Merge master.kernel.org:/pub/scm/linux/kernel/git/wim/linux-2.6-watchdog
[pandora-kernel.git] / drivers / block / pktcdvd.c
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/kthread.h>
52 #include <linux/errno.h>
53 #include <linux/spinlock.h>
54 #include <linux/file.h>
55 #include <linux/proc_fs.h>
56 #include <linux/seq_file.h>
57 #include <linux/miscdevice.h>
58 #include <linux/freezer.h>
59 #include <linux/mutex.h>
60 #include <scsi/scsi_cmnd.h>
61 #include <scsi/scsi_ioctl.h>
62 #include <scsi/scsi.h>
63 #include <linux/debugfs.h>
64 #include <linux/device.h>
65
66 #include <asm/uaccess.h>
67
68 #define DRIVER_NAME     "pktcdvd"
69
70 #if PACKET_DEBUG
71 #define DPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
72 #else
73 #define DPRINTK(fmt, args...)
74 #endif
75
76 #if PACKET_DEBUG > 1
77 #define VPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
78 #else
79 #define VPRINTK(fmt, args...)
80 #endif
81
82 #define MAX_SPEED 0xffff
83
84 #define ZONE(sector, pd) (((sector) + (pd)->offset) & ~((pd)->settings.size - 1))
85
86 static struct pktcdvd_device *pkt_devs[MAX_WRITERS];
87 static struct proc_dir_entry *pkt_proc;
88 static int pktdev_major;
89 static int write_congestion_on  = PKT_WRITE_CONGESTION_ON;
90 static int write_congestion_off = PKT_WRITE_CONGESTION_OFF;
91 static struct mutex ctl_mutex;  /* Serialize open/close/setup/teardown */
92 static mempool_t *psd_pool;
93
94 static struct class     *class_pktcdvd = NULL;    /* /sys/class/pktcdvd */
95 static struct dentry    *pkt_debugfs_root = NULL; /* /debug/pktcdvd */
96
97 /* forward declaration */
98 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev);
99 static int pkt_remove_dev(dev_t pkt_dev);
100 static int pkt_seq_show(struct seq_file *m, void *p);
101
102
103
104 /*
105  * create and register a pktcdvd kernel object.
106  */
107 static struct pktcdvd_kobj* pkt_kobj_create(struct pktcdvd_device *pd,
108                                         const char* name,
109                                         struct kobject* parent,
110                                         struct kobj_type* ktype)
111 {
112         struct pktcdvd_kobj *p;
113         p = kzalloc(sizeof(*p), GFP_KERNEL);
114         if (!p)
115                 return NULL;
116         kobject_set_name(&p->kobj, "%s", name);
117         p->kobj.parent = parent;
118         p->kobj.ktype = ktype;
119         p->pd = pd;
120         if (kobject_register(&p->kobj) != 0)
121                 return NULL;
122         return p;
123 }
124 /*
125  * remove a pktcdvd kernel object.
126  */
127 static void pkt_kobj_remove(struct pktcdvd_kobj *p)
128 {
129         if (p)
130                 kobject_unregister(&p->kobj);
131 }
132 /*
133  * default release function for pktcdvd kernel objects.
134  */
135 static void pkt_kobj_release(struct kobject *kobj)
136 {
137         kfree(to_pktcdvdkobj(kobj));
138 }
139
140
141 /**********************************************************
142  *
143  * sysfs interface for pktcdvd
144  * by (C) 2006  Thomas Maier <balagi@justmail.de>
145  *
146  **********************************************************/
147
148 #define DEF_ATTR(_obj,_name,_mode) \
149         static struct attribute _obj = { \
150                 .name = _name, .owner = THIS_MODULE, .mode = _mode }
151
152 /**********************************************************
153   /sys/class/pktcdvd/pktcdvd[0-7]/
154                      stat/reset
155                      stat/packets_started
156                      stat/packets_finished
157                      stat/kb_written
158                      stat/kb_read
159                      stat/kb_read_gather
160                      write_queue/size
161                      write_queue/congestion_off
162                      write_queue/congestion_on
163  **********************************************************/
164
165 DEF_ATTR(kobj_pkt_attr_st1, "reset", 0200);
166 DEF_ATTR(kobj_pkt_attr_st2, "packets_started", 0444);
167 DEF_ATTR(kobj_pkt_attr_st3, "packets_finished", 0444);
168 DEF_ATTR(kobj_pkt_attr_st4, "kb_written", 0444);
169 DEF_ATTR(kobj_pkt_attr_st5, "kb_read", 0444);
170 DEF_ATTR(kobj_pkt_attr_st6, "kb_read_gather", 0444);
171
172 static struct attribute *kobj_pkt_attrs_stat[] = {
173         &kobj_pkt_attr_st1,
174         &kobj_pkt_attr_st2,
175         &kobj_pkt_attr_st3,
176         &kobj_pkt_attr_st4,
177         &kobj_pkt_attr_st5,
178         &kobj_pkt_attr_st6,
179         NULL
180 };
181
182 DEF_ATTR(kobj_pkt_attr_wq1, "size", 0444);
183 DEF_ATTR(kobj_pkt_attr_wq2, "congestion_off", 0644);
184 DEF_ATTR(kobj_pkt_attr_wq3, "congestion_on",  0644);
185
186 static struct attribute *kobj_pkt_attrs_wqueue[] = {
187         &kobj_pkt_attr_wq1,
188         &kobj_pkt_attr_wq2,
189         &kobj_pkt_attr_wq3,
190         NULL
191 };
192
193 static ssize_t kobj_pkt_show(struct kobject *kobj,
194                         struct attribute *attr, char *data)
195 {
196         struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
197         int n = 0;
198         int v;
199         if (strcmp(attr->name, "packets_started") == 0) {
200                 n = sprintf(data, "%lu\n", pd->stats.pkt_started);
201
202         } else if (strcmp(attr->name, "packets_finished") == 0) {
203                 n = sprintf(data, "%lu\n", pd->stats.pkt_ended);
204
205         } else if (strcmp(attr->name, "kb_written") == 0) {
206                 n = sprintf(data, "%lu\n", pd->stats.secs_w >> 1);
207
208         } else if (strcmp(attr->name, "kb_read") == 0) {
209                 n = sprintf(data, "%lu\n", pd->stats.secs_r >> 1);
210
211         } else if (strcmp(attr->name, "kb_read_gather") == 0) {
212                 n = sprintf(data, "%lu\n", pd->stats.secs_rg >> 1);
213
214         } else if (strcmp(attr->name, "size") == 0) {
215                 spin_lock(&pd->lock);
216                 v = pd->bio_queue_size;
217                 spin_unlock(&pd->lock);
218                 n = sprintf(data, "%d\n", v);
219
220         } else if (strcmp(attr->name, "congestion_off") == 0) {
221                 spin_lock(&pd->lock);
222                 v = pd->write_congestion_off;
223                 spin_unlock(&pd->lock);
224                 n = sprintf(data, "%d\n", v);
225
226         } else if (strcmp(attr->name, "congestion_on") == 0) {
227                 spin_lock(&pd->lock);
228                 v = pd->write_congestion_on;
229                 spin_unlock(&pd->lock);
230                 n = sprintf(data, "%d\n", v);
231         }
232         return n;
233 }
234
235 static void init_write_congestion_marks(int* lo, int* hi)
236 {
237         if (*hi > 0) {
238                 *hi = max(*hi, 500);
239                 *hi = min(*hi, 1000000);
240                 if (*lo <= 0)
241                         *lo = *hi - 100;
242                 else {
243                         *lo = min(*lo, *hi - 100);
244                         *lo = max(*lo, 100);
245                 }
246         } else {
247                 *hi = -1;
248                 *lo = -1;
249         }
250 }
251
252 static ssize_t kobj_pkt_store(struct kobject *kobj,
253                         struct attribute *attr,
254                         const char *data, size_t len)
255 {
256         struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
257         int val;
258
259         if (strcmp(attr->name, "reset") == 0 && len > 0) {
260                 pd->stats.pkt_started = 0;
261                 pd->stats.pkt_ended = 0;
262                 pd->stats.secs_w = 0;
263                 pd->stats.secs_rg = 0;
264                 pd->stats.secs_r = 0;
265
266         } else if (strcmp(attr->name, "congestion_off") == 0
267                    && sscanf(data, "%d", &val) == 1) {
268                 spin_lock(&pd->lock);
269                 pd->write_congestion_off = val;
270                 init_write_congestion_marks(&pd->write_congestion_off,
271                                         &pd->write_congestion_on);
272                 spin_unlock(&pd->lock);
273
274         } else if (strcmp(attr->name, "congestion_on") == 0
275                    && sscanf(data, "%d", &val) == 1) {
276                 spin_lock(&pd->lock);
277                 pd->write_congestion_on = val;
278                 init_write_congestion_marks(&pd->write_congestion_off,
279                                         &pd->write_congestion_on);
280                 spin_unlock(&pd->lock);
281         }
282         return len;
283 }
284
285 static struct sysfs_ops kobj_pkt_ops = {
286         .show = kobj_pkt_show,
287         .store = kobj_pkt_store
288 };
289 static struct kobj_type kobj_pkt_type_stat = {
290         .release = pkt_kobj_release,
291         .sysfs_ops = &kobj_pkt_ops,
292         .default_attrs = kobj_pkt_attrs_stat
293 };
294 static struct kobj_type kobj_pkt_type_wqueue = {
295         .release = pkt_kobj_release,
296         .sysfs_ops = &kobj_pkt_ops,
297         .default_attrs = kobj_pkt_attrs_wqueue
298 };
299
300 static void pkt_sysfs_dev_new(struct pktcdvd_device *pd)
301 {
302         if (class_pktcdvd) {
303                 pd->clsdev = class_device_create(class_pktcdvd,
304                                         NULL, pd->pkt_dev,
305                                         NULL, "%s", pd->name);
306                 if (IS_ERR(pd->clsdev))
307                         pd->clsdev = NULL;
308         }
309         if (pd->clsdev) {
310                 pd->kobj_stat = pkt_kobj_create(pd, "stat",
311                                         &pd->clsdev->kobj,
312                                         &kobj_pkt_type_stat);
313                 pd->kobj_wqueue = pkt_kobj_create(pd, "write_queue",
314                                         &pd->clsdev->kobj,
315                                         &kobj_pkt_type_wqueue);
316         }
317 }
318
319 static void pkt_sysfs_dev_remove(struct pktcdvd_device *pd)
320 {
321         pkt_kobj_remove(pd->kobj_stat);
322         pkt_kobj_remove(pd->kobj_wqueue);
323         if (class_pktcdvd)
324                 class_device_destroy(class_pktcdvd, pd->pkt_dev);
325 }
326
327
328 /********************************************************************
329   /sys/class/pktcdvd/
330                      add            map block device
331                      remove         unmap packet dev
332                      device_map     show mappings
333  *******************************************************************/
334
335 static void class_pktcdvd_release(struct class *cls)
336 {
337         kfree(cls);
338 }
339 static ssize_t class_pktcdvd_show_map(struct class *c, char *data)
340 {
341         int n = 0;
342         int idx;
343         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
344         for (idx = 0; idx < MAX_WRITERS; idx++) {
345                 struct pktcdvd_device *pd = pkt_devs[idx];
346                 if (!pd)
347                         continue;
348                 n += sprintf(data+n, "%s %u:%u %u:%u\n",
349                         pd->name,
350                         MAJOR(pd->pkt_dev), MINOR(pd->pkt_dev),
351                         MAJOR(pd->bdev->bd_dev),
352                         MINOR(pd->bdev->bd_dev));
353         }
354         mutex_unlock(&ctl_mutex);
355         return n;
356 }
357
358 static ssize_t class_pktcdvd_store_add(struct class *c, const char *buf,
359                                         size_t count)
360 {
361         unsigned int major, minor;
362         if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
363                 pkt_setup_dev(MKDEV(major, minor), NULL);
364                 return count;
365         }
366         return -EINVAL;
367 }
368
369 static ssize_t class_pktcdvd_store_remove(struct class *c, const char *buf,
370                                         size_t count)
371 {
372         unsigned int major, minor;
373         if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
374                 pkt_remove_dev(MKDEV(major, minor));
375                 return count;
376         }
377         return -EINVAL;
378 }
379
380 static struct class_attribute class_pktcdvd_attrs[] = {
381  __ATTR(add,            0200, NULL, class_pktcdvd_store_add),
382  __ATTR(remove,         0200, NULL, class_pktcdvd_store_remove),
383  __ATTR(device_map,     0444, class_pktcdvd_show_map, NULL),
384  __ATTR_NULL
385 };
386
387
388 static int pkt_sysfs_init(void)
389 {
390         int ret = 0;
391
392         /*
393          * create control files in sysfs
394          * /sys/class/pktcdvd/...
