Driver Core: block: add nodename support for block drivers.
[safe/jmp/linux-2.6] / drivers / block / pktcdvd.c
1 /*
2  * Copyright (C) 2000 Jens Axboe <axboe@suse.de>
3  * Copyright (C) 2001-2004 Peter Osterlund <petero2@telia.com>
4  * Copyright (C) 2006 Thomas Maier <balagi@justmail.de>
5  *
6  * May be copied or modified under the terms of the GNU General Public
7  * License.  See linux/COPYING for more information.
8  *
9  * Packet writing layer for ATAPI and SCSI CD-RW, DVD+RW, DVD-RW and
10  * DVD-RAM devices.
11  *
12  * Theory of operation:
13  *
14  * At the lowest level, there is the standard driver for the CD/DVD device,
15  * typically ide-cd.c or sr.c. This driver can handle read and write requests,
16  * but it doesn't know anything about the special restrictions that apply to
17  * packet writing. One restriction is that write requests must be aligned to
18  * packet boundaries on the physical media, and the size of a write request
19  * must be equal to the packet size. Another restriction is that a
20  * GPCMD_FLUSH_CACHE command has to be issued to the drive before a read
21  * command, if the previous command was a write.
22  *
23  * The purpose of the packet writing driver is to hide these restrictions from
24  * higher layers, such as file systems, and present a block device that can be
25  * randomly read and written using 2kB-sized blocks.
26  *
27  * The lowest layer in the packet writing driver is the packet I/O scheduler.
28  * Its data is defined by the struct packet_iosched and includes two bio
29  * queues with pending read and write requests. These queues are processed
30  * by the pkt_iosched_process_queue() function. The write requests in this
31  * queue are already properly aligned and sized. This layer is responsible for
32  * issuing the flush cache commands and scheduling the I/O in a good order.
33  *
34  * The next layer transforms unaligned write requests to aligned writes. This
35  * transformation requires reading missing pieces of data from the underlying
36  * block device, assembling the pieces to full packets and queuing them to the
37  * packet I/O scheduler.
38  *
39  * At the top layer there is a custom make_request_fn function that forwards
40  * read requests directly to the iosched queue and puts write requests in the
41  * unaligned write queue. A kernel thread performs the necessary read
42  * gathering to convert the unaligned writes to aligned writes and then feeds
43  * them to the packet I/O scheduler.
44  *
45  *************************************************************************/
46
47 #include <linux/pktcdvd.h>
48 #include <linux/module.h>
49 #include <linux/types.h>
50 #include <linux/kernel.h>
51 #include <linux/kthread.h>
52 #include <linux/errno.h>
53 #include <linux/spinlock.h>
54 #include <linux/file.h>
55 #include <linux/proc_fs.h>
56 #include <linux/seq_file.h>
57 #include <linux/miscdevice.h>
58 #include <linux/freezer.h>
59 #include <linux/mutex.h>
60 #include <scsi/scsi_cmnd.h>
61 #include <scsi/scsi_ioctl.h>
62 #include <scsi/scsi.h>
63 #include <linux/debugfs.h>
64 #include <linux/device.h>
65
66 #include <asm/uaccess.h>
67
68 #define DRIVER_NAME     "pktcdvd"
69
70 #if PACKET_DEBUG
71 #define DPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
72 #else
73 #define DPRINTK(fmt, args...)
74 #endif
75
76 #if PACKET_DEBUG > 1
77 #define VPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
78 #else
79 #define VPRINTK(fmt, args...)
80 #endif
81
82 #define MAX_SPEED 0xffff
83
84 #define ZONE(sector, pd) (((sector) + (pd)->offset) & ~((pd)->settings.size - 1))
85
86 static struct pktcdvd_device *pkt_devs[MAX_WRITERS];
87 static struct proc_dir_entry *pkt_proc;
88 static int pktdev_major;
89 static int write_congestion_on  = PKT_WRITE_CONGESTION_ON;
90 static int write_congestion_off = PKT_WRITE_CONGESTION_OFF;
91 static struct mutex ctl_mutex;  /* Serialize open/close/setup/teardown */
92 static mempool_t *psd_pool;
93
94 static struct class     *class_pktcdvd = NULL;    /* /sys/class/pktcdvd */
95 static struct dentry    *pkt_debugfs_root = NULL; /* /debug/pktcdvd */
96
97 /* forward declaration */
98 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev);
99 static int pkt_remove_dev(dev_t pkt_dev);
100 static int pkt_seq_show(struct seq_file *m, void *p);
101
102
103
104 /*
105  * create and register a pktcdvd kernel object.
106  */
107 static struct pktcdvd_kobj* pkt_kobj_create(struct pktcdvd_device *pd,
108                                         const char* name,
109                                         struct kobject* parent,
110                                         struct kobj_type* ktype)
111 {
112         struct pktcdvd_kobj *p;
113         int error;
114
115         p = kzalloc(sizeof(*p), GFP_KERNEL);
116         if (!p)
117                 return NULL;
118         p->pd = pd;
119         error = kobject_init_and_add(&p->kobj, ktype, parent, "%s", name);
120         if (error) {
121                 kobject_put(&p->kobj);
122                 return NULL;
123         }
124         kobject_uevent(&p->kobj, KOBJ_ADD);
125         return p;
126 }
127 /*
128  * remove a pktcdvd kernel object.
129  */
130 static void pkt_kobj_remove(struct pktcdvd_kobj *p)
131 {
132         if (p)
133                 kobject_put(&p->kobj);
134 }
135 /*
136  * default release function for pktcdvd kernel objects.
137  */
138 static void pkt_kobj_release(struct kobject *kobj)
139 {
140         kfree(to_pktcdvdkobj(kobj));
141 }
142
143
144 /**********************************************************
145  *
146  * sysfs interface for pktcdvd
147  * by (C) 2006  Thomas Maier <balagi@justmail.de>
148  *
149  **********************************************************/
150
151 #define DEF_ATTR(_obj,_name,_mode) \
152         static struct attribute _obj = { .name = _name, .mode = _mode }
153
154 /**********************************************************
155   /sys/class/pktcdvd/pktcdvd[0-7]/
156                      stat/reset
157                      stat/packets_started
158                      stat/packets_finished
159                      stat/kb_written
160                      stat/kb_read
161                      stat/kb_read_gather
162                      write_queue/size
163                      write_queue/congestion_off
164                      write_queue/congestion_on
165  **********************************************************/
166
167 DEF_ATTR(kobj_pkt_attr_st1, "reset", 0200);
168 DEF_ATTR(kobj_pkt_attr_st2, "packets_started", 0444);
169 DEF_ATTR(kobj_pkt_attr_st3, "packets_finished", 0444);
170 DEF_ATTR(kobj_pkt_attr_st4, "kb_written", 0444);
171 DEF_ATTR(kobj_pkt_attr_st5, "kb_read", 0444);
172 DEF_ATTR(kobj_pkt_attr_st6, "kb_read_gather", 0444);
173
174 static struct attribute *kobj_pkt_attrs_stat[] = {
175         &kobj_pkt_attr_st1,
176         &kobj_pkt_attr_st2,
177         &kobj_pkt_attr_st3,
178         &kobj_pkt_attr_st4,
179         &kobj_pkt_attr_st5,
180         &kobj_pkt_attr_st6,
181         NULL
182 };
183
184 DEF_ATTR(kobj_pkt_attr_wq1, "size", 0444);
185 DEF_ATTR(kobj_pkt_attr_wq2, "congestion_off", 0644);
186 DEF_ATTR(kobj_pkt_attr_wq3, "congestion_on",  0644);
187
188 static struct attribute *kobj_pkt_attrs_wqueue[] = {
189         &kobj_pkt_attr_wq1,
190         &kobj_pkt_attr_wq2,
191         &kobj_pkt_attr_wq3,
192         NULL
193 };
194
195 static ssize_t kobj_pkt_show(struct kobject *kobj,
196                         struct attribute *attr, char *data)
197 {
198         struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
199         int n = 0;
200         int v;
201         if (strcmp(attr->name, "packets_started") == 0) {
202                 n = sprintf(data, "%lu\n", pd->stats.pkt_started);
203
204         } else if (strcmp(attr->name, "packets_finished") == 0) {
205                 n = sprintf(data, "%lu\n", pd->stats.pkt_ended);
206
207         } else if (strcmp(attr->name, "kb_written") == 0) {
208                 n = sprintf(data, "%lu\n", pd->stats.secs_w >> 1);
209
210         } else if (strcmp(attr->name, "kb_read") == 0) {
211                 n = sprintf(data, "%lu\n", pd->stats.secs_r >> 1);
212
213         } else if (strcmp(attr->name, "kb_read_gather") == 0) {
214                 n = sprintf(data, "%lu\n", pd->stats.secs_rg >> 1);
215
216         } else if (strcmp(attr->name, "size") == 0) {
217                 spin_lock(&pd->lock);
218                 v = pd->bio_queue_size;
219                 spin_unlock(&pd->lock);
220                 n = sprintf(data, "%d\n", v);
221
222         } else if (strcmp(attr->name, "congestion_off") == 0) {
223                 spin_lock(&pd->lock);
224                 v = pd->write_congestion_off;
225                 spin_unlock(&pd->lock);
226                 n = sprintf(data, "%d\n", v);
227
228         } else if (strcmp(attr->name, "congestion_on") == 0) {
229                 spin_lock(&pd->lock);
230                 v = pd->write_congestion_on;
231                 spin_unlock(&pd->lock);
232                 n = sprintf(data, "%d\n", v);
233         }
234         return n;
235 }
236
237 static void init_write_congestion_marks(int* lo, int* hi)
238 {
239         if (*hi > 0) {
240                 *hi = max(*hi, 500);
241                 *hi = min(*hi, 1000000);
242                 if (*lo <= 0)
243                         *lo = *hi - 100;
244                 else {
245                         *lo = min(*lo, *hi - 100);
246                         *lo = max(*lo, 100);
247                 }
248         } else {
249                 *hi = -1;
250                 *lo = -1;
251         }
252 }
253
254 static ssize_t kobj_pkt_store(struct kobject *kobj,
255                         struct attribute *attr,
256                         const char *data, size_t len)
257 {
258         struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
259         int val;
260
261         if (strcmp(attr->name, "reset") == 0 && len > 0) {
262                 pd->stats.pkt_started = 0;
263                 pd->stats.pkt_ended = 0;
264                 pd->stats.secs_w = 0;
265                 pd->stats.secs_rg = 0;
266                 pd->stats.secs_r = 0;
267
268         } else if (strcmp(attr->name, "congestion_off") == 0
269                    && sscanf(data, "%d", &val) == 1) {
270                 spin_lock(&pd->lock);
271                 pd->write_congestion_off = val;
272                 init_write_congestion_marks(&pd->write_congestion_off,
273                                         &pd->write_congestion_on);
274                 spin_unlock(&pd->lock);
275
276         } else if (strcmp(attr->name, "congestion_on") == 0
277                    && sscanf(data, "%d", &val) == 1) {
278                 spin_lock(&pd->lock);
279                 pd->write_congestion_on = val;
280                 init_write_congestion_marks(&pd->write_congestion_off,
281                                         &pd->write_congestion_on);
282                 spin_unlock(&pd->lock);
283         }
284         return len;
285 }
286
287 static struct sysfs_ops kobj_pkt_ops = {
288         .show = kobj_pkt_show,
289         .store = kobj_pkt_store
290 };
291 static struct kobj_type kobj_pkt_type_stat = {
292         .release = pkt_kobj_release,
293         .sysfs_ops = &kobj_pkt_ops,
294         .default_attrs = kobj_pkt_attrs_stat
295 };
296 static struct kobj_type kobj_pkt_type_wqueue = {
297         .release = pkt_kobj_release,
298         .sysfs_ops = &kobj_pkt_ops,
299         .default_attrs = kobj_pkt_attrs_wqueue
300 };
301
302 static void pkt_sysfs_dev_new(struct pktcdvd_device *pd)
303 {
304         if (class_pktcdvd) {
305                 pd->dev = device_create(class_pktcdvd, NULL, 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
995             <= queue_max_phys_segments(q)) {
996                 /*
997                  * The cdrom device can handle one segment/frame
998                  */
999                 clear_bit(PACKET_MERGE_SEGS, &pd->flags);
1000                 return 0;
1001         } else if ((pd->settings.size << 9) / PAGE_SIZE
1002                    <= queue_max_phys_segments(q)) {
1003                 /*
1004                  * We can handle this case at the expense of some extra memory
1005                  * copies during write operations
1006                  */
1007                 set_bit(PACKET_MERGE_SEGS, &pd->flags);
1008                 return 0;
1009         } else {
1010                 printk(DRIVER_NAME": cdrom max_phys_segments too small\n");
1011                 return -EIO;
1012         }
1013 }
1014
1015 /*
1016  * Copy CD_FRAMESIZE bytes from src_bio into a destination page
1017  */
1018 static void pkt_copy_bio_data(struct bio *src_bio, int seg, int offs, struct page *dst_page, int dst_offs)
1019 {
1020         unsigned int copy_size = CD_FRAMESIZE;
1021
1022         while (copy_size > 0) {
1023                 struct bio_vec *src_bvl = bio_iovec_idx(src_bio, seg);
1024                 void *vfrom = kmap_atomic(src_bvl->bv_page, KM_USER0) +
1025                         src_bvl->bv_offset + offs;
1026                 void *vto = page_address(dst_page) + dst_offs;
1027                 int len = min_t(int, copy_size, src_bvl->bv_len - offs);
1028
1029                 BUG_ON(len < 0);
1030                 memcpy(vto, vfrom, len);
1031                 kunmap_atomic(vfrom, KM_USER0);
1032
1033                 seg++;
1034                 offs = 0;
1035                 dst_offs += len;
1036                 copy_size -= len;
1037         }
1038 }
1039
1040 /*
1041  * Copy all data for this packet to pkt->pages[], so that
1042  * a) The number of required segments for the write bio is minimized, which
1043  *    is necessary for some scsi controllers.
