 			continue;
2676 		if (pd2->bdev->bd_dev == dev) {
2677 			printk(DRIVER_NAME": %s already setup\n", bdevname(pd2->bdev, b));
2678 			return -EBUSY;
2679 		}
2680 		if (pd2->pkt_dev == dev) {
2681 			printk(DRIVER_NAME": Can't chain pktcdvd devices\n");
2682 			return -EBUSY;
2683 		}
2684 	}
2685 
2686 	bdev = bdget(dev);
2687 	if (!bdev)
2688 		return -ENOMEM;
2689 	ret = blkdev_get(bdev, FMODE_READ | FMODE_NDELAY, NULL);
2690 	if (ret)
2691 		return ret;
2692 
2693 	/* This is safe, since we have a reference from open(). */
2694 	__module_get(THIS_MODULE);
2695 
2696 	pd->bdev = bdev;
2697 	set_blocksize(bdev, CD_FRAMESIZE);
2698 
2699 	pkt_init_queue(pd);
2700 
2701 	atomic_set(&pd->cdrw.pending_bios, 0);
2702 	pd->cdrw.thread = kthread_run(kcdrwd, pd, "%s", pd->name);
2703 	if (IS_ERR(pd->cdrw.thread)) {
2704 		printk(DRIVER_NAME": can't start kernel thread\n");
2705 		ret = -ENOMEM;
2706 		goto out_mem;
2707 	}
2708 
2709 	proc_create_data(pd->name, 0, pkt_proc, &pkt_proc_fops, pd);
2710 	DPRINTK(DRIVER_NAME": writer %s mapped to %s\n", pd->name, bdevname(bdev, b));
2711 	return 0;
2712 
2713 out_mem:
2714 	blkdev_put(bdev, FMODE_READ | FMODE_NDELAY);
2715 	/* This is safe: open() is still holding a reference. */
2716 	module_put(THIS_MODULE);
2717 	return ret;
2718 }
2719 
2720 static int pkt_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, unsigned long arg)
2721 {
2722 	struct pktcdvd_device *pd = bdev->bd_disk->private_data;
2723 	int ret;
2724 
2725 	VPRINTK("pkt_ioctl: cmd %x, dev %d:%d\n", cmd,
2726 		MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev));
2727 
2728 	mutex_lock(&pktcdvd_mutex);
2729 	switch (cmd) {
2730 	case CDROMEJECT:
2731 		/*
2732 		 * The door gets locked when the device is opened, so we
2733 		 * have to unlock it or else the eject command fails.
2734 		 */
2735 		if (pd->refcnt == 1)
2736 			pkt_lock_door(pd, 0);
2737 		/* fallthru */
2738 	/*
2739 	 * forward selected CDROM ioctls to CD-ROM, for UDF
2740 	 */
2741 	case CDROMMULTISESSION:
2742 	case CDROMREADTOCENTRY:
2743 	case CDROM_LAST_WRITTEN:
2744 	case CDROM_SEND_PACKET:
2745 	case SCSI_IOCTL_SEND_COMMAND:
2746 		ret = __blkdev_driver_ioctl(pd->bdev, mode, cmd, arg);
2747 		break;
2748 
2749 	default:
2750 		VPRINTK(DRIVER_NAME": Unknown ioctl for %s (%x)\n", pd->name, cmd);
2751 		ret = -ENOTTY;
2752 	}
2753 	mutex_unlock(&pktcdvd_mutex);
2754 
2755 	return ret;
2756 }
2757 
2758 static unsigned int pkt_check_events(struct gendisk *disk,
2759 				     unsigned int clearing)
2760 {
2761 	struct pktcdvd_device *pd = disk->private_data;
2762 	struct gendisk *attached_disk;
2763 
2764 	if (!pd)
2765 		return 0;
2766 	if (!pd->bdev)
2767 		return 0;
2768 	attached_disk = pd->bdev->bd_disk;
2769 	if (!attached_disk || !attached_disk->fops->check_events)
2770 		return 0;
2771 	return attached_disk->fops->check_events(attached_disk, clearing);
2772 }
2773 
2774 static const struct block_device_operations pktcdvd_ops = {
2775 	.owner =		THIS_MODULE,
2776 	.open =			pkt_open,
2777 	.release =		pkt_close,
2778 	.ioctl =		pkt_ioctl,
2779 	.check_events =		pkt_check_events,
2780 };
2781 
2782 static char *pktcdvd_devnode(struct gendisk *gd, umode_t *mode)
2783 {
2784 	return kasprintf(GFP_KERNEL, "pktcdvd/%s", gd->disk_name);
2785 }
2786 
2787 /*
2788  * Set up mapping from pktcdvd device to CD-ROM device.
