Merge branch 'master' of git://git.kernel.org/pub/scm/linux/kernel/git/jk/spufs into...
[linux-2.6] / 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         int error;
114
115         p = kzalloc(sizeof(*p), GFP_KERNEL);
116         if (!p)
117                 return NULL;
118         p->pd = pd;
119         error = kobject_init_and_add(&p->kobj, ktype, parent, "%s", name);
120         if (error) {
121                 kobject_put(&p->kobj);
122                 return NULL;
123         }
124         kobject_uevent(&p->kobj, KOBJ_ADD);
125         return p;
126 }
127 /*
128  * remove a pktcdvd kernel object.
129  */
130 static void pkt_kobj_remove(struct pktcdvd_kobj *p)
131 {
132         if (p)
133                 kobject_put(&p->kobj);
134 }
135 /*
136  * default release function for pktcdvd kernel objects.
137  */
138 static void pkt_kobj_release(struct kobject *kobj)
139 {
140         kfree(to_pktcdvdkobj(kobj));
141 }
142
143
144 /**********************************************************
145  *
146  * sysfs interface for pktcdvd
147  * by (C) 2006  Thomas Maier <balagi@justmail.de>
148  *
149  **********************************************************/
150
151 #define DEF_ATTR(_obj,_name,_mode) \
152         static struct attribute _obj = { .name = _name, .mode = _mode }
153
154 /**********************************************************
155   /sys/class/pktcdvd/pktcdvd[0-7]/
156                      stat/reset
157                      stat/packets_started
158                      stat/packets_finished
159                      stat/kb_written
160                      stat/kb_read
161                      stat/kb_read_gather
162                      write_queue/size
163                      write_queue/congestion_off
164                      write_queue/congestion_on
165  **********************************************************/
166
167 DEF_ATTR(kobj_pkt_attr_st1, "reset", 0200);
168 DEF_ATTR(kobj_pkt_attr_st2, "packets_started", 0444);
169 DEF_ATTR(kobj_pkt_attr_st3, "packets_finished", 0444);
170 DEF_ATTR(kobj_pkt_attr_st4, "kb_written", 0444);
171 DEF_ATTR(kobj_pkt_attr_st5, "kb_read", 0444);
172 DEF_ATTR(kobj_pkt_attr_st6, "kb_read_gather", 0444);
173
174 static struct attribute *kobj_pkt_attrs_stat[] = {
175         &kobj_pkt_attr_st1,
176         &kobj_pkt_attr_st2,
177         &kobj_pkt_attr_st3,
178         &kobj_pkt_attr_st4,
179         &kobj_pkt_attr_st5,
180         &kobj_pkt_attr_st6,
181         NULL
182 };
183
184 DEF_ATTR(kobj_pkt_attr_wq1, "size", 0444);
185 DEF_ATTR(kobj_pkt_attr_wq2, "congestion_off", 0644);
186 DEF_ATTR(kobj_pkt_attr_wq3, "congestion_on",  0644);
187
188 static struct attribute *kobj_pkt_attrs_wqueue[] = {
189         &kobj_pkt_attr_wq1,
190         &kobj_pkt_attr_wq2,
191         &kobj_pkt_attr_wq3,
192         NULL
193 };
194
195 static ssize_t kobj_pkt_show(struct kobject *kobj,
196                         struct attribute *attr, char *data)
197 {
198         struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
199         int n = 0;
200         int v;
201         if (strcmp(attr->name, "packets_started") == 0) {
202                 n = sprintf(data, "%lu\n", pd->stats.pkt_started);
203
204         } else if (strcmp(attr->name, "packets_finished") == 0) {
205                 n = sprintf(data, "%lu\n", pd->stats.pkt_ended);
206
207         } else if (strcmp(attr->name, "kb_written") == 0) {
208                 n = sprintf(data, "%lu\n", pd->stats.secs_w >> 1);
209
210         } else if (strcmp(attr->name, "kb_read") == 0) {
211                 n = sprintf(data, "%lu\n", pd->stats.secs_r >> 1);
212
213         } else if (strcmp(attr->name, "kb_read_gather") == 0) {
214                 n = sprintf(data, "%lu\n", pd->stats.secs_rg >> 1);
215
216         } else if (strcmp(attr->name, "size") == 0) {
217                 spin_lock(&pd->lock);
218                 v = pd->bio_queue_size;
219                 spin_unlock(&pd->lock);
220                 n = sprintf(data, "%d\n", v);
221
222         } else if (strcmp(attr->name, "congestion_off") == 0) {
223                 spin_lock(&pd->lock);
224                 v = pd->write_congestion_off;
225                 spin_unlock(&pd->lock);
226                 n = sprintf(data, "%d\n", v);
227
228         } else if (strcmp(attr->name, "congestion_on") == 0) {
229                 spin_lock(&pd->lock);
230                 v = pd->write_congestion_on;
231                 spin_unlock(&pd->lock);
232                 n = sprintf(data, "%d\n", v);
233         }
234         return n;
235 }
236
237 static void init_write_congestion_marks(int* lo, int* hi)
238 {
239         if (*hi > 0) {
240                 *hi = max(*hi, 500);
241                 *hi = min(*hi, 1000000);
242                 if (*lo <= 0)
243                         *lo = *hi - 100;
244                 else {
245                         *lo = min(*lo, *hi - 100);
246                         *lo = max(*lo, 100);
247                 }
248         } else {
249                 *hi = -1;
250                 *lo = -1;
251         }
252 }
253
254 static ssize_t kobj_pkt_store(struct kobject *kobj,
255                         struct attribute *attr,
256                         const char *data, size_t len)
257 {
258         struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
259         int val;
260
261         if (strcmp(attr->name, "reset") == 0 && len > 0) {
262                 pd->stats.pkt_started = 0;
263                 pd->stats.pkt_ended = 0;
264                 pd->stats.secs_w = 0;
265                 pd->stats.secs_rg = 0;
266                 pd->stats.secs_r = 0;
267
268         } else if (strcmp(attr->name, "congestion_off") == 0
269                    && sscanf(data, "%d", &val) == 1) {
270                 spin_lock(&pd->lock);
271                 pd->write_congestion_off = val;
272                 init_write_congestion_marks(&pd->write_congestion_off,
273                                         &pd->write_congestion_on);
274                 spin_unlock(&pd->lock);
275
276         } else if (strcmp(attr->name, "congestion_on") == 0
277                    && sscanf(data, "%d", &val) == 1) {
278                 spin_lock(&pd->lock);
279                 pd->write_congestion_on = val;
280                 init_write_congestion_marks(&pd->write_congestion_off,
281                                         &pd->write_congestion_on);
282                 spin_unlock(&pd->lock);
283         }
284         return len;
285 }
286
287 static struct sysfs_ops kobj_pkt_ops = {
288         .show = kobj_pkt_show,
289         .store = kobj_pkt_store
290 };
291 static struct kobj_type kobj_pkt_type_stat = {
292         .release = pkt_kobj_release,
293         .sysfs_ops = &kobj_pkt_ops,
294         .default_attrs = kobj_pkt_attrs_stat
295 };
296 static struct kobj_type kobj_pkt_type_wqueue = {
297         .release = pkt_kobj_release,
298         .sysfs_ops = &kobj_pkt_ops,
299         .default_attrs = kobj_pkt_attrs_wqueue
300 };
301
302 static void pkt_sysfs_dev_new(struct pktcdvd_device *pd)
303 {
304         if (class_pktcdvd) {
305                 pd->dev = device_create(class_pktcdvd, NULL, pd->pkt_dev, "%s", pd->name);
306                 if (IS_ERR(pd->dev))
307                         pd->dev = NULL;
308         }
309         if (pd->dev) {
310                 pd->kobj_stat = pkt_kobj_create(pd, "stat",
311                                         &pd->dev->kobj,
312                                         &kobj_pkt_type_stat);
313                 pd->kobj_wqueue = pkt_kobj_create(pd, "write_queue",
314                                         &pd->dev->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                 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
363         if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
364                 /* pkt_setup_dev() expects caller to hold reference to self */
365                 if (!try_module_get(THIS_MODULE))
366                         return -ENODEV;
367
368                 pkt_setup_dev(MKDEV(major, minor), NULL);
369
370                 module_put(THIS_MODULE);
371
372                 return count;
373         }
374
375         return -EINVAL;
376 }
377
378 static ssize_t class_pktcdvd_store_remove(struct class *c, const char *buf,
379                                         size_t count)
380 {
381         unsigned int major, minor;
382         if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
383                 pkt_remove_dev(MKDEV(major, minor));
384                 return count;
385         }
386         return -EINVAL;
387 }
388
389 static struct class_attribute class_pktcdvd_attrs[] = {
390  __ATTR(add,            0200, NULL, class_pktcdvd_store_add),
391  __ATTR(remove,         0200, NULL, class_pktcdvd_store_remove),
392  __ATTR(device_map,     0444, class_pktcdvd_show_map, NULL),
393  __ATTR_NULL
394 };
395
396
397 static int pkt_sysfs_init(void)
398 {
399         int ret = 0;
400
401         /*
402          * create control files in sysfs
403          * /sys/class/pktcdvd/...
404          */
405         class_pktcdvd = kzalloc(sizeof(*class_pktcdvd), GFP_KERNEL);
406         if (!class_pktcdvd)
407                 return -ENOMEM;
408         class_pktcdvd->name = DRIVER_NAME;
409         class_pktcdvd->owner = THIS_MODULE;
410         class_pktcdvd->class_release = class_pktcdvd_release;
411         class_pktcdvd->class_attrs = class_pktcdvd_attrs;
412         ret = class_register(class_pktcdvd);
413         if (ret) {
414                 kfree(class_pktcdvd);
415                 class_pktcdvd = NULL;
416                 printk(DRIVER_NAME": failed to create class pktcdvd\n");
417                 return ret;
418         }
419         return 0;
420 }
421
422 static void pkt_sysfs_cleanup(void)
423 {
424         if (class_pktcdvd)
425                 class_destroy(class_pktcdvd);
426         class_pktcdvd = NULL;
427 }
428
429 /********************************************************************
430   entries in debugfs
431
432   /debugfs/pktcdvd[0-7]/
433                         info
434
435  *******************************************************************/
436
437 static int pkt_debugfs_seq_show(struct seq_file *m, void *p)
438 {
439         return pkt_seq_show(m, p);
440 }
441
442 static int pkt_debugfs_fops_open(struct inode *inode, struct file *file)
443 {
444         return single_open(file, pkt_debugfs_seq_show, inode->i_private);
445 }
446
447 static const struct file_operations debug_fops = {
448         .open           = pkt_debugfs_fops_open,
449         .read           = seq_read,
450         .llseek         = seq_lseek,
451         .release        = single_release,
452         .owner          = THIS_MODULE,
453 };
454
455 static void pkt_debugfs_dev_new(struct pktcdvd_device *pd)
456 {
457         if (!pkt_debugfs_root)
458                 return;
459         pd->dfs_f_info = NULL;
460         pd->dfs_d_root = debugfs_create_dir(pd->name, pkt_debugfs_root);
461         if (IS_ERR(pd->dfs_d_root)) {
462                 pd->dfs_d_root = NULL;
463                 return;
464         }
465         pd->dfs_f_info = debugfs_create_file("info", S_IRUGO,
466                                 pd->dfs_d_root, pd, &debug_fops);
467         if (IS_ERR(pd->dfs_f_info)) {
468                 pd->dfs_f_info = NULL;
469                 return;
470         }
471 }
472
473 static void pkt_debugfs_dev_remove(struct pktcdvd_device *pd)
474 {
475         if (!pkt_debugfs_root)
476                 return;
477         if (pd->dfs_f_info)
478                 debugfs_remove(pd->dfs_f_info);
479         pd->dfs_f_info = NULL;
480         if (pd->dfs_d_root)
481                 debugfs_remove(pd->dfs_d_root);
482         pd->dfs_d_root = NULL;
483 }
484
485 static void pkt_debugfs_init(void)
486 {
487         pkt_debugfs_root = debugfs_create_dir(DRIVER_NAME, NULL);
488         if (IS_ERR(pkt_debugfs_root)) {
489                 pkt_debugfs_root = NULL;
490                 return;
491         }
492 }
493
494 static void pkt_debugfs_cleanup(void)
495 {
496         if (!pkt_debugfs_root)
497                 return;
498         debugfs_remove(pkt_debugfs_root);
499         pkt_debugfs_root = NULL;
500 }
501
502 /* ----------------------------------------------------------*/
503
504
505 static void pkt_bio_finished(struct pktcdvd_device *pd)
506 {
507         BUG_ON(atomic_read(&pd->cdrw.pending_bios) <= 0);
508         if (atomic_dec_and_test(&pd->cdrw.pending_bios)) {
509                 VPRINTK(DRIVER_NAME": queue empty\n");
510                 atomic_set(&pd->iosched.attention, 1);
511                 wake_up(&pd->wqueue);
512         }
513 }
514
515 static void pkt_bio_destructor(struct bio *bio)
516 {
517         kfree(bio->bi_io_vec);
518         kfree(bio);
519 }
520
521 static struct bio *pkt_bio_alloc(int nr_iovecs)
522 {
523         struct bio_vec *bvl = NULL;
524         struct bio *bio;
525
526         bio = kmalloc(sizeof(struct bio), GFP_KERNEL);
527         if (!bio)
528                 goto no_bio;
529         bio_init(bio);
530
531         bvl = kcalloc(nr_iovecs, sizeof(struct bio_vec), GFP_KERNEL);
532         if (!bvl)
533                 goto no_bvl;
534
535         bio->bi_max_vecs = nr_iovecs;
536         bio->bi_io_vec = bvl;
537         bio->bi_destructor = pkt_bio_destructor;
538
539         return bio;
540
541  no_bvl:
542         kfree(bio);
543  no_bio:
544         return NULL;
545 }
546
547 /*
548  * Allocate a packet_data struct
549  */
550 static struct packet_data *pkt_alloc_packet_data(int frames)
551 {
552         int i;
553         struct packet_data *pkt;
554
555         pkt = kzalloc(sizeof(struct packet_data), GFP_KERNEL);
556         if (!pkt)
557                 goto no_pkt;
558
559         pkt->frames = frames;
560         pkt->w_bio = pkt_bio_alloc(frames);
561         if (!pkt->w_bio)
562                 goto no_bio;
563
564         for (i = 0; i < frames / FRAMES_PER_PAGE; i++) {
565                 pkt->pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
566                 if (!pkt->pages[i])
567                         goto no_page;
568         }
569
570         spin_lock_init(&pkt->lock);
571
572         for (i = 0; i < frames; i++) {
573                 struct bio *bio = pkt_bio_alloc(1);
574                 if (!bio)
575                         goto no_rd_bio;
576                 pkt->r_bios[i] = bio;
577         }
578
579         return pkt;
580
581 no_rd_bio:
582         for (i = 0; i < frames; i++) {
583                 struct bio *bio = pkt->r_bios[i];
584                 if (bio)
585                         bio_put(bio);
586         }
587
588 no_page:
589         for (i = 0; i < frames / FRAMES_PER_PAGE; i++)
590                 if (pkt->pages[i])
591                         __free_page(pkt->pages[i]);
592         bio_put(pkt->w_bio);
593 no_bio:
594         kfree(pkt);
595 no_pkt:
596         return NULL;
597 }
598
599 /*
600  * Free a packet_data struct
601  */
602 static void pkt_free_packet_data(struct packet_data *pkt)
603 {
604         int i;
605
606         for (i = 0; i < pkt->frames; i++) {
607                 struct bio *bio = pkt->r_bios[i];
608                 if (bio)
609                         bio_put(bio);
610         }
611         for (i = 0; i < pkt->frames / FRAMES_PER_PAGE; i++)
612                 __free_page(pkt->pages[i]);
613         bio_put(pkt->w_bio);
614         kfree(pkt);
615 }
616
617 static void pkt_shrink_pktlist(struct pktcdvd_device *pd)
618 {
619         struct packet_data *pkt, *next;
620
621         BUG_ON(!list_empty(&pd->cdrw.pkt_active_list));
622
623         list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_free_list, list) {
624                 pkt_free_packet_data(pkt);
625         }
626         INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
627 }
628
629 static int pkt_grow_pktlist(struct pktcdvd_device *pd, int nr_packets)
630 {
631         struct packet_data *pkt;
632
633         BUG_ON(!list_empty(&pd->cdrw.pkt_free_list));
634
635         while (nr_packets > 0) {
636                 pkt = pkt_alloc_packet_data(pd->settings.size >> 2);
637                 if (!pkt) {
638                         pkt_shrink_pktlist(pd);
639                         return 0;
640                 }
641                 pkt->id = nr_packets;
642                 pkt->pd = pd;
643                 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
644                 nr_packets--;
645         }
646         return 1;
647 }
648
649 static inline struct pkt_rb_node *pkt_rbtree_next(struct pkt_rb_node *node)
650 {
651         struct rb_node *n = rb_next(&node->rb_node);
652         if (!n)
653                 return NULL;
654         return rb_entry(n, struct pkt_rb_node, rb_node);
655 }
656
657 static void pkt_rbtree_erase(struct pktcdvd_device *pd, struct pkt_rb_node *node)
658 {
659         rb_erase(&node->rb_node, &pd->bio_queue);
660         mempool_free(node, pd->rb_pool);
661         pd->bio_queue_size--;
662         BUG_ON(pd->bio_queue_size < 0);
663 }
664
665 /*
666  * Find the first node in the pd->bio_queue rb tree with a starting sector >= s.
