dc7a8c352da2e7c5b44a3d9a4e7a02bfd029d0c2
[linux-2.6.git] / drivers / block / pktcdvd.c
1 /*
2  * Copyright (C) 2000 Jens Axboe <axboe@suse.de>
3  * Copyright (C) 2001-2004 Peter Osterlund <petero2@telia.com>
4  * Copyright (C) 2006 Thomas Maier <balagi@justmail.de>
5  *
6  * May be copied or modified under the terms of the GNU General Public
7  * License.  See linux/COPYING for more information.
8  *
9  * Packet writing layer for ATAPI and SCSI CD-RW, DVD+RW, DVD-RW and
10  * DVD-RAM devices.
11  *
12  * Theory of operation:
13  *
14  * At the lowest level, there is the standard driver for the CD/DVD device,
15  * typically ide-cd.c or sr.c. This driver can handle read and write requests,
16  * but it doesn't know anything about the special restrictions that apply to
17  * packet writing. One restriction is that write requests must be aligned to
18  * packet boundaries on the physical media, and the size of a write request
19  * must be equal to the packet size. Another restriction is that a
20  * GPCMD_FLUSH_CACHE command has to be issued to the drive before a read
21  * command, if the previous command was a write.
22  *
23  * The purpose of the packet writing driver is to hide these restrictions from
24  * higher layers, such as file systems, and present a block device that can be
25  * randomly read and written using 2kB-sized blocks.
26  *
27  * The lowest layer in the packet writing driver is the packet I/O scheduler.
28  * Its data is defined by the struct packet_iosched and includes two bio
29  * queues with pending read and write requests. These queues are processed
30  * by the pkt_iosched_process_queue() function. The write requests in this
31  * queue are already properly aligned and sized. This layer is responsible for
32  * issuing the flush cache commands and scheduling the I/O in a good order.
33  *
34  * The next layer transforms unaligned write requests to aligned writes. This
35  * transformation requires reading missing pieces of data from the underlying
36  * block device, assembling the pieces to full packets and queuing them to the
37  * packet I/O scheduler.
38  *
39  * At the top layer there is a custom make_request_fn function that forwards
40  * read requests directly to the iosched queue and puts write requests in the
41  * unaligned write queue. A kernel thread performs the necessary read
42  * gathering to convert the unaligned writes to aligned writes and then feeds
43  * them to the packet I/O scheduler.
44  *
45  *************************************************************************/
46
47 #include <linux/pktcdvd.h>
48 #include <linux/module.h>
49 #include <linux/types.h>
50 #include <linux/kernel.h>
51 #include <linux/kthread.h>
52 #include <linux/errno.h>
53 #include <linux/spinlock.h>
54 #include <linux/file.h>
55 #include <linux/proc_fs.h>
56 #include <linux/seq_file.h>
57 #include <linux/miscdevice.h>
58 #include <linux/freezer.h>
59 #include <linux/mutex.h>
60 #include <scsi/scsi_cmnd.h>
61 #include <scsi/scsi_ioctl.h>
62 #include <scsi/scsi.h>
63 #include <linux/debugfs.h>
64 #include <linux/device.h>
65
66 #include <asm/uaccess.h>
67
68 #define DRIVER_NAME     "pktcdvd"
69
70 #if PACKET_DEBUG
71 #define DPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
72 #else
73 #define DPRINTK(fmt, args...)
74 #endif
75
76 #if PACKET_DEBUG > 1
77 #define VPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
78 #else
79 #define VPRINTK(fmt, args...)
80 #endif
81
82 #define MAX_SPEED 0xffff
83
84 #define ZONE(sector, pd) (((sector) + (pd)->offset) & ~((pd)->settings.size - 1))
85
86 static struct pktcdvd_device *pkt_devs[MAX_WRITERS];
87 static struct proc_dir_entry *pkt_proc;
88 static int pktdev_major;
89 static int write_congestion_on  = PKT_WRITE_CONGESTION_ON;
90 static int write_congestion_off = PKT_WRITE_CONGESTION_OFF;
91 static struct mutex ctl_mutex;  /* Serialize open/close/setup/teardown */
92 static mempool_t *psd_pool;
93
94 static struct class     *class_pktcdvd = NULL;    /* /sys/class/pktcdvd */
95 static struct dentry    *pkt_debugfs_root = NULL; /* /debug/pktcdvd */
96
97 /* forward declaration */
98 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev);
99 static int pkt_remove_dev(dev_t pkt_dev);
100 static int pkt_seq_show(struct seq_file *m, void *p);
101
102
103
104 /*
105  * create and register a pktcdvd kernel object.
106  */
107 static struct pktcdvd_kobj* pkt_kobj_create(struct pktcdvd_device *pd,
108                                         const char* name,
109                                         struct kobject* parent,
110                                         struct kobj_type* ktype)
111 {
112         struct pktcdvd_kobj *p;
113         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, MKDEV(0, 0), NULL,
306                                         "%s", pd->name);
307                 if (IS_ERR(pd->dev))
308                         pd->dev = NULL;
309         }
310         if (pd->dev) {
311                 pd->kobj_stat = pkt_kobj_create(pd, "stat",
312                                         &pd->dev->kobj,
313                                         &kobj_pkt_type_stat);
314                 pd->kobj_wqueue = pkt_kobj_create(pd, "write_queue",
315                                         &pd->dev->kobj,
316                                         &kobj_pkt_type_wqueue);
317         }
318 }
319
320 static void pkt_sysfs_dev_remove(struct pktcdvd_device *pd)
321 {
322         pkt_kobj_remove(pd->kobj_stat);
323         pkt_kobj_remove(pd->kobj_wqueue);
324         if (class_pktcdvd)
325                 device_destroy(class_pktcdvd, pd->pkt_dev);
326 }
327
328
329 /********************************************************************
330   /sys/class/pktcdvd/
331                      add            map block device
332                      remove         unmap packet dev
333                      device_map     show mappings
334  *******************************************************************/
335
336 static void class_pktcdvd_release(struct class *cls)
337 {
338         kfree(cls);
339 }
340 static ssize_t class_pktcdvd_show_map(struct class *c, char *data)
341 {
342         int n = 0;
343         int idx;
344         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
345         for (idx = 0; idx < MAX_WRITERS; idx++) {
346                 struct pktcdvd_device *pd = pkt_devs[idx];
347                 if (!pd)
348                         continue;
349                 n += sprintf(data+n, "%s %u:%u %u:%u\n",
350                         pd->name,
351                         MAJOR(pd->pkt_dev), MINOR(pd->pkt_dev),
352                         MAJOR(pd->bdev->bd_dev),
353                         MINOR(pd->bdev->bd_dev));
354         }
355         mutex_unlock(&ctl_mutex);
356         return n;
357 }
358
359 static ssize_t class_pktcdvd_store_add(struct class *c, const char *buf,
360                                         size_t count)
361 {
362         unsigned int major, minor;
363
364         if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
365                 /* pkt_setup_dev() expects caller to hold reference to self */
366                 if (!try_module_get(THIS_MODULE))
367                         return -ENODEV;
368
369                 pkt_setup_dev(MKDEV(major, minor), NULL);
370
371                 module_put(THIS_MODULE);
372
373                 return count;
374         }
375
376         return -EINVAL;
377 }
378
379 static ssize_t class_pktcdvd_store_remove(struct class *c, const char *buf,
380                                         size_t count)
381 {
382         unsigned int major, minor;
383         if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
384                 pkt_remove_dev(MKDEV(major, minor));
385                 return count;
386         }
387         return -EINVAL;
388 }
389
390 static struct class_attribute class_pktcdvd_attrs[] = {
391  __ATTR(add,            0200, NULL, class_pktcdvd_store_add),
392  __ATTR(remove,         0200, NULL, class_pktcdvd_store_remove),
393  __ATTR(device_map,     0444, class_pktcdvd_show_map, NULL),
394  __ATTR_NULL
395 };
396
397
398 static int pkt_sysfs_init(void)
399 {
400         int ret = 0;
401
402         /*
403          * create control files in sysfs
404          * /sys/class/pktcdvd/...
405          */
406         class_pktcdvd = kzalloc(sizeof(*class_pktcdvd), GFP_KERNEL);
407         if (!class_pktcdvd)
408                 return -ENOMEM;
409         class_pktcdvd->name = DRIVER_NAME;
410         class_pktcdvd->owner = THIS_MODULE;
411         class_pktcdvd->class_release = class_pktcdvd_release;
412         class_pktcdvd->class_attrs = class_pktcdvd_attrs;
413         ret = class_register(class_pktcdvd);
414         if (ret) {
415                 kfree(class_pktcdvd);
416                 class_pktcdvd = NULL;
417                 printk(DRIVER_NAME": failed to create class pktcdvd\n");
418                 return ret;
419         }
420         return 0;
421 }
422
423 static void pkt_sysfs_cleanup(void)
424 {
425         if (class_pktcdvd)
426                 class_destroy(class_pktcdvd);
427         class_pktcdvd = NULL;
428 }
429
430 /********************************************************************
431   entries in debugfs
432
433   /debugfs/pktcdvd[0-7]/
434                         info
435
436  *******************************************************************/
437
438 static int pkt_debugfs_seq_show(struct seq_file *m, void *p)
439 {
440         return pkt_seq_show(m, p);
441 }
442
443 static int pkt_debugfs_fops_open(struct inode *inode, struct file *file)
444 {
445         return single_open(file, pkt_debugfs_seq_show, inode->i_private);
446 }
447
448 static const struct file_operations debug_fops = {
449         .open           = pkt_debugfs_fops_open,
450         .read           = seq_read,
451         .llseek         = seq_lseek,
452         .release        = single_release,
453         .owner          = THIS_MODULE,
454 };
455
456 static void pkt_debugfs_dev_new(struct pktcdvd_device *pd)
457 {
458         if (!pkt_debugfs_root)
459                 return;
460         pd->dfs_f_info = NULL;
461         pd->dfs_d_root = debugfs_create_dir(pd->name, pkt_debugfs_root);
462         if (IS_ERR(pd->dfs_d_root)) {
463                 pd->dfs_d_root = NULL;
464                 return;
465         }
466         pd->dfs_f_info = debugfs_create_file("info", S_IRUGO,
467                                 pd->dfs_d_root, pd, &debug_fops);
468         if (IS_ERR(pd->dfs_f_info)) {
469                 pd->dfs_f_info = NULL;
470                 return;
471         }
472 }
473
474 static void pkt_debugfs_dev_remove(struct pktcdvd_device *pd)
475 {
476         if (!pkt_debugfs_root)
477                 return;
478         if (pd->dfs_f_info)
479                 debugfs_remove(pd->dfs_f_info);
480         pd->dfs_f_info = NULL;
481         if (pd->dfs_d_root)
482                 debugfs_remove(pd->dfs_d_root);
483         pd->dfs_d_root = NULL;
484 }
485
486 static void pkt_debugfs_init(void)
487 {
488         pkt_debugfs_root = debugfs_create_dir(DRIVER_NAME, NULL);
489         if (IS_ERR(pkt_debugfs_root)) {
490                 pkt_debugfs_root = NULL;
491                 return;
492         }
493 }
494
495 static void pkt_debugfs_cleanup(void)
496 {
497         if (!pkt_debugfs_root)
498                 return;
499         debugfs_remove(pkt_debugfs_root);
500         pkt_debugfs_root = NULL;
501 }
502
503 /* ----------------------------------------------------------*/
504
505
506 static void pkt_bio_finished(struct pktcdvd_device *pd)
507 {
508         BUG_ON(atomic_read(&pd->cdrw.pending_bios) <= 0);
509         if (atomic_dec_and_test(&pd->cdrw.pending_bios)) {
510                 VPRINTK(DRIVER_NAME": queue empty\n");
511                 atomic_set(&pd->iosched.attention, 1);
512                 wake_up(&pd->wqueue);
513         }
514 }
515
516 static void pkt_bio_destructor(struct bio *bio)
517 {
518         kfree(bio->bi_io_vec);
519         kfree(bio);
520 }
521
522 static struct bio *pkt_bio_alloc(int nr_iovecs)
523 {
524         struct bio_vec *bvl = NULL;
525         struct bio *bio;
526
527         bio = kmalloc(sizeof(struct bio), GFP_KERNEL);
528         if (!bio)
529                 goto no_bio;
530         bio_init(bio);
531
532         bvl = kcalloc(nr_iovecs, sizeof(struct bio_vec), GFP_KERNEL);
533         if (!bvl)
534                 goto no_bvl;
535
536         bio->bi_max_vecs = nr_iovecs;
537         bio->bi_io_vec = bvl;
538         bio->bi_destructor = pkt_bio_destructor;
539
540         return bio;
541
542  no_bvl:
543         kfree(bio);
544  no_bio:
545         return NULL;
546 }
547
548 /*
549  * Allocate a packet_data struct
550  */
551 static struct packet_data *pkt_alloc_packet_data(int frames)
552 {
553         int i;
554         struct packet_data *pkt;
555
556         pkt = kzalloc(sizeof(struct packet_data), GFP_KERNEL);
557         if (!pkt)
558                 goto no_pkt;
559
560         pkt->frames = frames;
561         pkt->w_bio = pkt_bio_alloc(frames);
562         if (!pkt->w_bio)
563                 goto no_bio;
564
565         for (i = 0; i < frames / FRAMES_PER_PAGE; i++) {
566                 pkt->pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
567                 if (!pkt->pages[i])
568                         goto no_page;
569         }
570
571         spin_lock_init(&pkt->lock);
572
573         for (i = 0; i < frames; i++) {
574                 struct bio *bio = pkt_bio_alloc(1);
575                 if (!bio)
576                         goto no_rd_bio;
577                 pkt->r_bios[i] = bio;
578         }
579
580         return pkt;
581
582 no_rd_bio:
583         for (i = 0; i < frames; i++) {
584                 struct bio *bio = pkt->r_bios[i];
585                 if (bio)
586                         bio_put(bio);
587         }
588
589 no_page:
590         for (i = 0; i < frames / FRAMES_PER_PAGE; i++)
591                 if (pkt->pages[i])
592                         __free_page(pkt->pages[i]);
593         bio_put(pkt->w_bio);
594 no_bio:
595         kfree(pkt);
596 no_pkt:
597         return NULL;
598 }
599
600 /*
601  * Free a packet_data struct
602  */
603 static void pkt_free_packet_data(struct packet_data *pkt)
604 {
605         int i;
606
607         for (i = 0; i < pkt->frames; i++) {
608                 struct bio *bio = pkt->r_bios[i];
609                 if (bio)
610                         bio_put(bio);
611         }
612         for (i = 0; i < pkt->frames / FRAMES_PER_PAGE; i++)
613                 __free_page(pkt->pages[i]);
614         bio_put(pkt->w_bio);
615         kfree(pkt);
616 }
617
618 static void pkt_shrink_pktlist(struct pktcdvd_device *pd)
619 {
620         struct packet_data *pkt, *next;
621
622         BUG_ON(!list_empty(&pd->cdrw.pkt_active_list));
623
624         list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_free_list, list) {
625                 pkt_free_packet_data(pkt);
626         }
627         INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
628 }
629
630 static int pkt_grow_pktlist(struct pktcdvd_device *pd, int nr_packets)
631 {
632         struct packet_data *pkt;
633
634         BUG_ON(!list_empty(&pd->cdrw.pkt_free_list));
635
636         while (nr_packets > 0) {
637                 pkt = pkt_alloc_packet_data(pd->settings.size >> 2);
638                 if (!pkt) {
639                         pkt_shrink_pktlist(pd);
640                         return 0;
641                 }
642                 pkt->id = nr_packets;
643                 pkt->pd = pd;
644                 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
645                 nr_packets--;
646         }
647         return 1;
648 }
649
650 static inline struct pkt_rb_node *pkt_rbtree_next(struct pkt_rb_node *node)
651 {
652         struct rb_node *n = rb_next(&node->rb_node);
653         if (!n)
654                 return NULL;
655         return rb_entry(n, struct pkt_rb_node, rb_node);
656 }
657
658 static void pkt_rbtree_erase(struct pktcdvd_device *pd, struct pkt_rb_node *node)
659 {
660         rb_erase(&node->rb_node, &pd->bio_queue);
661         mempool_free(node, pd->rb_pool);
662         pd->bio_queue_size--;
663         BUG_ON(pd->bio_queue_size < 0);
664 }
665
666 /*
667  * Find the first node in the pd->bio_queue rb tree with a starting sector >= s.