2789  */
2790 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev)
2791 {
2792 	int idx;
2793 	int ret = -ENOMEM;
2794 	struct pktcdvd_device *pd;
2795 	struct gendisk *disk;
2796 
2797 	mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2798 
2799 	for (idx = 0; idx < MAX_WRITERS; idx++)
2800 		if (!pkt_devs[idx])
2801 			break;
2802 	if (idx == MAX_WRITERS) {
2803 		printk(DRIVER_NAME": max %d writers supported\n", MAX_WRITERS);
2804 		ret = -EBUSY;
2805 		goto out_mutex;
2806 	}
2807 
2808 	pd = kzalloc(sizeof(struct pktcdvd_device), GFP_KERNEL);
2809 	if (!pd)
2810 		goto out_mutex;
2811 
2812 	pd->rb_pool = mempool_create_kmalloc_pool(PKT_RB_POOL_SIZE,
2813 						  sizeof(struct pkt_rb_node));
2814 	if (!pd->rb_pool)
2815 		goto out_mem;
2816 
2817 	INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
2818 	INIT_LIST_HEAD(&pd->cdrw.pkt_active_list);
2819 	spin_lock_init(&pd->cdrw.active_list_lock);
2820 
2821 	spin_lock_init(&pd->lock);
2822 	spin_lock_init(&pd->iosched.lock);
2823 	bio_list_init(&pd->iosched.read_queue);
2824 	bio_list_init(&pd->iosched.write_queue);
2825 	sprintf(pd->name, DRIVER_NAME"%d", idx);
2826 	init_waitqueue_head(&pd->wqueue);
2827 	pd->bio_queue = RB_ROOT;
2828 
2829 	pd->write_congestion_on  = write_congestion_on;
2830 	pd->write_congestion_off = write_congestion_off;
2831 
2832 	disk = alloc_disk(1);
2833 	if (!disk)
2834 		goto out_mem;
2835 	pd->disk = disk;
2836 	disk->major = pktdev_major;
2837 	disk->first_minor = idx;
2838 	disk->fops = &pktcdvd_ops;
2839 	disk->flags = GENHD_FL_REMOVABLE;
2840 	strcpy(disk->disk_name, pd->name);
2841 	disk->devnode = pktcdvd_devnode;
2842 	disk->private_data = pd;
2843 	disk->queue = blk_alloc_queue(GFP_KERNEL);
2844 	if (!disk->queue)
2845 		goto out_mem2;
2846 
2847 	pd->pkt_dev = MKDEV(pktdev_major, idx);
2848 	ret = pkt_new_dev(pd, dev);
2849 	if (ret)
2850 		goto out_new_dev;
2851 
2852 	/* inherit events of the host device */
2853 	disk->events = pd->bdev->bd_disk->events;
2854 	disk->async_events = pd->bdev->bd_disk->async_events;
2855 
2856 	add_disk(disk);
2857 
2858 	pkt_sysfs_dev_new(pd);
2859 	pkt_debugfs_dev_new(pd);
2860 
2861 	pkt_devs[idx] = pd;
2862 	if (pkt_dev)
2863 		*pkt_dev = pd->pkt_dev;
2864 
2865 	mutex_unlock(&ctl_mutex);
2866 	return 0;
2867 
2868 out_new_dev:
2869 	blk_cleanup_queue(disk->queue);
2870 out_mem2:
2871 	put_disk(disk);
2872 out_mem:
2873 	if (pd->rb_pool)
2874 		mempool_destroy(pd->rb_pool);
2875 	kfree(pd);
2876 out_mutex:
2877 	mutex_unlock(&ctl_mutex);
2878 	printk(DRIVER_NAME": setup of pktcdvd device failed\n");
2879 	return ret;
2880 }
2881 
2882 /*
2883  * Tear down mapping from pktcdvd device to CD-ROM device.