667  */
668 static struct pkt_rb_node *pkt_rbtree_find(struct pktcdvd_device *pd, sector_t s)
669 {
670         struct rb_node *n = pd->bio_queue.rb_node;
671         struct rb_node *next;
672         struct pkt_rb_node *tmp;
673
674         if (!n) {
675                 BUG_ON(pd->bio_queue_size > 0);
676                 return NULL;
677         }
678
679         for (;;) {
680                 tmp = rb_entry(n, struct pkt_rb_node, rb_node);
681                 if (s <= tmp->bio->bi_sector)
682                         next = n->rb_left;
683                 else
684                         next = n->rb_right;
685                 if (!next)
686                         break;
687                 n = next;
688         }
689
690         if (s > tmp->bio->bi_sector) {
691                 tmp = pkt_rbtree_next(tmp);
692                 if (!tmp)
693                         return NULL;
694         }
695         BUG_ON(s > tmp->bio->bi_sector);
696         return tmp;
697 }
698
699 /*
700  * Insert a node into the pd->bio_queue rb tree.
701  */
702 static void pkt_rbtree_insert(struct pktcdvd_device *pd, struct pkt_rb_node *node)
703 {
704         struct rb_node **p = &pd->bio_queue.rb_node;
705         struct rb_node *parent = NULL;
706         sector_t s = node->bio->bi_sector;
707         struct pkt_rb_node *tmp;
708
709         while (*p) {
710                 parent = *p;
711                 tmp = rb_entry(parent, struct pkt_rb_node, rb_node);
712                 if (s < tmp->bio->bi_sector)
713                         p = &(*p)->rb_left;
714                 else
715                         p = &(*p)->rb_right;
716         }
717         rb_link_node(&node->rb_node, parent, p);
718         rb_insert_color(&node->rb_node, &pd->bio_queue);
719         pd->bio_queue_size++;
720 }
721
722 /*
723  * Add a bio to a single linked list defined by its head and tail pointers.
724  */
725 static void pkt_add_list_last(struct bio *bio, struct bio **list_head, struct bio **list_tail)
726 {
727         bio->bi_next = NULL;
728         if (*list_tail) {
729                 BUG_ON((*list_head) == NULL);
730                 (*list_tail)->bi_next = bio;
731                 (*list_tail) = bio;
732         } else {
733                 BUG_ON((*list_head) != NULL);
734                 (*list_head) = bio;
735                 (*list_tail) = bio;
736         }
737 }
738
739 /*
740  * Remove and return the first bio from a single linked list defined by its
741  * head and tail pointers.
742  */
743 static inline struct bio *pkt_get_list_first(struct bio **list_head, struct bio **list_tail)
744 {
745         struct bio *bio;
746
747         if (*list_head == NULL)
748                 return NULL;
749
750         bio = *list_head;
751         *list_head = bio->bi_next;
752         if (*list_head == NULL)
753                 *list_tail = NULL;
754
755         bio->bi_next = NULL;
756         return bio;
757 }
758
759 /*
760  * Send a packet_command to the underlying block device and
761  * wait for completion.
762  */
763 static int pkt_generic_packet(struct pktcdvd_device *pd, struct packet_command *cgc)
764 {
765         struct request_queue *q = bdev_get_queue(pd->bdev);
766         struct request *rq;
767         int ret = 0;
768
769         rq = blk_get_request(q, (cgc->data_direction == CGC_DATA_WRITE) ?
770                              WRITE : READ, __GFP_WAIT);
771
772         if (cgc->buflen) {
773                 if (blk_rq_map_kern(q, rq, cgc->buffer, cgc->buflen, __GFP_WAIT))
774                         goto out;
775         }
776
777         rq->cmd_len = COMMAND_SIZE(cgc->cmd[0]);
778         memcpy(rq->cmd, cgc->cmd, CDROM_PACKET_SIZE);
779         if (sizeof(rq->cmd) > CDROM_PACKET_SIZE)
780                 memset(rq->cmd + CDROM_PACKET_SIZE, 0, sizeof(rq->cmd) - CDROM_PACKET_SIZE);
781
782         rq->timeout = 60*HZ;
783         rq->cmd_type = REQ_TYPE_BLOCK_PC;
784         rq->cmd_flags |= REQ_HARDBARRIER;
785         if (cgc->quiet)
786                 rq->cmd_flags |= REQ_QUIET;
787
788         blk_execute_rq(rq->q, pd->bdev->bd_disk, rq, 0);
789         if (rq->errors)
790                 ret = -EIO;
791 out:
792         blk_put_request(rq);
793         return ret;
794 }
795
796 /*
797  * A generic sense dump / resolve mechanism should be implemented across
798  * all ATAPI + SCSI devices.
799  */
800 static void pkt_dump_sense(struct packet_command *cgc)
801 {
802         static char *info[9] = { "No sense", "Recovered error", "Not ready",
803                                  "Medium error", "Hardware error", "Illegal request",
804                                  "Unit attention", "Data protect", "Blank check" };
805         int i;
806         struct request_sense *sense = cgc->sense;
807
808         printk(DRIVER_NAME":");
809         for (i = 0; i < CDROM_PACKET_SIZE; i++)
810                 printk(" %02x", cgc->cmd[i]);
811         printk(" - ");
812
813         if (sense == NULL) {
814                 printk("no sense\n");
815                 return;
816         }
817
818         printk("sense %02x.%02x.%02x", sense->sense_key, sense->asc, sense->ascq);
819
820         if (sense->sense_key > 8) {
821                 printk(" (INVALID)\n");
822                 return;
823         }
824
825         printk(" (%s)\n", info[sense->sense_key]);
826 }
827
828 /*
829  * flush the drive cache to media
830  */
831 static int pkt_flush_cache(struct pktcdvd_device *pd)
832 {
833         struct packet_command cgc;
834
835         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
836         cgc.cmd[0] = GPCMD_FLUSH_CACHE;
837         cgc.quiet = 1;
838
839         /*
840          * the IMMED bit -- we default to not setting it, although that
841          * would allow a much faster close, this is safer
842          */
843 #if 0
844         cgc.cmd[1] = 1 << 1;
845 #endif
846         return pkt_generic_packet(pd, &cgc);
847 }
848
849 /*
850  * speed is given as the normal factor, e.g. 4 for 4x
851  */
852 static noinline_for_stack int pkt_set_speed(struct pktcdvd_device *pd,
853                                 unsigned write_speed, unsigned read_speed)
854 {
855         struct packet_command cgc;
856         struct request_sense sense;
857         int ret;
858
859         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
860         cgc.sense = &sense;
861         cgc.cmd[0] = GPCMD_SET_SPEED;
862         cgc.cmd[2] = (read_speed >> 8) & 0xff;
863         cgc.cmd[3] = read_speed & 0xff;
864         cgc.cmd[4] = (write_speed >> 8) & 0xff;
865         cgc.cmd[5] = write_speed & 0xff;
866
867         if ((ret = pkt_generic_packet(pd, &cgc)))
868                 pkt_dump_sense(&cgc);
869
870         return ret;
871 }
872
873 /*
874  * Queue a bio for processing by the low-level CD device. Must be called
875  * from process context.
876  */
877 static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio)
878 {
879         spin_lock(&pd->iosched.lock);
880         if (bio_data_dir(bio) == READ) {
881                 pkt_add_list_last(bio, &pd->iosched.read_queue,
882                                   &pd->iosched.read_queue_tail);
883         } else {
884                 pkt_add_list_last(bio, &pd->iosched.write_queue,
885                                   &pd->iosched.write_queue_tail);
886         }
887         spin_unlock(&pd->iosched.lock);
888
889         atomic_set(&pd->iosched.attention, 1);
890         wake_up(&pd->wqueue);
891 }
892
893 /*
894  * Process the queued read/write requests. This function handles special
895  * requirements for CDRW drives:
896  * - A cache flush command must be inserted before a read request if the
897  *   previous request was a write.
898  * - Switching between reading and writing is slow, so don't do it more often
899  *   than necessary.
900  * - Optimize for throughput at the expense of latency. This means that streaming
901  *   writes will never be interrupted by a read, but if the drive has to seek
902  *   before the next write, switch to reading instead if there are any pending
903  *   read requests.
904  * - Set the read speed according to current usage pattern. When only reading
905  *   from the device, it's best to use the highest possible read speed, but
906  *   when switching often between reading and writing, it's better to have the
907  *   same read and write speeds.
908  */
909 static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
910 {
911
912         if (atomic_read(&pd->iosched.attention) == 0)
913                 return;
914         atomic_set(&pd->iosched.attention, 0);
915
916         for (;;) {
917                 struct bio *bio;
918                 int reads_queued, writes_queued;
919
920                 spin_lock(&pd->iosched.lock);
921                 reads_queued = (pd->iosched.read_queue != NULL);
922                 writes_queued = (pd->iosched.write_queue != NULL);
923                 spin_unlock(&pd->iosched.lock);
924
925                 if (!reads_queued && !writes_queued)
926                         break;
927
928                 if (pd->iosched.writing) {
929                         int need_write_seek = 1;
930                         spin_lock(&pd->iosched.lock);
931                         bio = pd->iosched.write_queue;
932                         spin_unlock(&pd->iosched.lock);
933                         if (bio && (bio->bi_sector == pd->iosched.last_write))
934                                 need_write_seek = 0;
935                         if (need_write_seek && reads_queued) {
936                                 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
937                                         VPRINTK(DRIVER_NAME": write, waiting\n");
938                                         break;
939                                 }
940                                 pkt_flush_cache(pd);
941                                 pd->iosched.writing = 0;
942                         }
943                 } else {
944                         if (!reads_queued && writes_queued) {
945                                 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
946                                         VPRINTK(DRIVER_NAME": read, waiting\n");
947                                         break;
948                                 }
949                                 pd->iosched.writing = 1;
950                         }
951                 }
952
953                 spin_lock(&pd->iosched.lock);
954                 if (pd->iosched.writing) {
955                         bio = pkt_get_list_first(&pd->iosched.write_queue,
956                                                  &pd->iosched.write_queue_tail);
957                 } else {
958                         bio = pkt_get_list_first(&pd->iosched.read_queue,
959                                                  &pd->iosched.read_queue_tail);
960                 }
961                 spin_unlock(&pd->iosched.lock);
962
963                 if (!bio)
964                         continue;
965
966                 if (bio_data_dir(bio) == READ)
967                         pd->iosched.successive_reads += bio->bi_size >> 10;
968                 else {
969                         pd->iosched.successive_reads = 0;
970                         pd->iosched.last_write = bio->bi_sector + bio_sectors(bio);
971                 }
972                 if (pd->iosched.successive_reads >= HI_SPEED_SWITCH) {
973                         if (pd->read_speed == pd->write_speed) {
974                                 pd->read_speed = MAX_SPEED;
975                                 pkt_set_speed(pd, pd->write_speed, pd->read_speed);
976                         }
977                 } else {
978                         if (pd->read_speed != pd->write_speed) {
979                                 pd->read_speed = pd->write_speed;
980                                 pkt_set_speed(pd, pd->write_speed, pd->read_speed);
981                         }
982                 }
983
984                 atomic_inc(&pd->cdrw.pending_bios);
985                 generic_make_request(bio);
986         }
987 }
988
989 /*
990  * Special care is needed if the underlying block device has a small
991  * max_phys_segments value.