668  */
669 static struct pkt_rb_node *pkt_rbtree_find(struct pktcdvd_device *pd, sector_t s)
670 {
671         struct rb_node *n = pd->bio_queue.rb_node;
672         struct rb_node *next;
673         struct pkt_rb_node *tmp;
674
675         if (!n) {
676                 BUG_ON(pd->bio_queue_size > 0);
677                 return NULL;
678         }
679
680         for (;;) {
681                 tmp = rb_entry(n, struct pkt_rb_node, rb_node);
682                 if (s <= tmp->bio->bi_sector)
683                         next = n->rb_left;
684                 else
685                         next = n->rb_right;
686                 if (!next)
687                         break;
688                 n = next;
689         }
690
691         if (s > tmp->bio->bi_sector) {
692                 tmp = pkt_rbtree_next(tmp);
693                 if (!tmp)
694                         return NULL;
695         }
696         BUG_ON(s > tmp->bio->bi_sector);
697         return tmp;
698 }
699
700 /*
701  * Insert a node into the pd->bio_queue rb tree.
702  */
703 static void pkt_rbtree_insert(struct pktcdvd_device *pd, struct pkt_rb_node *node)
704 {
705         struct rb_node **p = &pd->bio_queue.rb_node;
706         struct rb_node *parent = NULL;
707         sector_t s = node->bio->bi_sector;
708         struct pkt_rb_node *tmp;
709
710         while (*p) {
711                 parent = *p;
712                 tmp = rb_entry(parent, struct pkt_rb_node, rb_node);
713                 if (s < tmp->bio->bi_sector)
714                         p = &(*p)->rb_left;
715                 else
716                         p = &(*p)->rb_right;
717         }
718         rb_link_node(&node->rb_node, parent, p);
719         rb_insert_color(&node->rb_node, &pd->bio_queue);
720         pd->bio_queue_size++;
721 }
722
723 /*
724  * Add a bio to a single linked list defined by its head and tail pointers.
725  */
726 static void pkt_add_list_last(struct bio *bio, struct bio **list_head, struct bio **list_tail)
727 {
728         bio->bi_next = NULL;
729         if (*list_tail) {
730                 BUG_ON((*list_head) == NULL);
731                 (*list_tail)->bi_next = bio;
732                 (*list_tail) = bio;
733         } else {
734                 BUG_ON((*list_head) != NULL);
735                 (*list_head) = bio;
736                 (*list_tail) = bio;
737         }
738 }
739
740 /*
741  * Remove and return the first bio from a single linked list defined by its
742  * head and tail pointers.
743  */
744 static inline struct bio *pkt_get_list_first(struct bio **list_head, struct bio **list_tail)
745 {
746         struct bio *bio;
747
748         if (*list_head == NULL)
749                 return NULL;
750
751         bio = *list_head;
752         *list_head = bio->bi_next;
753         if (*list_head == NULL)
754                 *list_tail = NULL;
755
756         bio->bi_next = NULL;
757         return bio;
758 }
759
760 /*
761  * Send a packet_command to the underlying block device and
762  * wait for completion.
763  */
764 static int pkt_generic_packet(struct pktcdvd_device *pd, struct packet_command *cgc)
765 {
766         struct request_queue *q = bdev_get_queue(pd->bdev);
767         struct request *rq;
768         int ret = 0;
769
770         rq = blk_get_request(q, (cgc->data_direction == CGC_DATA_WRITE) ?
771                              WRITE : READ, __GFP_WAIT);
772
773         if (cgc->buflen) {
774                 if (blk_rq_map_kern(q, rq, cgc->buffer, cgc->buflen, __GFP_WAIT))
775                         goto out;
776         }
777
778         rq->cmd_len = COMMAND_SIZE(cgc->cmd[0]);
779         memcpy(rq->cmd, cgc->cmd, CDROM_PACKET_SIZE);
780
781         rq->timeout = 60*HZ;
782         rq->cmd_type = REQ_TYPE_BLOCK_PC;
783         rq->cmd_flags |= REQ_HARDBARRIER;
784         if (cgc->quiet)
785                 rq->cmd_flags |= REQ_QUIET;
786
787         blk_execute_rq(rq->q, pd->bdev->bd_disk, rq, 0);
788         if (rq->errors)
789                 ret = -EIO;
790 out:
791         blk_put_request(rq);
792         return ret;
793 }
794
795 /*
796  * A generic sense dump / resolve mechanism should be implemented across
797  * all ATAPI + SCSI devices.
798  */
799 static void pkt_dump_sense(struct packet_command *cgc)
800 {
801         static char *info[9] = { "No sense", "Recovered error", "Not ready",
802                                  "Medium error", "Hardware error", "Illegal request",
803                                  "Unit attention", "Data protect", "Blank check" };
804         int i;
805         struct request_sense *sense = cgc->sense;
806
807         printk(DRIVER_NAME":");
808         for (i = 0; i < CDROM_PACKET_SIZE; i++)
809                 printk(" %02x", cgc->cmd[i]);
810         printk(" - ");
811
812         if (sense == NULL) {
813                 printk("no sense\n");
814                 return;
815         }
816
817         printk("sense %02x.%02x.%02x", sense->sense_key, sense->asc, sense->ascq);
818
819         if (sense->sense_key > 8) {
820                 printk(" (INVALID)\n");
821                 return;
822         }
823
824         printk(" (%s)\n", info[sense->sense_key]);
825 }
826
827 /*
828  * flush the drive cache to media
829  */
830 static int pkt_flush_cache(struct pktcdvd_device *pd)
831 {
832         struct packet_command cgc;
833
834         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
835         cgc.cmd[0] = GPCMD_FLUSH_CACHE;
836         cgc.quiet = 1;
837
838         /*
839          * the IMMED bit -- we default to not setting it, although that
840          * would allow a much faster close, this is safer
841          */
842 #if 0
843         cgc.cmd[1] = 1 << 1;
844 #endif
845         return pkt_generic_packet(pd, &cgc);
846 }
847
848 /*
849  * speed is given as the normal factor, e.g. 4 for 4x
850  */
851 static noinline_for_stack int pkt_set_speed(struct pktcdvd_device *pd,
852                                 unsigned write_speed, unsigned read_speed)
853 {
854         struct packet_command cgc;
855         struct request_sense sense;
856         int ret;
857
858         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
859         cgc.sense = &sense;
860         cgc.cmd[0] = GPCMD_SET_SPEED;
861         cgc.cmd[2] = (read_speed >> 8) & 0xff;
862         cgc.cmd[3] = read_speed & 0xff;
863         cgc.cmd[4] = (write_speed >> 8) & 0xff;
864         cgc.cmd[5] = write_speed & 0xff;
865
866         if ((ret = pkt_generic_packet(pd, &cgc)))
867                 pkt_dump_sense(&cgc);
868
869         return ret;
870 }
871
872 /*
873  * Queue a bio for processing by the low-level CD device. Must be called
874  * from process context.
875  */
876 static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio)
877 {
878         spin_lock(&pd->iosched.lock);
879         if (bio_data_dir(bio) == READ) {
880                 pkt_add_list_last(bio, &pd->iosched.read_queue,
881                                   &pd->iosched.read_queue_tail);
882         } else {
883                 pkt_add_list_last(bio, &pd->iosched.write_queue,
884                                   &pd->iosched.write_queue_tail);
885         }
886         spin_unlock(&pd->iosched.lock);
887
888         atomic_set(&pd->iosched.attention, 1);
889         wake_up(&pd->wqueue);
890 }
891
892 /*
893  * Process the queued read/write requests. This function handles special
894  * requirements for CDRW drives:
895  * - A cache flush command must be inserted before a read request if the
896  *   previous request was a write.
897  * - Switching between reading and writing is slow, so don't do it more often
898  *   than necessary.
899  * - Optimize for throughput at the expense of latency. This means that streaming
900  *   writes will never be interrupted by a read, but if the drive has to seek
901  *   before the next write, switch to reading instead if there are any pending
902  *   read requests.
903  * - Set the read speed according to current usage pattern. When only reading
904  *   from the device, it's best to use the highest possible read speed, but
905  *   when switching often between reading and writing, it's better to have the
906  *   same read and write speeds.
907  */
908 static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
909 {
910
911         if (atomic_read(&pd->iosched.attention) == 0)
912                 return;
913         atomic_set(&pd->iosched.attention, 0);
914
915         for (;;) {
916                 struct bio *bio;
917                 int reads_queued, writes_queued;
918
919                 spin_lock(&pd->iosched.lock);
920                 reads_queued = (pd->iosched.read_queue != NULL);
921                 writes_queued = (pd->iosched.write_queue != NULL);
922                 spin_unlock(&pd->iosched.lock);
923
924                 if (!reads_queued && !writes_queued)
925                         break;
926
927                 if (pd->iosched.writing) {
928                         int need_write_seek = 1;
929                         spin_lock(&pd->iosched.lock);
930                         bio = pd->iosched.write_queue;
931                         spin_unlock(&pd->iosched.lock);
932                         if (bio && (bio->bi_sector == pd->iosched.last_write))
933                                 need_write_seek = 0;
934                         if (need_write_seek && reads_queued) {
935                                 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
936                                         VPRINTK(DRIVER_NAME": write, waiting\n");
937                                         break;
938                                 }
939                                 pkt_flush_cache(pd);
940                                 pd->iosched.writing = 0;
941                         }
942                 } else {
943                         if (!reads_queued && writes_queued) {
944                                 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
945                                         VPRINTK(DRIVER_NAME": read, waiting\n");
946                                         break;
947                                 }
948                                 pd->iosched.writing = 1;
949                         }
950                 }
951
952                 spin_lock(&pd->iosched.lock);
953                 if (pd->iosched.writing) {
954                         bio = pkt_get_list_first(&pd->iosched.write_queue,
955                                                  &pd->iosched.write_queue_tail);
956                 } else {
957                         bio = pkt_get_list_first(&pd->iosched.read_queue,
958                                                  &pd->iosched.read_queue_tail);
959                 }
960                 spin_unlock(&pd->iosched.lock);
961
962                 if (!bio)
963                         continue;
964
965                 if (bio_data_dir(bio) == READ)
966                         pd->iosched.successive_reads += bio->bi_size >> 10;
967                 else {
968                         pd->iosched.successive_reads = 0;
969                         pd->iosched.last_write = bio->bi_sector + bio_sectors(bio);
970                 }
971                 if (pd->iosched.successive_reads >= HI_SPEED_SWITCH) {
972                         if (pd->read_speed == pd->write_speed) {
973                                 pd->read_speed = MAX_SPEED;
974                                 pkt_set_speed(pd, pd->write_speed, pd->read_speed);
975                         }
976                 } else {
977                         if (pd->read_speed != pd->write_speed) {
978                                 pd->read_speed = pd->write_speed;
979                                 pkt_set_speed(pd, pd->write_speed, pd->read_speed);
980                         }
981                 }
982
983                 atomic_inc(&pd->cdrw.pending_bios);
984                 generic_make_request(bio);
985         }
986 }
987
988 /*
989  * Special care is needed if the underlying block device has a small
990  * max_phys_segments value.