395          */
396         class_pktcdvd = kzalloc(sizeof(*class_pktcdvd), GFP_KERNEL);
397         if (!class_pktcdvd)
398                 return -ENOMEM;
399         class_pktcdvd->name = DRIVER_NAME;
400         class_pktcdvd->owner = THIS_MODULE;
401         class_pktcdvd->class_release = class_pktcdvd_release;
402         class_pktcdvd->class_attrs = class_pktcdvd_attrs;
403         ret = class_register(class_pktcdvd);
404         if (ret) {
405                 kfree(class_pktcdvd);
406                 class_pktcdvd = NULL;
407                 printk(DRIVER_NAME": failed to create class pktcdvd\n");
408                 return ret;
409         }
410         return 0;
411 }
412
413 static void pkt_sysfs_cleanup(void)
414 {
415         if (class_pktcdvd)
416                 class_destroy(class_pktcdvd);
417         class_pktcdvd = NULL;
418 }
419
420 /********************************************************************
421   entries in debugfs
422
423   /debugfs/pktcdvd[0-7]/
424                         info
425
426  *******************************************************************/
427
428 static int pkt_debugfs_seq_show(struct seq_file *m, void *p)
429 {
430         return pkt_seq_show(m, p);
431 }
432
433 static int pkt_debugfs_fops_open(struct inode *inode, struct file *file)
434 {
435         return single_open(file, pkt_debugfs_seq_show, inode->i_private);
436 }
437
438 static const struct file_operations debug_fops = {
439         .open           = pkt_debugfs_fops_open,
440         .read           = seq_read,
441         .llseek         = seq_lseek,
442         .release        = single_release,
443         .owner          = THIS_MODULE,
444 };
445
446 static void pkt_debugfs_dev_new(struct pktcdvd_device *pd)
447 {
448         if (!pkt_debugfs_root)
449                 return;
450         pd->dfs_f_info = NULL;
451         pd->dfs_d_root = debugfs_create_dir(pd->name, pkt_debugfs_root);
452         if (IS_ERR(pd->dfs_d_root)) {
453                 pd->dfs_d_root = NULL;
454                 return;
455         }
456         pd->dfs_f_info = debugfs_create_file("info", S_IRUGO,
457                                 pd->dfs_d_root, pd, &debug_fops);
458         if (IS_ERR(pd->dfs_f_info)) {
459                 pd->dfs_f_info = NULL;
460                 return;
461         }
462 }
463
464 static void pkt_debugfs_dev_remove(struct pktcdvd_device *pd)
465 {
466         if (!pkt_debugfs_root)
467                 return;
468         if (pd->dfs_f_info)
469                 debugfs_remove(pd->dfs_f_info);
470         pd->dfs_f_info = NULL;
471         if (pd->dfs_d_root)
472                 debugfs_remove(pd->dfs_d_root);
473         pd->dfs_d_root = NULL;
474 }
475
476 static void pkt_debugfs_init(void)
477 {
478         pkt_debugfs_root = debugfs_create_dir(DRIVER_NAME, NULL);
479         if (IS_ERR(pkt_debugfs_root)) {
480                 pkt_debugfs_root = NULL;
481                 return;
482         }
483 }
484
485 static void pkt_debugfs_cleanup(void)
486 {
487         if (!pkt_debugfs_root)
488                 return;
489         debugfs_remove(pkt_debugfs_root);
490         pkt_debugfs_root = NULL;
491 }
492
493 /* ----------------------------------------------------------*/
494
495
496 static void pkt_bio_finished(struct pktcdvd_device *pd)
497 {
498         BUG_ON(atomic_read(&pd->cdrw.pending_bios) <= 0);
499         if (atomic_dec_and_test(&pd->cdrw.pending_bios)) {
500                 VPRINTK(DRIVER_NAME": queue empty\n");
501                 atomic_set(&pd->iosched.attention, 1);
502                 wake_up(&pd->wqueue);
503         }
504 }
505
506 static void pkt_bio_destructor(struct bio *bio)
507 {
508         kfree(bio->bi_io_vec);
509         kfree(bio);
510 }
511
512 static struct bio *pkt_bio_alloc(int nr_iovecs)
513 {
514         struct bio_vec *bvl = NULL;
515         struct bio *bio;
516
517         bio = kmalloc(sizeof(struct bio), GFP_KERNEL);
518         if (!bio)
519                 goto no_bio;
520         bio_init(bio);
521
522         bvl = kcalloc(nr_iovecs, sizeof(struct bio_vec), GFP_KERNEL);
523         if (!bvl)
524                 goto no_bvl;
525
526         bio->bi_max_vecs = nr_iovecs;
527         bio->bi_io_vec = bvl;
528         bio->bi_destructor = pkt_bio_destructor;
529
530         return bio;
531
532  no_bvl:
533         kfree(bio);
534  no_bio:
535         return NULL;
536 }
537
538 /*
539  * Allocate a packet_data struct
540  */
541 static struct packet_data *pkt_alloc_packet_data(int frames)
542 {
543         int i;
544         struct packet_data *pkt;
545
546         pkt = kzalloc(sizeof(struct packet_data), GFP_KERNEL);
547         if (!pkt)
548                 goto no_pkt;
549
550         pkt->frames = frames;
551         pkt->w_bio = pkt_bio_alloc(frames);
552         if (!pkt->w_bio)
553                 goto no_bio;
554
555         for (i = 0; i < frames / FRAMES_PER_PAGE; i++) {
556                 pkt->pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
557                 if (!pkt->pages[i])
558                         goto no_page;
559         }
560
561         spin_lock_init(&pkt->lock);
562
563         for (i = 0; i < frames; i++) {
564                 struct bio *bio = pkt_bio_alloc(1);
565                 if (!bio)
566                         goto no_rd_bio;
567                 pkt->r_bios[i] = bio;
568         }
569
570         return pkt;
571
572 no_rd_bio:
573         for (i = 0; i < frames; i++) {
574                 struct bio *bio = pkt->r_bios[i];
575                 if (bio)
576                         bio_put(bio);
577         }
578
579 no_page:
580         for (i = 0; i < frames / FRAMES_PER_PAGE; i++)
581                 if (pkt->pages[i])
582                         __free_page(pkt->pages[i]);
583         bio_put(pkt->w_bio);
584 no_bio:
585         kfree(pkt);
586 no_pkt:
587         return NULL;
588 }
589
590 /*
591  * Free a packet_data struct
592  */
593 static void pkt_free_packet_data(struct packet_data *pkt)
594 {
595         int i;
596
597         for (i = 0; i < pkt->frames; i++) {
598                 struct bio *bio = pkt->r_bios[i];
599                 if (bio)
600                         bio_put(bio);
601         }
602         for (i = 0; i < pkt->frames / FRAMES_PER_PAGE; i++)
603                 __free_page(pkt->pages[i]);
604         bio_put(pkt->w_bio);
605         kfree(pkt);
606 }
607
608 static void pkt_shrink_pktlist(struct pktcdvd_device *pd)
609 {
610         struct packet_data *pkt, *next;
611
612         BUG_ON(!list_empty(&pd->cdrw.pkt_active_list));
613
614         list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_free_list, list) {
615                 pkt_free_packet_data(pkt);
616         }
617         INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
618 }
619
620 static int pkt_grow_pktlist(struct pktcdvd_device *pd, int nr_packets)
621 {
622         struct packet_data *pkt;
623
624         BUG_ON(!list_empty(&pd->cdrw.pkt_free_list));
625
626         while (nr_packets > 0) {
627                 pkt = pkt_alloc_packet_data(pd->settings.size >> 2);
628                 if (!pkt) {
629                         pkt_shrink_pktlist(pd);
630                         return 0;
631                 }
632                 pkt->id = nr_packets;
633                 pkt->pd = pd;
634                 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
635                 nr_packets--;
636         }
637         return 1;
638 }
639
640 static inline struct pkt_rb_node *pkt_rbtree_next(struct pkt_rb_node *node)
641 {
642         struct rb_node *n = rb_next(&node->rb_node);
643         if (!n)
644                 return NULL;
645         return rb_entry(n, struct pkt_rb_node, rb_node);
646 }
647
648 static void pkt_rbtree_erase(struct pktcdvd_device *pd, struct pkt_rb_node *node)
649 {
650         rb_erase(&node->rb_node, &pd->bio_queue);
651         mempool_free(node, pd->rb_pool);
652         pd->bio_queue_size--;
653         BUG_ON(pd->bio_queue_size < 0);
654 }
655
656 /*
657  * Find the first node in the pd->bio_queue rb tree with a starting sector >= s.
658  */
659 static struct pkt_rb_node *pkt_rbtree_find(struct pktcdvd_device *pd, sector_t s)
660 {
661         struct rb_node *n = pd->bio_queue.rb_node;
662         struct rb_node *next;
663         struct pkt_rb_node *tmp;
664
665         if (!n) {
666                 BUG_ON(pd->bio_queue_size > 0);
667                 return NULL;
668         }
669
670         for (;;) {
671                 tmp = rb_entry(n, struct pkt_rb_node, rb_node);
672                 if (s <= tmp->bio->bi_sector)
673                         next = n->rb_left;
674                 else
675                         next = n->rb_right;
676                 if (!next)
677                         break;
678                 n = next;
679         }
680
681         if (s > tmp->bio->bi_sector) {
682                 tmp = pkt_rbtree_next(tmp);
683                 if (!tmp)
684                         return NULL;
685         }
686         BUG_ON(s > tmp->bio->bi_sector);
687         return tmp;
688 }
689
690 /*
691  * Insert a node into the pd->bio_queue rb tree.
692  */
693 static void pkt_rbtree_insert(struct pktcdvd_device *pd, struct pkt_rb_node *node)
694 {
695         struct rb_node **p = &pd->bio_queue.rb_node;
696         struct rb_node *parent = NULL;
697         sector_t s = node->bio->bi_sector;
698         struct pkt_rb_node *tmp;
699
700         while (*p) {
701                 parent = *p;
702                 tmp = rb_entry(parent, struct pkt_rb_node, rb_node);
703                 if (s < tmp->bio->bi_sector)
704                         p = &(*p)->rb_left;
705                 else
706                         p = &(*p)->rb_right;
707         }
708         rb_link_node(&node->rb_node, parent, p);
709         rb_insert_color(&node->rb_node, &pd->bio_queue);
710         pd->bio_queue_size++;
711 }
712
713 /*
714  * Add a bio to a single linked list defined by its head and tail pointers.
715  */
716 static void pkt_add_list_last(struct bio *bio, struct bio **list_head, struct bio **list_tail)
717 {
718         bio->bi_next = NULL;
719         if (*list_tail) {
720                 BUG_ON((*list_head) == NULL);
721                 (*list_tail)->bi_next = bio;
722                 (*list_tail) = bio;
723         } else {
724                 BUG_ON((*list_head) != NULL);
725                 (*list_head) = bio;
726                 (*list_tail) = bio;
727         }
728 }
729
730 /*
731  * Remove and return the first bio from a single linked list defined by its
732  * head and tail pointers.
733  */
734 static inline struct bio *pkt_get_list_first(struct bio **list_head, struct bio **list_tail)
735 {
736         struct bio *bio;
737
738         if (*list_head == NULL)
739                 return NULL;
740
741         bio = *list_head;
742         *list_head = bio->bi_next;
743         if (*list_head == NULL)
744                 *list_tail = NULL;
745
746         bio->bi_next = NULL;
747         return bio;
748 }
749
750 /*
751  * Send a packet_command to the underlying block device and
752  * wait for completion.
753  */
754 static int pkt_generic_packet(struct pktcdvd_device *pd, struct packet_command *cgc)
755 {
756         request_queue_t *q = bdev_get_queue(pd->bdev);
757         struct request *rq;
758         int ret = 0;
759
760         rq = blk_get_request(q, (cgc->data_direction == CGC_DATA_WRITE) ?
761                              WRITE : READ, __GFP_WAIT);
762
763         if (cgc->buflen) {
764                 if (blk_rq_map_kern(q, rq, cgc->buffer, cgc->buflen, __GFP_WAIT))
765                         goto out;
766         }
767
768         rq->cmd_len = COMMAND_SIZE(rq->cmd[0]);
769         memcpy(rq->cmd, cgc->cmd, CDROM_PACKET_SIZE);
770         if (sizeof(rq->cmd) > CDROM_PACKET_SIZE)
771                 memset(rq->cmd + CDROM_PACKET_SIZE, 0, sizeof(rq->cmd) - CDROM_PACKET_SIZE);
772
773         rq->timeout = 60*HZ;
774         rq->cmd_type = REQ_TYPE_BLOCK_PC;
775         rq->cmd_flags |= REQ_HARDBARRIER;
776         if (cgc->quiet)
777                 rq->cmd_flags |= REQ_QUIET;
778
779         blk_execute_rq(rq->q, pd->bdev->bd_disk, rq, 0);
780         ret = rq->errors;
781 out:
782         blk_put_request(rq);
783         return ret;
784 }
785
786 /*
787  * A generic sense dump / resolve mechanism should be implemented across
788  * all ATAPI + SCSI devices.
789  */
790 static void pkt_dump_sense(struct packet_command *cgc)
791 {
792         static char *info[9] = { "No sense", "Recovered error", "Not ready",
793                                  "Medium error", "Hardware error", "Illegal request",
794                                  "Unit attention", "Data protect", "Blank check" };
795         int i;
796         struct request_sense *sense = cgc->sense;
797
798         printk(DRIVER_NAME":");
799         for (i = 0; i < CDROM_PACKET_SIZE; i++)
800                 printk(" %02x", cgc->cmd[i]);
801         printk(" - ");
802
803         if (sense == NULL) {
804                 printk("no sense\n");
805                 return;
806         }
807
808         printk("sense %02x.%02x.%02x", sense->sense_key, sense->asc, sense->ascq);
809
810         if (sense->sense_key > 8) {
811                 printk(" (INVALID)\n");
812                 return;
813         }
814
815         printk(" (%s)\n", info[sense->sense_key]);
816 }
817
818 /*
819  * flush the drive cache to media
820  */
821 static int pkt_flush_cache(struct pktcdvd_device *pd)
822 {
823         struct packet_command cgc;
824
825         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
826         cgc.cmd[0] = GPCMD_FLUSH_CACHE;
827         cgc.quiet = 1;
828
829         /*
830          * the IMMED bit -- we default to not setting it, although that
831          * would allow a much faster close, this is safer
832          */
833 #if 0
834         cgc.cmd[1] = 1 << 1;
835 #endif
836         return pkt_generic_packet(pd, &cgc);
837 }
838
839 /*
840  * speed is given as the normal factor, e.g. 4 for 4x
841  */
842 static int pkt_set_speed(struct pktcdvd_device *pd, unsigned write_speed, unsigned read_speed)
843 {
844         struct packet_command cgc;
845         struct request_sense sense;
846         int ret;
847
848         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
849         cgc.sense = &sense;
850         cgc.cmd[0] = GPCMD_SET_SPEED;
851         cgc.cmd[2] = (read_speed >> 8) & 0xff;
852         cgc.cmd[3] = read_speed & 0xff;
853         cgc.cmd[4] = (write_speed >> 8) & 0xff;
854         cgc.cmd[5] = write_speed & 0xff;
855
856         if ((ret = pkt_generic_packet(pd, &cgc)))
857                 pkt_dump_sense(&cgc);
858
859         return ret;
860 }
861
862 /*
863  * Queue a bio for processing by the low-level CD device. Must be called
864  * from process context.
865  */
866 static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio)
867 {
868         spin_lock(&pd->iosched.lock);
869         if (bio_data_dir(bio) == READ) {
870                 pkt_add_list_last(bio, &pd->iosched.read_queue,
871                                   &pd->iosched.read_queue_tail);
872         } else {
873                 pkt_add_list_last(bio, &pd->iosched.write_queue,
874                                   &pd->iosched.write_queue_tail);
875         }
876         spin_unlock(&pd->iosched.lock);
877
878         atomic_set(&pd->iosched.attention, 1);
879         wake_up(&pd->wqueue);
880 }
881
882 /*
883  * Process the queued read/write requests. This function handles special
884  * requirements for CDRW drives:
885  * - A cache flush command must be inserted before a read request if the
886  *   previous request was a write.