2884  */
2885 static int pkt_remove_dev(dev_t pkt_dev)
2886 {
2887 	struct pktcdvd_device *pd;
2888 	int idx;
2889 	int ret = 0;
2890 
2891 	mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2892 
2893 	for (idx = 0; idx < MAX_WRITERS; idx++) {
2894 		pd = pkt_devs[idx];
2895 		if (pd && (pd->pkt_dev == pkt_dev))
2896 			break;
2897 	}
2898 	if (idx == MAX_WRITERS) {
2899 		DPRINTK(DRIVER_NAME": dev not setup\n");
2900 		ret = -ENXIO;
2901 		goto out;
2902 	}
2903 
2904 	if (pd->refcnt > 0) {
2905 		ret = -EBUSY;
2906 		goto out;
2907 	}
2908 	if (!IS_ERR(pd->cdrw.thread))
2909 		kthread_stop(pd->cdrw.thread);
2910 
2911 	pkt_devs[idx] = NULL;
2912 
2913 	pkt_debugfs_dev_remove(pd);
2914 	pkt_sysfs_dev_remove(pd);
2915 
2916 	blkdev_put(pd->bdev, FMODE_READ | FMODE_NDELAY);
2917 
2918 	remove_proc_entry(pd->name, pkt_proc);
2919 	DPRINTK(DRIVER_NAME": writer %s unmapped\n", pd->name);
2920 
2921 	del_gendisk(pd->disk);
2922 	blk_cleanup_queue(pd->disk->queue);
2923 	put_disk(pd->disk);
2924 
2925 	mempool_destroy(pd->rb_pool);
2926 	kfree(pd);
2927 
2928 	/* This is safe: open() is still holding a reference. */
2929 	module_put(THIS_MODULE);
2930 
2931 out:
2932 	mutex_unlock(&ctl_mutex);
2933 	return ret;
2934 }
2935 
2936 static void pkt_get_status(struct pkt_ctrl_command *ctrl_cmd)
2937 {
2938 	struct pktcdvd_device *pd;
2939 
2940 	mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2941 
2942 	pd = pkt_find_dev_from_minor(ctrl_cmd->dev_index);
2943 	if (pd) {
2944 		ctrl_cmd->dev = new_encode_dev(pd->bdev->bd_dev);
2945 		ctrl_cmd->pkt_dev = new_encode_dev(pd->pkt_dev);
2946 	} else {
2947 		ctrl_cmd->dev = 0;
2948 		ctrl_cmd->pkt_dev = 0;
2949 	}
2950 	ctrl_cmd->num_devices = MAX_WRITERS;
2951 
2952 	mutex_unlock(&ctl_mutex);
2953 }
2954 
2955 static long pkt_ctl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2956 {
2957 	void __user *argp = (void __user *)arg;
2958 	struct pkt_ctrl_command ctrl_cmd;
2959 	int ret = 0;
2960 	dev_t pkt_dev = 0;
2961 
2962 	if (cmd != PACKET_CTRL_CMD)
2963 		return -ENOTTY;
2964 
2965 	if (copy_from_user(&ctrl_cmd, argp, sizeof(struct pkt_ctrl_command)))
2966 		return -EFAULT;
2967 
2968 	switch (ctrl_cmd.command) {
2969 	case PKT_CTRL_CMD_SETUP:
2970 		if (!capable(CAP_SYS_ADMIN))
2971 			return -EPERM;
2972 		ret = pkt_setup_dev(new_decode_dev(ctrl_cmd.dev), &pkt_dev);
2973 		ctrl_cmd.pkt_dev = new_encode_dev(pkt_dev);
2974 		break;
2975 	case PKT_CTRL_CMD_TEARDOWN:
2976 		if (!capable(CAP_SYS_ADMIN))
2977 			return -EPERM;