991  */
992 static int pkt_set_segment_merging(struct pktcdvd_device *pd, struct request_queue *q)
993 {
994         if ((pd->settings.size << 9) / CD_FRAMESIZE <= q->max_phys_segments) {
995                 /*
996                  * The cdrom device can handle one segment/frame
997                  */
998                 clear_bit(PACKET_MERGE_SEGS, &pd->flags);
999                 return 0;
1000         } else if ((pd->settings.size << 9) / PAGE_SIZE <= q->max_phys_segments) {
1001                 /*
1002                  * We can handle this case at the expense of some extra memory
1003                  * copies during write operations
1004                  */
1005                 set_bit(PACKET_MERGE_SEGS, &pd->flags);
1006                 return 0;
1007         } else {
1008                 printk(DRIVER_NAME": cdrom max_phys_segments too small\n");
1009                 return -EIO;
1010         }
1011 }
1012
1013 /*
1014  * Copy CD_FRAMESIZE bytes from src_bio into a destination page
1015  */
1016 static void pkt_copy_bio_data(struct bio *src_bio, int seg, int offs, struct page *dst_page, int dst_offs)
1017 {
1018         unsigned int copy_size = CD_FRAMESIZE;
1019
1020         while (copy_size > 0) {
1021                 struct bio_vec *src_bvl = bio_iovec_idx(src_bio, seg);
1022                 void *vfrom = kmap_atomic(src_bvl->bv_page, KM_USER0) +
1023                         src_bvl->bv_offset + offs;
1024                 void *vto = page_address(dst_page) + dst_offs;
1025                 int len = min_t(int, copy_size, src_bvl->bv_len - offs);
1026
1027                 BUG_ON(len < 0);
1028                 memcpy(vto, vfrom, len);
1029                 kunmap_atomic(vfrom, KM_USER0);
1030
1031                 seg++;
1032                 offs = 0;
1033                 dst_offs += len;
1034                 copy_size -= len;
1035         }
1036 }
1037
1038 /*
1039  * Copy all data for this packet to pkt->pages[], so that
1040  * a) The number of required segments for the write bio is minimized, which
1041  *    is necessary for some scsi controllers.
1042  * b) The data can be used as cache to avoid read requests if we receive a
1043  *    new write request for the same zone.
1044  */
1045 static void pkt_make_local_copy(struct packet_data *pkt, struct bio_vec *bvec)
1046 {
1047         int f, p, offs;
1048
1049         /* Copy all data to pkt->pages[] */
1050         p = 0;
1051         offs = 0;
1052         for (f = 0; f < pkt->frames; f++) {
1053                 if (bvec[f].bv_page != pkt->pages[p]) {
1054                         void *vfrom = kmap_atomic(bvec[f].bv_page, KM_USER0) + bvec[f].bv_offset;
1055                         void *vto = page_address(pkt->pages[p]) + offs;
1056                         memcpy(vto, vfrom, CD_FRAMESIZE);
1057                         kunmap_atomic(vfrom, KM_USER0);
1058                         bvec[f].bv_page = pkt->pages[p];
1059                         bvec[f].bv_offset = offs;
1060                 } else {
1061                         BUG_ON(bvec[f].bv_offset != offs);
1062                 }
1063                 offs += CD_FRAMESIZE;
1064                 if (offs >= PAGE_SIZE) {
1065                         offs = 0;
1066                         p++;
1067                 }
1068         }
1069 }
1070
1071 static void pkt_end_io_read(struct bio *bio, int err)
1072 {
1073         struct packet_data *pkt = bio->bi_private;
1074         struct pktcdvd_device *pd = pkt->pd;
1075         BUG_ON(!pd);
1076
1077         VPRINTK("pkt_end_io_read: bio=%p sec0=%llx sec=%llx err=%d\n", bio,
1078                 (unsigned long long)pkt->sector, (unsigned long long)bio->bi_sector, err);
1079
1080         if (err)
1081                 atomic_inc(&pkt->io_errors);
1082         if (atomic_dec_and_test(&pkt->io_wait)) {
1083                 atomic_inc(&pkt->run_sm);
1084                 wake_up(&pd->wqueue);
1085         }
1086         pkt_bio_finished(pd);
1087 }
1088
1089 static void pkt_end_io_packet_write(struct bio *bio, int err)
1090 {
1091         struct packet_data *pkt = bio->bi_private;
1092         struct pktcdvd_device *pd = pkt->pd;
1093         BUG_ON(!pd);
1094
1095         VPRINTK("pkt_end_io_packet_write: id=%d, err=%d\n", pkt->id, err);
1096
1097         pd->stats.pkt_ended++;
1098
1099         pkt_bio_finished(pd);
1100         atomic_dec(&pkt->io_wait);
1101         atomic_inc(&pkt->run_sm);
1102         wake_up(&pd->wqueue);
1103 }
1104
1105 /*
1106  * Schedule reads for the holes in a packet
1107  */
1108 static void pkt_gather_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1109 {
1110         int frames_read = 0;
1111         struct bio *bio;
1112         int f;
1113         char written[PACKET_MAX_SIZE];
1114
1115         BUG_ON(!pkt->orig_bios);
1116
1117         atomic_set(&pkt->io_wait, 0);
1118         atomic_set(&pkt->io_errors, 0);
1119
1120         /*
1121          * Figure out which frames we need to read before we can write.
1122          */
1123         memset(written, 0, sizeof(written));
1124         spin_lock(&pkt->lock);
1125         for (bio = pkt->orig_bios; bio; bio = bio->bi_next) {
1126                 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1127                 int num_frames = bio->bi_size / CD_FRAMESIZE;
1128                 pd->stats.secs_w += num_frames * (CD_FRAMESIZE >> 9);
1129                 BUG_ON(first_frame < 0);
1130                 BUG_ON(first_frame + num_frames > pkt->frames);
1131                 for (f = first_frame; f < first_frame + num_frames; f++)
1132                         written[f] = 1;
1133         }
1134         spin_unlock(&pkt->lock);
1135
1136         if (pkt->cache_valid) {
1137                 VPRINTK("pkt_gather_data: zone %llx cached\n",
1138                         (unsigned long long)pkt->sector);
1139                 goto out_account;
1140         }
1141
1142         /*
1143          * Schedule reads for missing parts of the packet.
1144          */
1145         for (f = 0; f < pkt->frames; f++) {
1146                 struct bio_vec *vec;
1147
1148                 int p, offset;
1149                 if (written[f])
1150                         continue;
1151                 bio = pkt->r_bios[f];
1152                 vec = bio->bi_io_vec;
1153                 bio_init(bio);
1154                 bio->bi_max_vecs = 1;
1155                 bio->bi_sector = pkt->sector + f * (CD_FRAMESIZE >> 9);
1156                 bio->bi_bdev = pd->bdev;
1157                 bio->bi_end_io = pkt_end_io_read;
1158                 bio->bi_private = pkt;
1159                 bio->bi_io_vec = vec;
1160                 bio->bi_destructor = pkt_bio_destructor;
1161
1162                 p = (f * CD_FRAMESIZE) / PAGE_SIZE;
1163                 offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1164                 VPRINTK("pkt_gather_data: Adding frame %d, page:%p offs:%d\n",
1165                         f, pkt->pages[p], offset);
1166                 if (!bio_add_page(bio, pkt->pages[p], CD_FRAMESIZE, offset))
1167                         BUG();
1168
1169                 atomic_inc(&pkt->io_wait);
1170                 bio->bi_rw = READ;
1171                 pkt_queue_bio(pd, bio);
1172                 frames_read++;
1173         }
1174
1175 out_account:
1176         VPRINTK("pkt_gather_data: need %d frames for zone %llx\n",
1177                 frames_read, (unsigned long long)pkt->sector);
1178         pd->stats.pkt_started++;
1179         pd->stats.secs_rg += frames_read * (CD_FRAMESIZE >> 9);
1180 }
1181
1182 /*
1183  * Find a packet matching zone, or the least recently used packet if
1184  * there is no match.
1185  */
1186 static struct packet_data *pkt_get_packet_data(struct pktcdvd_device *pd, int zone)
1187 {
1188         struct packet_data *pkt;
1189
1190         list_for_each_entry(pkt, &pd->cdrw.pkt_free_list, list) {
1191                 if (pkt->sector == zone || pkt->list.next == &pd->cdrw.pkt_free_list) {
1192                         list_del_init(&pkt->list);
1193                         if (pkt->sector != zone)
1194                                 pkt->cache_valid = 0;
1195                         return pkt;
1196                 }
1197         }
1198         BUG();
1199         return NULL;
1200 }
1201
1202 static void pkt_put_packet_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1203 {
1204         if (pkt->cache_valid) {
1205                 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
1206         } else {
1207                 list_add_tail(&pkt->list, &pd->cdrw.pkt_free_list);
1208         }
1209 }
1210
1211 /*
1212  * recover a failed write, query for relocation if possible
1213  *
1214  * returns 1 if recovery is possible, or 0 if not
1215  *
1216  */
1217 static int pkt_start_recovery(struct packet_data *pkt)
1218 {
1219         /*
1220          * FIXME. We need help from the file system to implement
1221          * recovery handling.
1222          */
1223         return 0;
1224 #if 0
1225         struct request *rq = pkt->rq;
1226         struct pktcdvd_device *pd = rq->rq_disk->private_data;
1227         struct block_device *pkt_bdev;
1228         struct super_block *sb = NULL;
1229         unsigned long old_block, new_block;
1230         sector_t new_sector;
1231
1232         pkt_bdev = bdget(kdev_t_to_nr(pd->pkt_dev));
1233         if (pkt_bdev) {
1234                 sb = get_super(pkt_bdev);
1235                 bdput(pkt_bdev);
1236         }
1237
1238         if (!sb)
1239                 return 0;
1240
1241         if (!sb->s_op || !sb->s_op->relocate_blocks)
1242                 goto out;
1243
1244         old_block = pkt->sector / (CD_FRAMESIZE >> 9);
1245         if (sb->s_op->relocate_blocks(sb, old_block, &new_block))
1246                 goto out;
1247
1248         new_sector = new_block * (CD_FRAMESIZE >> 9);
1249         pkt->sector = new_sector;
1250
1251         pkt->bio->bi_sector = new_sector;
1252         pkt->bio->bi_next = NULL;
1253         pkt->bio->bi_flags = 1 << BIO_UPTODATE;
1254         pkt->bio->bi_idx = 0;
1255
1256         BUG_ON(pkt->bio->bi_rw != (1 << BIO_RW));
1257         BUG_ON(pkt->bio->bi_vcnt != pkt->frames);
1258         BUG_ON(pkt->bio->bi_size != pkt->frames * CD_FRAMESIZE);
1259         BUG_ON(pkt->bio->bi_end_io != pkt_end_io_packet_write);
1260         BUG_ON(pkt->bio->bi_private != pkt);
1261
1262         drop_super(sb);
1263         return 1;
1264
1265 out:
1266         drop_super(sb);
1267         return 0;
1268 #endif
1269 }
1270
1271 static inline void pkt_set_state(struct packet_data *pkt, enum packet_data_state state)
1272 {
1273 #if PACKET_DEBUG > 1
1274         static const char *state_name[] = {
1275                 "IDLE", "WAITING", "READ_WAIT", "WRITE_WAIT", "RECOVERY", "FINISHED"
1276         };
1277         enum packet_data_state old_state = pkt->state;
1278         VPRINTK("pkt %2d : s=%6llx %s -> %s\n", pkt->id, (unsigned long long)pkt->sector,
1279                 state_name[old_state], state_name[state]);
1280 #endif
1281         pkt->state = state;
1282 }
1283
1284 /*
1285  * Scan the work queue to see if we can start a new packet.
1286  * returns non-zero if any work was done.
1287  */
1288 static int pkt_handle_queue(struct pktcdvd_device *pd)
1289 {
1290         struct packet_data *pkt, *p;
1291         struct bio *bio = NULL;
1292         sector_t zone = 0; /* Suppress gcc warning */
1293         struct pkt_rb_node *node, *first_node;
1294         struct rb_node *n;
1295         int wakeup;
1296
1297         VPRINTK("handle_queue\n");
1298
1299         atomic_set(&pd->scan_queue, 0);
1300
1301         if (list_empty(&pd->cdrw.pkt_free_list)) {
1302                 VPRINTK("handle_queue: no pkt\n");
1303                 return 0;
1304         }
1305
1306         /*
1307          * Try to find a zone we are not already working on.
1308          */
1309         spin_lock(&pd->lock);
1310         first_node = pkt_rbtree_find(pd, pd->current_sector);
1311         if (!first_node) {
1312                 n = rb_first(&pd->bio_queue);
1313                 if (n)
1314                         first_node = rb_entry(n, struct pkt_rb_node, rb_node);
1315         }
1316         node = first_node;
1317         while (node) {
1318                 bio = node->bio;
1319                 zone = ZONE(bio->bi_sector, pd);
1320                 list_for_each_entry(p, &pd->cdrw.pkt_active_list, list) {
1321                         if (p->sector == zone) {
1322                                 bio = NULL;
1323                                 goto try_next_bio;
1324                         }
1325                 }
1326                 break;
1327 try_next_bio:
1328                 node = pkt_rbtree_next(node);
1329                 if (!node) {
1330                         n = rb_first(&pd->bio_queue);
1331                         if (n)
1332                                 node = rb_entry(n, struct pkt_rb_node, rb_node);
1333                 }
1334                 if (node == first_node)
1335                         node = NULL;
1336         }
1337         spin_unlock(&pd->lock);
1338         if (!bio) {
1339                 VPRINTK("handle_queue: no bio\n");
1340                 return 0;
1341         }
1342
1343         pkt = pkt_get_packet_data(pd, zone);
1344
1345         pd->current_sector = zone + pd->settings.size;
1346         pkt->sector = zone;
1347         BUG_ON(pkt->frames != pd->settings.size >> 2);
1348         pkt->write_size = 0;
1349
1350         /*
1351          * Scan work queue for bios in the same zone and link them
1352          * to this packet.