887  * - Switching between reading and writing is slow, so don't do it more often
888  *   than necessary.
889  * - Optimize for throughput at the expense of latency. This means that streaming
890  *   writes will never be interrupted by a read, but if the drive has to seek
891  *   before the next write, switch to reading instead if there are any pending
892  *   read requests.
893  * - Set the read speed according to current usage pattern. When only reading
894  *   from the device, it's best to use the highest possible read speed, but
895  *   when switching often between reading and writing, it's better to have the
896  *   same read and write speeds.
897  */
898 static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
899 {
900
901         if (atomic_read(&pd->iosched.attention) == 0)
902                 return;
903         atomic_set(&pd->iosched.attention, 0);
904
905         for (;;) {
906                 struct bio *bio;
907                 int reads_queued, writes_queued;
908
909                 spin_lock(&pd->iosched.lock);
910                 reads_queued = (pd->iosched.read_queue != NULL);
911                 writes_queued = (pd->iosched.write_queue != NULL);
912                 spin_unlock(&pd->iosched.lock);
913
914                 if (!reads_queued && !writes_queued)
915                         break;
916
917                 if (pd->iosched.writing) {
918                         int need_write_seek = 1;
919                         spin_lock(&pd->iosched.lock);
920                         bio = pd->iosched.write_queue;
921                         spin_unlock(&pd->iosched.lock);
922                         if (bio && (bio->bi_sector == pd->iosched.last_write))
923                                 need_write_seek = 0;
924                         if (need_write_seek && reads_queued) {
925                                 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
926                                         VPRINTK(DRIVER_NAME": write, waiting\n");
927                                         break;
928                                 }
929                                 pkt_flush_cache(pd);
930                                 pd->iosched.writing = 0;
931                         }
932                 } else {
933                         if (!reads_queued && writes_queued) {
934                                 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
935                                         VPRINTK(DRIVER_NAME": read, waiting\n");
936                                         break;
937                                 }
938                                 pd->iosched.writing = 1;
939                         }
940                 }
941
942                 spin_lock(&pd->iosched.lock);
943                 if (pd->iosched.writing) {
944                         bio = pkt_get_list_first(&pd->iosched.write_queue,
945                                                  &pd->iosched.write_queue_tail);
946                 } else {
947                         bio = pkt_get_list_first(&pd->iosched.read_queue,
948                                                  &pd->iosched.read_queue_tail);
949                 }
950                 spin_unlock(&pd->iosched.lock);
951
952                 if (!bio)
953                         continue;
954
955                 if (bio_data_dir(bio) == READ)
956                         pd->iosched.successive_reads += bio->bi_size >> 10;
957                 else {
958                         pd->iosched.successive_reads = 0;
959                         pd->iosched.last_write = bio->bi_sector + bio_sectors(bio);
960                 }
961                 if (pd->iosched.successive_reads >= HI_SPEED_SWITCH) {
962                         if (pd->read_speed == pd->write_speed) {
963                                 pd->read_speed = MAX_SPEED;
964                                 pkt_set_speed(pd, pd->write_speed, pd->read_speed);
965                         }
966                 } else {
967                         if (pd->read_speed != pd->write_speed) {
968                                 pd->read_speed = pd->write_speed;
969                                 pkt_set_speed(pd, pd->write_speed, pd->read_speed);
970                         }
971                 }
972
973                 atomic_inc(&pd->cdrw.pending_bios);
974                 generic_make_request(bio);
975         }
976 }
977
978 /*
979  * Special care is needed if the underlying block device has a small
980  * max_phys_segments value.
981  */
982 static int pkt_set_segment_merging(struct pktcdvd_device *pd, request_queue_t *q)
983 {
984         if ((pd->settings.size << 9) / CD_FRAMESIZE <= q->max_phys_segments) {
985                 /*
986                  * The cdrom device can handle one segment/frame
987                  */
988                 clear_bit(PACKET_MERGE_SEGS, &pd->flags);
989                 return 0;
990         } else if ((pd->settings.size << 9) / PAGE_SIZE <= q->max_phys_segments) {
991                 /*
992                  * We can handle this case at the expense of some extra memory
993                  * copies during write operations
994                  */
995                 set_bit(PACKET_MERGE_SEGS, &pd->flags);
996                 return 0;
997         } else {
998                 printk(DRIVER_NAME": cdrom max_phys_segments too small\n");
999                 return -EIO;
1000         }
1001 }
1002
1003 /*
1004  * Copy CD_FRAMESIZE bytes from src_bio into a destination page
1005  */
1006 static void pkt_copy_bio_data(struct bio *src_bio, int seg, int offs, struct page *dst_page, int dst_offs)
1007 {
1008         unsigned int copy_size = CD_FRAMESIZE;
1009
1010         while (copy_size > 0) {
1011                 struct bio_vec *src_bvl = bio_iovec_idx(src_bio, seg);
1012                 void *vfrom = kmap_atomic(src_bvl->bv_page, KM_USER0) +
1013                         src_bvl->bv_offset + offs;
1014                 void *vto = page_address(dst_page) + dst_offs;
1015                 int len = min_t(int, copy_size, src_bvl->bv_len - offs);
1016
1017                 BUG_ON(len < 0);
1018                 memcpy(vto, vfrom, len);
1019                 kunmap_atomic(vfrom, KM_USER0);
1020
1021                 seg++;
1022                 offs = 0;
1023                 dst_offs += len;
1024                 copy_size -= len;
1025         }
1026 }
1027
1028 /*
1029  * Copy all data for this packet to pkt->pages[], so that
1030  * a) The number of required segments for the write bio is minimized, which
1031  *    is necessary for some scsi controllers.
1032  * b) The data can be used as cache to avoid read requests if we receive a
1033  *    new write request for the same zone.
1034  */
1035 static void pkt_make_local_copy(struct packet_data *pkt, struct bio_vec *bvec)
1036 {
1037         int f, p, offs;
1038
1039         /* Copy all data to pkt->pages[] */
1040         p = 0;
1041         offs = 0;
1042         for (f = 0; f < pkt->frames; f++) {
1043                 if (bvec[f].bv_page != pkt->pages[p]) {
1044                         void *vfrom = kmap_atomic(bvec[f].bv_page, KM_USER0) + bvec[f].bv_offset;
1045                         void *vto = page_address(pkt->pages[p]) + offs;
1046                         memcpy(vto, vfrom, CD_FRAMESIZE);
1047                         kunmap_atomic(vfrom, KM_USER0);
1048                         bvec[f].bv_page = pkt->pages[p];
1049                         bvec[f].bv_offset = offs;
1050                 } else {
1051                         BUG_ON(bvec[f].bv_offset != offs);
1052                 }
1053                 offs += CD_FRAMESIZE;
1054                 if (offs >= PAGE_SIZE) {
1055                         offs = 0;
1056                         p++;
1057                 }
1058         }
1059 }
1060
1061 static int pkt_end_io_read(struct bio *bio, unsigned int bytes_done, int err)
1062 {
1063         struct packet_data *pkt = bio->bi_private;
1064         struct pktcdvd_device *pd = pkt->pd;
1065         BUG_ON(!pd);
1066
1067         if (bio->bi_size)
1068                 return 1;
1069
1070         VPRINTK("pkt_end_io_read: bio=%p sec0=%llx sec=%llx err=%d\n", bio,
1071                 (unsigned long long)pkt->sector, (unsigned long long)bio->bi_sector, err);
1072
1073         if (err)
1074                 atomic_inc(&pkt->io_errors);
1075         if (atomic_dec_and_test(&pkt->io_wait)) {
1076                 atomic_inc(&pkt->run_sm);
1077                 wake_up(&pd->wqueue);
1078         }
1079         pkt_bio_finished(pd);
1080
1081         return 0;
1082 }
1083
1084 static int pkt_end_io_packet_write(struct bio *bio, unsigned int bytes_done, int err)
1085 {
1086         struct packet_data *pkt = bio->bi_private;
1087         struct pktcdvd_device *pd = pkt->pd;
1088         BUG_ON(!pd);
1089
1090         if (bio->bi_size)
1091                 return 1;
1092
1093         VPRINTK("pkt_end_io_packet_write: id=%d, err=%d\n", pkt->id, err);
1094
1095         pd->stats.pkt_ended++;
1096
1097         pkt_bio_finished(pd);
1098         atomic_dec(&pkt->io_wait);
1099         atomic_inc(&pkt->run_sm);
1100         wake_up(&pd->wqueue);
1101         return 0;
1102 }
1103
1104 /*
1105  * Schedule reads for the holes in a packet
1106  */
1107 static void pkt_gather_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1108 {
1109         int frames_read = 0;
1110         struct bio *bio;
1111         int f;
1112         char written[PACKET_MAX_SIZE];
1113
1114         BUG_ON(!pkt->orig_bios);
1115
1116         atomic_set(&pkt->io_wait, 0);
1117         atomic_set(&pkt->io_errors, 0);
1118
1119         /*
1120          * Figure out which frames we need to read before we can write.
1121          */
1122         memset(written, 0, sizeof(written));
1123         spin_lock(&pkt->lock);
1124         for (bio = pkt->orig_bios; bio; bio = bio->bi_next) {
1125                 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1126                 int num_frames = bio->bi_size / CD_FRAMESIZE;
1127                 pd->stats.secs_w += num_frames * (CD_FRAMESIZE >> 9);
1128                 BUG_ON(first_frame < 0);
1129                 BUG_ON(first_frame + num_frames > pkt->frames);
1130                 for (f = first_frame; f < first_frame + num_frames; f++)
1131                         written[f] = 1;
1132         }
1133         spin_unlock(&pkt->lock);
1134
1135         if (pkt->cache_valid) {
1136                 VPRINTK("pkt_gather_data: zone %llx cached\n",
1137                         (unsigned long long)pkt->sector);
1138                 goto out_account;
1139         }
1140
1141         /*
1142          * Schedule reads for missing parts of the packet.
1143          */
1144         for (f = 0; f < pkt->frames; f++) {
1145                 int p, offset;
1146                 if (written[f])
1147                         continue;
1148                 bio = pkt->r_bios[f];
1149                 bio_init(bio);
1150                 bio->bi_max_vecs = 1;
1151                 bio->bi_sector = pkt->sector + f * (CD_FRAMESIZE >> 9);
1152                 bio->bi_bdev = pd->bdev;
1153                 bio->bi_end_io = pkt_end_io_read;
1154                 bio->bi_private = pkt;
1155
1156                 p = (f * CD_FRAMESIZE) / PAGE_SIZE;
1157                 offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1158                 VPRINTK("pkt_gather_data: Adding frame %d, page:%p offs:%d\n",
1159                         f, pkt->pages[p], offset);
1160                 if (!bio_add_page(bio, pkt->pages[p], CD_FRAMESIZE, offset))
1161                         BUG();
1162
1163                 atomic_inc(&pkt->io_wait);
1164                 bio->bi_rw = READ;
1165                 pkt_queue_bio(pd, bio);
1166                 frames_read++;
1167         }
1168
1169 out_account:
1170         VPRINTK("pkt_gather_data: need %d frames for zone %llx\n",
1171                 frames_read, (unsigned long long)pkt->sector);
1172         pd->stats.pkt_started++;
1173         pd->stats.secs_rg += frames_read * (CD_FRAMESIZE >> 9);
1174 }
1175
1176 /*
1177  * Find a packet matching zone, or the least recently used packet if
1178  * there is no match.
1179  */
1180 static struct packet_data *pkt_get_packet_data(struct pktcdvd_device *pd, int zone)
1181 {
1182         struct packet_data *pkt;
1183
1184         list_for_each_entry(pkt, &pd->cdrw.pkt_free_list, list) {
1185                 if (pkt->sector == zone || pkt->list.next == &pd->cdrw.pkt_free_list) {
1186                         list_del_init(&pkt->list);
1187                         if (pkt->sector != zone)
1188                                 pkt->cache_valid = 0;
1189                         return pkt;
1190                 }
1191         }
1192         BUG();
1193         return NULL;
1194 }
1195
1196 static void pkt_put_packet_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1197 {
1198         if (pkt->cache_valid) {
1199                 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
1200         } else {
1201                 list_add_tail(&pkt->list, &pd->cdrw.pkt_free_list);
1202         }
1203 }
1204
1205 /*
1206  * recover a failed write, query for relocation if possible
1207  *
1208  * returns 1 if recovery is possible, or 0 if not
1209  *
1210  */
1211 static int pkt_start_recovery(struct packet_data *pkt)
1212 {
1213         /*
1214          * FIXME. We need help from the file system to implement
1215          * recovery handling.
1216          */
1217         return 0;
1218 #if 0
1219         struct request *rq = pkt->rq;
1220         struct pktcdvd_device *pd = rq->rq_disk->private_data;
1221         struct block_device *pkt_bdev;
1222         struct super_block *sb = NULL;
1223         unsigned long old_block, new_block;
1224         sector_t new_sector;
1225
1226         pkt_bdev = bdget(kdev_t_to_nr(pd->pkt_dev));
1227         if (pkt_bdev) {
1228                 sb = get_super(pkt_bdev);
1229                 bdput(pkt_bdev);
1230         }
1231
1232         if (!sb)
1233                 return 0;
1234
1235         if (!sb->s_op || !sb->s_op->relocate_blocks)
1236                 goto out;
1237
1238         old_block = pkt->sector / (CD_FRAMESIZE >> 9);
1239         if (sb->s_op->relocate_blocks(sb, old_block, &new_block))
1240                 goto out;
1241
1242         new_sector = new_block * (CD_FRAMESIZE >> 9);
1243         pkt->sector = new_sector;
1244
1245         pkt->bio->bi_sector = new_sector;
1246         pkt->bio->bi_next = NULL;
1247         pkt->bio->bi_flags = 1 << BIO_UPTODATE;
1248         pkt->bio->bi_idx = 0;
1249
1250         BUG_ON(pkt->bio->bi_rw != (1 << BIO_RW));
1251         BUG_ON(pkt->bio->bi_vcnt != pkt->frames);
1252         BUG_ON(pkt->bio->bi_size != pkt->frames * CD_FRAMESIZE);
1253         BUG_ON(pkt->bio->bi_end_io != pkt_end_io_packet_write);
1254         BUG_ON(pkt->bio->bi_private != pkt);
1255
1256         drop_super(sb);
1257         return 1;
1258
1259 out:
1260         drop_super(sb);
1261         return 0;
1262 #endif
1263 }
1264
1265 static inline void pkt_set_state(struct packet_data *pkt, enum packet_data_state state)
1266 {
1267 #if PACKET_DEBUG > 1
1268         static const char *state_name[] = {
1269                 "IDLE", "WAITING", "READ_WAIT", "WRITE_WAIT", "RECOVERY", "FINISHED"
1270         };
1271         enum packet_data_state old_state = pkt->state;
1272         VPRINTK("pkt %2d : s=%6llx %s -> %s\n", pkt->id, (unsigned long long)pkt->sector,
1273                 state_name[old_state], state_name[state]);
1274 #endif
1275         pkt->state = state;
1276 }
1277
1278 /*
1279  * Scan the work queue to see if we can start a new packet.