992  */
993 static int pkt_set_segment_merging(struct pktcdvd_device *pd, struct request_queue *q)
994 {
995         if ((pd->settings.size << 9) / CD_FRAMESIZE <= q->max_phys_segments) {
996                 /*
997                  * The cdrom device can handle one segment/frame
998                  */
999                 clear_bit(PACKET_MERGE_SEGS, &pd->flags);
1000                 return 0;
1001         } else if ((pd->settings.size << 9) / PAGE_SIZE <= q->max_phys_segments) {
1002                 /*
1003                  * We can handle this case at the expense of some extra memory
1004                  * copies during write operations
1005                  */
1006                 set_bit(PACKET_MERGE_SEGS, &pd->flags);
1007                 return 0;
1008         } else {
1009                 printk(DRIVER_NAME": cdrom max_phys_segments too small\n");
1010                 return -EIO;
1011         }
1012 }
1013
1014 /*
1015  * Copy CD_FRAMESIZE bytes from src_bio into a destination page
1016  */
1017 static void pkt_copy_bio_data(struct bio *src_bio, int seg, int offs, struct page *dst_page, int dst_offs)
1018 {
1019         unsigned int copy_size = CD_FRAMESIZE;
1020
1021         while (copy_size > 0) {
1022                 struct bio_vec *src_bvl = bio_iovec_idx(src_bio, seg);
1023                 void *vfrom = kmap_atomic(src_bvl->bv_page, KM_USER0) +
1024                         src_bvl->bv_offset + offs;
1025                 void *vto = page_address(dst_page) + dst_offs;
1026                 int len = min_t(int, copy_size, src_bvl->bv_len - offs);
1027
1028                 BUG_ON(len < 0);
1029                 memcpy(vto, vfrom, len);
1030                 kunmap_atomic(vfrom, KM_USER0);
1031
1032                 seg++;
1033                 offs = 0;
1034                 dst_offs += len;
1035                 copy_size -= len;
1036         }
1037 }
1038
1039 /*
1040  * Copy all data for this packet to pkt->pages[], so that
1041  * a) The number of required segments for the write bio is minimized, which
1042  *    is necessary for some scsi controllers.
1043  * b) The data can be used as cache to avoid read requests if we receive a
1044  *    new write request for the same zone.
1045  */
1046 static void pkt_make_local_copy(struct packet_data *pkt, struct bio_vec *bvec)
1047 {
1048         int f, p, offs;
1049
1050         /* Copy all data to pkt->pages[] */
1051         p = 0;
1052         offs = 0;
1053         for (f = 0; f < pkt->frames; f++) {
1054                 if (bvec[f].bv_page != pkt->pages[p]) {
1055                         void *vfrom = kmap_atomic(bvec[f].bv_page, KM_USER0) + bvec[f].bv_offset;
1056                         void *vto = page_address(pkt->pages[p]) + offs;
1057                         memcpy(vto, vfrom, CD_FRAMESIZE);
1058                         kunmap_atomic(vfrom, KM_USER0);
1059                         bvec[f].bv_page = pkt->pages[p];
1060                         bvec[f].bv_offset = offs;
1061                 } else {
1062                         BUG_ON(bvec[f].bv_offset != offs);
1063                 }
1064                 offs += CD_FRAMESIZE;
1065                 if (offs >= PAGE_SIZE) {
1066                         offs = 0;
1067                         p++;
1068                 }
1069         }
1070 }
1071
1072 static void pkt_end_io_read(struct bio *bio, int err)
1073 {
1074         struct packet_data *pkt = bio->bi_private;
1075         struct pktcdvd_device *pd = pkt->pd;
1076         BUG_ON(!pd);
1077
1078         VPRINTK("pkt_end_io_read: bio=%p sec0=%llx sec=%llx err=%d\n", bio,
1079                 (unsigned long long)pkt->sector, (unsigned long long)bio->bi_sector, err);
1080
1081         if (err)
1082                 atomic_inc(&pkt->io_errors);
1083         if (atomic_dec_and_test(&pkt->io_wait)) {
1084                 atomic_inc(&pkt->run_sm);
1085                 wake_up(&pd->wqueue);
1086         }
1087         pkt_bio_finished(pd);
1088 }
1089
1090 static void pkt_end_io_packet_write(struct bio *bio, int err)
1091 {
1092         struct packet_data *pkt = bio->bi_private;
1093         struct pktcdvd_device *pd = pkt->pd;
1094         BUG_ON(!pd);
1095
1096         VPRINTK("pkt_end_io_packet_write: id=%d, err=%d\n", pkt->id, err);
1097
1098         pd->stats.pkt_ended++;
1099
1100         pkt_bio_finished(pd);
1101         atomic_dec(&pkt->io_wait);
1102         atomic_inc(&pkt->run_sm);
1103         wake_up(&pd->wqueue);
1104 }
1105
1106 /*
1107  * Schedule reads for the holes in a packet
1108  */
1109 static void pkt_gather_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1110 {
1111         int frames_read = 0;
1112         struct bio *bio;
1113         int f;
1114         char written[PACKET_MAX_SIZE];
1115
1116         BUG_ON(!pkt->orig_bios);
1117
1118         atomic_set(&pkt->io_wait, 0);
1119         atomic_set(&pkt->io_errors, 0);
1120
1121         /*
1122          * Figure out which frames we need to read before we can write.
1123          */
1124         memset(written, 0, sizeof(written));
1125         spin_lock(&pkt->lock);
1126         for (bio = pkt->orig_bios; bio; bio = bio->bi_next) {
1127                 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1128                 int num_frames = bio->bi_size / CD_FRAMESIZE;
1129                 pd->stats.secs_w += num_frames * (CD_FRAMESIZE >> 9);
1130                 BUG_ON(first_frame < 0);
1131                 BUG_ON(first_frame + num_frames > pkt->frames);
1132                 for (f = first_frame; f < first_frame + num_frames; f++)
1133                         written[f] = 1;
1134         }
1135         spin_unlock(&pkt->lock);
1136
1137         if (pkt->cache_valid) {
1138                 VPRINTK("pkt_gather_data: zone %llx cached\n",
1139                         (unsigned long long)pkt->sector);
1140                 goto out_account;
1141         }
1142
1143         /*
1144          * Schedule reads for missing parts of the packet.
1145          */
1146         for (f = 0; f < pkt->frames; f++) {
1147                 struct bio_vec *vec;
1148
1149                 int p, offset;
1150                 if (written[f])
1151                         continue;
1152                 bio = pkt->r_bios[f];
1153                 vec = bio->bi_io_vec;
1154                 bio_init(bio);
1155                 bio->bi_max_vecs = 1;
1156                 bio->bi_sector = pkt->sector + f * (CD_FRAMESIZE >> 9);
1157                 bio->bi_bdev = pd->bdev;
1158                 bio->bi_end_io = pkt_end_io_read;
1159                 bio->bi_private = pkt;
1160                 bio->bi_io_vec = vec;
1161                 bio->bi_destructor = pkt_bio_destructor;
1162
1163                 p = (f * CD_FRAMESIZE) / PAGE_SIZE;
1164                 offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1165                 VPRINTK("pkt_gather_data: Adding frame %d, page:%p offs:%d\n",
1166                         f, pkt->pages[p], offset);
1167                 if (!bio_add_page(bio, pkt->pages[p], CD_FRAMESIZE, offset))
1168                         BUG();
1169
1170                 atomic_inc(&pkt->io_wait);
1171                 bio->bi_rw = READ;
1172                 pkt_queue_bio(pd, bio);
1173                 frames_read++;
1174         }
1175
1176 out_account:
1177         VPRINTK("pkt_gather_data: need %d frames for zone %llx\n",
1178                 frames_read, (unsigned long long)pkt->sector);
1179         pd->stats.pkt_started++;
1180         pd->stats.secs_rg += frames_read * (CD_FRAMESIZE >> 9);
1181 }
1182
1183 /*
1184  * Find a packet matching zone, or the least recently used packet if
1185  * there is no match.
1186  */
1187 static struct packet_data *pkt_get_packet_data(struct pktcdvd_device *pd, int zone)
1188 {
1189         struct packet_data *pkt;
1190
1191         list_for_each_entry(pkt, &pd->cdrw.pkt_free_list, list) {
1192                 if (pkt->sector == zone || pkt->list.next == &pd->cdrw.pkt_free_list) {
1193                         list_del_init(&pkt->list);
1194                         if (pkt->sector != zone)
1195                                 pkt->cache_valid = 0;
1196                         return pkt;
1197                 }
1198         }
1199         BUG();
1200         return NULL;
1201 }
1202
1203 static void pkt_put_packet_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1204 {
1205         if (pkt->cache_valid) {
1206                 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
1207         } else {
1208                 list_add_tail(&pkt->list, &pd->cdrw.pkt_free_list);
1209         }
1210 }
1211
1212 /*
1213  * recover a failed write, query for relocation if possible
1214  *
1215  * returns 1 if recovery is possible, or 0 if not
1216  *
1217  */
1218 static int pkt_start_recovery(struct packet_data *pkt)
1219 {
1220         /*
1221          * FIXME. We need help from the file system to implement
1222          * recovery handling.
1223          */
1224         return 0;
1225 #if 0
1226         struct request *rq = pkt->rq;
1227         struct pktcdvd_device *pd = rq->rq_disk->private_data;
1228         struct block_device *pkt_bdev;
1229         struct super_block *sb = NULL;
1230         unsigned long old_block, new_block;
1231         sector_t new_sector;
1232
1233         pkt_bdev = bdget(kdev_t_to_nr(pd->pkt_dev));
1234         if (pkt_bdev) {
1235                 sb = get_super(pkt_bdev);
1236                 bdput(pkt_bdev);
1237         }
1238
1239         if (!sb)
1240                 return 0;
1241
1242         if (!sb->s_op || !sb->s_op->relocate_blocks)
1243                 goto out;
1244
1245         old_block = pkt->sector / (CD_FRAMESIZE >> 9);
1246         if (sb->s_op->relocate_blocks(sb, old_block, &new_block))
1247                 goto out;
1248
1249         new_sector = new_block * (CD_FRAMESIZE >> 9);
1250         pkt->sector = new_sector;
1251
1252         pkt->bio->bi_sector = new_sector;
1253         pkt->bio->bi_next = NULL;
1254         pkt->bio->bi_flags = 1 << BIO_UPTODATE;
1255         pkt->bio->bi_idx = 0;
1256
1257         BUG_ON(pkt->bio->bi_rw != (1 << BIO_RW));
1258         BUG_ON(pkt->bio->bi_vcnt != pkt->frames);
1259         BUG_ON(pkt->bio->bi_size != pkt->frames * CD_FRAMESIZE);
1260         BUG_ON(pkt->bio->bi_end_io != pkt_end_io_packet_write);
1261         BUG_ON(pkt->bio->bi_private != pkt);
1262
1263         drop_super(sb);
1264         return 1;
1265
1266 out:
1267         drop_super(sb);
1268         return 0;
1269 #endif
1270 }
1271
1272 static inline void pkt_set_state(struct packet_data *pkt, enum packet_data_state state)
1273 {
1274 #if PACKET_DEBUG > 1
1275         static const char *state_name[] = {
1276                 "IDLE", "WAITING", "READ_WAIT", "WRITE_WAIT", "RECOVERY", "FINISHED"
1277         };
1278         enum packet_data_state old_state = pkt->state;
1279         VPRINTK("pkt %2d : s=%6llx %s -> %s\n", pkt->id, (unsigned long long)pkt->sector,
1280                 state_name[old_state], state_name[state]);
1281 #endif
1282         pkt->state = state;
1283 }
1284
1285 /*
1286  * Scan the work queue to see if we can start a new packet.
1287  * returns non-zero if any work was done.
1288  */
1289 static int pkt_handle_queue(struct pktcdvd_device *pd)
1290 {
1291         struct packet_data *pkt, *p;
1292         struct bio *bio = NULL;
1293         sector_t zone = 0; /* Suppress gcc warning */
1294         struct pkt_rb_node *node, *first_node;
1295         struct rb_node *n;
1296         int wakeup;
1297
1298         VPRINTK("handle_queue\n");
1299
1300         atomic_set(&pd->scan_queue, 0);
1301
1302         if (list_empty(&pd->cdrw.pkt_free_list)) {
1303                 VPRINTK("handle_queue: no pkt\n");
1304                 return 0;
1305         }
1306
1307         /*
1308          * Try to find a zone we are not already working on.