1353          */
1354         spin_lock(&pd->lock);
1355         VPRINTK("pkt_handle_queue: looking for zone %llx\n", (unsigned long long)zone);
1356         while ((node = pkt_rbtree_find(pd, zone)) != NULL) {
1357                 bio = node->bio;
1358                 VPRINTK("pkt_handle_queue: found zone=%llx\n",
1359                         (unsigned long long)ZONE(bio->bi_sector, pd));
1360                 if (ZONE(bio->bi_sector, pd) != zone)
1361                         break;
1362                 pkt_rbtree_erase(pd, node);
1363                 spin_lock(&pkt->lock);
1364                 pkt_add_list_last(bio, &pkt->orig_bios, &pkt->orig_bios_tail);
1365                 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
1366                 spin_unlock(&pkt->lock);
1367         }
1368         /* check write congestion marks, and if bio_queue_size is
1369            below, wake up any waiters */
1370         wakeup = (pd->write_congestion_on > 0
1371                         && pd->bio_queue_size <= pd->write_congestion_off);
1372         spin_unlock(&pd->lock);
1373         if (wakeup)
1374                 clear_bdi_congested(&pd->disk->queue->backing_dev_info, WRITE);
1375
1376         pkt->sleep_time = max(PACKET_WAIT_TIME, 1);
1377         pkt_set_state(pkt, PACKET_WAITING_STATE);
1378         atomic_set(&pkt->run_sm, 1);
1379
1380         spin_lock(&pd->cdrw.active_list_lock);
1381         list_add(&pkt->list, &pd->cdrw.pkt_active_list);
1382         spin_unlock(&pd->cdrw.active_list_lock);
1383
1384         return 1;
1385 }
1386
1387 /*
1388  * Assemble a bio to write one packet and queue the bio for processing
1389  * by the underlying block device.
1390  */
1391 static void pkt_start_write(struct pktcdvd_device *pd, struct packet_data *pkt)
1392 {
1393         struct bio *bio;
1394         int f;
1395         int frames_write;
1396         struct bio_vec *bvec = pkt->w_bio->bi_io_vec;
1397
1398         for (f = 0; f < pkt->frames; f++) {
1399                 bvec[f].bv_page = pkt->pages[(f * CD_FRAMESIZE) / PAGE_SIZE];
1400                 bvec[f].bv_offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1401         }
1402
1403         /*
1404          * Fill-in bvec with data from orig_bios.
1405          */
1406         frames_write = 0;
1407         spin_lock(&pkt->lock);
1408         for (bio = pkt->orig_bios; bio; bio = bio->bi_next) {
1409                 int segment = bio->bi_idx;
1410                 int src_offs = 0;
1411                 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1412                 int num_frames = bio->bi_size / CD_FRAMESIZE;
1413                 BUG_ON(first_frame < 0);
1414                 BUG_ON(first_frame + num_frames > pkt->frames);
1415                 for (f = first_frame; f < first_frame + num_frames; f++) {
1416                         struct bio_vec *src_bvl = bio_iovec_idx(bio, segment);
1417
1418                         while (src_offs >= src_bvl->bv_len) {
1419                                 src_offs -= src_bvl->bv_len;
1420                                 segment++;
1421                                 BUG_ON(segment >= bio->bi_vcnt);
1422                                 src_bvl = bio_iovec_idx(bio, segment);
1423                         }
1424
1425                         if (src_bvl->bv_len - src_offs >= CD_FRAMESIZE) {
1426                                 bvec[f].bv_page = src_bvl->bv_page;
1427                                 bvec[f].bv_offset = src_bvl->bv_offset + src_offs;
1428                         } else {
1429                                 pkt_copy_bio_data(bio, segment, src_offs,
1430                                                   bvec[f].bv_page, bvec[f].bv_offset);
1431                         }
1432                         src_offs += CD_FRAMESIZE;
1433                         frames_write++;
1434                 }
1435         }
1436         pkt_set_state(pkt, PACKET_WRITE_WAIT_STATE);
1437         spin_unlock(&pkt->lock);
1438
1439         VPRINTK("pkt_start_write: Writing %d frames for zone %llx\n",
1440                 frames_write, (unsigned long long)pkt->sector);
1441         BUG_ON(frames_write != pkt->write_size);
1442
1443         if (test_bit(PACKET_MERGE_SEGS, &pd->flags) || (pkt->write_size < pkt->frames)) {
1444                 pkt_make_local_copy(pkt, bvec);
1445                 pkt->cache_valid = 1;
1446         } else {
1447                 pkt->cache_valid = 0;
1448         }
1449
1450         /* Start the write request */
1451         bio_init(pkt->w_bio);
1452         pkt->w_bio->bi_max_vecs = PACKET_MAX_SIZE;
1453         pkt->w_bio->bi_sector = pkt->sector;
1454         pkt->w_bio->bi_bdev = pd->bdev;
1455         pkt->w_bio->bi_end_io = pkt_end_io_packet_write;
1456         pkt->w_bio->bi_private = pkt;
1457         pkt->w_bio->bi_io_vec = bvec;
1458         pkt->w_bio->bi_destructor = pkt_bio_destructor;
1459         for (f = 0; f < pkt->frames; f++)
1460                 if (!bio_add_page(pkt->w_bio, bvec[f].bv_page, CD_FRAMESIZE, bvec[f].bv_offset))
1461                         BUG();
1462         VPRINTK(DRIVER_NAME": vcnt=%d\n", pkt->w_bio->bi_vcnt);
1463
1464         atomic_set(&pkt->io_wait, 1);
1465         pkt->w_bio->bi_rw = WRITE;
1466         pkt_queue_bio(pd, pkt->w_bio);
1467 }
1468
1469 static void pkt_finish_packet(struct packet_data *pkt, int uptodate)
1470 {
1471         struct bio *bio, *next;
1472
1473         if (!uptodate)
1474                 pkt->cache_valid = 0;
1475
1476         /* Finish all bios corresponding to this packet */
1477         bio = pkt->orig_bios;
1478         while (bio) {
1479                 next = bio->bi_next;
1480                 bio->bi_next = NULL;
1481                 bio_endio(bio, uptodate ? 0 : -EIO);
1482                 bio = next;
1483         }
1484         pkt->orig_bios = pkt->orig_bios_tail = NULL;
1485 }
1486
1487 static void pkt_run_state_machine(struct pktcdvd_device *pd, struct packet_data *pkt)
1488 {
1489         int uptodate;
1490
1491         VPRINTK("run_state_machine: pkt %d\n", pkt->id);
1492
1493         for (;;) {
1494                 switch (pkt->state) {
1495                 case PACKET_WAITING_STATE:
1496                         if ((pkt->write_size < pkt->frames) && (pkt->sleep_time > 0))
1497                                 return;
1498
1499                         pkt->sleep_time = 0;
1500                         pkt_gather_data(pd, pkt);
1501                         pkt_set_state(pkt, PACKET_READ_WAIT_STATE);
1502                         break;
1503
1504                 case PACKET_READ_WAIT_STATE:
1505                         if (atomic_read(&pkt->io_wait) > 0)
1506                                 return;
1507
1508                         if (atomic_read(&pkt->io_errors) > 0) {
1509                                 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1510                         } else {
1511                                 pkt_start_write(pd, pkt);
1512                         }
1513                         break;
1514
1515                 case PACKET_WRITE_WAIT_STATE:
1516                         if (atomic_read(&pkt->io_wait) > 0)
1517                                 return;
1518
1519                         if (test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags)) {
1520                                 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1521                         } else {
1522                                 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1523                         }
1524                         break;
1525
1526                 case PACKET_RECOVERY_STATE:
1527                         if (pkt_start_recovery(pkt)) {
1528                                 pkt_start_write(pd, pkt);
1529                         } else {
1530                                 VPRINTK("No recovery possible\n");
1531                                 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1532                         }
1533                         break;
1534
1535                 case PACKET_FINISHED_STATE:
1536                         uptodate = test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags);
1537                         pkt_finish_packet(pkt, uptodate);
1538                         return;
1539
1540                 default:
1541                         BUG();
1542                         break;
1543                 }
1544         }
1545 }
1546
1547 static void pkt_handle_packets(struct pktcdvd_device *pd)
1548 {
1549         struct packet_data *pkt, *next;
1550
1551         VPRINTK("pkt_handle_packets\n");
1552
1553         /*
1554          * Run state machine for active packets
1555          */
1556         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1557                 if (atomic_read(&pkt->run_sm) > 0) {
1558                         atomic_set(&pkt->run_sm, 0);
1559                         pkt_run_state_machine(pd, pkt);
1560                 }
1561         }
1562
1563         /*
1564          * Move no longer active packets to the free list
1565          */
1566         spin_lock(&pd->cdrw.active_list_lock);
1567         list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_active_list, list) {
1568                 if (pkt->state == PACKET_FINISHED_STATE) {
1569                         list_del(&pkt->list);
1570                         pkt_put_packet_data(pd, pkt);
1571                         pkt_set_state(pkt, PACKET_IDLE_STATE);
1572                         atomic_set(&pd->scan_queue, 1);
1573                 }
1574         }
1575         spin_unlock(&pd->cdrw.active_list_lock);
1576 }
1577
1578 static void pkt_count_states(struct pktcdvd_device *pd, int *states)
1579 {
1580         struct packet_data *pkt;
1581         int i;
1582
1583         for (i = 0; i < PACKET_NUM_STATES; i++)
1584                 states[i] = 0;
1585
1586         spin_lock(&pd->cdrw.active_list_lock);
1587         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1588                 states[pkt->state]++;
1589         }
1590         spin_unlock(&pd->cdrw.active_list_lock);
1591 }
1592
1593 /*
1594  * kcdrwd is woken up when writes have been queued for one of our
1595  * registered devices
1596  */
1597 static int kcdrwd(void *foobar)
1598 {
1599         struct pktcdvd_device *pd = foobar;
1600         struct packet_data *pkt;
1601         long min_sleep_time, residue;
1602
1603         set_user_nice(current, -20);
1604         set_freezable();
1605
1606         for (;;) {
1607                 DECLARE_WAITQUEUE(wait, current);
1608
1609                 /*
1610                  * Wait until there is something to do
1611                  */
1612                 add_wait_queue(&pd->wqueue, &wait);
1613                 for (;;) {
1614                         set_current_state(TASK_INTERRUPTIBLE);
1615
1616                         /* Check if we need to run pkt_handle_queue */
1617                         if (atomic_read(&pd->scan_queue) > 0)
1618                                 goto work_to_do;
1619
1620                         /* Check if we need to run the state machine for some packet */
1621                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1622                                 if (atomic_read(&pkt->run_sm) > 0)
1623                                         goto work_to_do;
1624                         }
1625
1626                         /* Check if we need to process the iosched queues */
1627                         if (atomic_read(&pd->iosched.attention) != 0)
1628                                 goto work_to_do;
1629
1630                         /* Otherwise, go to sleep */
1631                         if (PACKET_DEBUG > 1) {
1632                                 int states[PACKET_NUM_STATES];
1633                                 pkt_count_states(pd, states);
1634                                 VPRINTK("kcdrwd: i:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
1635                                         states[0], states[1], states[2], states[3],
1636                                         states[4], states[5]);
1637                         }
1638
1639                         min_sleep_time = MAX_SCHEDULE_TIMEOUT;
1640                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1641                                 if (pkt->sleep_time && pkt->sleep_time < min_sleep_time)
1642                                         min_sleep_time = pkt->sleep_time;
1643                         }
1644
1645                         generic_unplug_device(bdev_get_queue(pd->bdev));
1646
1647                         VPRINTK("kcdrwd: sleeping\n");
1648                         residue = schedule_timeout(min_sleep_time);
1649                         VPRINTK("kcdrwd: wake up\n");
1650
1651                         /* make swsusp happy with our thread */
1652                         try_to_freeze();
1653
1654                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1655                                 if (!pkt->sleep_time)
1656                                         continue;
1657                                 pkt->sleep_time -= min_sleep_time - residue;
1658                                 if (pkt->sleep_time <= 0) {
1659                                         pkt->sleep_time = 0;
1660                                         atomic_inc(&pkt->run_sm);
1661                                 }
1662                         }
1663
1664                         if (kthread_should_stop())
1665                                 break;
1666                 }
1667 work_to_do:
1668                 set_current_state(TASK_RUNNING);
1669                 remove_wait_queue(&pd->wqueue, &wait);
1670
1671                 if (kthread_should_stop())
1672                         break;
1673
1674                 /*
1675                  * if pkt_handle_queue returns true, we can queue
1676                  * another request.