1280  * returns non-zero if any work was done.
1281  */
1282 static int pkt_handle_queue(struct pktcdvd_device *pd)
1283 {
1284         struct packet_data *pkt, *p;
1285         struct bio *bio = NULL;
1286         sector_t zone = 0; /* Suppress gcc warning */
1287         struct pkt_rb_node *node, *first_node;
1288         struct rb_node *n;
1289         int wakeup;
1290
1291         VPRINTK("handle_queue\n");
1292
1293         atomic_set(&pd->scan_queue, 0);
1294
1295         if (list_empty(&pd->cdrw.pkt_free_list)) {
1296                 VPRINTK("handle_queue: no pkt\n");
1297                 return 0;
1298         }
1299
1300         /*
1301          * Try to find a zone we are not already working on.
1302          */
1303         spin_lock(&pd->lock);
1304         first_node = pkt_rbtree_find(pd, pd->current_sector);
1305         if (!first_node) {
1306                 n = rb_first(&pd->bio_queue);
1307                 if (n)
1308                         first_node = rb_entry(n, struct pkt_rb_node, rb_node);
1309         }
1310         node = first_node;
1311         while (node) {
1312                 bio = node->bio;
1313                 zone = ZONE(bio->bi_sector, pd);
1314                 list_for_each_entry(p, &pd->cdrw.pkt_active_list, list) {
1315                         if (p->sector == zone) {
1316                                 bio = NULL;
1317                                 goto try_next_bio;
1318                         }
1319                 }
1320                 break;
1321 try_next_bio:
1322                 node = pkt_rbtree_next(node);
1323                 if (!node) {
1324                         n = rb_first(&pd->bio_queue);
1325                         if (n)
1326                                 node = rb_entry(n, struct pkt_rb_node, rb_node);
1327                 }
1328                 if (node == first_node)
1329                         node = NULL;
1330         }
1331         spin_unlock(&pd->lock);
1332         if (!bio) {
1333                 VPRINTK("handle_queue: no bio\n");
1334                 return 0;
1335         }
1336
1337         pkt = pkt_get_packet_data(pd, zone);
1338
1339         pd->current_sector = zone + pd->settings.size;
1340         pkt->sector = zone;
1341         BUG_ON(pkt->frames != pd->settings.size >> 2);
1342         pkt->write_size = 0;
1343
1344         /*
1345          * Scan work queue for bios in the same zone and link them
1346          * to this packet.
1347          */
1348         spin_lock(&pd->lock);
1349         VPRINTK("pkt_handle_queue: looking for zone %llx\n", (unsigned long long)zone);
1350         while ((node = pkt_rbtree_find(pd, zone)) != NULL) {
1351                 bio = node->bio;
1352                 VPRINTK("pkt_handle_queue: found zone=%llx\n",
1353                         (unsigned long long)ZONE(bio->bi_sector, pd));
1354                 if (ZONE(bio->bi_sector, pd) != zone)
1355                         break;
1356                 pkt_rbtree_erase(pd, node);
1357                 spin_lock(&pkt->lock);
1358                 pkt_add_list_last(bio, &pkt->orig_bios, &pkt->orig_bios_tail);
1359                 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
1360                 spin_unlock(&pkt->lock);
1361         }
1362         /* check write congestion marks, and if bio_queue_size is
1363            below, wake up any waiters */
1364         wakeup = (pd->write_congestion_on > 0
1365                         && pd->bio_queue_size <= pd->write_congestion_off);
1366         spin_unlock(&pd->lock);
1367         if (wakeup)
1368                 clear_bdi_congested(&pd->disk->queue->backing_dev_info, WRITE);
1369
1370         pkt->sleep_time = max(PACKET_WAIT_TIME, 1);
1371         pkt_set_state(pkt, PACKET_WAITING_STATE);
1372         atomic_set(&pkt->run_sm, 1);
1373
1374         spin_lock(&pd->cdrw.active_list_lock);
1375         list_add(&pkt->list, &pd->cdrw.pkt_active_list);
1376         spin_unlock(&pd->cdrw.active_list_lock);
1377
1378         return 1;
1379 }
1380
1381 /*
1382  * Assemble a bio to write one packet and queue the bio for processing
1383  * by the underlying block device.
1384  */
1385 static void pkt_start_write(struct pktcdvd_device *pd, struct packet_data *pkt)
1386 {
1387         struct bio *bio;
1388         int f;
1389         int frames_write;
1390         struct bio_vec *bvec = pkt->w_bio->bi_io_vec;
1391
1392         for (f = 0; f < pkt->frames; f++) {
1393                 bvec[f].bv_page = pkt->pages[(f * CD_FRAMESIZE) / PAGE_SIZE];
1394                 bvec[f].bv_offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1395         }
1396
1397         /*
1398          * Fill-in bvec with data from orig_bios.
1399          */
1400         frames_write = 0;
1401         spin_lock(&pkt->lock);
1402         for (bio = pkt->orig_bios; bio; bio = bio->bi_next) {
1403                 int segment = bio->bi_idx;
1404                 int src_offs = 0;
1405                 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1406                 int num_frames = bio->bi_size / CD_FRAMESIZE;
1407                 BUG_ON(first_frame < 0);
1408                 BUG_ON(first_frame + num_frames > pkt->frames);
1409                 for (f = first_frame; f < first_frame + num_frames; f++) {
1410                         struct bio_vec *src_bvl = bio_iovec_idx(bio, segment);
1411
1412                         while (src_offs >= src_bvl->bv_len) {
1413                                 src_offs -= src_bvl->bv_len;
1414                                 segment++;
1415                                 BUG_ON(segment >= bio->bi_vcnt);
1416                                 src_bvl = bio_iovec_idx(bio, segment);
1417                         }
1418
1419                         if (src_bvl->bv_len - src_offs >= CD_FRAMESIZE) {
1420                                 bvec[f].bv_page = src_bvl->bv_page;
1421                                 bvec[f].bv_offset = src_bvl->bv_offset + src_offs;
1422                         } else {
1423                                 pkt_copy_bio_data(bio, segment, src_offs,
1424                                                   bvec[f].bv_page, bvec[f].bv_offset);
1425                         }
1426                         src_offs += CD_FRAMESIZE;
1427                         frames_write++;
1428                 }
1429         }
1430         pkt_set_state(pkt, PACKET_WRITE_WAIT_STATE);
1431         spin_unlock(&pkt->lock);
1432
1433         VPRINTK("pkt_start_write: Writing %d frames for zone %llx\n",
1434                 frames_write, (unsigned long long)pkt->sector);
1435         BUG_ON(frames_write != pkt->write_size);
1436
1437         if (test_bit(PACKET_MERGE_SEGS, &pd->flags) || (pkt->write_size < pkt->frames)) {
1438                 pkt_make_local_copy(pkt, bvec);
1439                 pkt->cache_valid = 1;
1440         } else {
1441                 pkt->cache_valid = 0;
1442         }
1443
1444         /* Start the write request */
1445         bio_init(pkt->w_bio);
1446         pkt->w_bio->bi_max_vecs = PACKET_MAX_SIZE;
1447         pkt->w_bio->bi_sector = pkt->sector;
1448         pkt->w_bio->bi_bdev = pd->bdev;
1449         pkt->w_bio->bi_end_io = pkt_end_io_packet_write;
1450         pkt->w_bio->bi_private = pkt;
1451         for (f = 0; f < pkt->frames; f++)
1452                 if (!bio_add_page(pkt->w_bio, bvec[f].bv_page, CD_FRAMESIZE, bvec[f].bv_offset))
1453                         BUG();
1454         VPRINTK(DRIVER_NAME": vcnt=%d\n", pkt->w_bio->bi_vcnt);
1455
1456         atomic_set(&pkt->io_wait, 1);
1457         pkt->w_bio->bi_rw = WRITE;
1458         pkt_queue_bio(pd, pkt->w_bio);
1459 }
1460
1461 static void pkt_finish_packet(struct packet_data *pkt, int uptodate)
1462 {
1463         struct bio *bio, *next;
1464
1465         if (!uptodate)
1466                 pkt->cache_valid = 0;
1467
1468         /* Finish all bios corresponding to this packet */
1469         bio = pkt->orig_bios;
1470         while (bio) {
1471                 next = bio->bi_next;
1472                 bio->bi_next = NULL;
1473                 bio_endio(bio, bio->bi_size, uptodate ? 0 : -EIO);
1474                 bio = next;
1475         }
1476         pkt->orig_bios = pkt->orig_bios_tail = NULL;
1477 }
1478
1479 static void pkt_run_state_machine(struct pktcdvd_device *pd, struct packet_data *pkt)
1480 {
1481         int uptodate;
1482
1483         VPRINTK("run_state_machine: pkt %d\n", pkt->id);
1484
1485         for (;;) {
1486                 switch (pkt->state) {
1487                 case PACKET_WAITING_STATE:
1488                         if ((pkt->write_size < pkt->frames) && (pkt->sleep_time > 0))
1489                                 return;
1490
1491                         pkt->sleep_time = 0;
1492                         pkt_gather_data(pd, pkt);
1493                         pkt_set_state(pkt, PACKET_READ_WAIT_STATE);
1494                         break;
1495
1496                 case PACKET_READ_WAIT_STATE:
1497                         if (atomic_read(&pkt->io_wait) > 0)
1498                                 return;
1499
1500                         if (atomic_read(&pkt->io_errors) > 0) {
1501                                 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1502                         } else {
1503                                 pkt_start_write(pd, pkt);
1504                         }
1505                         break;
1506
1507                 case PACKET_WRITE_WAIT_STATE:
1508                         if (atomic_read(&pkt->io_wait) > 0)
1509                                 return;
1510
1511                         if (test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags)) {
1512                                 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1513                         } else {
1514                                 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1515                         }
1516                         break;
1517
1518                 case PACKET_RECOVERY_STATE:
1519                         if (pkt_start_recovery(pkt)) {
1520                                 pkt_start_write(pd, pkt);
1521                         } else {
1522                                 VPRINTK("No recovery possible\n");
1523                                 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1524                         }
1525                         break;
1526
1527                 case PACKET_FINISHED_STATE:
1528                         uptodate = test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags);
1529                         pkt_finish_packet(pkt, uptodate);
1530                         return;
1531
1532                 default:
1533                         BUG();
1534                         break;
1535                 }
1536         }
1537 }
1538
1539 static void pkt_handle_packets(struct pktcdvd_device *pd)
1540 {
1541         struct packet_data *pkt, *next;
1542
1543         VPRINTK("pkt_handle_packets\n");
1544
1545         /*
1546          * Run state machine for active packets
1547          */
1548         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1549                 if (atomic_read(&pkt->run_sm) > 0) {
1550                         atomic_set(&pkt->run_sm, 0);
1551                         pkt_run_state_machine(pd, pkt);
1552                 }
1553         }
1554
1555         /*
1556          * Move no longer active packets to the free list
1557          */
1558         spin_lock(&pd->cdrw.active_list_lock);
1559         list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_active_list, list) {
1560                 if (pkt->state == PACKET_FINISHED_STATE) {
1561                         list_del(&pkt->list);
1562                         pkt_put_packet_data(pd, pkt);
1563                         pkt_set_state(pkt, PACKET_IDLE_STATE);
1564                         atomic_set(&pd->scan_queue, 1);
1565                 }
1566         }
1567         spin_unlock(&pd->cdrw.active_list_lock);
1568 }
1569
1570 static void pkt_count_states(struct pktcdvd_device *pd, int *states)
1571 {
1572         struct packet_data *pkt;
1573         int i;
1574
1575         for (i = 0; i < PACKET_NUM_STATES; i++)
1576                 states[i] = 0;
1577
1578         spin_lock(&pd->cdrw.active_list_lock);
1579         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1580                 states[pkt->state]++;
1581         }
1582         spin_unlock(&pd->cdrw.active_list_lock);
1583 }
1584
1585 /*
1586  * kcdrwd is woken up when writes have been queued for one of our
1587  * registered devices
1588  */
1589 static int kcdrwd(void *foobar)
1590 {
1591         struct pktcdvd_device *pd = foobar;
1592         struct packet_data *pkt;
1593         long min_sleep_time, residue;
1594
1595         set_user_nice(current, -20);
1596
1597         for (;;) {
1598                 DECLARE_WAITQUEUE(wait, current);
1599
1600                 /*
1601                  * Wait until there is something to do
1602                  */
1603                 add_wait_queue(&pd->wqueue, &wait);
1604                 for (;;) {
1605                         set_current_state(TASK_INTERRUPTIBLE);
1606
1607                         /* Check if we need to run pkt_handle_queue */
1608                         if (atomic_read(&pd->scan_queue) > 0)
1609                                 goto work_to_do;
1610
1611                         /* Check if we need to run the state machine for some packet */
1612                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1613                                 if (atomic_read(&pkt->run_sm) > 0)
1614                                         goto work_to_do;
1615                         }
1616
1617                         /* Check if we need to process the iosched queues */
1618                         if (atomic_read(&pd->iosched.attention) != 0)
1619                                 goto work_to_do;
1620
1621                         /* Otherwise, go to sleep */
1622                         if (PACKET_DEBUG > 1) {
1623                                 int states[PACKET_NUM_STATES];
1624                                 pkt_count_states(pd, states);
1625                                 VPRINTK("kcdrwd: i:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
1626                                         states[0], states[1], states[2], states[3],
1627                                         states[4], states[5]);
1628                         }
1629
1630                         min_sleep_time = MAX_SCHEDULE_TIMEOUT;
1631                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1632                                 if (pkt->sleep_time && pkt->sleep_time < min_sleep_time)
1633                                         min_sleep_time = pkt->sleep_time;
1634                         }
1635
1636                         generic_unplug_device(bdev_get_queue(pd->bdev));
1637
1638                         VPRINTK("kcdrwd: sleeping\n");
1639                         residue = schedule_timeout(min_sleep_time);
1640                         VPRINTK("kcdrwd: wake up\n");
1641
1642                         /* make swsusp happy with our thread */
1643                         try_to_freeze();
1644
1645                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1646                                 if (!pkt->sleep_time)
1647                                         continue;
1648                                 pkt->sleep_time -= min_sleep_time - residue;
1649                                 if (pkt->sleep_time <= 0) {
1650                                         pkt->sleep_time = 0;
1651                                         atomic_inc(&pkt->run_sm);
1652                                 }
1653                         }
1654
1655                         if (signal_pending(current)) {
1656                                 flush_signals(current);
1657                         }
1658                         if (kthread_should_stop())
1659                                 break;
1660                 }
1661 work_to_do:
1662                 set_current_state(TASK_RUNNING);
1663                 remove_wait_queue(&pd->wqueue, &wait);
1664
1665                 if (kthread_should_stop())
1666                         break;
1667
1668                 /*
1669                  * if pkt_handle_queue returns true, we can queue
1670                  * another request.