1044  * b) The data can be used as cache to avoid read requests if we receive a
1045  *    new write request for the same zone.
1046  */
1047 static void pkt_make_local_copy(struct packet_data *pkt, struct bio_vec *bvec)
1048 {
1049         int f, p, offs;
1050
1051         /* Copy all data to pkt->pages[] */
1052         p = 0;
1053         offs = 0;
1054         for (f = 0; f < pkt->frames; f++) {
1055                 if (bvec[f].bv_page != pkt->pages[p]) {
1056                         void *vfrom = kmap_atomic(bvec[f].bv_page, KM_USER0) + bvec[f].bv_offset;
1057                         void *vto = page_address(pkt->pages[p]) + offs;
1058                         memcpy(vto, vfrom, CD_FRAMESIZE);
1059                         kunmap_atomic(vfrom, KM_USER0);
1060                         bvec[f].bv_page = pkt->pages[p];
1061                         bvec[f].bv_offset = offs;
1062                 } else {
1063                         BUG_ON(bvec[f].bv_offset != offs);
1064                 }
1065                 offs += CD_FRAMESIZE;
1066                 if (offs >= PAGE_SIZE) {
1067                         offs = 0;
1068                         p++;
1069                 }
1070         }
1071 }
1072
1073 static void pkt_end_io_read(struct bio *bio, int err)
1074 {
1075         struct packet_data *pkt = bio->bi_private;
1076         struct pktcdvd_device *pd = pkt->pd;
1077         BUG_ON(!pd);
1078
1079         VPRINTK("pkt_end_io_read: bio=%p sec0=%llx sec=%llx err=%d\n", bio,
1080                 (unsigned long long)pkt->sector, (unsigned long long)bio->bi_sector, err);
1081
1082         if (err)
1083                 atomic_inc(&pkt->io_errors);
1084         if (atomic_dec_and_test(&pkt->io_wait)) {
1085                 atomic_inc(&pkt->run_sm);
1086                 wake_up(&pd->wqueue);
1087         }
1088         pkt_bio_finished(pd);
1089 }
1090
1091 static void pkt_end_io_packet_write(struct bio *bio, int err)
1092 {
1093         struct packet_data *pkt = bio->bi_private;
1094         struct pktcdvd_device *pd = pkt->pd;
1095         BUG_ON(!pd);
1096
1097         VPRINTK("pkt_end_io_packet_write: id=%d, err=%d\n", pkt->id, err);
1098
1099         pd->stats.pkt_ended++;
1100
1101         pkt_bio_finished(pd);
1102         atomic_dec(&pkt->io_wait);
1103         atomic_inc(&pkt->run_sm);
1104         wake_up(&pd->wqueue);
1105 }
1106
1107 /*
1108  * Schedule reads for the holes in a packet
1109  */
1110 static void pkt_gather_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1111 {
1112         int frames_read = 0;
1113         struct bio *bio;
1114         int f;
1115         char written[PACKET_MAX_SIZE];
1116
1117         BUG_ON(!pkt->orig_bios);
1118
1119         atomic_set(&pkt->io_wait, 0);
1120         atomic_set(&pkt->io_errors, 0);
1121
1122         /*
1123          * Figure out which frames we need to read before we can write.
1124          */
1125         memset(written, 0, sizeof(written));
1126         spin_lock(&pkt->lock);
1127         for (bio = pkt->orig_bios; bio; bio = bio->bi_next) {
1128                 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1129                 int num_frames = bio->bi_size / CD_FRAMESIZE;
1130                 pd->stats.secs_w += num_frames * (CD_FRAMESIZE >> 9);
1131                 BUG_ON(first_frame < 0);
1132                 BUG_ON(first_frame + num_frames > pkt->frames);
1133                 for (f = first_frame; f < first_frame + num_frames; f++)
1134                         written[f] = 1;
1135         }
1136         spin_unlock(&pkt->lock);
1137
1138         if (pkt->cache_valid) {
1139                 VPRINTK("pkt_gather_data: zone %llx cached\n",
1140                         (unsigned long long)pkt->sector);
1141                 goto out_account;
1142         }
1143
1144         /*
1145          * Schedule reads for missing parts of the packet.
1146          */
1147         for (f = 0; f < pkt->frames; f++) {
1148                 struct bio_vec *vec;
1149
1150                 int p, offset;
1151                 if (written[f])
1152                         continue;
1153                 bio = pkt->r_bios[f];
1154                 vec = bio->bi_io_vec;
1155                 bio_init(bio);
1156                 bio->bi_max_vecs = 1;
1157                 bio->bi_sector = pkt->sector + f * (CD_FRAMESIZE >> 9);
1158                 bio->bi_bdev = pd->bdev;
1159                 bio->bi_end_io = pkt_end_io_read;
1160                 bio->bi_private = pkt;
1161                 bio->bi_io_vec = vec;
1162                 bio->bi_destructor = pkt_bio_destructor;
1163
1164                 p = (f * CD_FRAMESIZE) / PAGE_SIZE;
1165                 offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1166                 VPRINTK("pkt_gather_data: Adding frame %d, page:%p offs:%d\n",
1167                         f, pkt->pages[p], offset);
1168                 if (!bio_add_page(bio, pkt->pages[p], CD_FRAMESIZE, offset))
1169                         BUG();
1170
1171                 atomic_inc(&pkt->io_wait);
1172                 bio->bi_rw = READ;
1173                 pkt_queue_bio(pd, bio);
1174                 frames_read++;
1175         }
1176
1177 out_account:
1178         VPRINTK("pkt_gather_data: need %d frames for zone %llx\n",
1179                 frames_read, (unsigned long long)pkt->sector);
1180         pd->stats.pkt_started++;
1181         pd->stats.secs_rg += frames_read * (CD_FRAMESIZE >> 9);
1182 }
1183
1184 /*
1185  * Find a packet matching zone, or the least recently used packet if
1186  * there is no match.
1187  */
1188 static struct packet_data *pkt_get_packet_data(struct pktcdvd_device *pd, int zone)
1189 {
1190         struct packet_data *pkt;
1191
1192         list_for_each_entry(pkt, &pd->cdrw.pkt_free_list, list) {
1193                 if (pkt->sector == zone || pkt->list.next == &pd->cdrw.pkt_free_list) {
1194                         list_del_init(&pkt->list);
1195                         if (pkt->sector != zone)
1196                                 pkt->cache_valid = 0;
1197                         return pkt;
1198                 }
1199         }
1200         BUG();
1201         return NULL;
1202 }
1203
1204 static void pkt_put_packet_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1205 {
1206         if (pkt->cache_valid) {
1207                 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
1208         } else {
1209                 list_add_tail(&pkt->list, &pd->cdrw.pkt_free_list);
1210         }
1211 }
1212
1213 /*
1214  * recover a failed write, query for relocation if possible
1215  *
1216  * returns 1 if recovery is possible, or 0 if not
1217  *
1218  */
1219 static int pkt_start_recovery(struct packet_data *pkt)
1220 {
1221         /*
1222          * FIXME. We need help from the file system to implement
1223          * recovery handling.