1309          */
1310         spin_lock(&pd->lock);
1311         first_node = pkt_rbtree_find(pd, pd->current_sector);
1312         if (!first_node) {
1313                 n = rb_first(&pd->bio_queue);
1314                 if (n)
1315                         first_node = rb_entry(n, struct pkt_rb_node, rb_node);
1316         }
1317         node = first_node;
1318         while (node) {
1319                 bio = node->bio;
1320                 zone = ZONE(bio->bi_sector, pd);
1321                 list_for_each_entry(p, &pd->cdrw.pkt_active_list, list) {
1322                         if (p->sector == zone) {
1323                                 bio = NULL;
1324                                 goto try_next_bio;
1325                         }
1326                 }
1327                 break;
1328 try_next_bio:
1329                 node = pkt_rbtree_next(node);
1330                 if (!node) {
1331                         n = rb_first(&pd->bio_queue);
1332                         if (n)
1333                                 node = rb_entry(n, struct pkt_rb_node, rb_node);
1334                 }
1335                 if (node == first_node)
1336                         node = NULL;
1337         }
1338         spin_unlock(&pd->lock);
1339         if (!bio) {
1340                 VPRINTK("handle_queue: no bio\n");
1341                 return 0;
1342         }
1343
1344         pkt = pkt_get_packet_data(pd, zone);
1345
1346         pd->current_sector = zone + pd->settings.size;
1347         pkt->sector = zone;
1348         BUG_ON(pkt->frames != pd->settings.size >> 2);
1349         pkt->write_size = 0;
1350
1351         /*
1352          * Scan work queue for bios in the same zone and link them
1353          * to this packet.
1354          */
1355         spin_lock(&pd->lock);
1356         VPRINTK("pkt_handle_queue: looking for zone %llx\n", (unsigned long long)zone);
1357         while ((node = pkt_rbtree_find(pd, zone)) != NULL) {
1358                 bio = node->bio;
1359                 VPRINTK("pkt_handle_queue: found zone=%llx\n",
1360                         (unsigned long long)ZONE(bio->bi_sector, pd));
1361                 if (ZONE(bio->bi_sector, pd) != zone)
1362                         break;
1363                 pkt_rbtree_erase(pd, node);
1364                 spin_lock(&pkt->lock);
1365                 pkt_add_list_last(bio, &pkt->orig_bios, &pkt->orig_bios_tail);
1366                 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
1367                 spin_unlock(&pkt->lock);
1368         }
1369         /* check write congestion marks, and if bio_queue_size is
1370            below, wake up any waiters */
1371         wakeup = (pd->write_congestion_on > 0
1372                         && pd->bio_queue_size <= pd->write_congestion_off);
1373         spin_unlock(&pd->lock);
1374         if (wakeup)
1375                 clear_bdi_congested(&pd->disk->queue->backing_dev_info, WRITE);
1376
1377         pkt->sleep_time = max(PACKET_WAIT_TIME, 1);
1378         pkt_set_state(pkt, PACKET_WAITING_STATE);
1379         atomic_set(&pkt->run_sm, 1);
1380
1381         spin_lock(&pd->cdrw.active_list_lock);
1382         list_add(&pkt->list, &pd->cdrw.pkt_active_list);
1383         spin_unlock(&pd->cdrw.active_list_lock);
1384
1385         return 1;
1386 }
1387
1388 /*
1389  * Assemble a bio to write one packet and queue the bio for processing
1390  * by the underlying block device.
1391  */
1392 static void pkt_start_write(struct pktcdvd_device *pd, struct packet_data *pkt)
1393 {
1394         struct bio *bio;
1395         int f;
1396         int frames_write;
1397         struct bio_vec *bvec = pkt->w_bio->bi_io_vec;
1398
1399         for (f = 0; f < pkt->frames; f++) {
1400                 bvec[f].bv_page = pkt->pages[(f * CD_FRAMESIZE) / PAGE_SIZE];
1401                 bvec[f].bv_offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1402         }
1403
1404         /*
1405          * Fill-in bvec with data from orig_bios.
1406          */
1407         frames_write = 0;
1408         spin_lock(&pkt->lock);
1409         for (bio = pkt->orig_bios; bio; bio = bio->bi_next) {
1410                 int segment = bio->bi_idx;
1411                 int src_offs = 0;
1412                 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1413                 int num_frames = bio->bi_size / CD_FRAMESIZE;
1414                 BUG_ON(first_frame < 0);
1415                 BUG_ON(first_frame + num_frames > pkt->frames);
1416                 for (f = first_frame; f < first_frame + num_frames; f++) {
1417                         struct bio_vec *src_bvl = bio_iovec_idx(bio, segment);
1418
1419                         while (src_offs >= src_bvl->bv_len) {
1420                                 src_offs -= src_bvl->bv_len;
1421                                 segment++;
1422                                 BUG_ON(segment >= bio->bi_vcnt);
1423                                 src_bvl = bio_iovec_idx(bio, segment);
1424                         }
1425
1426                         if (src_bvl->bv_len - src_offs >= CD_FRAMESIZE) {
1427                                 bvec[f].bv_page = src_bvl->bv_page;
1428                                 bvec[f].bv_offset = src_bvl->bv_offset + src_offs;
1429                         } else {
1430                                 pkt_copy_bio_data(bio, segment, src_offs,
1431                                                   bvec[f].bv_page, bvec[f].bv_offset);
1432                         }
1433                         src_offs += CD_FRAMESIZE;
1434                         frames_write++;
1435                 }
1436         }
1437         pkt_set_state(pkt, PACKET_WRITE_WAIT_STATE);
1438         spin_unlock(&pkt->lock);
1439
1440         VPRINTK("pkt_start_write: Writing %d frames for zone %llx\n",
1441                 frames_write, (unsigned long long)pkt->sector);
1442         BUG_ON(frames_write != pkt->write_size);
1443
1444         if (test_bit(PACKET_MERGE_SEGS, &pd->flags) || (pkt->write_size < pkt->frames)) {
1445                 pkt_make_local_copy(pkt, bvec);
1446                 pkt->cache_valid = 1;
1447         } else {
1448                 pkt->cache_valid = 0;
1449         }
1450
1451         /* Start the write request */
1452         bio_init(pkt->w_bio);
1453         pkt->w_bio->bi_max_vecs = PACKET_MAX_SIZE;
1454         pkt->w_bio->bi_sector = pkt->sector;
1455         pkt->w_bio->bi_bdev = pd->bdev;
1456         pkt->w_bio->bi_end_io = pkt_end_io_packet_write;
1457         pkt->w_bio->bi_private = pkt;
1458         pkt->w_bio->bi_io_vec = bvec;
1459         pkt->w_bio->bi_destructor = pkt_bio_destructor;
1460         for (f = 0; f < pkt->frames; f++)
1461                 if (!bio_add_page(pkt->w_bio, bvec[f].bv_page, CD_FRAMESIZE, bvec[f].bv_offset))
1462                         BUG();
1463         VPRINTK(DRIVER_NAME": vcnt=%d\n", pkt->w_bio->bi_vcnt);
1464
1465         atomic_set(&pkt->io_wait, 1);
1466         pkt->w_bio->bi_rw = WRITE;
1467         pkt_queue_bio(pd, pkt->w_bio);
1468 }
1469
1470 static void pkt_finish_packet(struct packet_data *pkt, int uptodate)
1471 {
1472         struct bio *bio, *next;
1473
1474         if (!uptodate)
1475                 pkt->cache_valid = 0;
1476
1477         /* Finish all bios corresponding to this packet */
1478         bio = pkt->orig_bios;
1479         while (bio) {
1480                 next = bio->bi_next;
1481                 bio->bi_next = NULL;
1482                 bio_endio(bio, uptodate ? 0 : -EIO);
1483                 bio = next;
1484         }
1485         pkt->orig_bios = pkt->orig_bios_tail = NULL;
1486 }
1487
1488 static void pkt_run_state_machine(struct pktcdvd_device *pd, struct packet_data *pkt)
1489 {
1490         int uptodate;
1491
1492         VPRINTK("run_state_machine: pkt %d\n", pkt->id);
1493
1494         for (;;) {
1495                 switch (pkt->state) {
1496                 case PACKET_WAITING_STATE:
1497                         if ((pkt->write_size < pkt->frames) && (pkt->sleep_time > 0))
1498                                 return;
1499
1500                         pkt->sleep_time = 0;
1501                         pkt_gather_data(pd, pkt);
1502                         pkt_set_state(pkt, PACKET_READ_WAIT_STATE);
1503                         break;
1504
1505                 case PACKET_READ_WAIT_STATE:
1506                         if (atomic_read(&pkt->io_wait) > 0)
1507                                 return;
1508
1509                         if (atomic_read(&pkt->io_errors) > 0) {
1510                                 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1511                         } else {
1512                                 pkt_start_write(pd, pkt);
1513                         }
1514                         break;
1515
1516                 case PACKET_WRITE_WAIT_STATE:
1517                         if (atomic_read(&pkt->io_wait) > 0)
1518                                 return;
1519
1520                         if (test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags)) {
1521                                 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1522                         } else {
1523                                 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1524                         }
1525                         break;
1526
1527                 case PACKET_RECOVERY_STATE:
1528                         if (pkt_start_recovery(pkt)) {
1529                                 pkt_start_write(pd, pkt);
1530                         } else {
1531                                 VPRINTK("No recovery possible\n");
1532                                 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1533                         }
1534                         break;
1535
1536                 case PACKET_FINISHED_STATE:
1537                         uptodate = test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags);
1538                         pkt_finish_packet(pkt, uptodate);
1539                         return;
1540
1541                 default:
1542                         BUG();
1543                         break;
1544                 }
1545         }
1546 }
1547
1548 static void pkt_handle_packets(struct pktcdvd_device *pd)
1549 {
1550         struct packet_data *pkt, *next;
1551
1552         VPRINTK("pkt_handle_packets\n");
1553
1554         /*
1555          * Run state machine for active packets
1556          */
1557         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1558                 if (atomic_read(&pkt->run_sm) > 0) {
1559                         atomic_set(&pkt->run_sm, 0);
1560                         pkt_run_state_machine(pd, pkt);
1561                 }
1562         }
1563
1564         /*
1565          * Move no longer active packets to the free list
1566          */
1567         spin_lock(&pd->cdrw.active_list_lock);
1568         list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_active_list, list) {
1569                 if (pkt->state == PACKET_FINISHED_STATE) {
1570                         list_del(&pkt->list);
1571                         pkt_put_packet_data(pd, pkt);
1572                         pkt_set_state(pkt, PACKET_IDLE_STATE);
1573                         atomic_set(&pd->scan_queue, 1);
1574                 }
1575         }
1576         spin_unlock(&pd->cdrw.active_list_lock);
1577 }
1578
1579 static void pkt_count_states(struct pktcdvd_device *pd, int *states)
1580 {
1581         struct packet_data *pkt;
1582         int i;
1583
1584         for (i = 0; i < PACKET_NUM_STATES; i++)
1585                 states[i] = 0;
1586
1587         spin_lock(&pd->cdrw.active_list_lock);
1588         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1589                 states[pkt->state]++;
1590         }
1591         spin_unlock(&pd->cdrw.active_list_lock);
1592 }
1593
1594 /*
1595  * kcdrwd is woken up when writes have been queued for one of our
1596  * registered devices
1597  */
1598 static int kcdrwd(void *foobar)
1599 {
1600         struct pktcdvd_device *pd = foobar;
1601         struct packet_data *pkt;
1602         long min_sleep_time, residue;
1603
1604         set_user_nice(current, -20);
1605         set_freezable();
1606
1607         for (;;) {
1608                 DECLARE_WAITQUEUE(wait, current);
1609
1610                 /*
1611                  * Wait until there is something to do
1612                  */
1613                 add_wait_queue(&pd->wqueue, &wait);
1614                 for (;;) {
1615                         set_current_state(TASK_INTERRUPTIBLE);
1616
1617                         /* Check if we need to run pkt_handle_queue */
1618                         if (atomic_read(&pd->scan_queue) > 0)
1619                                 goto work_to_do;
1620
1621                         /* Check if we need to run the state machine for some packet */
1622                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1623                                 if (atomic_read(&pkt->run_sm) > 0)
1624                                         goto work_to_do;
1625                         }
1626
1627                         /* Check if we need to process the iosched queues */
1628                         if (atomic_read(&pd->iosched.attention) != 0)
1629                                 goto work_to_do;
1630
1631                         /* Otherwise, go to sleep */
1632                         if (PACKET_DEBUG > 1) {
1633                                 int states[PACKET_NUM_STATES];
1634                                 pkt_count_states(pd, states);
1635                                 VPRINTK("kcdrwd: i:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
1636                                         states[0], states[1], states[2], states[3],
1637                                         states[4], states[5]);
1638                         }
1639
1640                         min_sleep_time = MAX_SCHEDULE_TIMEOUT;
1641                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1642                                 if (pkt->sleep_time && pkt->sleep_time < min_sleep_time)
1643                                         min_sleep_time = pkt->sleep_time;
1644                         }
1645
1646                         generic_unplug_device(bdev_get_queue(pd->bdev));
1647
1648                         VPRINTK("kcdrwd: sleeping\n");
1649                         residue = schedule_timeout(min_sleep_time);
1650                         VPRINTK("kcdrwd: wake up\n");
1651
1652                         /* make swsusp happy with our thread */
1653                         try_to_freeze();
1654
1655                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1656                                 if (!pkt->sleep_time)
1657                                         continue;
1658                                 pkt->sleep_time -= min_sleep_time - residue;
1659                                 if (pkt->sleep_time <= 0) {
1660                                         pkt->sleep_time = 0;
1661                                         atomic_inc(&pkt->run_sm);
1662                                 }
1663                         }
1664
1665                         if (kthread_should_stop())
1666                                 break;
1667                 }
1668 work_to_do:
1669                 set_current_state(TASK_RUNNING);
1670                 remove_wait_queue(&pd->wqueue, &wait);
1671
1672                 if (kthread_should_stop())
1673                         break;
1674
1675                 /*
1676                  * if pkt_handle_queue returns true, we can queue
1677                  * another request.