1671                  */
1672                 while (pkt_handle_queue(pd))
1673                         ;
1674
1675                 /*
1676                  * Handle packet state machine
1677                  */
1678                 pkt_handle_packets(pd);
1679
1680                 /*
1681                  * Handle iosched queues
1682                  */
1683                 pkt_iosched_process_queue(pd);
1684         }
1685
1686         return 0;
1687 }
1688
1689 static void pkt_print_settings(struct pktcdvd_device *pd)
1690 {
1691         printk(DRIVER_NAME": %s packets, ", pd->settings.fp ? "Fixed" : "Variable");
1692         printk("%u blocks, ", pd->settings.size >> 2);
1693         printk("Mode-%c disc\n", pd->settings.block_mode == 8 ? '1' : '2');
1694 }
1695
1696 static int pkt_mode_sense(struct pktcdvd_device *pd, struct packet_command *cgc, int page_code, int page_control)
1697 {
1698         memset(cgc->cmd, 0, sizeof(cgc->cmd));
1699
1700         cgc->cmd[0] = GPCMD_MODE_SENSE_10;
1701         cgc->cmd[2] = page_code | (page_control << 6);
1702         cgc->cmd[7] = cgc->buflen >> 8;
1703         cgc->cmd[8] = cgc->buflen & 0xff;
1704         cgc->data_direction = CGC_DATA_READ;
1705         return pkt_generic_packet(pd, cgc);
1706 }
1707
1708 static int pkt_mode_select(struct pktcdvd_device *pd, struct packet_command *cgc)
1709 {
1710         memset(cgc->cmd, 0, sizeof(cgc->cmd));
1711         memset(cgc->buffer, 0, 2);
1712         cgc->cmd[0] = GPCMD_MODE_SELECT_10;
1713         cgc->cmd[1] = 0x10;             /* PF */
1714         cgc->cmd[7] = cgc->buflen >> 8;
1715         cgc->cmd[8] = cgc->buflen & 0xff;
1716         cgc->data_direction = CGC_DATA_WRITE;
1717         return pkt_generic_packet(pd, cgc);
1718 }
1719
1720 static int pkt_get_disc_info(struct pktcdvd_device *pd, disc_information *di)
1721 {
1722         struct packet_command cgc;
1723         int ret;
1724
1725         /* set up command and get the disc info */
1726         init_cdrom_command(&cgc, di, sizeof(*di), CGC_DATA_READ);
1727         cgc.cmd[0] = GPCMD_READ_DISC_INFO;
1728         cgc.cmd[8] = cgc.buflen = 2;
1729         cgc.quiet = 1;
1730
1731         if ((ret = pkt_generic_packet(pd, &cgc)))
1732                 return ret;
1733
1734         /* not all drives have the same disc_info length, so requeue
1735          * packet with the length the drive tells us it can supply
1736          */
1737         cgc.buflen = be16_to_cpu(di->disc_information_length) +
1738                      sizeof(di->disc_information_length);
1739
1740         if (cgc.buflen > sizeof(disc_information))
1741                 cgc.buflen = sizeof(disc_information);
1742
1743         cgc.cmd[8] = cgc.buflen;
1744         return pkt_generic_packet(pd, &cgc);
1745 }
1746
1747 static int pkt_get_track_info(struct pktcdvd_device *pd, __u16 track, __u8 type, track_information *ti)
1748 {
1749         struct packet_command cgc;
1750         int ret;
1751
1752         init_cdrom_command(&cgc, ti, 8, CGC_DATA_READ);
1753         cgc.cmd[0] = GPCMD_READ_TRACK_RZONE_INFO;
1754         cgc.cmd[1] = type & 3;
1755         cgc.cmd[4] = (track & 0xff00) >> 8;
1756         cgc.cmd[5] = track & 0xff;
1757         cgc.cmd[8] = 8;
1758         cgc.quiet = 1;
1759
1760         if ((ret = pkt_generic_packet(pd, &cgc)))
1761                 return ret;
1762
1763         cgc.buflen = be16_to_cpu(ti->track_information_length) +
1764                      sizeof(ti->track_information_length);
1765
1766         if (cgc.buflen > sizeof(track_information))
1767                 cgc.buflen = sizeof(track_information);
1768
1769         cgc.cmd[8] = cgc.buflen;
1770         return pkt_generic_packet(pd, &cgc);
1771 }
1772
1773 static int pkt_get_last_written(struct pktcdvd_device *pd, long *last_written)
1774 {
1775         disc_information di;
1776         track_information ti;
1777         __u32 last_track;
1778         int ret = -1;
1779
1780         if ((ret = pkt_get_disc_info(pd, &di)))
1781                 return ret;
1782
1783         last_track = (di.last_track_msb << 8) | di.last_track_lsb;
1784         if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1785                 return ret;
1786
1787         /* if this track is blank, try the previous. */
1788         if (ti.blank) {
1789                 last_track--;
1790                 if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1791                         return ret;
1792         }
1793
1794         /* if last recorded field is valid, return it. */
1795         if (ti.lra_v) {
1796                 *last_written = be32_to_cpu(ti.last_rec_address);
1797         } else {
1798                 /* make it up instead */
1799                 *last_written = be32_to_cpu(ti.track_start) +
1800                                 be32_to_cpu(ti.track_size);
1801                 if (ti.free_blocks)
1802                         *last_written -= (be32_to_cpu(ti.free_blocks) + 7);
1803         }
1804         return 0;
1805 }
1806
1807 /*
1808  * write mode select package based on pd->settings
1809  */
1810 static int pkt_set_write_settings(struct pktcdvd_device *pd)
1811 {
1812         struct packet_command cgc;
1813         struct request_sense sense;
1814         write_param_page *wp;
1815         char buffer[128];
1816         int ret, size;
1817
1818         /* doesn't apply to DVD+RW or DVD-RAM */
1819         if ((pd->mmc3_profile == 0x1a) || (pd->mmc3_profile == 0x12))
1820                 return 0;
1821
1822         memset(buffer, 0, sizeof(buffer));
1823         init_cdrom_command(&cgc, buffer, sizeof(*wp), CGC_DATA_READ);
1824         cgc.sense = &sense;
1825         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1826                 pkt_dump_sense(&cgc);
1827                 return ret;
1828         }
1829
1830         size = 2 + ((buffer[0] << 8) | (buffer[1] & 0xff));
1831         pd->mode_offset = (buffer[6] << 8) | (buffer[7] & 0xff);
1832         if (size > sizeof(buffer))
1833                 size = sizeof(buffer);
1834
1835         /*
1836          * now get it all
1837          */
1838         init_cdrom_command(&cgc, buffer, size, CGC_DATA_READ);
1839         cgc.sense = &sense;
1840         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1841                 pkt_dump_sense(&cgc);
1842                 return ret;
1843         }
1844
1845         /*
1846          * write page is offset header + block descriptor length
1847          */
1848         wp = (write_param_page *) &buffer[sizeof(struct mode_page_header) + pd->mode_offset];
1849
1850         wp->fp = pd->settings.fp;
1851         wp->track_mode = pd->settings.track_mode;
1852         wp->write_type = pd->settings.write_type;
1853         wp->data_block_type = pd->settings.block_mode;
1854
1855         wp->multi_session = 0;
1856
1857 #ifdef PACKET_USE_LS
1858         wp->link_size = 7;
1859         wp->ls_v = 1;
1860 #endif
1861
1862         if (wp->data_block_type == PACKET_BLOCK_MODE1) {
1863                 wp->session_format = 0;
1864                 wp->subhdr2 = 0x20;
1865         } else if (wp->data_block_type == PACKET_BLOCK_MODE2) {
1866                 wp->session_format = 0x20;
1867                 wp->subhdr2 = 8;
1868 #if 0
1869                 wp->mcn[0] = 0x80;
1870                 memcpy(&wp->mcn[1], PACKET_MCN, sizeof(wp->mcn) - 1);
1871 #endif
1872         } else {
1873                 /*
1874                  * paranoia
1875                  */
1876                 printk(DRIVER_NAME": write mode wrong %d\n", wp->data_block_type);
1877                 return 1;
1878         }
1879         wp->packet_size = cpu_to_be32(pd->settings.size >> 2);
1880
1881         cgc.buflen = cgc.cmd[8] = size;
1882         if ((ret = pkt_mode_select(pd, &cgc))) {
1883                 pkt_dump_sense(&cgc);
1884                 return ret;
1885         }
1886
1887         pkt_print_settings(pd);
1888         return 0;
1889 }
1890
1891 /*
1892  * 1 -- we can write to this track, 0 -- we can't
1893  */
1894 static int pkt_writable_track(struct pktcdvd_device *pd, track_information *ti)
1895 {
1896         switch (pd->mmc3_profile) {
1897                 case 0x1a: /* DVD+RW */
1898                 case 0x12: /* DVD-RAM */
1899                         /* The track is always writable on DVD+RW/DVD-RAM */
1900                         return 1;
1901                 default:
1902                         break;
1903         }
1904
1905         if (!ti->packet || !ti->fp)
1906                 return 0;
1907
1908         /*
1909          * "good" settings as per Mt Fuji.
1910          */
1911         if (ti->rt == 0 && ti->blank == 0)
1912                 return 1;
1913
1914         if (ti->rt == 0 && ti->blank == 1)
1915                 return 1;
1916
1917         if (ti->rt == 1 && ti->blank == 0)
1918                 return 1;
1919
1920         printk(DRIVER_NAME": bad state %d-%d-%d\n", ti->rt, ti->blank, ti->packet);
1921         return 0;
1922 }
1923
1924 /*
1925  * 1 -- we can write to this disc, 0 -- we can't
1926  */
1927 static int pkt_writable_disc(struct pktcdvd_device *pd, disc_information *di)
1928 {
1929         switch (pd->mmc3_profile) {
1930                 case 0x0a: /* CD-RW */
1931                 case 0xffff: /* MMC3 not supported */
1932                         break;
1933                 case 0x1a: /* DVD+RW */
1934                 case 0x13: /* DVD-RW */
1935                 case 0x12: /* DVD-RAM */
1936                         return 1;
1937                 default:
1938                         VPRINTK(DRIVER_NAME": Wrong disc profile (%x)\n", pd->mmc3_profile);
1939                         return 0;
1940         }
1941
1942         /*
1943          * for disc type 0xff we should probably reserve a new track.
1944          * but i'm not sure, should we leave this to user apps? probably.
1945          */
1946         if (di->disc_type == 0xff) {
1947                 printk(DRIVER_NAME": Unknown disc. No track?\n");
1948                 return 0;
1949         }
1950
1951         if (di->disc_type != 0x20 && di->disc_type != 0) {
1952                 printk(DRIVER_NAME": Wrong disc type (%x)\n", di->disc_type);
1953                 return 0;
1954         }
1955
1956         if (di->erasable == 0) {
1957                 printk(DRIVER_NAME": Disc not erasable\n");
1958                 return 0;
1959         }
1960
1961         if (di->border_status == PACKET_SESSION_RESERVED) {
1962                 printk(DRIVER_NAME": Can't write to last track (reserved)\n");
1963                 return 0;
1964         }
1965
1966         return 1;
1967 }
1968
1969 static int pkt_probe_settings(struct pktcdvd_device *pd)
1970 {
1971         struct packet_command cgc;
1972         unsigned char buf[12];
1973         disc_information di;
1974         track_information ti;
1975         int ret, track;
1976
1977         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
1978         cgc.cmd[0] = GPCMD_GET_CONFIGURATION;
1979         cgc.cmd[8] = 8;
1980         ret = pkt_generic_packet(pd, &cgc);
1981         pd->mmc3_profile = ret ? 0xffff : buf[6] << 8 | buf[7];
1982
1983         memset(&di, 0, sizeof(disc_information));
1984         memset(&ti, 0, sizeof(track_information));
1985
1986         if ((ret = pkt_get_disc_info(pd, &di))) {
1987                 printk("failed get_disc\n");
1988                 return ret;
1989         }
1990
1991         if (!pkt_writable_disc(pd, &di))
1992                 return -EROFS;
1993
1994         pd->type = di.erasable ? PACKET_CDRW : PACKET_CDR;
1995
1996         track = 1; /* (di.last_track_msb << 8) | di.last_track_lsb; */
1997         if ((ret = pkt_get_track_info(pd, track, 1, &ti))) {
1998                 printk(DRIVER_NAME": failed get_track\n");
1999                 return ret;
2000         }
2001
2002         if (!pkt_writable_track(pd, &ti)) {
2003                 printk(DRIVER_NAME": can't write to this track\n");
2004                 return -EROFS;
2005         }
2006
2007         /*
2008          * we keep packet size in 512 byte units, makes it easier to
2009          * deal with request calculations.