1224          */
1225         return 0;
1226 #if 0
1227         struct request *rq = pkt->rq;
1228         struct pktcdvd_device *pd = rq->rq_disk->private_data;
1229         struct block_device *pkt_bdev;
1230         struct super_block *sb = NULL;
1231         unsigned long old_block, new_block;
1232         sector_t new_sector;
1233
1234         pkt_bdev = bdget(kdev_t_to_nr(pd->pkt_dev));
1235         if (pkt_bdev) {
1236                 sb = get_super(pkt_bdev);
1237                 bdput(pkt_bdev);
1238         }
1239
1240         if (!sb)
1241                 return 0;
1242
1243         if (!sb->s_op || !sb->s_op->relocate_blocks)
1244                 goto out;
1245
1246         old_block = pkt->sector / (CD_FRAMESIZE >> 9);
1247         if (sb->s_op->relocate_blocks(sb, old_block, &new_block))
1248                 goto out;
1249
1250         new_sector = new_block * (CD_FRAMESIZE >> 9);
1251         pkt->sector = new_sector;
1252
1253         pkt->bio->bi_sector = new_sector;
1254         pkt->bio->bi_next = NULL;
1255         pkt->bio->bi_flags = 1 << BIO_UPTODATE;
1256         pkt->bio->bi_idx = 0;
1257
1258         BUG_ON(pkt->bio->bi_rw != (1 << BIO_RW));
1259         BUG_ON(pkt->bio->bi_vcnt != pkt->frames);
1260         BUG_ON(pkt->bio->bi_size != pkt->frames * CD_FRAMESIZE);
1261         BUG_ON(pkt->bio->bi_end_io != pkt_end_io_packet_write);
1262         BUG_ON(pkt->bio->bi_private != pkt);
1263
1264         drop_super(sb);
1265         return 1;
1266
1267 out:
1268         drop_super(sb);
1269         return 0;
1270 #endif
1271 }
1272
1273 static inline void pkt_set_state(struct packet_data *pkt, enum packet_data_state state)
1274 {
1275 #if PACKET_DEBUG > 1
1276         static const char *state_name[] = {
1277                 "IDLE", "WAITING", "READ_WAIT", "WRITE_WAIT", "RECOVERY", "FINISHED"
1278         };
1279         enum packet_data_state old_state = pkt->state;
1280         VPRINTK("pkt %2d : s=%6llx %s -> %s\n", pkt->id, (unsigned long long)pkt->sector,
1281                 state_name[old_state], state_name[state]);
1282 #endif
1283         pkt->state = state;
1284 }
1285
1286 /*
1287  * Scan the work queue to see if we can start a new packet.
1288  * returns non-zero if any work was done.
1289  */
1290 static int pkt_handle_queue(struct pktcdvd_device *pd)
1291 {
1292         struct packet_data *pkt, *p;
1293         struct bio *bio = NULL;
1294         sector_t zone = 0; /* Suppress gcc warning */
1295         struct pkt_rb_node *node, *first_node;
1296         struct rb_node *n;
1297         int wakeup;
1298
1299         VPRINTK("handle_queue\n");
1300
1301         atomic_set(&pd->scan_queue, 0);
1302
1303         if (list_empty(&pd->cdrw.pkt_free_list)) {
1304                 VPRINTK("handle_queue: no pkt\n");
1305                 return 0;
1306         }
1307
1308         /*
1309          * Try to find a zone we are not already working on.
1310          */
1311         spin_lock(&pd->lock);
1312         first_node = pkt_rbtree_find(pd, pd->current_sector);
1313         if (!first_node) {
1314                 n = rb_first(&pd->bio_queue);
1315                 if (n)
1316                         first_node = rb_entry(n, struct pkt_rb_node, rb_node);
1317         }
1318         node = first_node;
1319         while (node) {
1320                 bio = node->bio;
1321                 zone = ZONE(bio->bi_sector, pd);
1322                 list_for_each_entry(p, &pd->cdrw.pkt_active_list, list) {
1323                         if (p->sector == zone) {
1324                                 bio = NULL;
1325                                 goto try_next_bio;
1326                         }
1327                 }
1328                 break;
1329 try_next_bio:
1330                 node = pkt_rbtree_next(node);
1331                 if (!node) {
1332                         n = rb_first(&pd->bio_queue);
1333                         if (n)
1334                                 node = rb_entry(n, struct pkt_rb_node, rb_node);
1335                 }
1336                 if (node == first_node)
1337                         node = NULL;
1338         }
1339         spin_unlock(&pd->lock);
1340         if (!bio) {
1341                 VPRINTK("handle_queue: no bio\n");
1342                 return 0;
1343         }
1344
1345         pkt = pkt_get_packet_data(pd, zone);
1346
1347         pd->current_sector = zone + pd->settings.size;
1348         pkt->sector = zone;
1349         BUG_ON(pkt->frames != pd->settings.size >> 2);
1350         pkt->write_size = 0;
1351
1352         /*
1353          * Scan work queue for bios in the same zone and link them
1354          * to this packet.
1355          */
1356         spin_lock(&pd->lock);
1357         VPRINTK("pkt_handle_queue: looking for zone %llx\n", (unsigned long long)zone);
1358         while ((node = pkt_rbtree_find(pd, zone)) != NULL) {
1359                 bio = node->bio;
1360                 VPRINTK("pkt_handle_queue: found zone=%llx\n",
1361                         (unsigned long long)ZONE(bio->bi_sector, pd));
1362                 if (ZONE(bio->bi_sector, pd) != zone)
1363                         break;
1364                 pkt_rbtree_erase(pd, node);
1365                 spin_lock(&pkt->lock);
1366                 pkt_add_list_last(bio, &pkt->orig_bios, &pkt->orig_bios_tail);
1367                 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
1368                 spin_unlock(&pkt->lock);
1369         }
1370         /* check write congestion marks, and if bio_queue_size is
1371            below, wake up any waiters */
1372         wakeup = (pd->write_congestion_on > 0
1373                         && pd->bio_queue_size <= pd->write_congestion_off);
1374         spin_unlock(&pd->lock);
1375         if (wakeup)
1376                 clear_bdi_congested(&pd->disk->queue->backing_dev_info, WRITE);
1377
1378         pkt->sleep_time = max(PACKET_WAIT_TIME, 1);
1379         pkt_set_state(pkt, PACKET_WAITING_STATE);
1380         atomic_set(&pkt->run_sm, 1);
1381
1382         spin_lock(&pd->cdrw.active_list_lock);
1383         list_add(&pkt->list, &pd->cdrw.pkt_active_list);
1384         spin_unlock(&pd->cdrw.active_list_lock);
1385
1386         return 1;
1387 }
1388
1389 /*
1390  * Assemble a bio to write one packet and queue the bio for processing
1391  * by the underlying block device.
1392  */
1393 static void pkt_start_write(struct pktcdvd_device *pd, struct packet_data *pkt)
1394 {
1395         struct bio *bio;
1396         int f;
1397         int frames_write;
1398         struct bio_vec *bvec = pkt->w_bio->bi_io_vec;
1399
1400         for (f = 0; f < pkt->frames; f++) {
1401                 bvec[f].bv_page = pkt->pages[(f * CD_FRAMESIZE) / PAGE_SIZE];
1402                 bvec[f].bv_offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1403         }
1404
1405         /*
1406          * Fill-in bvec with data from orig_bios.
1407          */
1408         frames_write = 0;
1409         spin_lock(&pkt->lock);
1410         for (bio = pkt->orig_bios; bio; bio = bio->bi_next) {
1411                 int segment = bio->bi_idx;
1412                 int src_offs = 0;
1413                 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1414                 int num_frames = bio->bi_size / CD_FRAMESIZE;
1415                 BUG_ON(first_frame < 0);
1416                 BUG_ON(first_frame + num_frames > pkt->frames);
1417                 for (f = first_frame; f < first_frame + num_frames; f++) {
1418                         struct bio_vec *src_bvl = bio_iovec_idx(bio, segment);
1419
1420                         while (src_offs >= src_bvl->bv_len) {
1421                                 src_offs -= src_bvl->bv_len;
1422                                 segment++;
1423                                 BUG_ON(segment >= bio->bi_vcnt);
1424                                 src_bvl = bio_iovec_idx(bio, segment);
1425                         }
1426
1427                         if (src_bvl->bv_len - src_offs >= CD_FRAMESIZE) {
1428                                 bvec[f].bv_page = src_bvl->bv_page;
1429                                 bvec[f].bv_offset = src_bvl->bv_offset + src_offs;
1430                         } else {
1431                                 pkt_copy_bio_data(bio, segment, src_offs,
1432                                                   bvec[f].bv_page, bvec[f].bv_offset);
1433                         }
1434                         src_offs += CD_FRAMESIZE;
1435                         frames_write++;
1436                 }
1437         }
1438         pkt_set_state(pkt, PACKET_WRITE_WAIT_STATE);
1439         spin_unlock(&pkt->lock);
1440
1441         VPRINTK("pkt_start_write: Writing %d frames for zone %llx\n",
1442                 frames_write, (unsigned long long)pkt->sector);
1443         BUG_ON(frames_write != pkt->write_size);
1444
1445         if (test_bit(PACKET_MERGE_SEGS, &pd->flags) || (pkt->write_size < pkt->frames)) {
1446                 pkt_make_local_copy(pkt, bvec);
1447                 pkt->cache_valid = 1;
1448         } else {
1449                 pkt->cache_valid = 0;
1450         }
1451
1452         /* Start the write request */
1453         bio_init(pkt->w_bio);
1454         pkt->w_bio->bi_max_vecs = PACKET_MAX_SIZE;
1455         pkt->w_bio->bi_sector = pkt->sector;
1456         pkt->w_bio->bi_bdev = pd->bdev;
1457         pkt->w_bio->bi_end_io = pkt_end_io_packet_write;
1458         pkt->w_bio->bi_private = pkt;
1459         pkt->w_bio->bi_io_vec = bvec;
1460         pkt->w_bio->bi_destructor = pkt_bio_destructor;
1461         for (f = 0; f < pkt->frames; f++)
1462                 if (!bio_add_page(pkt->w_bio, bvec[f].bv_page, CD_FRAMESIZE, bvec[f].bv_offset))
1463                         BUG();
1464         VPRINTK(DRIVER_NAME": vcnt=%d\n", pkt->w_bio->bi_vcnt);
1465
1466         atomic_set(&pkt->io_wait, 1);
1467         pkt->w_bio->bi_rw = WRITE;
1468         pkt_queue_bio(pd, pkt->w_bio);
1469 }
1470
1471 static void pkt_finish_packet(struct packet_data *pkt, int uptodate)
1472 {
1473         struct bio *bio, *next;
1474
1475         if (!uptodate)
1476                 pkt->cache_valid = 0;
1477
1478         /* Finish all bios corresponding to this packet */
1479         bio = pkt->orig_bios;
1480         while (bio) {
1481                 next = bio->bi_next;
1482                 bio->bi_next = NULL;
1483                 bio_endio(bio, uptodate ? 0 : -EIO);
1484                 bio = next;
1485         }
1486         pkt->orig_bios = pkt->orig_bios_tail = NULL;
1487 }
1488
1489 static void pkt_run_state_machine(struct pktcdvd_device *pd, struct packet_data *pkt)