1678                  */
1679                 while (pkt_handle_queue(pd))
1680                         ;
1681
1682                 /*
1683                  * Handle packet state machine
1684                  */
1685                 pkt_handle_packets(pd);
1686
1687                 /*
1688                  * Handle iosched queues
1689                  */
1690                 pkt_iosched_process_queue(pd);
1691         }
1692
1693         return 0;
1694 }
1695
1696 static void pkt_print_settings(struct pktcdvd_device *pd)
1697 {
1698         printk(DRIVER_NAME": %s packets, ", pd->settings.fp ? "Fixed" : "Variable");
1699         printk("%u blocks, ", pd->settings.size >> 2);
1700         printk("Mode-%c disc\n", pd->settings.block_mode == 8 ? '1' : '2');
1701 }
1702
1703 static int pkt_mode_sense(struct pktcdvd_device *pd, struct packet_command *cgc, int page_code, int page_control)
1704 {
1705         memset(cgc->cmd, 0, sizeof(cgc->cmd));
1706
1707         cgc->cmd[0] = GPCMD_MODE_SENSE_10;
1708         cgc->cmd[2] = page_code | (page_control << 6);
1709         cgc->cmd[7] = cgc->buflen >> 8;
1710         cgc->cmd[8] = cgc->buflen & 0xff;
1711         cgc->data_direction = CGC_DATA_READ;
1712         return pkt_generic_packet(pd, cgc);
1713 }
1714
1715 static int pkt_mode_select(struct pktcdvd_device *pd, struct packet_command *cgc)
1716 {
1717         memset(cgc->cmd, 0, sizeof(cgc->cmd));
1718         memset(cgc->buffer, 0, 2);
1719         cgc->cmd[0] = GPCMD_MODE_SELECT_10;
1720         cgc->cmd[1] = 0x10;             /* PF */
1721         cgc->cmd[7] = cgc->buflen >> 8;
1722         cgc->cmd[8] = cgc->buflen & 0xff;
1723         cgc->data_direction = CGC_DATA_WRITE;
1724         return pkt_generic_packet(pd, cgc);
1725 }
1726
1727 static int pkt_get_disc_info(struct pktcdvd_device *pd, disc_information *di)
1728 {
1729         struct packet_command cgc;
1730         int ret;
1731
1732         /* set up command and get the disc info */
1733         init_cdrom_command(&cgc, di, sizeof(*di), CGC_DATA_READ);
1734         cgc.cmd[0] = GPCMD_READ_DISC_INFO;
1735         cgc.cmd[8] = cgc.buflen = 2;
1736         cgc.quiet = 1;
1737
1738         if ((ret = pkt_generic_packet(pd, &cgc)))
1739                 return ret;
1740
1741         /* not all drives have the same disc_info length, so requeue
1742          * packet with the length the drive tells us it can supply
1743          */
1744         cgc.buflen = be16_to_cpu(di->disc_information_length) +
1745                      sizeof(di->disc_information_length);
1746
1747         if (cgc.buflen > sizeof(disc_information))
1748                 cgc.buflen = sizeof(disc_information);
1749
1750         cgc.cmd[8] = cgc.buflen;
1751         return pkt_generic_packet(pd, &cgc);
1752 }
1753
1754 static int pkt_get_track_info(struct pktcdvd_device *pd, __u16 track, __u8 type, track_information *ti)
1755 {
1756         struct packet_command cgc;
1757         int ret;
1758
1759         init_cdrom_command(&cgc, ti, 8, CGC_DATA_READ);
1760         cgc.cmd[0] = GPCMD_READ_TRACK_RZONE_INFO;
1761         cgc.cmd[1] = type & 3;
1762         cgc.cmd[4] = (track & 0xff00) >> 8;
1763         cgc.cmd[5] = track & 0xff;
1764         cgc.cmd[8] = 8;
1765         cgc.quiet = 1;
1766
1767         if ((ret = pkt_generic_packet(pd, &cgc)))
1768                 return ret;
1769
1770         cgc.buflen = be16_to_cpu(ti->track_information_length) +
1771                      sizeof(ti->track_information_length);
1772
1773         if (cgc.buflen > sizeof(track_information))
1774                 cgc.buflen = sizeof(track_information);
1775
1776         cgc.cmd[8] = cgc.buflen;
1777         return pkt_generic_packet(pd, &cgc);
1778 }
1779
1780 static noinline_for_stack int pkt_get_last_written(struct pktcdvd_device *pd,
1781                                                 long *last_written)
1782 {
1783         disc_information di;
1784         track_information ti;
1785         __u32 last_track;
1786         int ret = -1;
1787
1788         if ((ret = pkt_get_disc_info(pd, &di)))
1789                 return ret;
1790
1791         last_track = (di.last_track_msb << 8) | di.last_track_lsb;
1792         if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1793                 return ret;
1794
1795         /* if this track is blank, try the previous. */
1796         if (ti.blank) {
1797                 last_track--;
1798                 if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1799                         return ret;
1800         }
1801
1802         /* if last recorded field is valid, return it. */
1803         if (ti.lra_v) {
1804                 *last_written = be32_to_cpu(ti.last_rec_address);
1805         } else {
1806                 /* make it up instead */
1807                 *last_written = be32_to_cpu(ti.track_start) +
1808                                 be32_to_cpu(ti.track_size);
1809                 if (ti.free_blocks)
1810                         *last_written -= (be32_to_cpu(ti.free_blocks) + 7);
1811         }
1812         return 0;
1813 }
1814
1815 /*
1816  * write mode select package based on pd->settings
1817  */
1818 static noinline_for_stack int pkt_set_write_settings(struct pktcdvd_device *pd)
1819 {
1820         struct packet_command cgc;
1821         struct request_sense sense;
1822         write_param_page *wp;
1823         char buffer[128];
1824         int ret, size;
1825
1826         /* doesn't apply to DVD+RW or DVD-RAM */
1827         if ((pd->mmc3_profile == 0x1a) || (pd->mmc3_profile == 0x12))
1828                 return 0;
1829
1830         memset(buffer, 0, sizeof(buffer));
1831         init_cdrom_command(&cgc, buffer, sizeof(*wp), CGC_DATA_READ);
1832         cgc.sense = &sense;
1833         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1834                 pkt_dump_sense(&cgc);
1835                 return ret;
1836         }
1837
1838         size = 2 + ((buffer[0] << 8) | (buffer[1] & 0xff));
1839         pd->mode_offset = (buffer[6] << 8) | (buffer[7] & 0xff);
1840         if (size > sizeof(buffer))
1841                 size = sizeof(buffer);
1842
1843         /*
1844          * now get it all
1845          */
1846         init_cdrom_command(&cgc, buffer, size, CGC_DATA_READ);
1847         cgc.sense = &sense;
1848         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1849                 pkt_dump_sense(&cgc);
1850                 return ret;
1851         }
1852
1853         /*
1854          * write page is offset header + block descriptor length
1855          */
1856         wp = (write_param_page *) &buffer[sizeof(struct mode_page_header) + pd->mode_offset];
1857
1858         wp->fp = pd->settings.fp;
1859         wp->track_mode = pd->settings.track_mode;
1860         wp->write_type = pd->settings.write_type;
1861         wp->data_block_type = pd->settings.block_mode;
1862
1863         wp->multi_session = 0;
1864
1865 #ifdef PACKET_USE_LS
1866         wp->link_size = 7;
1867         wp->ls_v = 1;
1868 #endif
1869
1870         if (wp->data_block_type == PACKET_BLOCK_MODE1) {
1871                 wp->session_format = 0;
1872                 wp->subhdr2 = 0x20;
1873         } else if (wp->data_block_type == PACKET_BLOCK_MODE2) {
1874                 wp->session_format = 0x20;
1875                 wp->subhdr2 = 8;
1876 #if 0
1877                 wp->mcn[0] = 0x80;
1878                 memcpy(&wp->mcn[1], PACKET_MCN, sizeof(wp->mcn) - 1);
1879 #endif
1880         } else {
1881                 /*
1882                  * paranoia
1883                  */
1884                 printk(DRIVER_NAME": write mode wrong %d\n", wp->data_block_type);
1885                 return 1;
1886         }
1887         wp->packet_size = cpu_to_be32(pd->settings.size >> 2);
1888
1889         cgc.buflen = cgc.cmd[8] = size;
1890         if ((ret = pkt_mode_select(pd, &cgc))) {
1891                 pkt_dump_sense(&cgc);
1892                 return ret;
1893         }
1894
1895         pkt_print_settings(pd);
1896         return 0;
1897 }
1898
1899 /*
1900  * 1 -- we can write to this track, 0 -- we can't
1901  */
1902 static int pkt_writable_track(struct pktcdvd_device *pd, track_information *ti)
1903 {
1904         switch (pd->mmc3_profile) {
1905                 case 0x1a: /* DVD+RW */
1906                 case 0x12: /* DVD-RAM */
1907                         /* The track is always writable on DVD+RW/DVD-RAM */
1908                         return 1;
1909                 default:
1910                         break;
1911         }
1912
1913         if (!ti->packet || !ti->fp)
1914                 return 0;
1915
1916         /*
1917          * "good" settings as per Mt Fuji.
1918          */
1919         if (ti->rt == 0 && ti->blank == 0)
1920                 return 1;
1921
1922         if (ti->rt == 0 && ti->blank == 1)
1923                 return 1;
1924
1925         if (ti->rt == 1 && ti->blank == 0)
1926                 return 1;
1927
1928         printk(DRIVER_NAME": bad state %d-%d-%d\n", ti->rt, ti->blank, ti->packet);
1929         return 0;
1930 }
1931
1932 /*
1933  * 1 -- we can write to this disc, 0 -- we can't
1934  */
1935 static int pkt_writable_disc(struct pktcdvd_device *pd, disc_information *di)
1936 {
1937         switch (pd->mmc3_profile) {
1938                 case 0x0a: /* CD-RW */
1939                 case 0xffff: /* MMC3 not supported */
1940                         break;
1941                 case 0x1a: /* DVD+RW */
1942                 case 0x13: /* DVD-RW */
1943                 case 0x12: /* DVD-RAM */
1944                         return 1;
1945                 default:
1946                         VPRINTK(DRIVER_NAME": Wrong disc profile (%x)\n", pd->mmc3_profile);
1947                         return 0;
1948         }
1949
1950         /*
1951          * for disc type 0xff we should probably reserve a new track.
1952          * but i'm not sure, should we leave this to user apps? probably.
1953          */
1954         if (di->disc_type == 0xff) {
1955                 printk(DRIVER_NAME": Unknown disc. No track?\n");
1956                 return 0;
1957         }
1958
1959         if (di->disc_type != 0x20 && di->disc_type != 0) {
1960                 printk(DRIVER_NAME": Wrong disc type (%x)\n", di->disc_type);
1961                 return 0;
1962         }
1963
1964         if (di->erasable == 0) {
1965                 printk(DRIVER_NAME": Disc not erasable\n");
1966                 return 0;
1967         }
1968
1969         if (di->border_status == PACKET_SESSION_RESERVED) {
1970                 printk(DRIVER_NAME": Can't write to last track (reserved)\n");
1971                 return 0;
1972         }
1973
1974         return 1;
1975 }
1976
1977 static noinline_for_stack int pkt_probe_settings(struct pktcdvd_device *pd)
1978 {
1979         struct packet_command cgc;
1980         unsigned char buf[12];
1981         disc_information di;
1982         track_information ti;
1983         int ret, track;
1984
1985         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
1986         cgc.cmd[0] = GPCMD_GET_CONFIGURATION;
1987         cgc.cmd[8] = 8;
1988         ret = pkt_generic_packet(pd, &cgc);
1989         pd->mmc3_profile = ret ? 0xffff : buf[6] << 8 | buf[7];
1990
1991         memset(&di, 0, sizeof(disc_information));
1992         memset(&ti, 0, sizeof(track_information));
1993
1994         if ((ret = pkt_get_disc_info(pd, &di))) {
1995                 printk("failed get_disc\n");
1996                 return ret;
1997         }
1998
1999         if (!pkt_writable_disc(pd, &di))
2000                 return -EROFS;
2001
2002         pd->type = di.erasable ? PACKET_CDRW : PACKET_CDR;
2003
2004         track = 1; /* (di.last_track_msb << 8) | di.last_track_lsb; */
2005         if ((ret = pkt_get_track_info(pd, track, 1, &ti))) {
2006                 printk(DRIVER_NAME": failed get_track\n");
2007                 return ret;
2008         }
2009
2010         if (!pkt_writable_track(pd, &ti)) {
2011                 printk(DRIVER_NAME": can't write to this track\n");
2012                 return -EROFS;
2013         }
2014
2015         /*
2016          * we keep packet size in 512 byte units, makes it easier to
2017          * deal with request calculations.
2018          */
2019         pd->settings.size = be32_to_cpu(ti.fixed_packet_size) << 2;
2020         if (pd->settings.size == 0) {
2021                 printk(DRIVER_NAME": detected zero packet size!\n");
2022                 return -ENXIO;
2023         }
2024         if (pd->settings.size > PACKET_MAX_SECTORS) {
2025                 printk(DRIVER_NAME": packet size is too big\n");
2026                 return -EROFS;
2027         }
2028         pd->settings.fp = ti.fp;
2029         pd->offset = (be32_to_cpu(ti.track_start) << 2) & (pd->settings.size - 1);
2030
2031         if (ti.nwa_v) {
2032                 pd->nwa = be32_to_cpu(ti.next_writable);
2033                 set_bit(PACKET_NWA_VALID, &pd->flags);
2034         }
2035
2036         /*
2037          * in theory we could use lra on -RW media as well and just zero
2038          * blocks that haven't been written yet, but in practice that
2039          * is just a no-go. we'll use that for -R, naturally.