2010          */
2011         pd->settings.size = be32_to_cpu(ti.fixed_packet_size) << 2;
2012         if (pd->settings.size == 0) {
2013                 printk(DRIVER_NAME": detected zero packet size!\n");
2014                 return -ENXIO;
2015         }
2016         if (pd->settings.size > PACKET_MAX_SECTORS) {
2017                 printk(DRIVER_NAME": packet size is too big\n");
2018                 return -EROFS;
2019         }
2020         pd->settings.fp = ti.fp;
2021         pd->offset = (be32_to_cpu(ti.track_start) << 2) & (pd->settings.size - 1);
2022
2023         if (ti.nwa_v) {
2024                 pd->nwa = be32_to_cpu(ti.next_writable);
2025                 set_bit(PACKET_NWA_VALID, &pd->flags);
2026         }
2027
2028         /*
2029          * in theory we could use lra on -RW media as well and just zero
2030          * blocks that haven't been written yet, but in practice that
2031          * is just a no-go. we'll use that for -R, naturally.
2032          */
2033         if (ti.lra_v) {
2034                 pd->lra = be32_to_cpu(ti.last_rec_address);
2035                 set_bit(PACKET_LRA_VALID, &pd->flags);
2036         } else {
2037                 pd->lra = 0xffffffff;
2038                 set_bit(PACKET_LRA_VALID, &pd->flags);
2039         }
2040
2041         /*
2042          * fine for now
2043          */
2044         pd->settings.link_loss = 7;
2045         pd->settings.write_type = 0;    /* packet */
2046         pd->settings.track_mode = ti.track_mode;
2047
2048         /*
2049          * mode1 or mode2 disc
2050          */
2051         switch (ti.data_mode) {
2052                 case PACKET_MODE1:
2053                         pd->settings.block_mode = PACKET_BLOCK_MODE1;
2054                         break;
2055                 case PACKET_MODE2:
2056                         pd->settings.block_mode = PACKET_BLOCK_MODE2;
2057                         break;
2058                 default:
2059                         printk(DRIVER_NAME": unknown data mode\n");
2060                         return -EROFS;
2061         }
2062         return 0;
2063 }
2064
2065 /*
2066  * enable/disable write caching on drive
2067  */
2068 static int pkt_write_caching(struct pktcdvd_device *pd, int set)
2069 {
2070         struct packet_command cgc;
2071         struct request_sense sense;
2072         unsigned char buf[64];
2073         int ret;
2074
2075         memset(buf, 0, sizeof(buf));
2076         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
2077         cgc.sense = &sense;
2078         cgc.buflen = pd->mode_offset + 12;
2079
2080         /*
2081          * caching mode page might not be there, so quiet this command
2082          */
2083         cgc.quiet = 1;
2084
2085         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WCACHING_PAGE, 0)))
2086                 return ret;
2087
2088         buf[pd->mode_offset + 10] |= (!!set << 2);
2089
2090         cgc.buflen = cgc.cmd[8] = 2 + ((buf[0] << 8) | (buf[1] & 0xff));
2091         ret = pkt_mode_select(pd, &cgc);
2092         if (ret) {
2093                 printk(DRIVER_NAME": write caching control failed\n");
2094                 pkt_dump_sense(&cgc);
2095         } else if (!ret && set)
2096                 printk(DRIVER_NAME": enabled write caching on %s\n", pd->name);
2097         return ret;
2098 }
2099
2100 static int pkt_lock_door(struct pktcdvd_device *pd, int lockflag)
2101 {
2102         struct packet_command cgc;
2103
2104         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2105         cgc.cmd[0] = GPCMD_PREVENT_ALLOW_MEDIUM_REMOVAL;
2106         cgc.cmd[4] = lockflag ? 1 : 0;
2107         return pkt_generic_packet(pd, &cgc);
2108 }
2109
2110 /*
2111  * Returns drive maximum write speed
2112  */
2113 static int pkt_get_max_speed(struct pktcdvd_device *pd, unsigned *write_speed)
2114 {
2115         struct packet_command cgc;
2116         struct request_sense sense;
2117         unsigned char buf[256+18];
2118         unsigned char *cap_buf;
2119         int ret, offset;
2120
2121         memset(buf, 0, sizeof(buf));
2122         cap_buf = &buf[sizeof(struct mode_page_header) + pd->mode_offset];
2123         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_UNKNOWN);
2124         cgc.sense = &sense;
2125
2126         ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2127         if (ret) {
2128                 cgc.buflen = pd->mode_offset + cap_buf[1] + 2 +
2129                              sizeof(struct mode_page_header);
2130                 ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2131                 if (ret) {
2132                         pkt_dump_sense(&cgc);
2133                         return ret;
2134                 }
2135         }
2136
2137         offset = 20;                        /* Obsoleted field, used by older drives */
2138         if (cap_buf[1] >= 28)
2139                 offset = 28;                /* Current write speed selected */
2140         if (cap_buf[1] >= 30) {
2141                 /* If the drive reports at least one "Logical Unit Write
2142                  * Speed Performance Descriptor Block", use the information
2143                  * in the first block. (contains the highest speed)
2144                  */
2145                 int num_spdb = (cap_buf[30] << 8) + cap_buf[31];
2146                 if (num_spdb > 0)
2147                         offset = 34;
2148         }
2149
2150         *write_speed = (cap_buf[offset] << 8) | cap_buf[offset + 1];
2151         return 0;
2152 }
2153
2154 /* These tables from cdrecord - I don't have orange book */
2155 /* standard speed CD-RW (1-4x) */
2156 static char clv_to_speed[16] = {
2157         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2158            0, 2, 4, 6, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2159 };
2160 /* high speed CD-RW (-10x) */
2161 static char hs_clv_to_speed[16] = {
2162         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2163            0, 2, 4, 6, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2164 };
2165 /* ultra high speed CD-RW */
2166 static char us_clv_to_speed[16] = {
2167         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2168            0, 2, 4, 8, 0, 0,16, 0,24,32,40,48, 0, 0, 0, 0
2169 };
2170
2171 /*
2172  * reads the maximum media speed from ATIP
2173  */
2174 static int pkt_media_speed(struct pktcdvd_device *pd, unsigned *speed)
2175 {
2176         struct packet_command cgc;
2177         struct request_sense sense;
2178         unsigned char buf[64];
2179         unsigned int size, st, sp;
2180         int ret;
2181
2182         init_cdrom_command(&cgc, buf, 2, CGC_DATA_READ);
2183         cgc.sense = &sense;
2184         cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2185         cgc.cmd[1] = 2;
2186         cgc.cmd[2] = 4; /* READ ATIP */
2187         cgc.cmd[8] = 2;
2188         ret = pkt_generic_packet(pd, &cgc);
2189         if (ret) {
2190                 pkt_dump_sense(&cgc);
2191                 return ret;
2192         }
2193         size = ((unsigned int) buf[0]<<8) + buf[1] + 2;
2194         if (size > sizeof(buf))
2195                 size = sizeof(buf);
2196
2197         init_cdrom_command(&cgc, buf, size, CGC_DATA_READ);
2198         cgc.sense = &sense;
2199         cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2200         cgc.cmd[1] = 2;
2201         cgc.cmd[2] = 4;
2202         cgc.cmd[8] = size;
2203         ret = pkt_generic_packet(pd, &cgc);
2204         if (ret) {
2205                 pkt_dump_sense(&cgc);
2206                 return ret;
2207         }
2208
2209         if (!buf[6] & 0x40) {
2210                 printk(DRIVER_NAME": Disc type is not CD-RW\n");
2211                 return 1;
2212         }
2213         if (!buf[6] & 0x4) {
2214                 printk(DRIVER_NAME": A1 values on media are not valid, maybe not CDRW?\n");
2215                 return 1;
2216         }
2217
2218         st = (buf[6] >> 3) & 0x7; /* disc sub-type */
2219
2220         sp = buf[16] & 0xf; /* max speed from ATIP A1 field */
2221
2222         /* Info from cdrecord */
2223         switch (st) {
2224                 case 0: /* standard speed */
2225                         *speed = clv_to_speed[sp];
2226                         break;
2227                 case 1: /* high speed */
2228                         *speed = hs_clv_to_speed[sp];
2229                         break;
2230                 case 2: /* ultra high speed */
2231                         *speed = us_clv_to_speed[sp];
2232                         break;
2233                 default:
2234                         printk(DRIVER_NAME": Unknown disc sub-type %d\n",st);
2235                         return 1;
2236         }
2237         if (*speed) {
2238                 printk(DRIVER_NAME": Max. media speed: %d\n",*speed);
2239                 return 0;
2240         } else {
2241                 printk(DRIVER_NAME": Unknown speed %d for sub-type %d\n",sp,st);
2242                 return 1;
2243         }
2244 }
2245
2246 static int pkt_perform_opc(struct pktcdvd_device *pd)
2247 {
2248         struct packet_command cgc;
2249         struct request_sense sense;
2250         int ret;
2251
2252         VPRINTK(DRIVER_NAME": Performing OPC\n");
2253
2254         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2255         cgc.sense = &sense;
2256         cgc.timeout = 60*HZ;
2257         cgc.cmd[0] = GPCMD_SEND_OPC;
2258         cgc.cmd[1] = 1;
2259         if ((ret = pkt_generic_packet(pd, &cgc)))
2260                 pkt_dump_sense(&cgc);
2261         return ret;
2262 }
2263
2264 static int pkt_open_write(struct pktcdvd_device *pd)
2265 {
2266         int ret;
2267         unsigned int write_speed, media_write_speed, read_speed;
2268
2269         if ((ret = pkt_probe_settings(pd))) {
2270                 VPRINTK(DRIVER_NAME": %s failed probe\n", pd->name);
2271                 return ret;
2272         }
2273
2274         if ((ret = pkt_set_write_settings(pd))) {
2275                 DPRINTK(DRIVER_NAME": %s failed saving write settings\n", pd->name);
2276                 return -EIO;
2277         }
2278
2279         pkt_write_caching(pd, USE_WCACHING);
2280
2281         if ((ret = pkt_get_max_speed(pd, &write_speed)))
2282                 write_speed = 16 * 177;
2283         switch (pd->mmc3_profile) {
2284                 case 0x13: /* DVD-RW */
2285                 case 0x1a: /* DVD+RW */
2286                 case 0x12: /* DVD-RAM */
2287                         DPRINTK(DRIVER_NAME": write speed %ukB/s\n", write_speed);
2288                         break;
2289                 default:
2290                         if ((ret = pkt_media_speed(pd, &media_write_speed)))
2291                                 media_write_speed = 16;
2292                         write_speed = min(write_speed, media_write_speed * 177);
2293                         DPRINTK(DRIVER_NAME": write speed %ux\n", write_speed / 176);
2294                         break;
2295         }
2296         read_speed = write_speed;
2297
2298         if ((ret = pkt_set_speed(pd, write_speed, read_speed))) {
2299                 DPRINTK(DRIVER_NAME": %s couldn't set write speed\n", pd->name);
2300                 return -EIO;
2301         }
2302         pd->write_speed = write_speed;
2303         pd->read_speed = read_speed;
2304
2305         if ((ret = pkt_perform_opc(pd))) {
2306                 DPRINTK(DRIVER_NAME": %s Optimum Power Calibration failed\n", pd->name);
2307         }
2308
2309         return 0;
2310 }
2311
2312 /*
2313  * called at open time.
2314  */
2315 static int pkt_open_dev(struct pktcdvd_device *pd, int write)
2316 {
2317         int ret;
2318         long lba;
2319         request_queue_t *q;
2320
2321         /*
2322          * We need to re-open the cdrom device without O_NONBLOCK to be able
2323          * to read/write from/to it. It is already opened in O_NONBLOCK mode
2324          * so bdget() can't fail.
2325          */
2326         bdget(pd->bdev->bd_dev);
2327         if ((ret = blkdev_get(pd->bdev, FMODE_READ, O_RDONLY)))
2328                 goto out;
2329
2330         if ((ret = bd_claim(pd->bdev, pd)))
2331                 goto out_putdev;
2332
2333         if ((ret = pkt_get_last_written(pd, &lba))) {
2334                 printk(DRIVER_NAME": pkt_get_last_written failed\n");
2335                 goto out_unclaim;
2336         }
2337
2338         set_capacity(pd->disk, lba << 2);
2339         set_capacity(pd->bdev->bd_disk, lba << 2);
2340         bd_set_size(pd->bdev, (loff_t)lba << 11);
2341
2342         q = bdev_get_queue(pd->bdev);
2343         if (write) {
2344                 if ((ret = pkt_open_write(pd)))
2345                         goto out_unclaim;
2346                 /*
2347                  * Some CDRW drives can not handle writes larger than one packet,
2348                  * even if the size is a multiple of the packet size.
2349                  */
2350                 spin_lock_irq(q->queue_lock);
2351                 blk_queue_max_sectors(q, pd->settings.size);
2352                 spin_unlock_irq(q->queue_lock);
2353                 set_bit(PACKET_WRITABLE, &pd->flags);
2354         } else {
2355                 pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2356                 clear_bit(PACKET_WRITABLE, &pd->flags);
2357         }
2358
2359         if ((ret = pkt_set_segment_merging(pd, q)))
2360                 goto out_unclaim;
2361
2362         if (write) {
2363                 if (!pkt_grow_pktlist(pd, CONFIG_CDROM_PKTCDVD_BUFFERS)) {
2364                         printk(DRIVER_NAME": not enough memory for buffers\n");
2365                         ret = -ENOMEM;
2366                         goto out_unclaim;
2367                 }
2368                 printk(DRIVER_NAME": %lukB available on disc\n", lba << 1);
2369         }
2370
2371         return 0;
2372
2373 out_unclaim:
2374         bd_release(pd->bdev);
2375 out_putdev:
2376         blkdev_put(pd->bdev);
2377 out:
2378         return ret;
2379 }
2380
2381 /*
2382  * called when the device is closed. makes sure that the device flushes
2383  * the internal cache before we close.