1490 {
1491         int uptodate;
1492
1493         VPRINTK("run_state_machine: pkt %d\n", pkt->id);
1494
1495         for (;;) {
1496                 switch (pkt->state) {
1497                 case PACKET_WAITING_STATE:
1498                         if ((pkt->write_size < pkt->frames) && (pkt->sleep_time > 0))
1499                                 return;
1500
1501                         pkt->sleep_time = 0;
1502                         pkt_gather_data(pd, pkt);
1503                         pkt_set_state(pkt, PACKET_READ_WAIT_STATE);
1504                         break;
1505
1506                 case PACKET_READ_WAIT_STATE:
1507                         if (atomic_read(&pkt->io_wait) > 0)
1508                                 return;
1509
1510                         if (atomic_read(&pkt->io_errors) > 0) {
1511                                 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1512                         } else {
1513                                 pkt_start_write(pd, pkt);
1514                         }
1515                         break;
1516
1517                 case PACKET_WRITE_WAIT_STATE:
1518                         if (atomic_read(&pkt->io_wait) > 0)
1519                                 return;
1520
1521                         if (test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags)) {
1522                                 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1523                         } else {
1524                                 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1525                         }
1526                         break;
1527
1528                 case PACKET_RECOVERY_STATE:
1529                         if (pkt_start_recovery(pkt)) {
1530                                 pkt_start_write(pd, pkt);
1531                         } else {
1532                                 VPRINTK("No recovery possible\n");
1533                                 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1534                         }
1535                         break;
1536
1537                 case PACKET_FINISHED_STATE:
1538                         uptodate = test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags);
1539                         pkt_finish_packet(pkt, uptodate);
1540                         return;
1541
1542                 default:
1543                         BUG();
1544                         break;
1545                 }
1546         }
1547 }
1548
1549 static void pkt_handle_packets(struct pktcdvd_device *pd)
1550 {
1551         struct packet_data *pkt, *next;
1552
1553         VPRINTK("pkt_handle_packets\n");
1554
1555         /*
1556          * Run state machine for active packets
1557          */
1558         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1559                 if (atomic_read(&pkt->run_sm) > 0) {
1560                         atomic_set(&pkt->run_sm, 0);
1561                         pkt_run_state_machine(pd, pkt);
1562                 }
1563         }
1564
1565         /*
1566          * Move no longer active packets to the free list
1567          */
1568         spin_lock(&pd->cdrw.active_list_lock);
1569         list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_active_list, list) {
1570                 if (pkt->state == PACKET_FINISHED_STATE) {
1571                         list_del(&pkt->list);
1572                         pkt_put_packet_data(pd, pkt);
1573                         pkt_set_state(pkt, PACKET_IDLE_STATE);
1574                         atomic_set(&pd->scan_queue, 1);
1575                 }
1576         }
1577         spin_unlock(&pd->cdrw.active_list_lock);
1578 }
1579
1580 static void pkt_count_states(struct pktcdvd_device *pd, int *states)
1581 {
1582         struct packet_data *pkt;
1583         int i;
1584
1585         for (i = 0; i < PACKET_NUM_STATES; i++)
1586                 states[i] = 0;
1587
1588         spin_lock(&pd->cdrw.active_list_lock);
1589         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1590                 states[pkt->state]++;
1591         }
1592         spin_unlock(&pd->cdrw.active_list_lock);
1593 }
1594
1595 /*
1596  * kcdrwd is woken up when writes have been queued for one of our
1597  * registered devices
1598  */
1599 static int kcdrwd(void *foobar)
1600 {
1601         struct pktcdvd_device *pd = foobar;
1602         struct packet_data *pkt;
1603         long min_sleep_time, residue;
1604
1605         set_user_nice(current, -20);
1606         set_freezable();
1607
1608         for (;;) {
1609                 DECLARE_WAITQUEUE(wait, current);
1610
1611                 /*
1612                  * Wait until there is something to do
1613                  */
1614                 add_wait_queue(&pd->wqueue, &wait);
1615                 for (;;) {
1616                         set_current_state(TASK_INTERRUPTIBLE);
1617
1618                         /* Check if we need to run pkt_handle_queue */
1619                         if (atomic_read(&pd->scan_queue) > 0)
1620                                 goto work_to_do;
1621
1622                         /* Check if we need to run the state machine for some packet */
1623                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1624                                 if (atomic_read(&pkt->run_sm) > 0)
1625                                         goto work_to_do;
1626                         }
1627
1628                         /* Check if we need to process the iosched queues */
1629                         if (atomic_read(&pd->iosched.attention) != 0)
1630                                 goto work_to_do;
1631
1632                         /* Otherwise, go to sleep */
1633                         if (PACKET_DEBUG > 1) {
1634                                 int states[PACKET_NUM_STATES];
1635                                 pkt_count_states(pd, states);
1636                                 VPRINTK("kcdrwd: i:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
1637                                         states[0], states[1], states[2], states[3],
1638                                         states[4], states[5]);
1639                         }
1640
1641                         min_sleep_time = MAX_SCHEDULE_TIMEOUT;
1642                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1643                                 if (pkt->sleep_time && pkt->sleep_time < min_sleep_time)
1644                                         min_sleep_time = pkt->sleep_time;
1645                         }
1646
1647                         generic_unplug_device(bdev_get_queue(pd->bdev));
1648
1649                         VPRINTK("kcdrwd: sleeping\n");
1650                         residue = schedule_timeout(min_sleep_time);
1651                         VPRINTK("kcdrwd: wake up\n");
1652
1653                         /* make swsusp happy with our thread */
1654                         try_to_freeze();
1655
1656                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1657                                 if (!pkt->sleep_time)
1658                                         continue;
1659                                 pkt->sleep_time -= min_sleep_time - residue;
1660                                 if (pkt->sleep_time <= 0) {
1661                                         pkt->sleep_time = 0;
1662                                         atomic_inc(&pkt->run_sm);
1663                                 }
1664                         }
1665
1666                         if (kthread_should_stop())
1667                                 break;
1668                 }
1669 work_to_do:
1670                 set_current_state(TASK_RUNNING);
1671                 remove_wait_queue(&pd->wqueue, &wait);
1672
1673                 if (kthread_should_stop())
1674                         break;
1675
1676                 /*
1677                  * if pkt_handle_queue returns true, we can queue
1678                  * another request.
1679                  */
1680                 while (pkt_handle_queue(pd))
1681                         ;
1682
1683                 /*
1684                  * Handle packet state machine
1685                  */
1686                 pkt_handle_packets(pd);
1687
1688                 /*
1689                  * Handle iosched queues
1690                  */
1691                 pkt_iosched_process_queue(pd);
1692         }
1693
1694         return 0;
1695 }
1696
1697 static void pkt_print_settings(struct pktcdvd_device *pd)
1698 {
1699         printk(DRIVER_NAME": %s packets, ", pd->settings.fp ? "Fixed" : "Variable");
1700         printk("%u blocks, ", pd->settings.size >> 2);
1701         printk("Mode-%c disc\n", pd->settings.block_mode == 8 ? '1' : '2');
1702 }
1703
1704 static int pkt_mode_sense(struct pktcdvd_device *pd, struct packet_command *cgc, int page_code, int page_control)
1705 {
1706         memset(cgc->cmd, 0, sizeof(cgc->cmd));
1707
1708         cgc->cmd[0] = GPCMD_MODE_SENSE_10;
1709         cgc->cmd[2] = page_code | (page_control << 6);
1710         cgc->cmd[7] = cgc->buflen >> 8;
1711         cgc->cmd[8] = cgc->buflen & 0xff;
1712         cgc->data_direction = CGC_DATA_READ;
1713         return pkt_generic_packet(pd, cgc);
1714 }
1715
1716 static int pkt_mode_select(struct pktcdvd_device *pd, struct packet_command *cgc)
1717 {
1718         memset(cgc->cmd, 0, sizeof(cgc->cmd));
1719         memset(cgc->buffer, 0, 2);
1720         cgc->cmd[0] = GPCMD_MODE_SELECT_10;
1721         cgc->cmd[1] = 0x10;             /* PF */
1722         cgc->cmd[7] = cgc->buflen >> 8;
1723         cgc->cmd[8] = cgc->buflen & 0xff;
1724         cgc->data_direction = CGC_DATA_WRITE;
1725         return pkt_generic_packet(pd, cgc);
1726 }
1727
1728 static int pkt_get_disc_info(struct pktcdvd_device *pd, disc_information *di)
1729 {
1730         struct packet_command cgc;
1731         int ret;
1732
1733         /* set up command and get the disc info */
1734         init_cdrom_command(&cgc, di, sizeof(*di), CGC_DATA_READ);
1735         cgc.cmd[0] = GPCMD_READ_DISC_INFO;
1736         cgc.cmd[8] = cgc.buflen = 2;
1737         cgc.quiet = 1;
1738
1739         if ((ret = pkt_generic_packet(pd, &cgc)))
1740                 return ret;
1741
1742         /* not all drives have the same disc_info length, so requeue
1743          * packet with the length the drive tells us it can supply
1744          */
1745         cgc.buflen = be16_to_cpu(di->disc_information_length) +
1746                      sizeof(di->disc_information_length);
1747
1748         if (cgc.buflen > sizeof(disc_information))