2040          */
2041         if (ti.lra_v) {
2042                 pd->lra = be32_to_cpu(ti.last_rec_address);
2043                 set_bit(PACKET_LRA_VALID, &pd->flags);
2044         } else {
2045                 pd->lra = 0xffffffff;
2046                 set_bit(PACKET_LRA_VALID, &pd->flags);
2047         }
2048
2049         /*
2050          * fine for now
2051          */
2052         pd->settings.link_loss = 7;
2053         pd->settings.write_type = 0;    /* packet */
2054         pd->settings.track_mode = ti.track_mode;
2055
2056         /*
2057          * mode1 or mode2 disc
2058          */
2059         switch (ti.data_mode) {
2060                 case PACKET_MODE1:
2061                         pd->settings.block_mode = PACKET_BLOCK_MODE1;
2062                         break;
2063                 case PACKET_MODE2:
2064                         pd->settings.block_mode = PACKET_BLOCK_MODE2;
2065                         break;
2066                 default:
2067                         printk(DRIVER_NAME": unknown data mode\n");
2068                         return -EROFS;
2069         }
2070         return 0;
2071 }
2072
2073 /*
2074  * enable/disable write caching on drive
2075  */
2076 static noinline_for_stack int pkt_write_caching(struct pktcdvd_device *pd,
2077                                                 int set)
2078 {
2079         struct packet_command cgc;
2080         struct request_sense sense;
2081         unsigned char buf[64];
2082         int ret;
2083
2084         memset(buf, 0, sizeof(buf));
2085         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
2086         cgc.sense = &sense;
2087         cgc.buflen = pd->mode_offset + 12;
2088
2089         /*
2090          * caching mode page might not be there, so quiet this command
2091          */
2092         cgc.quiet = 1;
2093
2094         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WCACHING_PAGE, 0)))
2095                 return ret;
2096
2097         buf[pd->mode_offset + 10] |= (!!set << 2);
2098
2099         cgc.buflen = cgc.cmd[8] = 2 + ((buf[0] << 8) | (buf[1] & 0xff));
2100         ret = pkt_mode_select(pd, &cgc);
2101         if (ret) {
2102                 printk(DRIVER_NAME": write caching control failed\n");
2103                 pkt_dump_sense(&cgc);
2104         } else if (!ret && set)
2105                 printk(DRIVER_NAME": enabled write caching on %s\n", pd->name);
2106         return ret;
2107 }
2108
2109 static int pkt_lock_door(struct pktcdvd_device *pd, int lockflag)
2110 {
2111         struct packet_command cgc;
2112
2113         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2114         cgc.cmd[0] = GPCMD_PREVENT_ALLOW_MEDIUM_REMOVAL;
2115         cgc.cmd[4] = lockflag ? 1 : 0;
2116         return pkt_generic_packet(pd, &cgc);
2117 }
2118
2119 /*
2120  * Returns drive maximum write speed
2121  */
2122 static noinline_for_stack int pkt_get_max_speed(struct pktcdvd_device *pd,
2123                                                 unsigned *write_speed)
2124 {
2125         struct packet_command cgc;
2126         struct request_sense sense;
2127         unsigned char buf[256+18];
2128         unsigned char *cap_buf;
2129         int ret, offset;
2130
2131         memset(buf, 0, sizeof(buf));
2132         cap_buf = &buf[sizeof(struct mode_page_header) + pd->mode_offset];
2133         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_UNKNOWN);
2134         cgc.sense = &sense;
2135
2136         ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2137         if (ret) {
2138                 cgc.buflen = pd->mode_offset + cap_buf[1] + 2 +
2139                              sizeof(struct mode_page_header);
2140                 ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2141                 if (ret) {
2142                         pkt_dump_sense(&cgc);
2143                         return ret;
2144                 }
2145         }
2146
2147         offset = 20;                        /* Obsoleted field, used by older drives */
2148         if (cap_buf[1] >= 28)
2149                 offset = 28;                /* Current write speed selected */
2150         if (cap_buf[1] >= 30) {
2151                 /* If the drive reports at least one "Logical Unit Write
2152                  * Speed Performance Descriptor Block", use the information
2153                  * in the first block. (contains the highest speed)
2154                  */
2155                 int num_spdb = (cap_buf[30] << 8) + cap_buf[31];
2156                 if (num_spdb > 0)
2157                         offset = 34;
2158         }
2159
2160         *write_speed = (cap_buf[offset] << 8) | cap_buf[offset + 1];
2161         return 0;
2162 }
2163
2164 /* These tables from cdrecord - I don't have orange book */
2165 /* standard speed CD-RW (1-4x) */
2166 static char clv_to_speed[16] = {
2167         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2168            0, 2, 4, 6, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2169 };
2170 /* high speed CD-RW (-10x) */
2171 static char hs_clv_to_speed[16] = {
2172         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2173            0, 2, 4, 6, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2174 };
2175 /* ultra high speed CD-RW */
2176 static char us_clv_to_speed[16] = {
2177         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2178            0, 2, 4, 8, 0, 0,16, 0,24,32,40,48, 0, 0, 0, 0
2179 };
2180
2181 /*
2182  * reads the maximum media speed from ATIP
2183  */
2184 static noinline_for_stack int pkt_media_speed(struct pktcdvd_device *pd,
2185                                                 unsigned *speed)
2186 {
2187         struct packet_command cgc;
2188         struct request_sense sense;
2189         unsigned char buf[64];
2190         unsigned int size, st, sp;
2191         int ret;
2192
2193         init_cdrom_command(&cgc, buf, 2, CGC_DATA_READ);
2194         cgc.sense = &sense;
2195         cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2196         cgc.cmd[1] = 2;
2197         cgc.cmd[2] = 4; /* READ ATIP */
2198         cgc.cmd[8] = 2;
2199         ret = pkt_generic_packet(pd, &cgc);
2200         if (ret) {
2201                 pkt_dump_sense(&cgc);
2202                 return ret;
2203         }
2204         size = ((unsigned int) buf[0]<<8) + buf[1] + 2;
2205         if (size > sizeof(buf))
2206                 size = sizeof(buf);
2207
2208         init_cdrom_command(&cgc, buf, size, CGC_DATA_READ);
2209         cgc.sense = &sense;
2210         cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2211         cgc.cmd[1] = 2;
2212         cgc.cmd[2] = 4;
2213         cgc.cmd[8] = size;
2214         ret = pkt_generic_packet(pd, &cgc);
2215         if (ret) {
2216                 pkt_dump_sense(&cgc);
2217                 return ret;
2218         }
2219
2220         if (!(buf[6] & 0x40)) {
2221                 printk(DRIVER_NAME": Disc type is not CD-RW\n");
2222                 return 1;
2223         }
2224         if (!(buf[6] & 0x4)) {
2225                 printk(DRIVER_NAME": A1 values on media are not valid, maybe not CDRW?\n");
2226                 return 1;
2227         }
2228
2229         st = (buf[6] >> 3) & 0x7; /* disc sub-type */
2230
2231         sp = buf[16] & 0xf; /* max speed from ATIP A1 field */
2232
2233         /* Info from cdrecord */
2234         switch (st) {
2235                 case 0: /* standard speed */
2236                         *speed = clv_to_speed[sp];
2237                         break;
2238                 case 1: /* high speed */
2239                         *speed = hs_clv_to_speed[sp];
2240                         break;
2241                 case 2: /* ultra high speed */
2242                         *speed = us_clv_to_speed[sp];
2243                         break;
2244                 default:
2245                         printk(DRIVER_NAME": Unknown disc sub-type %d\n",st);
2246                         return 1;
2247         }
2248         if (*speed) {
2249                 printk(DRIVER_NAME": Max. media speed: %d\n",*speed);
2250                 return 0;
2251         } else {
2252                 printk(DRIVER_NAME": Unknown speed %d for sub-type %d\n",sp,st);
2253                 return 1;
2254         }
2255 }
2256
2257 static noinline_for_stack int pkt_perform_opc(struct pktcdvd_device *pd)
2258 {
2259         struct packet_command cgc;
2260         struct request_sense sense;
2261         int ret;
2262
2263         VPRINTK(DRIVER_NAME": Performing OPC\n");
2264
2265         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2266         cgc.sense = &sense;
2267         cgc.timeout = 60*HZ;
2268         cgc.cmd[0] = GPCMD_SEND_OPC;
2269         cgc.cmd[1] = 1;
2270         if ((ret = pkt_generic_packet(pd, &cgc)))
2271                 pkt_dump_sense(&cgc);
2272         return ret;
2273 }
2274
2275 static int pkt_open_write(struct pktcdvd_device *pd)
2276 {
2277         int ret;
2278         unsigned int write_speed, media_write_speed, read_speed;
2279
2280         if ((ret = pkt_probe_settings(pd))) {
2281                 VPRINTK(DRIVER_NAME": %s failed probe\n", pd->name);
2282                 return ret;
2283         }
2284
2285         if ((ret = pkt_set_write_settings(pd))) {
2286                 DPRINTK(DRIVER_NAME": %s failed saving write settings\n", pd->name);
2287                 return -EIO;
2288         }
2289
2290         pkt_write_caching(pd, USE_WCACHING);
2291
2292         if ((ret = pkt_get_max_speed(pd, &write_speed)))
2293                 write_speed = 16 * 177;
2294         switch (pd->mmc3_profile) {
2295                 case 0x13: /* DVD-RW */
2296                 case 0x1a: /* DVD+RW */
2297                 case 0x12: /* DVD-RAM */
2298                         DPRINTK(DRIVER_NAME": write speed %ukB/s\n", write_speed);
2299                         break;
2300                 default:
2301                         if ((ret = pkt_media_speed(pd, &media_write_speed)))
2302                                 media_write_speed = 16;
2303                         write_speed = min(write_speed, media_write_speed * 177);
2304                         DPRINTK(DRIVER_NAME": write speed %ux\n", write_speed / 176);
2305                         break;
2306         }
2307         read_speed = write_speed;
2308
2309         if ((ret = pkt_set_speed(pd, write_speed, read_speed))) {
2310                 DPRINTK(DRIVER_NAME": %s couldn't set write speed\n", pd->name);
2311                 return -EIO;
2312         }
2313         pd->write_speed = write_speed;
2314         pd->read_speed = read_speed;
2315
2316         if ((ret = pkt_perform_opc(pd))) {
2317                 DPRINTK(DRIVER_NAME": %s Optimum Power Calibration failed\n", pd->name);
2318         }
2319
2320         return 0;
2321 }
2322
2323 /*
2324  * called at open time.
2325  */
2326 static int pkt_open_dev(struct pktcdvd_device *pd, int write)
2327 {
2328         int ret;
2329         long lba;
2330         struct request_queue *q;
2331
2332         /*
2333          * We need to re-open the cdrom device without O_NONBLOCK to be able
2334          * to read/write from/to it. It is already opened in O_NONBLOCK mode
2335          * so bdget() can't fail.
2336          */
2337         bdget(pd->bdev->bd_dev);
2338         if ((ret = blkdev_get(pd->bdev, FMODE_READ, O_RDONLY)))
2339                 goto out;
2340
2341         if ((ret = bd_claim(pd->bdev, pd)))
2342                 goto out_putdev;
2343
2344         if ((ret = pkt_get_last_written(pd, &lba))) {
2345                 printk(DRIVER_NAME": pkt_get_last_written failed\n");
2346                 goto out_unclaim;
2347         }
2348
2349         set_capacity(pd->disk, lba << 2);
2350         set_capacity(pd->bdev->bd_disk, lba << 2);
2351         bd_set_size(pd->bdev, (loff_t)lba << 11);
2352
2353         q = bdev_get_queue(pd->bdev);
2354         if (write) {
2355                 if ((ret = pkt_open_write(pd)))
2356                         goto out_unclaim;
2357                 /*
2358                  * Some CDRW drives can not handle writes larger than one packet,
2359                  * even if the size is a multiple of the packet size.
2360                  */
2361                 spin_lock_irq(q->queue_lock);
2362                 blk_queue_max_sectors(q, pd->settings.size);
2363                 spin_unlock_irq(q->queue_lock);
2364                 set_bit(PACKET_WRITABLE, &pd->flags);
2365         } else {
2366                 pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2367                 clear_bit(PACKET_WRITABLE, &pd->flags);
2368         }
2369
2370         if ((ret = pkt_set_segment_merging(pd, q)))
2371                 goto out_unclaim;
2372
2373         if (write) {
2374                 if (!pkt_grow_pktlist(pd, CONFIG_CDROM_PKTCDVD_BUFFERS)) {
2375                         printk(DRIVER_NAME": not enough memory for buffers\n");
2376                         ret = -ENOMEM;
2377                         goto out_unclaim;
2378                 }
2379                 printk(DRIVER_NAME": %lukB available on disc\n", lba << 1);
2380         }
2381
2382         return 0;
2383
2384 out_unclaim:
2385         bd_release(pd->bdev);
2386 out_putdev:
2387         blkdev_put(pd->bdev);
2388 out:
2389         return ret;
2390 }
2391
2392 /*
2393  * called when the device is closed. makes sure that the device flushes
2394  * the internal cache before we close.