1677                  */
1678                 while (pkt_handle_queue(pd))
1679                         ;
1680
1681                 /*
1682                  * Handle packet state machine
1683                  */
1684                 pkt_handle_packets(pd);
1685
1686                 /*
1687                  * Handle iosched queues
1688                  */
1689                 pkt_iosched_process_queue(pd);
1690         }
1691
1692         return 0;
1693 }
1694
1695 static void pkt_print_settings(struct pktcdvd_device *pd)
1696 {
1697         printk(DRIVER_NAME": %s packets, ", pd->settings.fp ? "Fixed" : "Variable");
1698         printk("%u blocks, ", pd->settings.size >> 2);
1699         printk("Mode-%c disc\n", pd->settings.block_mode == 8 ? '1' : '2');
1700 }
1701
1702 static int pkt_mode_sense(struct pktcdvd_device *pd, struct packet_command *cgc, int page_code, int page_control)
1703 {
1704         memset(cgc->cmd, 0, sizeof(cgc->cmd));
1705
1706         cgc->cmd[0] = GPCMD_MODE_SENSE_10;
1707         cgc->cmd[2] = page_code | (page_control << 6);
1708         cgc->cmd[7] = cgc->buflen >> 8;
1709         cgc->cmd[8] = cgc->buflen & 0xff;
1710         cgc->data_direction = CGC_DATA_READ;
1711         return pkt_generic_packet(pd, cgc);
1712 }
1713
1714 static int pkt_mode_select(struct pktcdvd_device *pd, struct packet_command *cgc)
1715 {
1716         memset(cgc->cmd, 0, sizeof(cgc->cmd));
1717         memset(cgc->buffer, 0, 2);
1718         cgc->cmd[0] = GPCMD_MODE_SELECT_10;
1719         cgc->cmd[1] = 0x10;             /* PF */
1720         cgc->cmd[7] = cgc->buflen >> 8;
1721         cgc->cmd[8] = cgc->buflen & 0xff;
1722         cgc->data_direction = CGC_DATA_WRITE;
1723         return pkt_generic_packet(pd, cgc);
1724 }
1725
1726 static int pkt_get_disc_info(struct pktcdvd_device *pd, disc_information *di)
1727 {
1728         struct packet_command cgc;
1729         int ret;
1730
1731         /* set up command and get the disc info */
1732         init_cdrom_command(&cgc, di, sizeof(*di), CGC_DATA_READ);
1733         cgc.cmd[0] = GPCMD_READ_DISC_INFO;
1734         cgc.cmd[8] = cgc.buflen = 2;
1735         cgc.quiet = 1;
1736
1737         if ((ret = pkt_generic_packet(pd, &cgc)))
1738                 return ret;
1739
1740         /* not all drives have the same disc_info length, so requeue
1741          * packet with the length the drive tells us it can supply
1742          */
1743         cgc.buflen = be16_to_cpu(di->disc_information_length) +
1744                      sizeof(di->disc_information_length);
1745
1746         if (cgc.buflen > sizeof(disc_information))
1747                 cgc.buflen = sizeof(disc_information);
1748
1749         cgc.cmd[8] = cgc.buflen;
1750         return pkt_generic_packet(pd, &cgc);
1751 }
1752
1753 static int pkt_get_track_info(struct pktcdvd_device *pd, __u16 track, __u8 type, track_information *ti)
1754 {
1755         struct packet_command cgc;
1756         int ret;
1757
1758         init_cdrom_command(&cgc, ti, 8, CGC_DATA_READ);
1759         cgc.cmd[0] = GPCMD_READ_TRACK_RZONE_INFO;
1760         cgc.cmd[1] = type & 3;
1761         cgc.cmd[4] = (track & 0xff00) >> 8;
1762         cgc.cmd[5] = track & 0xff;
1763         cgc.cmd[8] = 8;
1764         cgc.quiet = 1;
1765
1766         if ((ret = pkt_generic_packet(pd, &cgc)))
1767                 return ret;
1768
1769         cgc.buflen = be16_to_cpu(ti->track_information_length) +
1770                      sizeof(ti->track_information_length);
1771
1772         if (cgc.buflen > sizeof(track_information))
1773                 cgc.buflen = sizeof(track_information);
1774
1775         cgc.cmd[8] = cgc.buflen;
1776         return pkt_generic_packet(pd, &cgc);
1777 }
1778
1779 static noinline_for_stack int pkt_get_last_written(struct pktcdvd_device *pd,
1780                                                 long *last_written)
1781 {
1782         disc_information di;
1783         track_information ti;
1784         __u32 last_track;
1785         int ret = -1;
1786
1787         if ((ret = pkt_get_disc_info(pd, &di)))
1788                 return ret;
1789
1790         last_track = (di.last_track_msb << 8) | di.last_track_lsb;
1791         if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1792                 return ret;
1793
1794         /* if this track is blank, try the previous. */
1795         if (ti.blank) {
1796                 last_track--;
1797                 if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1798                         return ret;
1799         }
1800
1801         /* if last recorded field is valid, return it. */
1802         if (ti.lra_v) {
1803                 *last_written = be32_to_cpu(ti.last_rec_address);
1804         } else {
1805                 /* make it up instead */
1806                 *last_written = be32_to_cpu(ti.track_start) +
1807                                 be32_to_cpu(ti.track_size);
1808                 if (ti.free_blocks)
1809                         *last_written -= (be32_to_cpu(ti.free_blocks) + 7);
1810         }
1811         return 0;
1812 }
1813
1814 /*
1815  * write mode select package based on pd->settings
1816  */
1817 static noinline_for_stack int pkt_set_write_settings(struct pktcdvd_device *pd)
1818 {
1819         struct packet_command cgc;
1820         struct request_sense sense;
1821         write_param_page *wp;
1822         char buffer[128];
1823         int ret, size;
1824
1825         /* doesn't apply to DVD+RW or DVD-RAM */
1826         if ((pd->mmc3_profile == 0x1a) || (pd->mmc3_profile == 0x12))
1827                 return 0;
1828
1829         memset(buffer, 0, sizeof(buffer));
1830         init_cdrom_command(&cgc, buffer, sizeof(*wp), CGC_DATA_READ);
1831         cgc.sense = &sense;
1832         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1833                 pkt_dump_sense(&cgc);
1834                 return ret;
1835         }
1836
1837         size = 2 + ((buffer[0] << 8) | (buffer[1] & 0xff));
1838         pd->mode_offset = (buffer[6] << 8) | (buffer[7] & 0xff);
1839         if (size > sizeof(buffer))
1840                 size = sizeof(buffer);
1841
1842         /*
1843          * now get it all
1844          */
1845         init_cdrom_command(&cgc, buffer, size, CGC_DATA_READ);
1846         cgc.sense = &sense;
1847         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1848                 pkt_dump_sense(&cgc);
1849                 return ret;
1850         }
1851
1852         /*
1853          * write page is offset header + block descriptor length
1854          */
1855         wp = (write_param_page *) &buffer[sizeof(struct mode_page_header) + pd->mode_offset];
1856
1857         wp->fp = pd->settings.fp;
1858         wp->track_mode = pd->settings.track_mode;
1859         wp->write_type = pd->settings.write_type;
1860         wp->data_block_type = pd->settings.block_mode;
1861
1862         wp->multi_session = 0;
1863
1864 #ifdef PACKET_USE_LS
1865         wp->link_size = 7;
1866         wp->ls_v = 1;
1867 #endif
1868
1869         if (wp->data_block_type == PACKET_BLOCK_MODE1) {
1870                 wp->session_format = 0;
1871                 wp->subhdr2 = 0x20;
1872         } else if (wp->data_block_type == PACKET_BLOCK_MODE2) {
1873                 wp->session_format = 0x20;
1874                 wp->subhdr2 = 8;
1875 #if 0
1876                 wp->mcn[0] = 0x80;
1877                 memcpy(&wp->mcn[1], PACKET_MCN, sizeof(wp->mcn) - 1);
1878 #endif
1879         } else {
1880                 /*
1881                  * paranoia
1882                  */
1883                 printk(DRIVER_NAME": write mode wrong %d\n", wp->data_block_type);
1884                 return 1;
1885         }
1886         wp->packet_size = cpu_to_be32(pd->settings.size >> 2);
1887
1888         cgc.buflen = cgc.cmd[8] = size;
1889         if ((ret = pkt_mode_select(pd, &cgc))) {
1890                 pkt_dump_sense(&cgc);
1891                 return ret;
1892         }
1893
1894         pkt_print_settings(pd);
1895         return 0;
1896 }
1897
1898 /*
1899  * 1 -- we can write to this track, 0 -- we can't
1900  */
1901 static int pkt_writable_track(struct pktcdvd_device *pd, track_information *ti)
1902 {
1903         switch (pd->mmc3_profile) {
1904                 case 0x1a: /* DVD+RW */
1905                 case 0x12: /* DVD-RAM */
1906                         /* The track is always writable on DVD+RW/DVD-RAM */
1907                         return 1;
1908                 default:
1909                         break;
1910         }
1911
1912         if (!ti->packet || !ti->fp)
1913                 return 0;
1914
1915         /*
1916          * "good" settings as per Mt Fuji.
1917          */
1918         if (ti->rt == 0 && ti->blank == 0)
1919                 return 1;
1920
1921         if (ti->rt == 0 && ti->blank == 1)
1922                 return 1;
1923
1924         if (ti->rt == 1 && ti->blank == 0)
1925                 return 1;
1926
1927         printk(DRIVER_NAME": bad state %d-%d-%d\n", ti->rt, ti->blank, ti->packet);
1928         return 0;
1929 }
1930
1931 /*
1932  * 1 -- we can write to this disc, 0 -- we can't
1933  */
1934 static int pkt_writable_disc(struct pktcdvd_device *pd, disc_information *di)
1935 {
1936         switch (pd->mmc3_profile) {
1937                 case 0x0a: /* CD-RW */
1938                 case 0xffff: /* MMC3 not supported */
1939                         break;
1940                 case 0x1a: /* DVD+RW */
1941                 case 0x13: /* DVD-RW */
1942                 case 0x12: /* DVD-RAM */
1943                         return 1;
1944                 default:
1945                         VPRINTK(DRIVER_NAME": Wrong disc profile (%x)\n", pd->mmc3_profile);
1946                         return 0;
1947         }
1948
1949         /*
1950          * for disc type 0xff we should probably reserve a new track.
1951          * but i'm not sure, should we leave this to user apps? probably.
1952          */
1953         if (di->disc_type == 0xff) {
1954                 printk(DRIVER_NAME": Unknown disc. No track?\n");
1955                 return 0;
1956         }
1957
1958         if (di->disc_type != 0x20 && di->disc_type != 0) {
1959                 printk(DRIVER_NAME": Wrong disc type (%x)\n", di->disc_type);
1960                 return 0;
1961         }
1962
1963         if (di->erasable == 0) {
1964                 printk(DRIVER_NAME": Disc not erasable\n");
1965                 return 0;
1966         }
1967
1968         if (di->border_status == PACKET_SESSION_RESERVED) {
1969                 printk(DRIVER_NAME": Can't write to last track (reserved)\n");
1970                 return 0;
1971         }
1972
1973         return 1;
1974 }
1975
1976 static noinline_for_stack int pkt_probe_settings(struct pktcdvd_device *pd)
1977 {
1978         struct packet_command cgc;
1979         unsigned char buf[12];
1980         disc_information di;
1981         track_information ti;
1982         int ret, track;
1983
1984         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
1985         cgc.cmd[0] = GPCMD_GET_CONFIGURATION;
1986         cgc.cmd[8] = 8;
1987         ret = pkt_generic_packet(pd, &cgc);
1988         pd->mmc3_profile = ret ? 0xffff : buf[6] << 8 | buf[7];
1989
1990         memset(&di, 0, sizeof(disc_information));
1991         memset(&ti, 0, sizeof(track_information));
1992
1993         if ((ret = pkt_get_disc_info(pd, &di))) {
1994                 printk("failed get_disc\n");
1995                 return ret;
1996         }
1997
1998         if (!pkt_writable_disc(pd, &di))
1999                 return -EROFS;
2000
2001         pd->type = di.erasable ? PACKET_CDRW : PACKET_CDR;
2002
2003         track = 1; /* (di.last_track_msb << 8) | di.last_track_lsb; */
2004         if ((ret = pkt_get_track_info(pd, track, 1, &ti))) {
2005                 printk(DRIVER_NAME": failed get_track\n");
2006                 return ret;
2007         }
2008
2009         if (!pkt_writable_track(pd, &ti)) {
2010                 printk(DRIVER_NAME": can't write to this track\n");
2011                 return -EROFS;
2012         }
2013
2014         /*
2015          * we keep packet size in 512 byte units, makes it easier to
2016          * deal with request calculations.
2017          */
2018         pd->settings.size = be32_to_cpu(ti.fixed_packet_size) << 2;
2019         if (pd->settings.size == 0) {
2020                 printk(DRIVER_NAME": detected zero packet size!\n");
2021                 return -ENXIO;
2022         }
2023         if (pd->settings.size > PACKET_MAX_SECTORS) {
2024                 printk(DRIVER_NAME": packet size is too big\n");
2025                 return -EROFS;
2026         }
2027         pd->settings.fp = ti.fp;
2028         pd->offset = (be32_to_cpu(ti.track_start) << 2) & (pd->settings.size - 1);
2029
2030         if (ti.nwa_v) {
2031                 pd->nwa = be32_to_cpu(ti.next_writable);
2032                 set_bit(PACKET_NWA_VALID, &pd->flags);
2033         }
2034
2035         /*
2036          * in theory we could use lra on -RW media as well and just zero
2037          * blocks that haven't been written yet, but in practice that
2038          * is just a no-go. we'll use that for -R, naturally.