1749                 cgc.buflen = sizeof(disc_information);
1750
1751         cgc.cmd[8] = cgc.buflen;
1752         return pkt_generic_packet(pd, &cgc);
1753 }
1754
1755 static int pkt_get_track_info(struct pktcdvd_device *pd, __u16 track, __u8 type, track_information *ti)
1756 {
1757         struct packet_command cgc;
1758         int ret;
1759
1760         init_cdrom_command(&cgc, ti, 8, CGC_DATA_READ);
1761         cgc.cmd[0] = GPCMD_READ_TRACK_RZONE_INFO;
1762         cgc.cmd[1] = type & 3;
1763         cgc.cmd[4] = (track & 0xff00) >> 8;
1764         cgc.cmd[5] = track & 0xff;
1765         cgc.cmd[8] = 8;
1766         cgc.quiet = 1;
1767
1768         if ((ret = pkt_generic_packet(pd, &cgc)))
1769                 return ret;
1770
1771         cgc.buflen = be16_to_cpu(ti->track_information_length) +
1772                      sizeof(ti->track_information_length);
1773
1774         if (cgc.buflen > sizeof(track_information))
1775                 cgc.buflen = sizeof(track_information);
1776
1777         cgc.cmd[8] = cgc.buflen;
1778         return pkt_generic_packet(pd, &cgc);
1779 }
1780
1781 static noinline_for_stack int pkt_get_last_written(struct pktcdvd_device *pd,
1782                                                 long *last_written)
1783 {
1784         disc_information di;
1785         track_information ti;
1786         __u32 last_track;
1787         int ret = -1;
1788
1789         if ((ret = pkt_get_disc_info(pd, &di)))
1790                 return ret;
1791
1792         last_track = (di.last_track_msb << 8) | di.last_track_lsb;
1793         if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1794                 return ret;
1795
1796         /* if this track is blank, try the previous. */
1797         if (ti.blank) {
1798                 last_track--;
1799                 if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1800                         return ret;
1801         }
1802
1803         /* if last recorded field is valid, return it. */
1804         if (ti.lra_v) {
1805                 *last_written = be32_to_cpu(ti.last_rec_address);
1806         } else {
1807                 /* make it up instead */
1808                 *last_written = be32_to_cpu(ti.track_start) +
1809                                 be32_to_cpu(ti.track_size);
1810                 if (ti.free_blocks)
1811                         *last_written -= (be32_to_cpu(ti.free_blocks) + 7);
1812         }
1813         return 0;
1814 }
1815
1816 /*
1817  * write mode select package based on pd->settings
1818  */
1819 static noinline_for_stack int pkt_set_write_settings(struct pktcdvd_device *pd)
1820 {
1821         struct packet_command cgc;
1822         struct request_sense sense;
1823         write_param_page *wp;
1824         char buffer[128];
1825         int ret, size;
1826
1827         /* doesn't apply to DVD+RW or DVD-RAM */
1828         if ((pd->mmc3_profile == 0x1a) || (pd->mmc3_profile == 0x12))
1829                 return 0;
1830
1831         memset(buffer, 0, sizeof(buffer));
1832         init_cdrom_command(&cgc, buffer, sizeof(*wp), CGC_DATA_READ);
1833         cgc.sense = &sense;
1834         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1835                 pkt_dump_sense(&cgc);
1836                 return ret;
1837         }
1838
1839         size = 2 + ((buffer[0] << 8) | (buffer[1] & 0xff));
1840         pd->mode_offset = (buffer[6] << 8) | (buffer[7] & 0xff);
1841         if (size > sizeof(buffer))
1842                 size = sizeof(buffer);
1843
1844         /*
1845          * now get it all
1846          */
1847         init_cdrom_command(&cgc, buffer, size, CGC_DATA_READ);
1848         cgc.sense = &sense;
1849         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1850                 pkt_dump_sense(&cgc);
1851                 return ret;
1852         }
1853
1854         /*
1855          * write page is offset header + block descriptor length
1856          */
1857         wp = (write_param_page *) &buffer[sizeof(struct mode_page_header) + pd->mode_offset];
1858
1859         wp->fp = pd->settings.fp;
1860         wp->track_mode = pd->settings.track_mode;
1861         wp->write_type = pd->settings.write_type;
1862         wp->data_block_type = pd->settings.block_mode;
1863
1864         wp->multi_session = 0;
1865
1866 #ifdef PACKET_USE_LS
1867         wp->link_size = 7;
1868         wp->ls_v = 1;
1869 #endif
1870
1871         if (wp->data_block_type == PACKET_BLOCK_MODE1) {
1872                 wp->session_format = 0;
1873                 wp->subhdr2 = 0x20;
1874         } else if (wp->data_block_type == PACKET_BLOCK_MODE2) {
1875                 wp->session_format = 0x20;
1876                 wp->subhdr2 = 8;
1877 #if 0
1878                 wp->mcn[0] = 0x80;
1879                 memcpy(&wp->mcn[1], PACKET_MCN, sizeof(wp->mcn) - 1);
1880 #endif
1881         } else {
1882                 /*
1883                  * paranoia
1884                  */
1885                 printk(DRIVER_NAME": write mode wrong %d\n", wp->data_block_type);
1886                 return 1;
1887         }
1888         wp->packet_size = cpu_to_be32(pd->settings.size >> 2);
1889
1890         cgc.buflen = cgc.cmd[8] = size;
1891         if ((ret = pkt_mode_select(pd, &cgc))) {
1892                 pkt_dump_sense(&cgc);
1893                 return ret;
1894         }
1895
1896         pkt_print_settings(pd);
1897         return 0;
1898 }
1899
1900 /*
1901  * 1 -- we can write to this track, 0 -- we can't
1902  */
1903 static int pkt_writable_track(struct pktcdvd_device *pd, track_information *ti)
1904 {
1905         switch (pd->mmc3_profile) {
1906                 case 0x1a: /* DVD+RW */
1907                 case 0x12: /* DVD-RAM */
1908                         /* The track is always writable on DVD+RW/DVD-RAM */
1909                         return 1;
1910                 default:
1911                         break;
1912         }
1913
1914         if (!ti->packet || !ti->fp)
1915                 return 0;
1916
1917         /*
1918          * "good" settings as per Mt Fuji.
1919          */
1920         if (ti->rt == 0 && ti->blank == 0)
1921                 return 1;
1922
1923         if (ti->rt == 0 && ti->blank == 1)
1924                 return 1;
1925
1926         if (ti->rt == 1 && ti->blank == 0)
1927                 return 1;
1928
1929         printk(DRIVER_NAME": bad state %d-%d-%d\n", ti->rt, ti->blank, ti->packet);
1930         return 0;
1931 }
1932
1933 /*
1934  * 1 -- we can write to this disc, 0 -- we can't
1935  */
1936 static int pkt_writable_disc(struct pktcdvd_device *pd, disc_information *di)
1937 {
1938         switch (pd->mmc3_profile) {
1939                 case 0x0a: /* CD-RW */
1940                 case 0xffff: /* MMC3 not supported */
1941                         break;
1942                 case 0x1a: /* DVD+RW */
1943                 case 0x13: /* DVD-RW */
1944                 case 0x12: /* DVD-RAM */
1945                         return 1;
1946                 default:
1947                         VPRINTK(DRIVER_NAME": Wrong disc profile (%x)\n", pd->mmc3_profile);
1948                         return 0;
1949         }
1950
1951         /*
1952          * for disc type 0xff we should probably reserve a new track.
1953          * but i'm not sure, should we leave this to user apps? probably.
1954          */
1955         if (di->disc_type == 0xff) {
1956                 printk(DRIVER_NAME": Unknown disc. No track?\n");
1957                 return 0;
1958         }
1959
1960         if (di->disc_type != 0x20 && di->disc_type != 0) {
1961                 printk(DRIVER_NAME": Wrong disc type (%x)\n", di->disc_type);
1962                 return 0;
1963         }
1964
1965         if (di->erasable == 0) {
1966                 printk(DRIVER_NAME": Disc not erasable\n");
1967                 return 0;
1968         }
1969
1970         if (di->border_status == PACKET_SESSION_RESERVED) {
1971                 printk(DRIVER_NAME": Can't write to last track (reserved)\n");
1972                 return 0;
1973         }
1974
1975         return 1;
1976 }
1977
1978 static noinline_for_stack int pkt_probe_settings(struct pktcdvd_device *pd)
1979 {
1980         struct packet_command cgc;
1981         unsigned char buf[12];
1982         disc_information di;
1983         track_information ti;
1984         int ret, track;
1985
1986         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
1987         cgc.cmd[0] = GPCMD_GET_CONFIGURATION;
1988         cgc.cmd[8] = 8;
1989         ret = pkt_generic_packet(pd, &cgc);
1990         pd->mmc3_profile = ret ? 0xffff : buf[6] << 8 | buf[7];
1991
1992         memset(&di, 0, sizeof(disc_information));
1993         memset(&ti, 0, sizeof(track_information));
1994
1995         if ((ret = pkt_get_disc_info(pd, &di))) {
1996                 printk("failed get_disc\n");
1997                 return ret;
1998         }
1999
2000         if (!pkt_writable_disc(pd, &di))
2001                 return -EROFS;
2002
2003         pd->type = di.erasable ? PACKET_CDRW : PACKET_CDR;
2004
2005         track = 1; /* (di.last_track_msb << 8) | di.last_track_lsb; */
2006         if ((ret = pkt_get_track_info(pd, track, 1, &ti))) {
2007                 printk(DRIVER_NAME": failed get_track\n");
2008                 return ret;
2009         }
2010
2011         if (!pkt_writable_track(pd, &ti)) {
2012                 printk(DRIVER_NAME": can't write to this track\n");
2013                 return -EROFS;
2014         }
2015
2016         /*
2017          * we keep packet size in 512 byte units, makes it easier to
2018          * deal with request calculations.
2019          */
2020         pd->settings.size = be32_to_cpu(ti.fixed_packet_size) << 2;
2021         if (pd->settings.size == 0) {
2022                 printk(DRIVER_NAME": detected zero packet size!\n");
2023                 return -ENXIO;
2024         }
2025         if (pd->settings.size > PACKET_MAX_SECTORS) {
2026                 printk(DRIVER_NAME": packet size is too big\n");
2027                 return -EROFS;
2028         }
2029         pd->settings.fp = ti.fp;
2030         pd->offset = (be32_to_cpu(ti.track_start) << 2) & (pd->settings.size - 1);
2031
2032         if (ti.nwa_v) {
2033                 pd->nwa = be32_to_cpu(ti.next_writable);
2034                 set_bit(PACKET_NWA_VALID, &pd->flags);
2035         }
2036
2037         /*
2038          * in theory we could use lra on -RW media as well and just zero
2039          * blocks that haven't been written yet, but in practice that
2040          * is just a no-go. we'll use that for -R, naturally.