2395  */
2396 static void pkt_release_dev(struct pktcdvd_device *pd, int flush)
2397 {
2398         if (flush && pkt_flush_cache(pd))
2399                 DPRINTK(DRIVER_NAME": %s not flushing cache\n", pd->name);
2400
2401         pkt_lock_door(pd, 0);
2402
2403         pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2404         bd_release(pd->bdev);
2405         blkdev_put(pd->bdev);
2406
2407         pkt_shrink_pktlist(pd);
2408 }
2409
2410 static struct pktcdvd_device *pkt_find_dev_from_minor(int dev_minor)
2411 {
2412         if (dev_minor >= MAX_WRITERS)
2413                 return NULL;
2414         return pkt_devs[dev_minor];
2415 }
2416
2417 static int pkt_open(struct inode *inode, struct file *file)
2418 {
2419         struct pktcdvd_device *pd = NULL;
2420         int ret;
2421
2422         VPRINTK(DRIVER_NAME": entering open\n");
2423
2424         mutex_lock(&ctl_mutex);
2425         pd = pkt_find_dev_from_minor(iminor(inode));
2426         if (!pd) {
2427                 ret = -ENODEV;
2428                 goto out;
2429         }
2430         BUG_ON(pd->refcnt < 0);
2431
2432         pd->refcnt++;
2433         if (pd->refcnt > 1) {
2434                 if ((file->f_mode & FMODE_WRITE) &&
2435                     !test_bit(PACKET_WRITABLE, &pd->flags)) {
2436                         ret = -EBUSY;
2437                         goto out_dec;
2438                 }
2439         } else {
2440                 ret = pkt_open_dev(pd, file->f_mode & FMODE_WRITE);
2441                 if (ret)
2442                         goto out_dec;
2443                 /*
2444                  * needed here as well, since ext2 (among others) may change
2445                  * the blocksize at mount time
2446                  */
2447                 set_blocksize(inode->i_bdev, CD_FRAMESIZE);
2448         }
2449
2450         mutex_unlock(&ctl_mutex);
2451         return 0;
2452
2453 out_dec:
2454         pd->refcnt--;
2455 out:
2456         VPRINTK(DRIVER_NAME": failed open (%d)\n", ret);
2457         mutex_unlock(&ctl_mutex);
2458         return ret;
2459 }
2460
2461 static int pkt_close(struct inode *inode, struct file *file)
2462 {
2463         struct pktcdvd_device *pd = inode->i_bdev->bd_disk->private_data;
2464         int ret = 0;
2465
2466         mutex_lock(&ctl_mutex);
2467         pd->refcnt--;
2468         BUG_ON(pd->refcnt < 0);
2469         if (pd->refcnt == 0) {
2470                 int flush = test_bit(PACKET_WRITABLE, &pd->flags);
2471                 pkt_release_dev(pd, flush);
2472         }
2473         mutex_unlock(&ctl_mutex);
2474         return ret;
2475 }
2476
2477
2478 static void pkt_end_io_read_cloned(struct bio *bio, int err)
2479 {
2480         struct packet_stacked_data *psd = bio->bi_private;
2481         struct pktcdvd_device *pd = psd->pd;
2482
2483         bio_put(bio);
2484         bio_endio(psd->bio, err);
2485         mempool_free(psd, psd_pool);
2486         pkt_bio_finished(pd);
2487 }
2488
2489 static int pkt_make_request(struct request_queue *q, struct bio *bio)
2490 {
2491         struct pktcdvd_device *pd;
2492         char b[BDEVNAME_SIZE];
2493         sector_t zone;
2494         struct packet_data *pkt;
2495         int was_empty, blocked_bio;
2496         struct pkt_rb_node *node;
2497
2498         pd = q->queuedata;
2499         if (!pd) {
2500                 printk(DRIVER_NAME": %s incorrect request queue\n", bdevname(bio->bi_bdev, b));
2501                 goto end_io;
2502         }
2503
2504         /*
2505          * Clone READ bios so we can have our own bi_end_io callback.
2506          */
2507         if (bio_data_dir(bio) == READ) {
2508                 struct bio *cloned_bio = bio_clone(bio, GFP_NOIO);
2509                 struct packet_stacked_data *psd = mempool_alloc(psd_pool, GFP_NOIO);
2510
2511                 psd->pd = pd;
2512                 psd->bio = bio;
2513                 cloned_bio->bi_bdev = pd->bdev;
2514                 cloned_bio->bi_private = psd;
2515                 cloned_bio->bi_end_io = pkt_end_io_read_cloned;
2516                 pd->stats.secs_r += bio->bi_size >> 9;
2517                 pkt_queue_bio(pd, cloned_bio);
2518                 return 0;
2519         }
2520
2521         if (!test_bit(PACKET_WRITABLE, &pd->flags)) {
2522                 printk(DRIVER_NAME": WRITE for ro device %s (%llu)\n",
2523                         pd->name, (unsigned long long)bio->bi_sector);
2524                 goto end_io;
2525         }
2526
2527         if (!bio->bi_size || (bio->bi_size % CD_FRAMESIZE)) {
2528                 printk(DRIVER_NAME": wrong bio size\n");
2529                 goto end_io;
2530         }
2531
2532         blk_queue_bounce(q, &bio);
2533
2534         zone = ZONE(bio->bi_sector, pd);
2535         VPRINTK("pkt_make_request: start = %6llx stop = %6llx\n",
2536                 (unsigned long long)bio->bi_sector,
2537                 (unsigned long long)(bio->bi_sector + bio_sectors(bio)));
2538
2539         /* Check if we have to split the bio */
2540         {
2541                 struct bio_pair *bp;
2542                 sector_t last_zone;
2543                 int first_sectors;
2544
2545                 last_zone = ZONE(bio->bi_sector + bio_sectors(bio) - 1, pd);
2546                 if (last_zone != zone) {
2547                         BUG_ON(last_zone != zone + pd->settings.size);
2548                         first_sectors = last_zone - bio->bi_sector;
2549                         bp = bio_split(bio, bio_split_pool, first_sectors);
2550                         BUG_ON(!bp);
2551                         pkt_make_request(q, &bp->bio1);
2552                         pkt_make_request(q, &bp->bio2);
2553                         bio_pair_release(bp);
2554                         return 0;
2555                 }
2556         }
2557
2558         /*
2559          * If we find a matching packet in state WAITING or READ_WAIT, we can
2560          * just append this bio to that packet.
2561          */
2562         spin_lock(&pd->cdrw.active_list_lock);
2563         blocked_bio = 0;
2564         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
2565                 if (pkt->sector == zone) {
2566                         spin_lock(&pkt->lock);
2567                         if ((pkt->state == PACKET_WAITING_STATE) ||
2568                             (pkt->state == PACKET_READ_WAIT_STATE)) {
2569                                 pkt_add_list_last(bio, &pkt->orig_bios,
2570                                                   &pkt->orig_bios_tail);
2571                                 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
2572                                 if ((pkt->write_size >= pkt->frames) &&
2573                                     (pkt->state == PACKET_WAITING_STATE)) {
2574                                         atomic_inc(&pkt->run_sm);
2575                                         wake_up(&pd->wqueue);
2576                                 }
2577                                 spin_unlock(&pkt->lock);
2578                                 spin_unlock(&pd->cdrw.active_list_lock);
2579                                 return 0;
2580                         } else {
2581                                 blocked_bio = 1;
2582                         }
2583                         spin_unlock(&pkt->lock);
2584                 }
2585         }
2586         spin_unlock(&pd->cdrw.active_list_lock);
2587
2588         /*
2589          * Test if there is enough room left in the bio work queue
2590          * (queue size >= congestion on mark).
2591          * If not, wait till the work queue size is below the congestion off mark.
2592          */
2593         spin_lock(&pd->lock);
2594         if (pd->write_congestion_on > 0
2595             && pd->bio_queue_size >= pd->write_congestion_on) {
2596                 set_bdi_congested(&q->backing_dev_info, WRITE);
2597                 do {
2598                         spin_unlock(&pd->lock);
2599                         congestion_wait(WRITE, HZ);
2600                         spin_lock(&pd->lock);
2601                 } while(pd->bio_queue_size > pd->write_congestion_off);
2602         }
2603         spin_unlock(&pd->lock);
2604
2605         /*
2606          * No matching packet found. Store the bio in the work queue.
2607          */
2608         node = mempool_alloc(pd->rb_pool, GFP_NOIO);
2609         node->bio = bio;
2610         spin_lock(&pd->lock);
2611         BUG_ON(pd->bio_queue_size < 0);
2612         was_empty = (pd->bio_queue_size == 0);
2613         pkt_rbtree_insert(pd, node);
2614         spin_unlock(&pd->lock);
2615
2616         /*
2617          * Wake up the worker thread.
2618          */
2619         atomic_set(&pd->scan_queue, 1);
2620         if (was_empty) {
2621                 /* This wake_up is required for correct operation */
2622                 wake_up(&pd->wqueue);
2623         } else if (!list_empty(&pd->cdrw.pkt_free_list) && !blocked_bio) {
2624                 /*
2625                  * This wake up is not required for correct operation,
2626                  * but improves performance in some cases.
2627                  */
2628                 wake_up(&pd->wqueue);
2629         }
2630         return 0;
2631 end_io:
2632         bio_io_error(bio);
2633         return 0;
2634 }
2635
2636
2637
2638 static int pkt_merge_bvec(struct request_queue *q, struct bio *bio, struct bio_vec *bvec)
2639 {
2640         struct pktcdvd_device *pd = q->queuedata;
2641         sector_t zone = ZONE(bio->bi_sector, pd);
2642         int used = ((bio->bi_sector - zone) << 9) + bio->bi_size;
2643         int remaining = (pd->settings.size << 9) - used;
2644         int remaining2;
2645
2646         /*
2647          * A bio <= PAGE_SIZE must be allowed. If it crosses a packet
2648          * boundary, pkt_make_request() will split the bio.
2649          */
2650         remaining2 = PAGE_SIZE - bio->bi_size;
2651         remaining = max(remaining, remaining2);
2652
2653         BUG_ON(remaining < 0);
2654         return remaining;
2655 }
2656
2657 static void pkt_init_queue(struct pktcdvd_device *pd)
2658 {
2659         struct request_queue *q = pd->disk->queue;
2660
2661         blk_queue_make_request(q, pkt_make_request);
2662         blk_queue_hardsect_size(q, CD_FRAMESIZE);
2663         blk_queue_max_sectors(q, PACKET_MAX_SECTORS);
2664         blk_queue_merge_bvec(q, pkt_merge_bvec);
2665         q->queuedata = pd;
2666 }
2667
2668 static int pkt_seq_show(struct seq_file *m, void *p)
2669 {
2670         struct pktcdvd_device *pd = m->private;
2671         char *msg;
2672         char bdev_buf[BDEVNAME_SIZE];
2673         int states[PACKET_NUM_STATES];
2674
2675         seq_printf(m, "Writer %s mapped to %s:\n", pd->name,
2676                    bdevname(pd->bdev, bdev_buf));
2677
2678         seq_printf(m, "\nSettings:\n");
2679         seq_printf(m, "\tpacket size:\t\t%dkB\n", pd->settings.size / 2);
2680
2681         if (pd->settings.write_type == 0)
2682                 msg = "Packet";
2683         else
2684                 msg = "Unknown";
2685         seq_printf(m, "\twrite type:\t\t%s\n", msg);
2686
2687         seq_printf(m, "\tpacket type:\t\t%s\n", pd->settings.fp ? "Fixed" : "Variable");
2688         seq_printf(m, "\tlink loss:\t\t%d\n", pd->settings.link_loss);
2689
2690         seq_printf(m, "\ttrack mode:\t\t%d\n", pd->settings.track_mode);
2691
2692         if (pd->settings.block_mode == PACKET_BLOCK_MODE1)
2693                 msg = "Mode 1";
2694         else if (pd->settings.block_mode == PACKET_BLOCK_MODE2)
2695                 msg = "Mode 2";
2696         else
2697                 msg = "Unknown";
2698         seq_printf(m, "\tblock mode:\t\t%s\n", msg);
2699
2700         seq_printf(m, "\nStatistics:\n");
2701         seq_printf(m, "\tpackets started:\t%lu\n", pd->stats.pkt_started);
2702         seq_printf(m, "\tpackets ended:\t\t%lu\n", pd->stats.pkt_ended);
2703         seq_printf(m, "\twritten:\t\t%lukB\n", pd->stats.secs_w >> 1);
2704         seq_printf(m, "\tread gather:\t\t%lukB\n", pd->stats.secs_rg >> 1);
2705         seq_printf(m, "\tread:\t\t\t%lukB\n", pd->stats.secs_r >> 1);
2706
2707         seq_printf(m, "\nMisc:\n");
2708         seq_printf(m, "\treference count:\t%d\n", pd->refcnt);
2709         seq_printf(m, "\tflags:\t\t\t0x%lx\n", pd->flags);
2710         seq_printf(m, "\tread speed:\t\t%ukB/s\n", pd->read_speed);
2711         seq_printf(m, "\twrite speed:\t\t%ukB/s\n", pd->write_speed);
2712         seq_printf(m, "\tstart offset:\t\t%lu\n", pd->offset);
2713         seq_printf(m, "\tmode page offset:\t%u\n", pd->mode_offset);
2714
2715         seq_printf(m, "\nQueue state:\n");
2716         seq_printf(m, "\tbios queued:\t\t%d\n", pd->bio_queue_size);
2717         seq_printf(m, "\tbios pending:\t\t%d\n", atomic_read(&pd->cdrw.pending_bios));
2718         seq_printf(m, "\tcurrent sector:\t\t0x%llx\n", (unsigned long long)pd->current_sector);
2719
2720         pkt_count_states(pd, states);
2721         seq_printf(m, "\tstate:\t\t\ti:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
2722                    states[0], states[1], states[2], states[3], states[4], states[5]);
2723
2724         seq_printf(m, "\twrite congestion marks:\toff=%d on=%d\n",
2725                         pd->write_congestion_off,
2726                         pd->write_congestion_on);
2727         return 0;
2728 }
2729
2730 static int pkt_seq_open(struct inode *inode, struct file *file)
2731 {
2732         return single_open(file, pkt_seq_show, PDE(inode)->data);
2733 }
2734
2735 static const struct file_operations pkt_proc_fops = {
2736         .open   = pkt_seq_open,
2737         .read   = seq_read,
2738         .llseek = seq_lseek,
2739         .release = single_release
2740 };
2741
2742 static int pkt_new_dev(struct pktcdvd_device *pd, dev_t dev)
2743 {
2744         int i;
2745         int ret = 0;
2746         char b[BDEVNAME_SIZE];
2747         struct proc_dir_entry *proc;
2748         struct block_device *bdev;
2749
2750         if (pd->pkt_dev == dev) {
2751                 printk(DRIVER_NAME": Recursive setup not allowed\n");
2752                 return -EBUSY;
2753         }
2754         for (i = 0; i < MAX_WRITERS; i++) {
2755                 struct pktcdvd_device *pd2 = pkt_devs[i];
2756                 if (!pd2)
2757                         continue;
2758                 if (pd2->bdev->bd_dev == dev) {
2759                         printk(DRIVER_NAME": %s already setup\n", bdevname(pd2->bdev, b));
2760                         return -EBUSY;
2761                 }
2762                 if (pd2->pkt_dev == dev) {
2763                         printk(DRIVER_NAME": Can't chain pktcdvd devices\n");
2764                         return -EBUSY;
2765                 }
2766         }
2767
2768         bdev = bdget(dev);
2769         if (!bdev)
2770                 return -ENOMEM;
2771         ret = blkdev_get(bdev, FMODE_READ, O_RDONLY | O_NONBLOCK);
2772         if (ret)
2773                 return ret;
2774
2775         /* This is safe, since we have a reference from open(). */
2776         __module_get(THIS_MODULE);
2777
2778         pd->bdev = bdev;
2779         set_blocksize(bdev, CD_FRAMESIZE);
2780
2781         pkt_init_queue(pd);
2782
2783         atomic_set(&pd->cdrw.pending_bios, 0);
2784         pd->cdrw.thread = kthread_run(kcdrwd, pd, "%s", pd->name);
2785         if (IS_ERR(pd->cdrw.thread)) {
2786                 printk(DRIVER_NAME": can't start kernel thread\n");
2787                 ret = -ENOMEM;
2788                 goto out_mem;
2789         }
2790
2791         proc = create_proc_entry(pd->name, 0, pkt_proc);
2792         if (proc) {
2793                 proc->data = pd;
2794                 proc->proc_fops = &pkt_proc_fops;
2795         }
2796         DPRINTK(DRIVER_NAME": writer %s mapped to %s\n", pd->name, bdevname(bdev, b));
2797         return 0;
2798
2799 out_mem:
2800         blkdev_put(bdev);
2801         /* This is safe: open() is still holding a reference. */
2802         module_put(THIS_MODULE);
2803         return ret;
2804 }
2805
2806 static int pkt_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg)
2807 {
2808         struct pktcdvd_device *pd = inode->i_bdev->bd_disk->private_data;
2809
2810         VPRINTK("pkt_ioctl: cmd %x, dev %d:%d\n", cmd, imajor(inode), iminor(inode));
2811
2812         switch (cmd) {
2813         /*
2814          * forward selected CDROM ioctls to CD-ROM, for UDF
2815          */
2816         case CDROMMULTISESSION:
2817         case CDROMREADTOCENTRY:
2818         case CDROM_LAST_WRITTEN:
2819         case CDROM_SEND_PACKET:
2820         case SCSI_IOCTL_SEND_COMMAND:
2821                 return blkdev_ioctl(pd->bdev->bd_inode, file, cmd, arg);
2822
2823         case CDROMEJECT:
2824                 /*
2825                  * The door gets locked when the device is opened, so we
2826                  * have to unlock it or else the eject command fails.