2039          */
2040         if (ti.lra_v) {
2041                 pd->lra = be32_to_cpu(ti.last_rec_address);
2042                 set_bit(PACKET_LRA_VALID, &pd->flags);
2043         } else {
2044                 pd->lra = 0xffffffff;
2045                 set_bit(PACKET_LRA_VALID, &pd->flags);
2046         }
2047
2048         /*
2049          * fine for now
2050          */
2051         pd->settings.link_loss = 7;
2052         pd->settings.write_type = 0;    /* packet */
2053         pd->settings.track_mode = ti.track_mode;
2054
2055         /*
2056          * mode1 or mode2 disc
2057          */
2058         switch (ti.data_mode) {
2059                 case PACKET_MODE1:
2060                         pd->settings.block_mode = PACKET_BLOCK_MODE1;
2061                         break;
2062                 case PACKET_MODE2:
2063                         pd->settings.block_mode = PACKET_BLOCK_MODE2;
2064                         break;
2065                 default:
2066                         printk(DRIVER_NAME": unknown data mode\n");
2067                         return -EROFS;
2068         }
2069         return 0;
2070 }
2071
2072 /*
2073  * enable/disable write caching on drive
2074  */
2075 static noinline_for_stack int pkt_write_caching(struct pktcdvd_device *pd,
2076                                                 int set)
2077 {
2078         struct packet_command cgc;
2079         struct request_sense sense;
2080         unsigned char buf[64];
2081         int ret;
2082
2083         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
2084         cgc.sense = &sense;
2085         cgc.buflen = pd->mode_offset + 12;
2086
2087         /*
2088          * caching mode page might not be there, so quiet this command
2089          */
2090         cgc.quiet = 1;
2091
2092         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WCACHING_PAGE, 0)))
2093                 return ret;
2094
2095         buf[pd->mode_offset + 10] |= (!!set << 2);
2096
2097         cgc.buflen = cgc.cmd[8] = 2 + ((buf[0] << 8) | (buf[1] & 0xff));
2098         ret = pkt_mode_select(pd, &cgc);
2099         if (ret) {
2100                 printk(DRIVER_NAME": write caching control failed\n");
2101                 pkt_dump_sense(&cgc);
2102         } else if (!ret && set)
2103                 printk(DRIVER_NAME": enabled write caching on %s\n", pd->name);
2104         return ret;
2105 }
2106
2107 static int pkt_lock_door(struct pktcdvd_device *pd, int lockflag)
2108 {
2109         struct packet_command cgc;
2110
2111         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2112         cgc.cmd[0] = GPCMD_PREVENT_ALLOW_MEDIUM_REMOVAL;
2113         cgc.cmd[4] = lockflag ? 1 : 0;
2114         return pkt_generic_packet(pd, &cgc);
2115 }
2116
2117 /*
2118  * Returns drive maximum write speed
2119  */
2120 static noinline_for_stack int pkt_get_max_speed(struct pktcdvd_device *pd,
2121                                                 unsigned *write_speed)
2122 {
2123         struct packet_command cgc;
2124         struct request_sense sense;
2125         unsigned char buf[256+18];
2126         unsigned char *cap_buf;
2127         int ret, offset;
2128
2129         cap_buf = &buf[sizeof(struct mode_page_header) + pd->mode_offset];
2130         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_UNKNOWN);
2131         cgc.sense = &sense;
2132
2133         ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2134         if (ret) {
2135                 cgc.buflen = pd->mode_offset + cap_buf[1] + 2 +
2136                              sizeof(struct mode_page_header);
2137                 ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2138                 if (ret) {
2139                         pkt_dump_sense(&cgc);
2140                         return ret;
2141                 }
2142         }
2143
2144         offset = 20;                        /* Obsoleted field, used by older drives */
2145         if (cap_buf[1] >= 28)
2146                 offset = 28;                /* Current write speed selected */
2147         if (cap_buf[1] >= 30) {
2148                 /* If the drive reports at least one "Logical Unit Write
2149                  * Speed Performance Descriptor Block", use the information
2150                  * in the first block. (contains the highest speed)
2151                  */
2152                 int num_spdb = (cap_buf[30] << 8) + cap_buf[31];
2153                 if (num_spdb > 0)
2154                         offset = 34;
2155         }
2156
2157         *write_speed = (cap_buf[offset] << 8) | cap_buf[offset + 1];
2158         return 0;
2159 }
2160
2161 /* These tables from cdrecord - I don't have orange book */
2162 /* standard speed CD-RW (1-4x) */
2163 static char clv_to_speed[16] = {
2164         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2165            0, 2, 4, 6, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2166 };
2167 /* high speed CD-RW (-10x) */
2168 static char hs_clv_to_speed[16] = {
2169         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2170            0, 2, 4, 6, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2171 };
2172 /* ultra high speed CD-RW */
2173 static char us_clv_to_speed[16] = {
2174         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2175            0, 2, 4, 8, 0, 0,16, 0,24,32,40,48, 0, 0, 0, 0
2176 };
2177
2178 /*
2179  * reads the maximum media speed from ATIP
2180  */
2181 static noinline_for_stack int pkt_media_speed(struct pktcdvd_device *pd,
2182                                                 unsigned *speed)
2183 {
2184         struct packet_command cgc;
2185         struct request_sense sense;
2186         unsigned char buf[64];
2187         unsigned int size, st, sp;
2188         int ret;
2189
2190         init_cdrom_command(&cgc, buf, 2, CGC_DATA_READ);
2191         cgc.sense = &sense;
2192         cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2193         cgc.cmd[1] = 2;
2194         cgc.cmd[2] = 4; /* READ ATIP */
2195         cgc.cmd[8] = 2;
2196         ret = pkt_generic_packet(pd, &cgc);
2197         if (ret) {
2198                 pkt_dump_sense(&cgc);
2199                 return ret;
2200         }
2201         size = ((unsigned int) buf[0]<<8) + buf[1] + 2;
2202         if (size > sizeof(buf))
2203                 size = sizeof(buf);
2204
2205         init_cdrom_command(&cgc, buf, size, CGC_DATA_READ);
2206         cgc.sense = &sense;
2207         cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2208         cgc.cmd[1] = 2;
2209         cgc.cmd[2] = 4;
2210         cgc.cmd[8] = size;
2211         ret = pkt_generic_packet(pd, &cgc);
2212         if (ret) {
2213                 pkt_dump_sense(&cgc);
2214                 return ret;
2215         }
2216
2217         if (!(buf[6] & 0x40)) {
2218                 printk(DRIVER_NAME": Disc type is not CD-RW\n");
2219                 return 1;
2220         }
2221         if (!(buf[6] & 0x4)) {
2222                 printk(DRIVER_NAME": A1 values on media are not valid, maybe not CDRW?\n");
2223                 return 1;
2224         }
2225
2226         st = (buf[6] >> 3) & 0x7; /* disc sub-type */
2227
2228         sp = buf[16] & 0xf; /* max speed from ATIP A1 field */
2229
2230         /* Info from cdrecord */
2231         switch (st) {
2232                 case 0: /* standard speed */
2233                         *speed = clv_to_speed[sp];
2234                         break;
2235                 case 1: /* high speed */
2236                         *speed = hs_clv_to_speed[sp];
2237                         break;
2238                 case 2: /* ultra high speed */
2239                         *speed = us_clv_to_speed[sp];
2240                         break;
2241                 default:
2242                         printk(DRIVER_NAME": Unknown disc sub-type %d\n",st);
2243                         return 1;
2244         }
2245         if (*speed) {
2246                 printk(DRIVER_NAME": Max. media speed: %d\n",*speed);
2247                 return 0;
2248         } else {
2249                 printk(DRIVER_NAME": Unknown speed %d for sub-type %d\n",sp,st);
2250                 return 1;
2251         }
2252 }
2253
2254 static noinline_for_stack int pkt_perform_opc(struct pktcdvd_device *pd)
2255 {
2256         struct packet_command cgc;
2257         struct request_sense sense;
2258         int ret;
2259
2260         VPRINTK(DRIVER_NAME": Performing OPC\n");
2261
2262         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2263         cgc.sense = &sense;
2264         cgc.timeout = 60*HZ;
2265         cgc.cmd[0] = GPCMD_SEND_OPC;
2266         cgc.cmd[1] = 1;
2267         if ((ret = pkt_generic_packet(pd, &cgc)))
2268                 pkt_dump_sense(&cgc);
2269         return ret;
2270 }
2271
2272 static int pkt_open_write(struct pktcdvd_device *pd)
2273 {
2274         int ret;
2275         unsigned int write_speed, media_write_speed, read_speed;
2276
2277         if ((ret = pkt_probe_settings(pd))) {
2278                 VPRINTK(DRIVER_NAME": %s failed probe\n", pd->name);
2279                 return ret;
2280         }
2281
2282         if ((ret = pkt_set_write_settings(pd))) {
2283                 DPRINTK(DRIVER_NAME": %s failed saving write settings\n", pd->name);
2284                 return -EIO;
2285         }
2286
2287         pkt_write_caching(pd, USE_WCACHING);
2288
2289         if ((ret = pkt_get_max_speed(pd, &write_speed)))
2290                 write_speed = 16 * 177;
2291         switch (pd->mmc3_profile) {
2292                 case 0x13: /* DVD-RW */
2293                 case 0x1a: /* DVD+RW */
2294                 case 0x12: /* DVD-RAM */
2295                         DPRINTK(DRIVER_NAME": write speed %ukB/s\n", write_speed);
2296                         break;
2297                 default:
2298                         if ((ret = pkt_media_speed(pd, &media_write_speed)))
2299                                 media_write_speed = 16;
2300                         write_speed = min(write_speed, media_write_speed * 177);
2301                         DPRINTK(DRIVER_NAME": write speed %ux\n", write_speed / 176);
2302                         break;
2303         }
2304         read_speed = write_speed;
2305
2306         if ((ret = pkt_set_speed(pd, write_speed, read_speed))) {
2307                 DPRINTK(DRIVER_NAME": %s couldn't set write speed\n", pd->name);
2308                 return -EIO;
2309         }
2310         pd->write_speed = write_speed;
2311         pd->read_speed = read_speed;
2312
2313         if ((ret = pkt_perform_opc(pd))) {
2314                 DPRINTK(DRIVER_NAME": %s Optimum Power Calibration failed\n", pd->name);
2315         }
2316
2317         return 0;
2318 }
2319
2320 /*
2321  * called at open time.
2322  */
2323 static int pkt_open_dev(struct pktcdvd_device *pd, fmode_t write)
2324 {
2325         int ret;
2326         long lba;
2327         struct request_queue *q;
2328
2329         /*
2330          * We need to re-open the cdrom device without O_NONBLOCK to be able
2331          * to read/write from/to it. It is already opened in O_NONBLOCK mode
2332          * so bdget() can't fail.
2333          */
2334         bdget(pd->bdev->bd_dev);
2335         if ((ret = blkdev_get(pd->bdev, FMODE_READ)))
2336                 goto out;
2337
2338         if ((ret = bd_claim(pd->bdev, pd)))
2339                 goto out_putdev;
2340
2341         if ((ret = pkt_get_last_written(pd, &lba))) {
2342                 printk(DRIVER_NAME": pkt_get_last_written failed\n");
2343                 goto out_unclaim;
2344         }
2345
2346         set_capacity(pd->disk, lba << 2);
2347         set_capacity(pd->bdev->bd_disk, lba << 2);
2348         bd_set_size(pd->bdev, (loff_t)lba << 11);
2349
2350         q = bdev_get_queue(pd->bdev);
2351         if (write) {
2352                 if ((ret = pkt_open_write(pd)))
2353                         goto out_unclaim;
2354                 /*
2355                  * Some CDRW drives can not handle writes larger than one packet,
2356                  * even if the size is a multiple of the packet size.
2357                  */
2358                 spin_lock_irq(q->queue_lock);
2359                 blk_queue_max_sectors(q, pd->settings.size);
2360                 spin_unlock_irq(q->queue_lock);
2361                 set_bit(PACKET_WRITABLE, &pd->flags);
2362         } else {
2363                 pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2364                 clear_bit(PACKET_WRITABLE, &pd->flags);
2365         }
2366
2367         if ((ret = pkt_set_segment_merging(pd, q)))
2368                 goto out_unclaim;
2369
2370         if (write) {
2371                 if (!pkt_grow_pktlist(pd, CONFIG_CDROM_PKTCDVD_BUFFERS)) {
2372                         printk(DRIVER_NAME": not enough memory for buffers\n");
2373                         ret = -ENOMEM;
2374                         goto out_unclaim;
2375                 }
2376                 printk(DRIVER_NAME": %lukB available on disc\n", lba << 1);
2377         }
2378
2379         return 0;
2380
2381 out_unclaim:
2382         bd_release(pd->bdev);
2383 out_putdev:
2384         blkdev_put(pd->bdev, FMODE_READ);
2385 out:
2386         return ret;
2387 }
2388
2389 /*
2390  * called when the device is closed. makes sure that the device flushes
2391  * the internal cache before we close.