2041          */
2042         if (ti.lra_v) {
2043                 pd->lra = be32_to_cpu(ti.last_rec_address);
2044                 set_bit(PACKET_LRA_VALID, &pd->flags);
2045         } else {
2046                 pd->lra = 0xffffffff;
2047                 set_bit(PACKET_LRA_VALID, &pd->flags);
2048         }
2049
2050         /*
2051          * fine for now
2052          */
2053         pd->settings.link_loss = 7;
2054         pd->settings.write_type = 0;    /* packet */
2055         pd->settings.track_mode = ti.track_mode;
2056
2057         /*
2058          * mode1 or mode2 disc
2059          */
2060         switch (ti.data_mode) {
2061                 case PACKET_MODE1:
2062                         pd->settings.block_mode = PACKET_BLOCK_MODE1;
2063                         break;
2064                 case PACKET_MODE2:
2065                         pd->settings.block_mode = PACKET_BLOCK_MODE2;
2066                         break;
2067                 default:
2068                         printk(DRIVER_NAME": unknown data mode\n");
2069                         return -EROFS;
2070         }
2071         return 0;
2072 }
2073
2074 /*
2075  * enable/disable write caching on drive
2076  */
2077 static noinline_for_stack int pkt_write_caching(struct pktcdvd_device *pd,
2078                                                 int set)
2079 {
2080         struct packet_command cgc;
2081         struct request_sense sense;
2082         unsigned char buf[64];
2083         int ret;
2084
2085         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
2086         cgc.sense = &sense;
2087         cgc.buflen = pd->mode_offset + 12;
2088
2089         /*
2090          * caching mode page might not be there, so quiet this command
2091          */
2092         cgc.quiet = 1;
2093
2094         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WCACHING_PAGE, 0)))
2095                 return ret;
2096
2097         buf[pd->mode_offset + 10] |= (!!set << 2);
2098
2099         cgc.buflen = cgc.cmd[8] = 2 + ((buf[0] << 8) | (buf[1] & 0xff));
2100         ret = pkt_mode_select(pd, &cgc);
2101         if (ret) {
2102                 printk(DRIVER_NAME": write caching control failed\n");
2103                 pkt_dump_sense(&cgc);
2104         } else if (!ret && set)
2105                 printk(DRIVER_NAME": enabled write caching on %s\n", pd->name);
2106         return ret;
2107 }
2108
2109 static int pkt_lock_door(struct pktcdvd_device *pd, int lockflag)
2110 {
2111         struct packet_command cgc;
2112
2113         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2114         cgc.cmd[0] = GPCMD_PREVENT_ALLOW_MEDIUM_REMOVAL;
2115         cgc.cmd[4] = lockflag ? 1 : 0;
2116         return pkt_generic_packet(pd, &cgc);
2117 }
2118
2119 /*
2120  * Returns drive maximum write speed
2121  */
2122 static noinline_for_stack int pkt_get_max_speed(struct pktcdvd_device *pd,
2123                                                 unsigned *write_speed)
2124 {
2125         struct packet_command cgc;
2126         struct request_sense sense;
2127         unsigned char buf[256+18];
2128         unsigned char *cap_buf;
2129         int ret, offset;
2130
2131         cap_buf = &buf[sizeof(struct mode_page_header) + pd->mode_offset];
2132         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_UNKNOWN);
2133         cgc.sense = &sense;
2134
2135         ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2136         if (ret) {
2137                 cgc.buflen = pd->mode_offset + cap_buf[1] + 2 +
2138                              sizeof(struct mode_page_header);
2139                 ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2140                 if (ret) {
2141                         pkt_dump_sense(&cgc);
2142                         return ret;
2143                 }
2144         }
2145
2146         offset = 20;                        /* Obsoleted field, used by older drives */
2147         if (cap_buf[1] >= 28)
2148                 offset = 28;                /* Current write speed selected */
2149         if (cap_buf[1] >= 30) {
2150                 /* If the drive reports at least one "Logical Unit Write
2151                  * Speed Performance Descriptor Block", use the information
2152                  * in the first block. (contains the highest speed)
2153                  */
2154                 int num_spdb = (cap_buf[30] << 8) + cap_buf[31];
2155                 if (num_spdb > 0)
2156                         offset = 34;
2157         }
2158
2159         *write_speed = (cap_buf[offset] << 8) | cap_buf[offset + 1];
2160         return 0;
2161 }
2162
2163 /* These tables from cdrecord - I don't have orange book */
2164 /* standard speed CD-RW (1-4x) */
2165 static char clv_to_speed[16] = {
2166         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2167            0, 2, 4, 6, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2168 };
2169 /* high speed CD-RW (-10x) */
2170 static char hs_clv_to_speed[16] = {
2171         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2172            0, 2, 4, 6, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2173 };
2174 /* ultra high speed CD-RW */
2175 static char us_clv_to_speed[16] = {
2176         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2177            0, 2, 4, 8, 0, 0,16, 0,24,32,40,48, 0, 0, 0, 0
2178 };
2179
2180 /*
2181  * reads the maximum media speed from ATIP
2182  */
2183 static noinline_for_stack int pkt_media_speed(struct pktcdvd_device *pd,
2184                                                 unsigned *speed)
2185 {
2186         struct packet_command cgc;
2187         struct request_sense sense;
2188         unsigned char buf[64];
2189         unsigned int size, st, sp;
2190         int ret;
2191
2192         init_cdrom_command(&cgc, buf, 2, CGC_DATA_READ);
2193         cgc.sense = &sense;
2194         cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2195         cgc.cmd[1] = 2;
2196         cgc.cmd[2] = 4; /* READ ATIP */
2197         cgc.cmd[8] = 2;
2198         ret = pkt_generic_packet(pd, &cgc);
2199         if (ret) {
2200                 pkt_dump_sense(&cgc);
2201                 return ret;
2202         }
2203         size = ((unsigned int) buf[0]<<8) + buf[1] + 2;
2204         if (size > sizeof(buf))
2205                 size = sizeof(buf);
2206
2207         init_cdrom_command(&cgc, buf, size, CGC_DATA_READ);
2208         cgc.sense = &sense;
2209         cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2210         cgc.cmd[1] = 2;
2211         cgc.cmd[2] = 4;
2212         cgc.cmd[8] = size;
2213         ret = pkt_generic_packet(pd, &cgc);
2214         if (ret) {
2215                 pkt_dump_sense(&cgc);
2216                 return ret;
2217         }
2218
2219         if (!(buf[6] & 0x40)) {
2220                 printk(DRIVER_NAME": Disc type is not CD-RW\n");
2221                 return 1;
2222         }
2223         if (!(buf[6] & 0x4)) {
2224                 printk(DRIVER_NAME": A1 values on media are not valid, maybe not CDRW?\n");
2225                 return 1;
2226         }
2227
2228         st = (buf[6] >> 3) & 0x7; /* disc sub-type */
2229
2230         sp = buf[16] & 0xf; /* max speed from ATIP A1 field */
2231
2232         /* Info from cdrecord */
2233         switch (st) {
2234                 case 0: /* standard speed */
2235                         *speed = clv_to_speed[sp];
2236                         break;
2237                 case 1: /* high speed */
2238                         *speed = hs_clv_to_speed[sp];
2239                         break;
2240                 case 2: /* ultra high speed */
2241                         *speed = us_clv_to_speed[sp];
2242                         break;
2243                 default:
2244                         printk(DRIVER_NAME": Unknown disc sub-type %d\n",st);
2245                         return 1;
2246         }
2247         if (*speed) {
2248                 printk(DRIVER_NAME": Max. media speed: %d\n",*speed);
2249                 return 0;
2250         } else {
2251                 printk(DRIVER_NAME": Unknown speed %d for sub-type %d\n",sp,st);
2252                 return 1;
2253         }
2254 }
2255
2256 static noinline_for_stack int pkt_perform_opc(struct pktcdvd_device *pd)
2257 {
2258         struct packet_command cgc;
2259         struct request_sense sense;
2260         int ret;
2261
2262         VPRINTK(DRIVER_NAME": Performing OPC\n");
2263
2264         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2265         cgc.sense = &sense;
2266         cgc.timeout = 60*HZ;
2267         cgc.cmd[0] = GPCMD_SEND_OPC;
2268         cgc.cmd[1] = 1;
2269         if ((ret = pkt_generic_packet(pd, &cgc)))
2270                 pkt_dump_sense(&cgc);
2271         return ret;
2272 }
2273
2274 static int pkt_open_write(struct pktcdvd_device *pd)
2275 {
2276         int ret;
2277         unsigned int write_speed, media_write_speed, read_speed;
2278
2279         if ((ret = pkt_probe_settings(pd))) {
2280                 VPRINTK(DRIVER_NAME": %s failed probe\n", pd->name);
2281                 return ret;
2282         }
2283
2284         if ((ret = pkt_set_write_settings(pd))) {
2285                 DPRINTK(DRIVER_NAME": %s failed saving write settings\n", pd->name);
2286                 return -EIO;
2287         }
2288
2289         pkt_write_caching(pd, USE_WCACHING);
2290
2291         if ((ret = pkt_get_max_speed(pd, &write_speed)))
2292                 write_speed = 16 * 177;
2293         switch (pd->mmc3_profile) {
2294                 case 0x13: /* DVD-RW */
2295                 case 0x1a: /* DVD+RW */
2296                 case 0x12: /* DVD-RAM */
2297                         DPRINTK(DRIVER_NAME": write speed %ukB/s\n", write_speed);
2298                         break;
2299                 default:
2300                         if ((ret = pkt_media_speed(pd, &media_write_speed)))
2301                                 media_write_speed = 16;
2302                         write_speed = min(write_speed, media_write_speed * 177);
2303                         DPRINTK(DRIVER_NAME": write speed %ux\n", write_speed / 176);
2304                         break;
2305         }
2306         read_speed = write_speed;
2307
2308         if ((ret = pkt_set_speed(pd, write_speed, read_speed))) {
2309                 DPRINTK(DRIVER_NAME": %s couldn't set write speed\n", pd->name);
2310                 return -EIO;
2311         }
2312         pd->write_speed = write_speed;
2313         pd->read_speed = read_speed;
2314
2315         if ((ret = pkt_perform_opc(pd))) {
2316                 DPRINTK(DRIVER_NAME": %s Optimum Power Calibration failed\n", pd->name);
2317         }
2318
2319         return 0;
2320 }
2321
2322 /*
2323  * called at open time.
2324  */
2325 static int pkt_open_dev(struct pktcdvd_device *pd, fmode_t write)
2326 {
2327         int ret;
2328         long lba;
2329         struct request_queue *q;
2330
2331         /*
2332          * We need to re-open the cdrom device without O_NONBLOCK to be able
2333          * to read/write from/to it. It is already opened in O_NONBLOCK mode
2334          * so bdget() can't fail.
2335          */
2336         bdget(pd->bdev->bd_dev);
2337         if ((ret = blkdev_get(pd->bdev, FMODE_READ)))
2338                 goto out;
2339
2340         if ((ret = bd_claim(pd->bdev, pd)))
2341                 goto out_putdev;
2342
2343         if ((ret = pkt_get_last_written(pd, &lba))) {
2344                 printk(DRIVER_NAME": pkt_get_last_written failed\n");
2345                 goto out_unclaim;
2346         }
2347
2348         set_capacity(pd->disk, lba << 2);
2349         set_capacity(pd->bdev->bd_disk, lba << 2);
2350         bd_set_size(pd->bdev, (loff_t)lba << 11);
2351
2352         q = bdev_get_queue(pd->bdev);
2353         if (write) {
2354                 if ((ret = pkt_open_write(pd)))
2355                         goto out_unclaim;
2356                 /*
2357                  * Some CDRW drives can not handle writes larger than one packet,
2358                  * even if the size is a multiple of the packet size.
2359                  */
2360                 spin_lock_irq(q->queue_lock);
2361                 blk_queue_max_sectors(q, pd->settings.size);
2362                 spin_unlock_irq(q->queue_lock);
2363                 set_bit(PACKET_WRITABLE, &pd->flags);
2364         } else {
2365                 pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2366                 clear_bit(PACKET_WRITABLE, &pd->flags);
2367         }
2368
2369         if ((ret = pkt_set_segment_merging(pd, q)))
2370                 goto out_unclaim;
2371
2372         if (write) {
2373                 if (!pkt_grow_pktlist(pd, CONFIG_CDROM_PKTCDVD_BUFFERS)) {
2374                         printk(DRIVER_NAME": not enough memory for buffers\n");
2375                         ret = -ENOMEM;
2376                         goto out_unclaim;
2377                 }
2378                 printk(DRIVER_NAME": %lukB available on disc\n", lba << 1);
2379         }
2380
2381         return 0;
2382
2383 out_unclaim:
2384         bd_release(pd->bdev);
2385 out_putdev:
2386         blkdev_put(pd->bdev, FMODE_READ);
2387 out:
2388         return ret;
2389 }
2390
2391 /*
2392  * called when the device is closed. makes sure that the device flushes
2393  * the internal cache before we close.