2827                  */
2828                 if (pd->refcnt == 1)
2829                         pkt_lock_door(pd, 0);
2830                 return blkdev_ioctl(pd->bdev->bd_inode, file, cmd, arg);
2831
2832         default:
2833                 VPRINTK(DRIVER_NAME": Unknown ioctl for %s (%x)\n", pd->name, cmd);
2834                 return -ENOTTY;
2835         }
2836
2837         return 0;
2838 }
2839
2840 static int pkt_media_changed(struct gendisk *disk)
2841 {
2842         struct pktcdvd_device *pd = disk->private_data;
2843         struct gendisk *attached_disk;
2844
2845         if (!pd)
2846                 return 0;
2847         if (!pd->bdev)
2848                 return 0;
2849         attached_disk = pd->bdev->bd_disk;
2850         if (!attached_disk)
2851                 return 0;
2852         return attached_disk->fops->media_changed(attached_disk);
2853 }
2854
2855 static struct block_device_operations pktcdvd_ops = {
2856         .owner =                THIS_MODULE,
2857         .open =                 pkt_open,
2858         .release =              pkt_close,
2859         .ioctl =                pkt_ioctl,
2860         .media_changed =        pkt_media_changed,
2861 };
2862
2863 /*
2864  * Set up mapping from pktcdvd device to CD-ROM device.
2865  */
2866 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev)
2867 {
2868         int idx;
2869         int ret = -ENOMEM;
2870         struct pktcdvd_device *pd;
2871         struct gendisk *disk;
2872
2873         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2874
2875         for (idx = 0; idx < MAX_WRITERS; idx++)
2876                 if (!pkt_devs[idx])
2877                         break;
2878         if (idx == MAX_WRITERS) {
2879                 printk(DRIVER_NAME": max %d writers supported\n", MAX_WRITERS);
2880                 ret = -EBUSY;
2881                 goto out_mutex;
2882         }
2883
2884         pd = kzalloc(sizeof(struct pktcdvd_device), GFP_KERNEL);
2885         if (!pd)
2886                 goto out_mutex;
2887
2888         pd->rb_pool = mempool_create_kmalloc_pool(PKT_RB_POOL_SIZE,
2889                                                   sizeof(struct pkt_rb_node));
2890         if (!pd->rb_pool)
2891                 goto out_mem;
2892
2893         INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
2894         INIT_LIST_HEAD(&pd->cdrw.pkt_active_list);
2895         spin_lock_init(&pd->cdrw.active_list_lock);
2896
2897         spin_lock_init(&pd->lock);
2898         spin_lock_init(&pd->iosched.lock);
2899         sprintf(pd->name, DRIVER_NAME"%d", idx);
2900         init_waitqueue_head(&pd->wqueue);
2901         pd->bio_queue = RB_ROOT;
2902
2903         pd->write_congestion_on  = write_congestion_on;
2904         pd->write_congestion_off = write_congestion_off;
2905
2906         disk = alloc_disk(1);
2907         if (!disk)
2908                 goto out_mem;
2909         pd->disk = disk;
2910         disk->major = pktdev_major;
2911         disk->first_minor = idx;
2912         disk->fops = &pktcdvd_ops;
2913         disk->flags = GENHD_FL_REMOVABLE;
2914         strcpy(disk->disk_name, pd->name);
2915         disk->private_data = pd;
2916         disk->queue = blk_alloc_queue(GFP_KERNEL);
2917         if (!disk->queue)
2918                 goto out_mem2;
2919
2920         pd->pkt_dev = MKDEV(disk->major, disk->first_minor);
2921         ret = pkt_new_dev(pd, dev);
2922         if (ret)
2923                 goto out_new_dev;
2924
2925         add_disk(disk);
2926
2927         pkt_sysfs_dev_new(pd);
2928         pkt_debugfs_dev_new(pd);
2929
2930         pkt_devs[idx] = pd;
2931         if (pkt_dev)
2932                 *pkt_dev = pd->pkt_dev;
2933
2934         mutex_unlock(&ctl_mutex);
2935         return 0;
2936
2937 out_new_dev:
2938         blk_cleanup_queue(disk->queue);
2939 out_mem2:
2940         put_disk(disk);
2941 out_mem:
2942         if (pd->rb_pool)
2943                 mempool_destroy(pd->rb_pool);
2944         kfree(pd);
2945 out_mutex:
2946         mutex_unlock(&ctl_mutex);
2947         printk(DRIVER_NAME": setup of pktcdvd device failed\n");
2948         return ret;
2949 }
2950
2951 /*
2952  * Tear down mapping from pktcdvd device to CD-ROM device.
2953  */
2954 static int pkt_remove_dev(dev_t pkt_dev)
2955 {
2956         struct pktcdvd_device *pd;
2957         int idx;
2958         int ret = 0;
2959
2960         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2961
2962         for (idx = 0; idx < MAX_WRITERS; idx++) {
2963                 pd = pkt_devs[idx];
2964                 if (pd && (pd->pkt_dev == pkt_dev))
2965                         break;
2966         }
2967         if (idx == MAX_WRITERS) {
2968                 DPRINTK(DRIVER_NAME": dev not setup\n");
2969                 ret = -ENXIO;
2970                 goto out;
2971         }
2972
2973         if (pd->refcnt > 0) {
2974                 ret = -EBUSY;
2975                 goto out;
2976         }
2977         if (!IS_ERR(pd->cdrw.thread))
2978                 kthread_stop(pd->cdrw.thread);
2979
2980         pkt_devs[idx] = NULL;
2981
2982         pkt_debugfs_dev_remove(pd);
2983         pkt_sysfs_dev_remove(pd);
2984
2985         blkdev_put(pd->bdev);
2986
2987         remove_proc_entry(pd->name, pkt_proc);
2988         DPRINTK(DRIVER_NAME": writer %s unmapped\n", pd->name);
2989
2990         del_gendisk(pd->disk);
2991         blk_cleanup_queue(pd->disk->queue);
2992         put_disk(pd->disk);
2993
2994         mempool_destroy(pd->rb_pool);
2995         kfree(pd);
2996
2997         /* This is safe: open() is still holding a reference. */
2998         module_put(THIS_MODULE);
2999
3000 out:
3001         mutex_unlock(&ctl_mutex);
3002         return ret;
3003 }
3004
3005 static void pkt_get_status(struct pkt_ctrl_command *ctrl_cmd)
3006 {
3007         struct pktcdvd_device *pd;
3008
3009         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3010
3011         pd = pkt_find_dev_from_minor(ctrl_cmd->dev_index);
3012         if (pd) {
3013                 ctrl_cmd->dev = new_encode_dev(pd->bdev->bd_dev);
3014                 ctrl_cmd->pkt_dev = new_encode_dev(pd->pkt_dev);
3015         } else {
3016                 ctrl_cmd->dev = 0;
3017                 ctrl_cmd->pkt_dev = 0;
3018         }
3019         ctrl_cmd->num_devices = MAX_WRITERS;
3020
3021         mutex_unlock(&ctl_mutex);
3022 }
3023
3024 static int pkt_ctl_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg)
3025 {
3026         void __user *argp = (void __user *)arg;
3027         struct pkt_ctrl_command ctrl_cmd;
3028         int ret = 0;
3029         dev_t pkt_dev = 0;
3030
3031         if (cmd != PACKET_CTRL_CMD)
3032                 return -ENOTTY;
3033
3034         if (copy_from_user(&ctrl_cmd, argp, sizeof(struct pkt_ctrl_command)))
3035                 return -EFAULT;
3036
3037         switch (ctrl_cmd.command) {
3038         case PKT_CTRL_CMD_SETUP:
3039                 if (!capable(CAP_SYS_ADMIN))
3040                         return -EPERM;
3041                 ret = pkt_setup_dev(new_decode_dev(ctrl_cmd.dev), &pkt_dev);
3042                 ctrl_cmd.pkt_dev = new_encode_dev(pkt_dev);
3043                 break;
3044         case PKT_CTRL_CMD_TEARDOWN:
3045                 if (!capable(CAP_SYS_ADMIN))
3046                         return -EPERM;
3047                 ret = pkt_remove_dev(new_decode_dev(ctrl_cmd.pkt_dev));
3048                 break;
3049         case PKT_CTRL_CMD_STATUS:
3050                 pkt_get_status(&ctrl_cmd);
3051                 break;
3052         default:
3053                 return -ENOTTY;
3054         }
3055
3056         if (copy_to_user(argp, &ctrl_cmd, sizeof(struct pkt_ctrl_command)))
3057                 return -EFAULT;
3058         return ret;
3059 }
3060
3061
3062 static const struct file_operations pkt_ctl_fops = {
3063         .ioctl   = pkt_ctl_ioctl,
3064         .owner   = THIS_MODULE,
3065 };
3066
3067 static struct miscdevice pkt_misc = {
3068         .minor          = MISC_DYNAMIC_MINOR,
3069         .name           = DRIVER_NAME,
3070         .fops           = &pkt_ctl_fops
3071 };
3072
3073 static int __init pkt_init(void)
3074 {
3075         int ret;
3076
3077         mutex_init(&ctl_mutex);
3078
3079         psd_pool = mempool_create_kmalloc_pool(PSD_POOL_SIZE,
3080                                         sizeof(struct packet_stacked_data));
3081         if (!psd_pool)
3082                 return -ENOMEM;
3083
3084         ret = register_blkdev(pktdev_major, DRIVER_NAME);
3085         if (ret < 0) {
3086                 printk(DRIVER_NAME": Unable to register block device\n");
3087                 goto out2;
3088         }
3089         if (!pktdev_major)
3090                 pktdev_major = ret;
3091
3092         ret = pkt_sysfs_init();
3093         if (ret)
3094                 goto out;
3095
3096         pkt_debugfs_init();
3097
3098         ret = misc_register(&pkt_misc);
3099         if (ret) {
3100                 printk(DRIVER_NAME": Unable to register misc device\n");
3101                 goto out_misc;
3102         }
3103
3104         pkt_proc = proc_mkdir(DRIVER_NAME, proc_root_driver);
3105
3106         return 0;
3107
3108 out_misc:
3109         pkt_debugfs_cleanup();
3110         pkt_sysfs_cleanup();
3111 out:
3112         unregister_blkdev(pktdev_major, DRIVER_NAME);
3113 out2:
3114         mempool_destroy(psd_pool);
3115         return ret;
3116 }
3117
3118 static void __exit pkt_exit(void)
3119 {
3120         remove_proc_entry(DRIVER_NAME, proc_root_driver);
3121         misc_deregister(&pkt_misc);
3122
3123         pkt_debugfs_cleanup();
3124         pkt_sysfs_cleanup();
3125
3126         unregister_blkdev(pktdev_major, DRIVER_NAME);
3127         mempool_destroy(psd_pool);
3128 }
3129
3130 MODULE_DESCRIPTION("Packet writing layer for CD/DVD drives");
3131 MODULE_AUTHOR("Jens Axboe <axboe@suse.de>");
3132 MODULE_LICENSE("GPL");
3133
3134 module_init(pkt_init);
3135 module_exit(pkt_exit);