2392  */
2393 static void pkt_release_dev(struct pktcdvd_device *pd, int flush)
2394 {
2395         if (flush && pkt_flush_cache(pd))
2396                 DPRINTK(DRIVER_NAME": %s not flushing cache\n", pd->name);
2397
2398         pkt_lock_door(pd, 0);
2399
2400         pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2401         bd_release(pd->bdev);
2402         blkdev_put(pd->bdev, FMODE_READ);
2403
2404         pkt_shrink_pktlist(pd);
2405 }
2406
2407 static struct pktcdvd_device *pkt_find_dev_from_minor(int dev_minor)
2408 {
2409         if (dev_minor >= MAX_WRITERS)
2410                 return NULL;
2411         return pkt_devs[dev_minor];
2412 }
2413
2414 static int pkt_open(struct block_device *bdev, fmode_t mode)
2415 {
2416         struct pktcdvd_device *pd = NULL;
2417         int ret;
2418
2419         VPRINTK(DRIVER_NAME": entering open\n");
2420
2421         mutex_lock(&ctl_mutex);
2422         pd = pkt_find_dev_from_minor(MINOR(bdev->bd_dev));
2423         if (!pd) {
2424                 ret = -ENODEV;
2425                 goto out;
2426         }
2427         BUG_ON(pd->refcnt < 0);
2428
2429         pd->refcnt++;
2430         if (pd->refcnt > 1) {
2431                 if ((mode & FMODE_WRITE) &&
2432                     !test_bit(PACKET_WRITABLE, &pd->flags)) {
2433                         ret = -EBUSY;
2434                         goto out_dec;
2435                 }
2436         } else {
2437                 ret = pkt_open_dev(pd, mode & FMODE_WRITE);
2438                 if (ret)
2439                         goto out_dec;
2440                 /*
2441                  * needed here as well, since ext2 (among others) may change
2442                  * the blocksize at mount time
2443                  */
2444                 set_blocksize(bdev, CD_FRAMESIZE);
2445         }
2446
2447         mutex_unlock(&ctl_mutex);
2448         return 0;
2449
2450 out_dec:
2451         pd->refcnt--;
2452 out:
2453         VPRINTK(DRIVER_NAME": failed open (%d)\n", ret);
2454         mutex_unlock(&ctl_mutex);
2455         return ret;
2456 }
2457
2458 static int pkt_close(struct gendisk *disk, fmode_t mode)
2459 {
2460         struct pktcdvd_device *pd = disk->private_data;
2461         int ret = 0;
2462
2463         mutex_lock(&ctl_mutex);
2464         pd->refcnt--;
2465         BUG_ON(pd->refcnt < 0);
2466         if (pd->refcnt == 0) {
2467                 int flush = test_bit(PACKET_WRITABLE, &pd->flags);
2468                 pkt_release_dev(pd, flush);
2469         }
2470         mutex_unlock(&ctl_mutex);
2471         return ret;
2472 }
2473
2474
2475 static void pkt_end_io_read_cloned(struct bio *bio, int err)
2476 {
2477         struct packet_stacked_data *psd = bio->bi_private;
2478         struct pktcdvd_device *pd = psd->pd;
2479
2480         bio_put(bio);
2481         bio_endio(psd->bio, err);
2482         mempool_free(psd, psd_pool);
2483         pkt_bio_finished(pd);
2484 }
2485
2486 static int pkt_make_request(struct request_queue *q, struct bio *bio)
2487 {
2488         struct pktcdvd_device *pd;
2489         char b[BDEVNAME_SIZE];
2490         sector_t zone;
2491         struct packet_data *pkt;
2492         int was_empty, blocked_bio;
2493         struct pkt_rb_node *node;
2494
2495         pd = q->queuedata;
2496         if (!pd) {
2497                 printk(DRIVER_NAME": %s incorrect request queue\n", bdevname(bio->bi_bdev, b));
2498                 goto end_io;
2499         }
2500
2501         /*
2502          * Clone READ bios so we can have our own bi_end_io callback.
2503          */
2504         if (bio_data_dir(bio) == READ) {
2505                 struct bio *cloned_bio = bio_clone(bio, GFP_NOIO);
2506                 struct packet_stacked_data *psd = mempool_alloc(psd_pool, GFP_NOIO);
2507
2508                 psd->pd = pd;
2509                 psd->bio = bio;
2510                 cloned_bio->bi_bdev = pd->bdev;
2511                 cloned_bio->bi_private = psd;
2512                 cloned_bio->bi_end_io = pkt_end_io_read_cloned;
2513                 pd->stats.secs_r += bio->bi_size >> 9;
2514                 pkt_queue_bio(pd, cloned_bio);
2515                 return 0;
2516         }
2517
2518         if (!test_bit(PACKET_WRITABLE, &pd->flags)) {
2519                 printk(DRIVER_NAME": WRITE for ro device %s (%llu)\n",
2520                         pd->name, (unsigned long long)bio->bi_sector);
2521                 goto end_io;
2522         }
2523
2524         if (!bio->bi_size || (bio->bi_size % CD_FRAMESIZE)) {
2525                 printk(DRIVER_NAME": wrong bio size\n");
2526                 goto end_io;
2527         }
2528
2529         blk_queue_bounce(q, &bio);
2530
2531         zone = ZONE(bio->bi_sector, pd);
2532         VPRINTK("pkt_make_request: start = %6llx stop = %6llx\n",
2533                 (unsigned long long)bio->bi_sector,
2534                 (unsigned long long)(bio->bi_sector + bio_sectors(bio)));
2535
2536         /* Check if we have to split the bio */
2537         {
2538                 struct bio_pair *bp;
2539                 sector_t last_zone;
2540                 int first_sectors;
2541
2542                 last_zone = ZONE(bio->bi_sector + bio_sectors(bio) - 1, pd);
2543                 if (last_zone != zone) {
2544                         BUG_ON(last_zone != zone + pd->settings.size);
2545                         first_sectors = last_zone - bio->bi_sector;
2546                         bp = bio_split(bio, first_sectors);
2547                         BUG_ON(!bp);
2548                         pkt_make_request(q, &bp->bio1);
2549                         pkt_make_request(q, &bp->bio2);
2550                         bio_pair_release(bp);
2551                         return 0;
2552                 }
2553         }
2554
2555         /*
2556          * If we find a matching packet in state WAITING or READ_WAIT, we can
2557          * just append this bio to that packet.
2558          */
2559         spin_lock(&pd->cdrw.active_list_lock);
2560         blocked_bio = 0;
2561         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
2562                 if (pkt->sector == zone) {
2563                         spin_lock(&pkt->lock);
2564                         if ((pkt->state == PACKET_WAITING_STATE) ||
2565                             (pkt->state == PACKET_READ_WAIT_STATE)) {
2566                                 pkt_add_list_last(bio, &pkt->orig_bios,
2567                                                   &pkt->orig_bios_tail);
2568                                 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
2569                                 if ((pkt->write_size >= pkt->frames) &&
2570                                     (pkt->state == PACKET_WAITING_STATE)) {
2571                                         atomic_inc(&pkt->run_sm);
2572                                         wake_up(&pd->wqueue);
2573                                 }
2574                                 spin_unlock(&pkt->lock);
2575                                 spin_unlock(&pd->cdrw.active_list_lock);
2576                                 return 0;
2577                         } else {
2578                                 blocked_bio = 1;
2579                         }
2580                         spin_unlock(&pkt->lock);
2581                 }
2582         }
2583         spin_unlock(&pd->cdrw.active_list_lock);
2584
2585         /*
2586          * Test if there is enough room left in the bio work queue
2587          * (queue size >= congestion on mark).
2588          * If not, wait till the work queue size is below the congestion off mark.
2589          */
2590         spin_lock(&pd->lock);
2591         if (pd->write_congestion_on > 0
2592             && pd->bio_queue_size >= pd->write_congestion_on) {
2593                 set_bdi_congested(&q->backing_dev_info, WRITE);
2594                 do {
2595                         spin_unlock(&pd->lock);
2596                         congestion_wait(WRITE, HZ);
2597                         spin_lock(&pd->lock);
2598                 } while(pd->bio_queue_size > pd->write_congestion_off);
2599         }
2600         spin_unlock(&pd->lock);
2601
2602         /*
2603          * No matching packet found. Store the bio in the work queue.
2604          */
2605         node = mempool_alloc(pd->rb_pool, GFP_NOIO);
2606         node->bio = bio;
2607         spin_lock(&pd->lock);
2608         BUG_ON(pd->bio_queue_size < 0);
2609         was_empty = (pd->bio_queue_size == 0);
2610         pkt_rbtree_insert(pd, node);
2611         spin_unlock(&pd->lock);
2612
2613         /*
2614          * Wake up the worker thread.
2615          */
2616         atomic_set(&pd->scan_queue, 1);
2617         if (was_empty) {
2618                 /* This wake_up is required for correct operation */
2619                 wake_up(&pd->wqueue);
2620         } else if (!list_empty(&pd->cdrw.pkt_free_list) && !blocked_bio) {
2621                 /*
2622                  * This wake up is not required for correct operation,
2623                  * but improves performance in some cases.
2624                  */
2625                 wake_up(&pd->wqueue);
2626         }
2627         return 0;
2628 end_io:
2629         bio_io_error(bio);
2630         return 0;
2631 }
2632
2633
2634
2635 static int pkt_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2636                           struct bio_vec *bvec)
2637 {
2638         struct pktcdvd_device *pd = q->queuedata;
2639         sector_t zone = ZONE(bmd->bi_sector, pd);
2640         int used = ((bmd->bi_sector - zone) << 9) + bmd->bi_size;
2641         int remaining = (pd->settings.size << 9) - used;
2642         int remaining2;
2643
2644         /*
2645          * A bio <= PAGE_SIZE must be allowed. If it crosses a packet
2646          * boundary, pkt_make_request() will split the bio.
2647          */
2648         remaining2 = PAGE_SIZE - bmd->bi_size;
2649         remaining = max(remaining, remaining2);
2650
2651         BUG_ON(remaining < 0);
2652         return remaining;
2653 }
2654
2655 static void pkt_init_queue(struct pktcdvd_device *pd)
2656 {
2657         struct request_queue *q = pd->disk->queue;
2658
2659         blk_queue_make_request(q, pkt_make_request);
2660         blk_queue_hardsect_size(q, CD_FRAMESIZE);
2661         blk_queue_max_sectors(q, PACKET_MAX_SECTORS);
2662         blk_queue_merge_bvec(q, pkt_merge_bvec);
2663         q->queuedata = pd;
2664 }
2665
2666 static int pkt_seq_show(struct seq_file *m, void *p)
2667 {
2668         struct pktcdvd_device *pd = m->private;
2669         char *msg;
2670         char bdev_buf[BDEVNAME_SIZE];
2671         int states[PACKET_NUM_STATES];
2672
2673         seq_printf(m, "Writer %s mapped to %s:\n", pd->name,
2674                    bdevname(pd->bdev, bdev_buf));
2675
2676         seq_printf(m, "\nSettings:\n");
2677         seq_printf(m, "\tpacket size:\t\t%dkB\n", pd->settings.size / 2);
2678
2679         if (pd->settings.write_type == 0)
2680                 msg = "Packet";
2681         else
2682                 msg = "Unknown";
2683         seq_printf(m, "\twrite type:\t\t%s\n", msg);
2684
2685         seq_printf(m, "\tpacket type:\t\t%s\n", pd->settings.fp ? "Fixed" : "Variable");
2686         seq_printf(m, "\tlink loss:\t\t%d\n", pd->settings.link_loss);
2687
2688         seq_printf(m, "\ttrack mode:\t\t%d\n", pd->settings.track_mode);
2689
2690         if (pd->settings.block_mode == PACKET_BLOCK_MODE1)
2691                 msg = "Mode 1";
2692         else if (pd->settings.block_mode == PACKET_BLOCK_MODE2)
2693                 msg = "Mode 2";
2694         else
2695                 msg = "Unknown";
2696         seq_printf(m, "\tblock mode:\t\t%s\n", msg);
2697
2698         seq_printf(m, "\nStatistics:\n");
2699         seq_printf(m, "\tpackets started:\t%lu\n", pd->stats.pkt_started);
2700         seq_printf(m, "\tpackets ended:\t\t%lu\n", pd->stats.pkt_ended);
2701         seq_printf(m, "\twritten:\t\t%lukB\n", pd->stats.secs_w >> 1);
2702         seq_printf(m, "\tread gather:\t\t%lukB\n", pd->stats.secs_rg >> 1);
2703         seq_printf(m, "\tread:\t\t\t%lukB\n", pd->stats.secs_r >> 1);
2704
2705         seq_printf(m, "\nMisc:\n");
2706         seq_printf(m, "\treference count:\t%d\n", pd->refcnt);
2707         seq_printf(m, "\tflags:\t\t\t0x%lx\n", pd->flags);
2708         seq_printf(m, "\tread speed:\t\t%ukB/s\n", pd->read_speed);
2709         seq_printf(m, "\twrite speed:\t\t%ukB/s\n", pd->write_speed);
2710         seq_printf(m, "\tstart offset:\t\t%lu\n", pd->offset);
2711         seq_printf(m, "\tmode page offset:\t%u\n", pd->mode_offset);
2712
2713         seq_printf(m, "\nQueue state:\n");
2714         seq_printf(m, "\tbios queued:\t\t%d\n", pd->bio_queue_size);
2715         seq_printf(m, "\tbios pending:\t\t%d\n", atomic_read(&pd->cdrw.pending_bios));
2716         seq_printf(m, "\tcurrent sector:\t\t0x%llx\n", (unsigned long long)pd->current_sector);
2717
2718         pkt_count_states(pd, states);
2719         seq_printf(m, "\tstate:\t\t\ti:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
2720                    states[0], states[1], states[2], states[3], states[4], states[5]);
2721
2722         seq_printf(m, "\twrite congestion marks:\toff=%d on=%d\n",
2723                         pd->write_congestion_off,
2724                         pd->write_congestion_on);
2725         return 0;
2726 }
2727
2728 static int pkt_seq_open(struct inode *inode, struct file *file)
2729 {
2730         return single_open(file, pkt_seq_show, PDE(inode)->data);
2731 }
2732
2733 static const struct file_operations pkt_proc_fops = {
2734         .open   = pkt_seq_open,
2735         .read   = seq_read,
2736         .llseek = seq_lseek,
2737         .release = single_release
2738 };
2739
2740 static int pkt_new_dev(struct pktcdvd_device *pd, dev_t dev)
2741 {
2742         int i;
2743         int ret = 0;
2744         char b[BDEVNAME_SIZE];
2745         struct block_device *bdev;
2746
2747         if (pd->pkt_dev == dev) {
2748                 printk(DRIVER_NAME": Recursive setup not allowed\n");
2749                 return -EBUSY;
2750         }
2751         for (i = 0; i < MAX_WRITERS; i++) {
2752                 struct pktcdvd_device *pd2 = pkt_devs[i];
2753                 if (!pd2)
2754                         continue;
2755                 if (pd2->bdev->bd_dev == dev) {
2756                         printk(DRIVER_NAME": %s already setup\n", bdevname(pd2->bdev, b));
2757                         return -EBUSY;
2758                 }
2759                 if (pd2->pkt_dev == dev) {
2760                         printk(DRIVER_NAME": Can't chain pktcdvd devices\n");
2761                         return -EBUSY;
2762                 }
2763         }
2764
2765         bdev = bdget(dev);
2766         if (!bdev)
2767                 return -ENOMEM;
2768         ret = blkdev_get(bdev, FMODE_READ | FMODE_NDELAY);
2769         if (ret)
2770                 return ret;
2771
2772         /* This is safe, since we have a reference from open(). */
2773         __module_get(THIS_MODULE);
2774
2775         pd->bdev = bdev;
2776         set_blocksize(bdev, CD_FRAMESIZE);
2777
2778         pkt_init_queue(pd);
2779
2780         atomic_set(&pd->cdrw.pending_bios, 0);
2781         pd->cdrw.thread = kthread_run(kcdrwd, pd, "%s", pd->name);
2782         if (IS_ERR(pd->cdrw.thread)) {
2783                 printk(DRIVER_NAME": can't start kernel thread\n");
2784                 ret = -ENOMEM;
2785                 goto out_mem;
2786         }
2787
2788         proc_create_data(pd->name, 0, pkt_proc, &pkt_proc_fops, pd);
2789         DPRINTK(DRIVER_NAME": writer %s mapped to %s\n", pd->name, bdevname(bdev, b));
2790         return 0;
2791
2792 out_mem:
2793         blkdev_put(bdev, FMODE_READ | FMODE_NDELAY);
2794         /* This is safe: open() is still holding a reference. */
2795         module_put(THIS_MODULE);
2796         return ret;
2797 }
2798
2799 static int pkt_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, unsigned long arg)
2800 {
2801         struct pktcdvd_device *pd = bdev->bd_disk->private_data;
2802
2803         VPRINTK("pkt_ioctl: cmd %x, dev %d:%d\n", cmd,
2804                 MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev));
2805
2806         switch (cmd) {
2807         case CDROMEJECT:
2808                 /*
2809                  * The door gets locked when the device is opened, so we
2810                  * have to unlock it or else the eject command fails.