2394  */
2395 static void pkt_release_dev(struct pktcdvd_device *pd, int flush)
2396 {
2397         if (flush && pkt_flush_cache(pd))
2398                 DPRINTK(DRIVER_NAME": %s not flushing cache\n", pd->name);
2399
2400         pkt_lock_door(pd, 0);
2401
2402         pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2403         bd_release(pd->bdev);
2404         blkdev_put(pd->bdev, FMODE_READ);
2405
2406         pkt_shrink_pktlist(pd);
2407 }
2408
2409 static struct pktcdvd_device *pkt_find_dev_from_minor(int dev_minor)
2410 {
2411         if (dev_minor >= MAX_WRITERS)
2412                 return NULL;
2413         return pkt_devs[dev_minor];
2414 }
2415
2416 static int pkt_open(struct block_device *bdev, fmode_t mode)
2417 {
2418         struct pktcdvd_device *pd = NULL;
2419         int ret;
2420
2421         VPRINTK(DRIVER_NAME": entering open\n");
2422
2423         mutex_lock(&ctl_mutex);
2424         pd = pkt_find_dev_from_minor(MINOR(bdev->bd_dev));
2425         if (!pd) {
2426                 ret = -ENODEV;
2427                 goto out;
2428         }
2429         BUG_ON(pd->refcnt < 0);
2430
2431         pd->refcnt++;
2432         if (pd->refcnt > 1) {
2433                 if ((mode & FMODE_WRITE) &&
2434                     !test_bit(PACKET_WRITABLE, &pd->flags)) {
2435                         ret = -EBUSY;
2436                         goto out_dec;
2437                 }
2438         } else {
2439                 ret = pkt_open_dev(pd, mode & FMODE_WRITE);
2440                 if (ret)
2441                         goto out_dec;
2442                 /*
2443                  * needed here as well, since ext2 (among others) may change
2444                  * the blocksize at mount time
2445                  */
2446                 set_blocksize(bdev, CD_FRAMESIZE);
2447         }
2448
2449         mutex_unlock(&ctl_mutex);
2450         return 0;
2451
2452 out_dec:
2453         pd->refcnt--;
2454 out:
2455         VPRINTK(DRIVER_NAME": failed open (%d)\n", ret);
2456         mutex_unlock(&ctl_mutex);
2457         return ret;
2458 }
2459
2460 static int pkt_close(struct gendisk *disk, fmode_t mode)
2461 {
2462         struct pktcdvd_device *pd = disk->private_data;
2463         int ret = 0;
2464
2465         mutex_lock(&ctl_mutex);
2466         pd->refcnt--;
2467         BUG_ON(pd->refcnt < 0);
2468         if (pd->refcnt == 0) {
2469                 int flush = test_bit(PACKET_WRITABLE, &pd->flags);
2470                 pkt_release_dev(pd, flush);
2471         }
2472         mutex_unlock(&ctl_mutex);
2473         return ret;
2474 }
2475
2476
2477 static void pkt_end_io_read_cloned(struct bio *bio, int err)
2478 {
2479         struct packet_stacked_data *psd = bio->bi_private;
2480         struct pktcdvd_device *pd = psd->pd;
2481
2482         bio_put(bio);
2483         bio_endio(psd->bio, err);
2484         mempool_free(psd, psd_pool);
2485         pkt_bio_finished(pd);
2486 }
2487
2488 static int pkt_make_request(struct request_queue *q, struct bio *bio)
2489 {
2490         struct pktcdvd_device *pd;
2491         char b[BDEVNAME_SIZE];
2492         sector_t zone;
2493         struct packet_data *pkt;
2494         int was_empty, blocked_bio;
2495         struct pkt_rb_node *node;
2496
2497         pd = q->queuedata;
2498         if (!pd) {
2499                 printk(DRIVER_NAME": %s incorrect request queue\n", bdevname(bio->bi_bdev, b));
2500                 goto end_io;
2501         }
2502
2503         /*
2504          * Clone READ bios so we can have our own bi_end_io callback.
2505          */
2506         if (bio_data_dir(bio) == READ) {
2507                 struct bio *cloned_bio = bio_clone(bio, GFP_NOIO);
2508                 struct packet_stacked_data *psd = mempool_alloc(psd_pool, GFP_NOIO);
2509
2510                 psd->pd = pd;
2511                 psd->bio = bio;
2512                 cloned_bio->bi_bdev = pd->bdev;
2513                 cloned_bio->bi_private = psd;
2514                 cloned_bio->bi_end_io = pkt_end_io_read_cloned;
2515                 pd->stats.secs_r += bio->bi_size >> 9;
2516                 pkt_queue_bio(pd, cloned_bio);
2517                 return 0;
2518         }
2519
2520         if (!test_bit(PACKET_WRITABLE, &pd->flags)) {
2521                 printk(DRIVER_NAME": WRITE for ro device %s (%llu)\n",
2522                         pd->name, (unsigned long long)bio->bi_sector);
2523                 goto end_io;
2524         }
2525
2526         if (!bio->bi_size || (bio->bi_size % CD_FRAMESIZE)) {
2527                 printk(DRIVER_NAME": wrong bio size\n");
2528                 goto end_io;
2529         }
2530
2531         blk_queue_bounce(q, &bio);
2532
2533         zone = ZONE(bio->bi_sector, pd);
2534         VPRINTK("pkt_make_request: start = %6llx stop = %6llx\n",
2535                 (unsigned long long)bio->bi_sector,
2536                 (unsigned long long)(bio->bi_sector + bio_sectors(bio)));
2537
2538         /* Check if we have to split the bio */
2539         {
2540                 struct bio_pair *bp;
2541                 sector_t last_zone;
2542                 int first_sectors;
2543
2544                 last_zone = ZONE(bio->bi_sector + bio_sectors(bio) - 1, pd);
2545                 if (last_zone != zone) {
2546                         BUG_ON(last_zone != zone + pd->settings.size);
2547                         first_sectors = last_zone - bio->bi_sector;
2548                         bp = bio_split(bio, first_sectors);
2549                         BUG_ON(!bp);
2550                         pkt_make_request(q, &bp->bio1);
2551                         pkt_make_request(q, &bp->bio2);
2552                         bio_pair_release(bp);
2553                         return 0;
2554                 }
2555         }
2556
2557         /*
2558          * If we find a matching packet in state WAITING or READ_WAIT, we can
2559          * just append this bio to that packet.
2560          */
2561         spin_lock(&pd->cdrw.active_list_lock);
2562         blocked_bio = 0;
2563         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
2564                 if (pkt->sector == zone) {
2565                         spin_lock(&pkt->lock);
2566                         if ((pkt->state == PACKET_WAITING_STATE) ||
2567                             (pkt->state == PACKET_READ_WAIT_STATE)) {
2568                                 pkt_add_list_last(bio, &pkt->orig_bios,
2569                                                   &pkt->orig_bios_tail);
2570                                 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
2571                                 if ((pkt->write_size >= pkt->frames) &&
2572                                     (pkt->state == PACKET_WAITING_STATE)) {
2573                                         atomic_inc(&pkt->run_sm);
2574                                         wake_up(&pd->wqueue);
2575                                 }
2576                                 spin_unlock(&pkt->lock);
2577                                 spin_unlock(&pd->cdrw.active_list_lock);
2578                                 return 0;
2579                         } else {
2580                                 blocked_bio = 1;
2581                         }
2582                         spin_unlock(&pkt->lock);
2583                 }
2584         }
2585         spin_unlock(&pd->cdrw.active_list_lock);
2586
2587         /*
2588          * Test if there is enough room left in the bio work queue
2589          * (queue size >= congestion on mark).
2590          * If not, wait till the work queue size is below the congestion off mark.
2591          */
2592         spin_lock(&pd->lock);
2593         if (pd->write_congestion_on > 0
2594             && pd->bio_queue_size >= pd->write_congestion_on) {
2595                 set_bdi_congested(&q->backing_dev_info, WRITE);
2596                 do {
2597                         spin_unlock(&pd->lock);
2598                         congestion_wait(WRITE, HZ);
2599                         spin_lock(&pd->lock);
2600                 } while(pd->bio_queue_size > pd->write_congestion_off);
2601         }
2602         spin_unlock(&pd->lock);
2603
2604         /*
2605          * No matching packet found. Store the bio in the work queue.
2606          */
2607         node = mempool_alloc(pd->rb_pool, GFP_NOIO);
2608         node->bio = bio;
2609         spin_lock(&pd->lock);
2610         BUG_ON(pd->bio_queue_size < 0);
2611         was_empty = (pd->bio_queue_size == 0);
2612         pkt_rbtree_insert(pd, node);
2613         spin_unlock(&pd->lock);
2614
2615         /*
2616          * Wake up the worker thread.
2617          */
2618         atomic_set(&pd->scan_queue, 1);
2619         if (was_empty) {
2620                 /* This wake_up is required for correct operation */
2621                 wake_up(&pd->wqueue);
2622         } else if (!list_empty(&pd->cdrw.pkt_free_list) && !blocked_bio) {
2623                 /*
2624                  * This wake up is not required for correct operation,
2625                  * but improves performance in some cases.
2626                  */
2627                 wake_up(&pd->wqueue);
2628         }
2629         return 0;
2630 end_io:
2631         bio_io_error(bio);
2632         return 0;
2633 }
2634
2635
2636
2637 static int pkt_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2638                           struct bio_vec *bvec)
2639 {
2640         struct pktcdvd_device *pd = q->queuedata;
2641         sector_t zone = ZONE(bmd->bi_sector, pd);
2642         int used = ((bmd->bi_sector - zone) << 9) + bmd->bi_size;
2643         int remaining = (pd->settings.size << 9) - used;
2644         int remaining2;
2645
2646         /*
2647          * A bio <= PAGE_SIZE must be allowed. If it crosses a packet
2648          * boundary, pkt_make_request() will split the bio.