2384  */
2385 static void pkt_release_dev(struct pktcdvd_device *pd, int flush)
2386 {
2387         if (flush && pkt_flush_cache(pd))
2388                 DPRINTK(DRIVER_NAME": %s not flushing cache\n", pd->name);
2389
2390         pkt_lock_door(pd, 0);
2391
2392         pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2393         bd_release(pd->bdev);
2394         blkdev_put(pd->bdev);
2395
2396         pkt_shrink_pktlist(pd);
2397 }
2398
2399 static struct pktcdvd_device *pkt_find_dev_from_minor(int dev_minor)
2400 {
2401         if (dev_minor >= MAX_WRITERS)
2402                 return NULL;
2403         return pkt_devs[dev_minor];
2404 }
2405
2406 static int pkt_open(struct inode *inode, struct file *file)
2407 {
2408         struct pktcdvd_device *pd = NULL;
2409         int ret;
2410
2411         VPRINTK(DRIVER_NAME": entering open\n");
2412
2413         mutex_lock(&ctl_mutex);
2414         pd = pkt_find_dev_from_minor(iminor(inode));
2415         if (!pd) {
2416                 ret = -ENODEV;
2417                 goto out;
2418         }
2419         BUG_ON(pd->refcnt < 0);
2420
2421         pd->refcnt++;
2422         if (pd->refcnt > 1) {
2423                 if ((file->f_mode & FMODE_WRITE) &&
2424                     !test_bit(PACKET_WRITABLE, &pd->flags)) {
2425                         ret = -EBUSY;
2426                         goto out_dec;
2427                 }
2428         } else {
2429                 ret = pkt_open_dev(pd, file->f_mode & FMODE_WRITE);
2430                 if (ret)
2431                         goto out_dec;
2432                 /*
2433                  * needed here as well, since ext2 (among others) may change
2434                  * the blocksize at mount time
2435                  */
2436                 set_blocksize(inode->i_bdev, CD_FRAMESIZE);
2437         }
2438
2439         mutex_unlock(&ctl_mutex);
2440         return 0;
2441
2442 out_dec:
2443         pd->refcnt--;
2444 out:
2445         VPRINTK(DRIVER_NAME": failed open (%d)\n", ret);
2446         mutex_unlock(&ctl_mutex);
2447         return ret;
2448 }
2449
2450 static int pkt_close(struct inode *inode, struct file *file)
2451 {
2452         struct pktcdvd_device *pd = inode->i_bdev->bd_disk->private_data;
2453         int ret = 0;
2454
2455         mutex_lock(&ctl_mutex);
2456         pd->refcnt--;
2457         BUG_ON(pd->refcnt < 0);
2458         if (pd->refcnt == 0) {
2459                 int flush = test_bit(PACKET_WRITABLE, &pd->flags);
2460                 pkt_release_dev(pd, flush);
2461         }
2462         mutex_unlock(&ctl_mutex);
2463         return ret;
2464 }
2465
2466
2467 static int pkt_end_io_read_cloned(struct bio *bio, unsigned int bytes_done, int err)
2468 {
2469         struct packet_stacked_data *psd = bio->bi_private;
2470         struct pktcdvd_device *pd = psd->pd;
2471
2472         if (bio->bi_size)
2473                 return 1;
2474
2475         bio_put(bio);
2476         bio_endio(psd->bio, psd->bio->bi_size, err);
2477         mempool_free(psd, psd_pool);
2478         pkt_bio_finished(pd);
2479         return 0;
2480 }
2481
2482 static int pkt_make_request(request_queue_t *q, struct bio *bio)
2483 {
2484         struct pktcdvd_device *pd;
2485         char b[BDEVNAME_SIZE];
2486         sector_t zone;
2487         struct packet_data *pkt;
2488         int was_empty, blocked_bio;
2489         struct pkt_rb_node *node;
2490
2491         pd = q->queuedata;
2492         if (!pd) {
2493                 printk(DRIVER_NAME": %s incorrect request queue\n", bdevname(bio->bi_bdev, b));
2494                 goto end_io;
2495         }
2496
2497         /*
2498          * Clone READ bios so we can have our own bi_end_io callback.
2499          */
2500         if (bio_data_dir(bio) == READ) {
2501                 struct bio *cloned_bio = bio_clone(bio, GFP_NOIO);
2502                 struct packet_stacked_data *psd = mempool_alloc(psd_pool, GFP_NOIO);
2503
2504                 psd->pd = pd;
2505                 psd->bio = bio;
2506                 cloned_bio->bi_bdev = pd->bdev;
2507                 cloned_bio->bi_private = psd;
2508                 cloned_bio->bi_end_io = pkt_end_io_read_cloned;
2509                 pd->stats.secs_r += bio->bi_size >> 9;
2510                 pkt_queue_bio(pd, cloned_bio);
2511                 return 0;
2512         }
2513
2514         if (!test_bit(PACKET_WRITABLE, &pd->flags)) {
2515                 printk(DRIVER_NAME": WRITE for ro device %s (%llu)\n",
2516                         pd->name, (unsigned long long)bio->bi_sector);
2517                 goto end_io;
2518         }
2519
2520         if (!bio->bi_size || (bio->bi_size % CD_FRAMESIZE)) {
2521                 printk(DRIVER_NAME": wrong bio size\n");
2522                 goto end_io;
2523         }
2524
2525         blk_queue_bounce(q, &bio);
2526
2527         zone = ZONE(bio->bi_sector, pd);
2528         VPRINTK("pkt_make_request: start = %6llx stop = %6llx\n",
2529                 (unsigned long long)bio->bi_sector,
2530                 (unsigned long long)(bio->bi_sector + bio_sectors(bio)));
2531
2532         /* Check if we have to split the bio */
2533         {
2534                 struct bio_pair *bp;
2535                 sector_t last_zone;
2536                 int first_sectors;
2537
2538                 last_zone = ZONE(bio->bi_sector + bio_sectors(bio) - 1, pd);
2539                 if (last_zone != zone) {
2540                         BUG_ON(last_zone != zone + pd->settings.size);
2541                         first_sectors = last_zone - bio->bi_sector;
2542                         bp = bio_split(bio, bio_split_pool, first_sectors);
2543                         BUG_ON(!bp);
2544                         pkt_make_request(q, &bp->bio1);
2545                         pkt_make_request(q, &bp->bio2);
2546                         bio_pair_release(bp);
2547                         return 0;
2548                 }
2549         }
2550
2551         /*
2552          * If we find a matching packet in state WAITING or READ_WAIT, we can
2553          * just append this bio to that packet.
2554          */
2555         spin_lock(&pd->cdrw.active_list_lock);
2556         blocked_bio = 0;
2557         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
2558                 if (pkt->sector == zone) {
2559                         spin_lock(&pkt->lock);
2560                         if ((pkt->state == PACKET_WAITING_STATE) ||
2561                             (pkt->state == PACKET_READ_WAIT_STATE)) {
2562                                 pkt_add_list_last(bio, &pkt->orig_bios,
2563                                                   &pkt->orig_bios_tail);
2564                                 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
2565                                 if ((pkt->write_size >= pkt->frames) &&
2566                                     (pkt->state == PACKET_WAITING_STATE)) {
2567                                         atomic_inc(&pkt->run_sm);
2568                                         wake_up(&pd->wqueue);
2569                                 }
2570                                 spin_unlock(&pkt->lock);
2571                                 spin_unlock(&pd->cdrw.active_list_lock);
2572                                 return 0;
2573                         } else {
2574                                 blocked_bio = 1;
2575                         }
2576                         spin_unlock(&pkt->lock);
2577                 }
2578         }
2579         spin_unlock(&pd->cdrw.active_list_lock);
2580
2581         /*
2582          * Test if there is enough room left in the bio work queue
2583          * (queue size >= congestion on mark).
2584          * If not, wait till the work queue size is below the congestion off mark.
2585          */
2586         spin_lock(&pd->lock);
2587         if (pd->write_congestion_on > 0
2588             && pd->bio_queue_size >= pd->write_congestion_on) {
2589                 set_bdi_congested(&q->backing_dev_info, WRITE);
2590                 do {
2591                         spin_unlock(&pd->lock);
2592                         congestion_wait(WRITE, HZ);
2593                         spin_lock(&pd->lock);
2594                 } while(pd->bio_queue_size > pd->write_congestion_off);
2595         }
2596         spin_unlock(&pd->lock);
2597
2598         /*
2599          * No matching packet found. Store the bio in the work queue.
2600          */
2601         node = mempool_alloc(pd->rb_pool, GFP_NOIO);
2602         node->bio = bio;
2603         spin_lock(&pd->lock);
2604         BUG_ON(pd->bio_queue_size < 0);
2605         was_empty = (pd->bio_queue_size == 0);
2606         pkt_rbtree_insert(pd, node);
2607         spin_unlock(&pd->lock);
2608
2609         /*
2610          * Wake up the worker thread.
2611          */
2612         atomic_set(&pd->scan_queue, 1);
2613         if (was_empty) {
2614                 /* This wake_up is required for correct operation */
2615                 wake_up(&pd->wqueue);
2616         } else if (!list_empty(&pd->cdrw.pkt_free_list) && !blocked_bio) {
2617                 /*
2618                  * This wake up is not required for correct operation,
2619                  * but improves performance in some cases.
2620                  */
2621                 wake_up(&pd->wqueue);
2622         }
2623         return 0;
2624 end_io:
2625         bio_io_error(bio, bio->bi_size);
2626         return 0;
2627 }
2628
2629
2630
2631 static int pkt_merge_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *bvec)
2632 {
2633         struct pktcdvd_device *pd = q->queuedata;
2634         sector_t zone = ZONE(bio->bi_sector, pd);
2635         int used = ((bio->bi_sector - zone) << 9) + bio->bi_size;
2636         int remaining = (pd->settings.size << 9) - used;
2637         int remaining2;
2638
2639         /*
2640          * A bio <= PAGE_SIZE must be allowed. If it crosses a packet
2641          * boundary, pkt_make_request() will split the bio.
2642          */
2643         remaining2 = PAGE_SIZE - bio->bi_size;
2644         remaining = max(remaining, remaining2);
2645
2646         BUG_ON(remaining < 0);
2647         return remaining;
2648 }
2649
2650 static void pkt_init_queue(struct pktcdvd_device *pd)
2651 {
2652         request_queue_t *q = pd->disk->queue;
2653
2654         blk_queue_make_request(q, pkt_make_request);
2655         blk_queue_hardsect_size(q, CD_FRAMESIZE);
2656         blk_queue_max_sectors(q, PACKET_MAX_SECTORS);
2657         blk_queue_merge_bvec(q, pkt_merge_bvec);
2658         q->queuedata = pd;
2659 }
2660
2661 static int pkt_seq_show(struct seq_file *m, void *p)
2662 {
2663         struct pktcdvd_device *pd = m->private;
2664         char *msg;
2665         char bdev_buf[BDEVNAME_SIZE];
2666         int states[PACKET_NUM_STATES];
2667
2668         seq_printf(m, "Writer %s mapped to %s:\n", pd->name,
2669                    bdevname(pd->bdev, bdev_buf));
2670
2671         seq_printf(m, "\nSettings:\n");
2672         seq_printf(m, "\tpacket size:\t\t%dkB\n", pd->settings.size / 2);
2673
2674         if (pd->settings.write_type == 0)
2675                 msg = "Packet";
2676         else
2677                 msg = "Unknown";
2678         seq_printf(m, "\twrite type:\t\t%s\n", msg);
2679
2680         seq_printf(m, "\tpacket type:\t\t%s\n", pd->settings.fp ? "Fixed" : "Variable");
2681         seq_printf(m, "\tlink loss:\t\t%d\n", pd->settings.link_loss);
2682
2683         seq_printf(m, "\ttrack mode:\t\t%d\n", pd->settings.track_mode);
2684
2685         if (pd->settings.block_mode == PACKET_BLOCK_MODE1)
2686                 msg = "Mode 1";
2687         else if (pd->settings.block_mode == PACKET_BLOCK_MODE2)
2688                 msg = "Mode 2";
2689         else
2690                 msg = "Unknown";
2691         seq_printf(m, "\tblock mode:\t\t%s\n", msg);
2692
2693         seq_printf(m, "\nStatistics:\n");
2694         seq_printf(m, "\tpackets started:\t%lu\n", pd->stats.pkt_started);
2695         seq_printf(m, "\tpackets ended:\t\t%lu\n", pd->stats.pkt_ended);
2696         seq_printf(m, "\twritten:\t\t%lukB\n", pd->stats.secs_w >> 1);
2697         seq_printf(m, "\tread gather:\t\t%lukB\n", pd->stats.secs_rg >> 1);
2698         seq_printf(m, "\tread:\t\t\t%lukB\n", pd->stats.secs_r >> 1);
2699
2700         seq_printf(m, "\nMisc:\n");
2701         seq_printf(m, "\treference count:\t%d\n", pd->refcnt);
2702         seq_printf(m, "\tflags:\t\t\t0x%lx\n", pd->flags);
2703         seq_printf(m, "\tread speed:\t\t%ukB/s\n", pd->read_speed);
2704         seq_printf(m, "\twrite speed:\t\t%ukB/s\n", pd->write_speed);
2705         seq_printf(m, "\tstart offset:\t\t%lu\n", pd->offset);
2706         seq_printf(m, "\tmode page offset:\t%u\n", pd->mode_offset);
2707
2708         seq_printf(m, "\nQueue state:\n");
2709         seq_printf(m, "\tbios queued:\t\t%d\n", pd->bio_queue_size);
2710         seq_printf(m, "\tbios pending:\t\t%d\n", atomic_read(&pd->cdrw.pending_bios));
2711         seq_printf(m, "\tcurrent sector:\t\t0x%llx\n", (unsigned long long)pd->current_sector);
2712
2713         pkt_count_states(pd, states);
2714         seq_printf(m, "\tstate:\t\t\ti:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
2715                    states[0], states[1], states[2], states[3], states[4], states[5]);
2716
2717         seq_printf(m, "\twrite congestion marks:\toff=%d on=%d\n",
2718                         pd->write_congestion_off,
2719                         pd->write_congestion_on);
2720         return 0;
2721 }
2722
2723 static int pkt_seq_open(struct inode *inode, struct file *file)
2724 {
2725         return single_open(file, pkt_seq_show, PDE(inode)->data);
2726 }
2727
2728 static const struct file_operations pkt_proc_fops = {
2729         .open   = pkt_seq_open,
2730         .read   = seq_read,
2731         .llseek = seq_lseek,
2732         .release = single_release
2733 };
2734
2735 static int pkt_new_dev(struct pktcdvd_device *pd, dev_t dev)
2736 {
2737         int i;
2738         int ret = 0;
2739         char b[BDEVNAME_SIZE];
2740         struct proc_dir_entry *proc;
2741         struct block_device *bdev;
2742
2743         if (pd->pkt_dev == dev) {
2744                 printk(DRIVER_NAME": Recursive setup not allowed\n");
2745                 return -EBUSY;
2746         }
2747         for (i = 0; i < MAX_WRITERS; i++) {
2748                 struct pktcdvd_device *pd2 = pkt_devs[i];
2749                 if (!pd2)
2750                         continue;
2751                 if (pd2->bdev->bd_dev == dev) {
2752                         printk(DRIVER_NAME": %s already setup\n", bdevname(pd2->bdev, b));
2753                         return -EBUSY;
2754                 }
2755                 if (pd2->pkt_dev == dev) {
2756                         printk(DRIVER_NAME": Can't chain pktcdvd devices\n");
2757                         return -EBUSY;
2758                 }
2759         }
2760
2761         bdev = bdget(dev);
2762         if (!bdev)
2763                 return -ENOMEM;
2764         ret = blkdev_get(bdev, FMODE_READ, O_RDONLY | O_NONBLOCK);
2765         if (ret)
2766                 return ret;
2767
2768         /* This is safe, since we have a reference from open(). */
2769         __module_get(THIS_MODULE);
2770
2771         pd->bdev = bdev;
2772         set_blocksize(bdev, CD_FRAMESIZE);
2773
2774         pkt_init_queue(pd);
2775
2776         atomic_set(&pd->cdrw.pending_bios, 0);
2777         pd->cdrw.thread = kthread_run(kcdrwd, pd, "%s", pd->name);
2778         if (IS_ERR(pd->cdrw.thread)) {
2779                 printk(DRIVER_NAME": can't start kernel thread\n");
2780                 ret = -ENOMEM;
2781                 goto out_mem;
2782         }
2783
2784         proc = create_proc_entry(pd->name, 0, pkt_proc);
2785         if (proc) {
2786                 proc->data = pd;
2787                 proc->proc_fops = &pkt_proc_fops;
2788         }
2789         DPRINTK(DRIVER_NAME": writer %s mapped to %s\n", pd->name, bdevname(bdev, b));
2790         return 0;
2791
2792 out_mem:
2793         blkdev_put(bdev);
2794         /* This is safe: open() is still holding a reference. */
2795         module_put(THIS_MODULE);
2796         return ret;
2797 }
2798
2799 static int pkt_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg)
2800 {
2801         struct pktcdvd_device *pd = inode->i_bdev->bd_disk->private_data;
2802
2803         VPRINTK("pkt_ioctl: cmd %x, dev %d:%d\n", cmd, imajor(inode), iminor(inode));
2804
2805         switch (cmd) {
2806         /*
2807          * forward selected CDROM ioctls to CD-ROM, for UDF
2808          */
2809         case CDROMMULTISESSION:
2810         case CDROMREADTOCENTRY:
2811         case CDROM_LAST_WRITTEN:
2812         case CDROM_SEND_PACKET:
2813         case SCSI_IOCTL_SEND_COMMAND:
2814                 return blkdev_ioctl(pd->bdev->bd_inode, file, cmd, arg);
2815
2816         case CDROMEJECT:
2817                 /*
2818                  * The door gets locked when the device is opened, so we
2819                  * have to unlock it or else the eject command fails.