2811                  */
2812                 if (pd->refcnt == 1)
2813                         pkt_lock_door(pd, 0);
2814                 /* fallthru */
2815         /*
2816          * forward selected CDROM ioctls to CD-ROM, for UDF
2817          */
2818         case CDROMMULTISESSION:
2819         case CDROMREADTOCENTRY:
2820         case CDROM_LAST_WRITTEN:
2821         case CDROM_SEND_PACKET:
2822         case SCSI_IOCTL_SEND_COMMAND:
2823                 return __blkdev_driver_ioctl(pd->bdev, mode, cmd, arg);
2824
2825         default:
2826                 VPRINTK(DRIVER_NAME": Unknown ioctl for %s (%x)\n", pd->name, cmd);
2827                 return -ENOTTY;
2828         }
2829
2830         return 0;
2831 }
2832
2833 static int pkt_media_changed(struct gendisk *disk)
2834 {
2835         struct pktcdvd_device *pd = disk->private_data;
2836         struct gendisk *attached_disk;
2837
2838         if (!pd)
2839                 return 0;
2840         if (!pd->bdev)
2841                 return 0;
2842         attached_disk = pd->bdev->bd_disk;
2843         if (!attached_disk)
2844                 return 0;
2845         return attached_disk->fops->media_changed(attached_disk);
2846 }
2847
2848 static struct block_device_operations pktcdvd_ops = {
2849         .owner =                THIS_MODULE,
2850         .open =                 pkt_open,
2851         .release =              pkt_close,
2852         .locked_ioctl =         pkt_ioctl,
2853         .media_changed =        pkt_media_changed,
2854 };
2855
2856 /*
2857  * Set up mapping from pktcdvd device to CD-ROM device.
2858  */
2859 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev)
2860 {
2861         int idx;
2862         int ret = -ENOMEM;
2863         struct pktcdvd_device *pd;
2864         struct gendisk *disk;
2865
2866         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2867
2868         for (idx = 0; idx < MAX_WRITERS; idx++)
2869                 if (!pkt_devs[idx])
2870                         break;
2871         if (idx == MAX_WRITERS) {
2872                 printk(DRIVER_NAME": max %d writers supported\n", MAX_WRITERS);
2873                 ret = -EBUSY;
2874                 goto out_mutex;
2875         }
2876
2877         pd = kzalloc(sizeof(struct pktcdvd_device), GFP_KERNEL);
2878         if (!pd)
2879                 goto out_mutex;
2880
2881         pd->rb_pool = mempool_create_kmalloc_pool(PKT_RB_POOL_SIZE,
2882                                                   sizeof(struct pkt_rb_node));
2883         if (!pd->rb_pool)
2884                 goto out_mem;
2885
2886         INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
2887         INIT_LIST_HEAD(&pd->cdrw.pkt_active_list);
2888         spin_lock_init(&pd->cdrw.active_list_lock);
2889
2890         spin_lock_init(&pd->lock);
2891         spin_lock_init(&pd->iosched.lock);
2892         sprintf(pd->name, DRIVER_NAME"%d", idx);
2893         init_waitqueue_head(&pd->wqueue);
2894         pd->bio_queue = RB_ROOT;
2895
2896         pd->write_congestion_on  = write_congestion_on;
2897         pd->write_congestion_off = write_congestion_off;
2898
2899         disk = alloc_disk(1);
2900         if (!disk)
2901                 goto out_mem;
2902         pd->disk = disk;
2903         disk->major = pktdev_major;
2904         disk->first_minor = idx;
2905         disk->fops = &pktcdvd_ops;
2906         disk->flags = GENHD_FL_REMOVABLE;
2907         strcpy(disk->disk_name, pd->name);
2908         disk->private_data = pd;
2909         disk->queue = blk_alloc_queue(GFP_KERNEL);
2910         if (!disk->queue)
2911                 goto out_mem2;
2912
2913         pd->pkt_dev = MKDEV(pktdev_major, idx);
2914         ret = pkt_new_dev(pd, dev);
2915         if (ret)
2916                 goto out_new_dev;
2917
2918         add_disk(disk);
2919
2920         pkt_sysfs_dev_new(pd);
2921         pkt_debugfs_dev_new(pd);
2922
2923         pkt_devs[idx] = pd;
2924         if (pkt_dev)
2925                 *pkt_dev = pd->pkt_dev;
2926
2927         mutex_unlock(&ctl_mutex);
2928         return 0;
2929
2930 out_new_dev:
2931         blk_cleanup_queue(disk->queue);
2932 out_mem2:
2933         put_disk(disk);
2934 out_mem:
2935         if (pd->rb_pool)
2936                 mempool_destroy(pd->rb_pool);
2937         kfree(pd);
2938 out_mutex:
2939         mutex_unlock(&ctl_mutex);
2940         printk(DRIVER_NAME": setup of pktcdvd device failed\n");
2941         return ret;
2942 }
2943
2944 /*
2945  * Tear down mapping from pktcdvd device to CD-ROM device.
2946  */
2947 static int pkt_remove_dev(dev_t pkt_dev)
2948 {
2949         struct pktcdvd_device *pd;
2950         int idx;
2951         int ret = 0;
2952
2953         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2954
2955         for (idx = 0; idx < MAX_WRITERS; idx++) {
2956                 pd = pkt_devs[idx];
2957                 if (pd && (pd->pkt_dev == pkt_dev))
2958                         break;
2959         }
2960         if (idx == MAX_WRITERS) {
2961                 DPRINTK(DRIVER_NAME": dev not setup\n");
2962                 ret = -ENXIO;
2963                 goto out;
2964         }
2965
2966         if (pd->refcnt > 0) {
2967                 ret = -EBUSY;
2968                 goto out;
2969         }
2970         if (!IS_ERR(pd->cdrw.thread))
2971                 kthread_stop(pd->cdrw.thread);
2972
2973         pkt_devs[idx] = NULL;
2974
2975         pkt_debugfs_dev_remove(pd);
2976         pkt_sysfs_dev_remove(pd);
2977
2978         blkdev_put(pd->bdev, FMODE_READ | FMODE_NDELAY);
2979
2980         remove_proc_entry(pd->name, pkt_proc);
2981         DPRINTK(DRIVER_NAME": writer %s unmapped\n", pd->name);
2982
2983         del_gendisk(pd->disk);
2984         blk_cleanup_queue(pd->disk->queue);
2985         put_disk(pd->disk);
2986
2987         mempool_destroy(pd->rb_pool);
2988         kfree(pd);
2989
2990         /* This is safe: open() is still holding a reference. */
2991         module_put(THIS_MODULE);
2992
2993 out:
2994         mutex_unlock(&ctl_mutex);
2995         return ret;
2996 }
2997
2998 static void pkt_get_status(struct pkt_ctrl_command *ctrl_cmd)
2999 {
3000         struct pktcdvd_device *pd;
3001
3002         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3003
3004         pd = pkt_find_dev_from_minor(ctrl_cmd->dev_index);
3005         if (pd) {
3006                 ctrl_cmd->dev = new_encode_dev(pd->bdev->bd_dev);
3007                 ctrl_cmd->pkt_dev = new_encode_dev(pd->pkt_dev);
3008         } else {
3009                 ctrl_cmd->dev = 0;
3010                 ctrl_cmd->pkt_dev = 0;
3011         }
3012         ctrl_cmd->num_devices = MAX_WRITERS;
3013
3014         mutex_unlock(&ctl_mutex);
3015 }
3016
3017 static int pkt_ctl_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg)
3018 {
3019         void __user *argp = (void __user *)arg;
3020         struct pkt_ctrl_command ctrl_cmd;
3021         int ret = 0;
3022         dev_t pkt_dev = 0;
3023
3024         if (cmd != PACKET_CTRL_CMD)
3025                 return -ENOTTY;
3026
3027         if (copy_from_user(&ctrl_cmd, argp, sizeof(struct pkt_ctrl_command)))
3028                 return -EFAULT;
3029
3030         switch (ctrl_cmd.command) {
3031         case PKT_CTRL_CMD_SETUP:
3032                 if (!capable(CAP_SYS_ADMIN))
3033                         return -EPERM;
3034                 ret = pkt_setup_dev(new_decode_dev(ctrl_cmd.dev), &pkt_dev);
3035                 ctrl_cmd.pkt_dev = new_encode_dev(pkt_dev);
3036                 break;
3037         case PKT_CTRL_CMD_TEARDOWN:
3038                 if (!capable(CAP_SYS_ADMIN))
3039                         return -EPERM;
3040                 ret = pkt_remove_dev(new_decode_dev(ctrl_cmd.pkt_dev));
3041                 break;
3042         case PKT_CTRL_CMD_STATUS:
3043                 pkt_get_status(&ctrl_cmd);
3044                 break;
3045         default:
3046                 return -ENOTTY;
3047         }
3048
3049         if (copy_to_user(argp, &ctrl_cmd, sizeof(struct pkt_ctrl_command)))
3050                 return -EFAULT;
3051         return ret;
3052 }
3053
3054
3055 static const struct file_operations pkt_ctl_fops = {
3056         .ioctl   = pkt_ctl_ioctl,
3057         .owner   = THIS_MODULE,
3058 };
3059
3060 static struct miscdevice pkt_misc = {
3061         .minor          = MISC_DYNAMIC_MINOR,
3062         .name           = DRIVER_NAME,
3063         .fops           = &pkt_ctl_fops
3064 };
3065
3066 static int __init pkt_init(void)
3067 {
3068         int ret;
3069
3070         mutex_init(&ctl_mutex);
3071
3072         psd_pool = mempool_create_kmalloc_pool(PSD_POOL_SIZE,
3073                                         sizeof(struct packet_stacked_data));
3074         if (!psd_pool)
3075                 return -ENOMEM;
3076
3077         ret = register_blkdev(pktdev_major, DRIVER_NAME);
3078         if (ret < 0) {
3079                 printk(DRIVER_NAME": Unable to register block device\n");
3080                 goto out2;
3081         }
3082         if (!pktdev_major)
3083                 pktdev_major = ret;
3084
3085         ret = pkt_sysfs_init();
3086         if (ret)
3087                 goto out;
3088
3089         pkt_debugfs_init();
3090
3091         ret = misc_register(&pkt_misc);
3092         if (ret) {
3093                 printk(DRIVER_NAME": Unable to register misc device\n");
3094                 goto out_misc;
3095         }
3096
3097         pkt_proc = proc_mkdir("driver/"DRIVER_NAME, NULL);
3098
3099         return 0;
3100
3101 out_misc:
3102         pkt_debugfs_cleanup();
3103         pkt_sysfs_cleanup();
3104 out:
3105         unregister_blkdev(pktdev_major, DRIVER_NAME);
3106 out2:
3107         mempool_destroy(psd_pool);
3108         return ret;
3109 }
3110
3111 static void __exit pkt_exit(void)
3112 {
3113         remove_proc_entry("driver/"DRIVER_NAME, NULL);
3114         misc_deregister(&pkt_misc);
3115
3116         pkt_debugfs_cleanup();
3117         pkt_sysfs_cleanup();
3118
3119         unregister_blkdev(pktdev_major, DRIVER_NAME);
3120         mempool_destroy(psd_pool);
3121 }
3122
3123 MODULE_DESCRIPTION("Packet writing layer for CD/DVD drives");
3124 MODULE_AUTHOR("Jens Axboe <axboe@suse.de>");
3125 MODULE_LICENSE("GPL");
3126
3127 module_init(pkt_init);
3128 module_exit(pkt_exit);