2978 		ret = pkt_remove_dev(new_decode_dev(ctrl_cmd.pkt_dev));
2979 		break;
2980 	case PKT_CTRL_CMD_STATUS:
2981 		pkt_get_status(&ctrl_cmd);
2982 		break;
2983 	default:
2984 		return -ENOTTY;
2985 	}
2986 
2987 	if (copy_to_user(argp, &ctrl_cmd, sizeof(struct pkt_ctrl_command)))
2988 		return -EFAULT;
2989 	return ret;
2990 }
2991 
2992 #ifdef CONFIG_COMPAT
2993 static long pkt_ctl_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2994 {
2995 	return pkt_ctl_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
2996 }
2997 #endif
2998 
2999 static const struct file_operations pkt_ctl_fops = {
3000 	.open		= nonseekable_open,
3001 	.unlocked_ioctl	= pkt_ctl_ioctl,
3002 #ifdef CONFIG_COMPAT
3003 	.compat_ioctl	= pkt_ctl_compat_ioctl,
3004 #endif
3005 	.owner		= THIS_MODULE,
3006 	.llseek		= no_llseek,
3007 };
3008 
3009 static struct miscdevice pkt_misc = {
3010 	.minor 		= MISC_DYNAMIC_MINOR,
3011 	.name  		= DRIVER_NAME,
3012 	.nodename	= "pktcdvd/control",
3013 	.fops  		= &pkt_ctl_fops
3014 };
3015 
3016 static int __init pkt_init(void)
3017 {
3018 	int ret;
3019 
3020 	mutex_init(&ctl_mutex);
3021 
3022 	psd_pool = mempool_create_kmalloc_pool(PSD_POOL_SIZE,
3023 					sizeof(struct packet_stacked_data));
3024 	if (!psd_pool)
3025 		return -ENOMEM;
3026 
3027 	ret = register_blkdev(pktdev_major, DRIVER_NAME);
3028 	if (ret < 0) {
3029 		printk(DRIVER_NAME": Unable to register block device\n");
3030 		goto out2;
3031 	}
3032 	if (!pktdev_major)
3033 		pktdev_major = ret;
3034 
3035 	ret = pkt_sysfs_init();
3036 	if (ret)
3037 		goto out;
3038 
3039 	pkt_debugfs_init();
3040 
3041 	ret = misc_register(&pkt_misc);
3042 	if (ret) {
3043 		printk(DRIVER_NAME": Unable to register misc device\n");
3044 		goto out_misc;
3045 	}
3046 
3047 	pkt_proc = proc_mkdir("driver/"DRIVER_NAME, NULL);
3048 
3049 	return 0;
3050 
3051 out_misc:
3052 	pkt_debugfs_cleanup();
3053 	pkt_sysfs_cleanup();
3054 out:
3055 	unregister_blkdev(pktdev_major, DRIVER_NAME);
3056 out2:
3057 	mempool_destroy(psd_pool);
3058 	return ret;
3059 }
3060 
3061 static void __exit pkt_exit(void)
3062 {
3063 	remove_proc_entry("driver/"DRIVER_NAME, NULL);
3064 	misc_deregister(&pkt_misc);
3065 
3066 	pkt_debugfs_cleanup();
3067 	pkt_sysfs_cleanup();
3068 
3069 	unregister_blkdev(pktdev_major, DRIVER_NAME);
3070 	mempool_destroy(psd_pool);
3071 }
3072 
3073 MODULE_DESCRIPTION("Packet writing layer for CD/DVD drives");
3074 MODULE_AUTHOR("Jens Axboe <axboe@suse.de>");
3075 MODULE_LICENSE("GPL");
3076 
3077 module_init(pkt_init);
3078 module_exit(pkt_exit);
3079