2820                  */
2821                 if (pd->refcnt == 1)
2822                         pkt_lock_door(pd, 0);
2823                 return blkdev_ioctl(pd->bdev->bd_inode, file, cmd, arg);
2824
2825         default:
2826                 VPRINTK(DRIVER_NAME": Unknown ioctl for %s (%x)\n", pd->name, cmd);
2827                 return -ENOTTY;
2828         }
2829
2830         return 0;
2831 }
2832
2833 static int pkt_media_changed(struct gendisk *disk)
2834 {
2835         struct pktcdvd_device *pd = disk->private_data;
2836         struct gendisk *attached_disk;
2837
2838         if (!pd)
2839                 return 0;
2840         if (!pd->bdev)
2841                 return 0;
2842         attached_disk = pd->bdev->bd_disk;
2843         if (!attached_disk)
2844                 return 0;
2845         return attached_disk->fops->media_changed(attached_disk);
2846 }
2847
2848 static struct block_device_operations pktcdvd_ops = {
2849         .owner =                THIS_MODULE,
2850         .open =                 pkt_open,
2851         .release =              pkt_close,
2852         .ioctl =                pkt_ioctl,
2853         .media_changed =        pkt_media_changed,
2854 };
2855
2856 /*
2857  * Set up mapping from pktcdvd device to CD-ROM device.
2858  */
2859 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev)
2860 {
2861         int idx;
2862         int ret = -ENOMEM;
2863         struct pktcdvd_device *pd;
2864         struct gendisk *disk;
2865
2866         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2867
2868         for (idx = 0; idx < MAX_WRITERS; idx++)
2869                 if (!pkt_devs[idx])
2870                         break;
2871         if (idx == MAX_WRITERS) {
2872                 printk(DRIVER_NAME": max %d writers supported\n", MAX_WRITERS);
2873                 ret = -EBUSY;
2874                 goto out_mutex;
2875         }
2876
2877         pd = kzalloc(sizeof(struct pktcdvd_device), GFP_KERNEL);
2878         if (!pd)
2879                 goto out_mutex;
2880
2881         pd->rb_pool = mempool_create_kmalloc_pool(PKT_RB_POOL_SIZE,
2882                                                   sizeof(struct pkt_rb_node));
2883         if (!pd->rb_pool)
2884                 goto out_mem;
2885
2886         INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
2887         INIT_LIST_HEAD(&pd->cdrw.pkt_active_list);
2888         spin_lock_init(&pd->cdrw.active_list_lock);
2889
2890         spin_lock_init(&pd->lock);
2891         spin_lock_init(&pd->iosched.lock);
2892         sprintf(pd->name, DRIVER_NAME"%d", idx);
2893         init_waitqueue_head(&pd->wqueue);
2894         pd->bio_queue = RB_ROOT;
2895
2896         pd->write_congestion_on  = write_congestion_on;
2897         pd->write_congestion_off = write_congestion_off;
2898
2899         disk = alloc_disk(1);
2900         if (!disk)
2901                 goto out_mem;
2902         pd->disk = disk;
2903         disk->major = pktdev_major;
2904         disk->first_minor = idx;
2905         disk->fops = &pktcdvd_ops;
2906         disk->flags = GENHD_FL_REMOVABLE;
2907         strcpy(disk->disk_name, pd->name);
2908         disk->private_data = pd;
2909         disk->queue = blk_alloc_queue(GFP_KERNEL);
2910         if (!disk->queue)
2911                 goto out_mem2;
2912
2913         pd->pkt_dev = MKDEV(disk->major, disk->first_minor);
2914         ret = pkt_new_dev(pd, dev);
2915         if (ret)
2916                 goto out_new_dev;
2917
2918         add_disk(disk);
2919
2920         pkt_sysfs_dev_new(pd);
2921         pkt_debugfs_dev_new(pd);
2922
2923         pkt_devs[idx] = pd;
2924         if (pkt_dev)
2925                 *pkt_dev = pd->pkt_dev;
2926
2927         mutex_unlock(&ctl_mutex);
2928         return 0;
2929
2930 out_new_dev:
2931         blk_cleanup_queue(disk->queue);
2932 out_mem2:
2933         put_disk(disk);
2934 out_mem:
2935         if (pd->rb_pool)
2936                 mempool_destroy(pd->rb_pool);
2937         kfree(pd);
2938 out_mutex:
2939         mutex_unlock(&ctl_mutex);
2940         printk(DRIVER_NAME": setup of pktcdvd device failed\n");
2941         return ret;
2942 }
2943
2944 /*
2945  * Tear down mapping from pktcdvd device to CD-ROM device.
2946  */
2947 static int pkt_remove_dev(dev_t pkt_dev)
2948 {
2949         struct pktcdvd_device *pd;
2950         int idx;
2951         int ret = 0;
2952
2953         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2954
2955         for (idx = 0; idx < MAX_WRITERS; idx++) {
2956                 pd = pkt_devs[idx];
2957                 if (pd && (pd->pkt_dev == pkt_dev))
2958                         break;
2959         }
2960         if (idx == MAX_WRITERS) {
2961                 DPRINTK(DRIVER_NAME": dev not setup\n");
2962                 ret = -ENXIO;
2963                 goto out;
2964         }
2965
2966         if (pd->refcnt > 0) {
2967                 ret = -EBUSY;
2968                 goto out;
2969         }
2970         if (!IS_ERR(pd->cdrw.thread))
2971                 kthread_stop(pd->cdrw.thread);
2972
2973         pkt_devs[idx] = NULL;
2974
2975         pkt_debugfs_dev_remove(pd);
2976         pkt_sysfs_dev_remove(pd);
2977
2978         blkdev_put(pd->bdev);
2979
2980         remove_proc_entry(pd->name, pkt_proc);
2981         DPRINTK(DRIVER_NAME": writer %s unmapped\n", pd->name);
2982
2983         del_gendisk(pd->disk);
2984         blk_cleanup_queue(pd->disk->queue);
2985         put_disk(pd->disk);
2986
2987         mempool_destroy(pd->rb_pool);
2988         kfree(pd);
2989
2990         /* This is safe: open() is still holding a reference. */
2991         module_put(THIS_MODULE);
2992
2993 out:
2994         mutex_unlock(&ctl_mutex);
2995         return ret;
2996 }
2997
2998 static void pkt_get_status(struct pkt_ctrl_command *ctrl_cmd)
2999 {
3000         struct pktcdvd_device *pd;
3001
3002         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3003
3004         pd = pkt_find_dev_from_minor(ctrl_cmd->dev_index);
3005         if (pd) {
3006                 ctrl_cmd->dev = new_encode_dev(pd->bdev->bd_dev);
3007                 ctrl_cmd->pkt_dev = new_encode_dev(pd->pkt_dev);
3008         } else {
3009                 ctrl_cmd->dev = 0;
3010                 ctrl_cmd->pkt_dev = 0;
3011         }
3012         ctrl_cmd->num_devices = MAX_WRITERS;
3013
3014         mutex_unlock(&ctl_mutex);
3015 }
3016
3017 static int pkt_ctl_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg)
3018 {
3019         void __user *argp = (void __user *)arg;
3020         struct pkt_ctrl_command ctrl_cmd;
3021         int ret = 0;
3022         dev_t pkt_dev = 0;
3023
3024         if (cmd != PACKET_CTRL_CMD)
3025                 return -ENOTTY;
3026
3027         if (copy_from_user(&ctrl_cmd, argp, sizeof(struct pkt_ctrl_command)))
3028                 return -EFAULT;
3029
3030         switch (ctrl_cmd.command) {
3031         case PKT_CTRL_CMD_SETUP:
3032                 if (!capable(CAP_SYS_ADMIN))
3033                         return -EPERM;
3034                 ret = pkt_setup_dev(new_decode_dev(ctrl_cmd.dev), &pkt_dev);
3035                 ctrl_cmd.pkt_dev = new_encode_dev(pkt_dev);
3036                 break;
3037         case PKT_CTRL_CMD_TEARDOWN:
3038                 if (!capable(CAP_SYS_ADMIN))
3039                         return -EPERM;
3040                 ret = pkt_remove_dev(new_decode_dev(ctrl_cmd.pkt_dev));
3041                 break;
3042         case PKT_CTRL_CMD_STATUS:
3043                 pkt_get_status(&ctrl_cmd);
3044                 break;
3045         default:
3046                 return -ENOTTY;
3047         }
3048
3049         if (copy_to_user(argp, &ctrl_cmd, sizeof(struct pkt_ctrl_command)))
3050                 return -EFAULT;
3051         return ret;
3052 }
3053
3054
3055 static const struct file_operations pkt_ctl_fops = {
3056         .ioctl   = pkt_ctl_ioctl,
3057         .owner   = THIS_MODULE,
3058 };
3059
3060 static struct miscdevice pkt_misc = {
3061         .minor          = MISC_DYNAMIC_MINOR,
3062         .name           = DRIVER_NAME,
3063         .fops           = &pkt_ctl_fops
3064 };
3065
3066 static int __init pkt_init(void)
3067 {
3068         int ret;
3069
3070         mutex_init(&ctl_mutex);
3071
3072         psd_pool = mempool_create_kmalloc_pool(PSD_POOL_SIZE,
3073                                         sizeof(struct packet_stacked_data));
3074         if (!psd_pool)
3075                 return -ENOMEM;
3076
3077         ret = register_blkdev(pktdev_major, DRIVER_NAME);
3078         if (ret < 0) {
3079                 printk(DRIVER_NAME": Unable to register block device\n");
3080                 goto out2;
3081         }
3082         if (!pktdev_major)
3083                 pktdev_major = ret;
3084
3085         ret = pkt_sysfs_init();
3086         if (ret)
3087                 goto out;
3088
3089         pkt_debugfs_init();
3090
3091         ret = misc_register(&pkt_misc);
3092         if (ret) {
3093                 printk(DRIVER_NAME": Unable to register misc device\n");
3094                 goto out_misc;
3095         }
3096
3097         pkt_proc = proc_mkdir(DRIVER_NAME, proc_root_driver);
3098
3099         return 0;
3100
3101 out_misc:
3102         pkt_debugfs_cleanup();
3103         pkt_sysfs_cleanup();
3104 out:
3105         unregister_blkdev(pktdev_major, DRIVER_NAME);
3106 out2:
3107         mempool_destroy(psd_pool);
3108         return ret;
3109 }
3110
3111 static void __exit pkt_exit(void)
3112 {
3113         remove_proc_entry(DRIVER_NAME, proc_root_driver);
3114         misc_deregister(&pkt_misc);
3115
3116         pkt_debugfs_cleanup();
3117         pkt_sysfs_cleanup();
3118
3119         unregister_blkdev(pktdev_major, DRIVER_NAME);
3120         mempool_destroy(psd_pool);
3121 }
3122
3123 MODULE_DESCRIPTION("Packet writing layer for CD/DVD drives");
3124 MODULE_AUTHOR("Jens Axboe <axboe@suse.de>");
3125 MODULE_LICENSE("GPL");
3126
3127 module_init(pkt_init);
3128 module_exit(pkt_exit);