2649          */
2650         remaining2 = PAGE_SIZE - bmd->bi_size;
2651         remaining = max(remaining, remaining2);
2652
2653         BUG_ON(remaining < 0);
2654         return remaining;
2655 }
2656
2657 static void pkt_init_queue(struct pktcdvd_device *pd)
2658 {
2659         struct request_queue *q = pd->disk->queue;
2660
2661         blk_queue_make_request(q, pkt_make_request);
2662         blk_queue_logical_block_size(q, CD_FRAMESIZE);
2663         blk_queue_max_sectors(q, PACKET_MAX_SECTORS);
2664         blk_queue_merge_bvec(q, pkt_merge_bvec);
2665         q->queuedata = pd;
2666 }
2667
2668 static int pkt_seq_show(struct seq_file *m, void *p)
2669 {
2670         struct pktcdvd_device *pd = m->private;
2671         char *msg;
2672         char bdev_buf[BDEVNAME_SIZE];
2673         int states[PACKET_NUM_STATES];
2674
2675         seq_printf(m, "Writer %s mapped to %s:\n", pd->name,
2676                    bdevname(pd->bdev, bdev_buf));
2677
2678         seq_printf(m, "\nSettings:\n");
2679         seq_printf(m, "\tpacket size:\t\t%dkB\n", pd->settings.size / 2);
2680
2681         if (pd->settings.write_type == 0)
2682                 msg = "Packet";
2683         else
2684                 msg = "Unknown";
2685         seq_printf(m, "\twrite type:\t\t%s\n", msg);
2686
2687         seq_printf(m, "\tpacket type:\t\t%s\n", pd->settings.fp ? "Fixed" : "Variable");
2688         seq_printf(m, "\tlink loss:\t\t%d\n", pd->settings.link_loss);
2689
2690         seq_printf(m, "\ttrack mode:\t\t%d\n", pd->settings.track_mode);
2691
2692         if (pd->settings.block_mode == PACKET_BLOCK_MODE1)
2693                 msg = "Mode 1";
2694         else if (pd->settings.block_mode == PACKET_BLOCK_MODE2)
2695                 msg = "Mode 2";
2696         else
2697                 msg = "Unknown";
2698         seq_printf(m, "\tblock mode:\t\t%s\n", msg);
2699
2700         seq_printf(m, "\nStatistics:\n");
2701         seq_printf(m, "\tpackets started:\t%lu\n", pd->stats.pkt_started);
2702         seq_printf(m, "\tpackets ended:\t\t%lu\n", pd->stats.pkt_ended);
2703         seq_printf(m, "\twritten:\t\t%lukB\n", pd->stats.secs_w >> 1);
2704         seq_printf(m, "\tread gather:\t\t%lukB\n", pd->stats.secs_rg >> 1);
2705         seq_printf(m, "\tread:\t\t\t%lukB\n", pd->stats.secs_r >> 1);
2706
2707         seq_printf(m, "\nMisc:\n");
2708         seq_printf(m, "\treference count:\t%d\n", pd->refcnt);
2709         seq_printf(m, "\tflags:\t\t\t0x%lx\n", pd->flags);
2710         seq_printf(m, "\tread speed:\t\t%ukB/s\n", pd->read_speed);
2711         seq_printf(m, "\twrite speed:\t\t%ukB/s\n", pd->write_speed);
2712         seq_printf(m, "\tstart offset:\t\t%lu\n", pd->offset);
2713         seq_printf(m, "\tmode page offset:\t%u\n", pd->mode_offset);
2714
2715         seq_printf(m, "\nQueue state:\n");
2716         seq_printf(m, "\tbios queued:\t\t%d\n", pd->bio_queue_size);
2717         seq_printf(m, "\tbios pending:\t\t%d\n", atomic_read(&pd->cdrw.pending_bios));
2718         seq_printf(m, "\tcurrent sector:\t\t0x%llx\n", (unsigned long long)pd->current_sector);
2719
2720         pkt_count_states(pd, states);
2721         seq_printf(m, "\tstate:\t\t\ti:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
2722                    states[0], states[1], states[2], states[3], states[4], states[5]);
2723
2724         seq_printf(m, "\twrite congestion marks:\toff=%d on=%d\n",
2725                         pd->write_congestion_off,
2726                         pd->write_congestion_on);
2727         return 0;
2728 }
2729
2730 static int pkt_seq_open(struct inode *inode, struct file *file)
2731 {
2732         return single_open(file, pkt_seq_show, PDE(inode)->data);
2733 }
2734
2735 static const struct file_operations pkt_proc_fops = {
2736         .open   = pkt_seq_open,
2737         .read   = seq_read,
2738         .llseek = seq_lseek,
2739         .release = single_release
2740 };
2741
2742 static int pkt_new_dev(struct pktcdvd_device *pd, dev_t dev)
2743 {
2744         int i;
2745         int ret = 0;
2746         char b[BDEVNAME_SIZE];
2747         struct block_device *bdev;
2748
2749         if (pd->pkt_dev == dev) {
2750                 printk(DRIVER_NAME": Recursive setup not allowed\n");
2751                 return -EBUSY;
2752         }
2753         for (i = 0; i < MAX_WRITERS; i++) {
2754                 struct pktcdvd_device *pd2 = pkt_devs[i];
2755                 if (!pd2)
2756                         continue;
2757                 if (pd2->bdev->bd_dev == dev) {
2758                         printk(DRIVER_NAME": %s already setup\n", bdevname(pd2->bdev, b));
2759                         return -EBUSY;
2760                 }
2761                 if (pd2->pkt_dev == dev) {
2762                         printk(DRIVER_NAME": Can't chain pktcdvd devices\n");
2763                         return -EBUSY;
2764                 }
2765         }
2766
2767         bdev = bdget(dev);
2768         if (!bdev)
2769                 return -ENOMEM;
2770         ret = blkdev_get(bdev, FMODE_READ | FMODE_NDELAY);
2771         if (ret)
2772                 return ret;
2773
2774         /* This is safe, since we have a reference from open(). */
2775         __module_get(THIS_MODULE);
2776
2777         pd->bdev = bdev;
2778         set_blocksize(bdev, CD_FRAMESIZE);
2779
2780         pkt_init_queue(pd);
2781
2782         atomic_set(&pd->cdrw.pending_bios, 0);
2783         pd->cdrw.thread = kthread_run(kcdrwd, pd, "%s", pd->name);
2784         if (IS_ERR(pd->cdrw.thread)) {
2785                 printk(DRIVER_NAME": can't start kernel thread\n");
2786                 ret = -ENOMEM;
2787                 goto out_mem;
2788         }
2789
2790         proc_create_data(pd->name, 0, pkt_proc, &pkt_proc_fops, pd);
2791         DPRINTK(DRIVER_NAME": writer %s mapped to %s\n", pd->name, bdevname(bdev, b));
2792         return 0;
2793
2794 out_mem:
2795         blkdev_put(bdev, FMODE_READ | FMODE_NDELAY);
2796         /* This is safe: open() is still holding a reference. */
2797         module_put(THIS_MODULE);
2798         return ret;
2799 }
2800
2801 static int pkt_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, unsigned long arg)
2802 {
2803         struct pktcdvd_device *pd = bdev->bd_disk->private_data;
2804
2805         VPRINTK("pkt_ioctl: cmd %x, dev %d:%d\n", cmd,
2806                 MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev));
2807
2808         switch (cmd) {
2809         case CDROMEJECT:
2810                 /*
2811                  * The door gets locked when the device is opened, so we
2812                  * have to unlock it or else the eject command fails.
2813                  */
2814                 if (pd->refcnt == 1)
2815                         pkt_lock_door(pd, 0);
2816                 /* fallthru */
2817         /*
2818          * forward selected CDROM ioctls to CD-ROM, for UDF
2819          */
2820         case CDROMMULTISESSION:
2821         case CDROMREADTOCENTRY:
2822         case CDROM_LAST_WRITTEN:
2823         case CDROM_SEND_PACKET:
2824         case SCSI_IOCTL_SEND_COMMAND:
2825                 return __blkdev_driver_ioctl(pd->bdev, mode, cmd, arg);
2826
2827         default:
2828                 VPRINTK(DRIVER_NAME": Unknown ioctl for %s (%x)\n", pd->name, cmd);
2829                 return -ENOTTY;
2830         }
2831
2832         return 0;
2833 }
2834
2835 static int pkt_media_changed(struct gendisk *disk)
2836 {
2837         struct pktcdvd_device *pd = disk->private_data;
2838         struct gendisk *attached_disk;
2839
2840         if (!pd)
2841                 return 0;
2842         if (!pd->bdev)
2843                 return 0;
2844         attached_disk = pd->bdev->bd_disk;
2845         if (!attached_disk)
2846                 return 0;
2847         return attached_disk->fops->media_changed(attached_disk);
2848 }
2849
2850 static struct block_device_operations pktcdvd_ops = {
2851         .owner =                THIS_MODULE,
2852         .open =                 pkt_open,
2853         .release =              pkt_close,
2854         .locked_ioctl =         pkt_ioctl,
2855         .media_changed =        pkt_media_changed,
2856 };
2857
2858 static char *pktcdvd_nodename(struct gendisk *gd)
2859 {
2860         return kasprintf(GFP_KERNEL, "pktcdvd/%s", gd->disk_name);
2861 }
2862
2863 /*
2864  * Set up mapping from pktcdvd device to CD-ROM device.
2865  */
2866 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev)
2867 {
2868         int idx;
2869         int ret = -ENOMEM;
2870         struct pktcdvd_device *pd;
2871         struct gendisk *disk;
2872
2873         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2874
2875         for (idx = 0; idx < MAX_WRITERS; idx++)
2876                 if (!pkt_devs[idx])
2877                         break;
2878         if (idx == MAX_WRITERS) {
2879                 printk(DRIVER_NAME": max %d writers supported\n", MAX_WRITERS);
2880                 ret = -EBUSY;
2881                 goto out_mutex;
2882         }
2883
2884         pd = kzalloc(sizeof(struct pktcdvd_device), GFP_KERNEL);
2885         if (!pd)
2886                 goto out_mutex;
2887
2888         pd->rb_pool = mempool_create_kmalloc_pool(PKT_RB_POOL_SIZE,
2889                                                   sizeof(struct pkt_rb_node));
2890         if (!pd->rb_pool)
2891                 goto out_mem;
2892
2893         INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
2894         INIT_LIST_HEAD(&pd->cdrw.pkt_active_list);
2895         spin_lock_init(&pd->cdrw.active_list_lock);
2896
2897         spin_lock_init(&pd->lock);
2898         spin_lock_init(&pd->iosched.lock);
2899         sprintf(pd->name, DRIVER_NAME"%d", idx);
2900         init_waitqueue_head(&pd->wqueue);
2901         pd->bio_queue = RB_ROOT;
2902
2903         pd->write_congestion_on  = write_congestion_on;
2904         pd->write_congestion_off = write_congestion_off;
2905
2906         disk = alloc_disk(1);
2907         if (!disk)
2908                 goto out_mem;
2909         pd->disk = disk;
2910         disk->major = pktdev_major;
2911         disk->first_minor = idx;
2912         disk->fops = &pktcdvd_ops;
2913         disk->flags = GENHD_FL_REMOVABLE;
2914         strcpy(disk->disk_name, pd->name);
2915         disk->nodename = pktcdvd_nodename;
2916         disk->private_data = pd;
2917         disk->queue = blk_alloc_queue(GFP_KERNEL);
2918         if (!disk->queue)
2919                 goto out_mem2;
2920
2921         pd->pkt_dev = MKDEV(pktdev_major, idx);
2922         ret = pkt_new_dev(pd, dev);
2923         if (ret)
2924                 goto out_new_dev;
2925
2926         add_disk(disk);
2927
2928         pkt_sysfs_dev_new(pd);
2929         pkt_debugfs_dev_new(pd);
2930
2931         pkt_devs[idx] = pd;
2932         if (pkt_dev)
2933                 *pkt_dev = pd->pkt_dev;
2934
2935         mutex_unlock(&ctl_mutex);
2936         return 0;
2937
2938 out_new_dev:
2939         blk_cleanup_queue(disk->queue);
2940 out_mem2:
2941         put_disk(disk);
2942 out_mem:
2943         if (pd->rb_pool)
2944                 mempool_destroy(pd->rb_pool);
2945         kfree(pd);
2946 out_mutex:
2947         mutex_unlock(&ctl_mutex);
2948         printk(DRIVER_NAME": setup of pktcdvd device failed\n");
2949         return ret;
2950 }
2951
2952 /*
2953  * Tear down mapping from pktcdvd device to CD-ROM device.
2954  */
2955 static int pkt_remove_dev(dev_t pkt_dev)
2956 {
2957         struct pktcdvd_device *pd;
2958         int idx;
2959         int ret = 0;
2960
2961         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2962
2963         for (idx = 0; idx < MAX_WRITERS; idx++) {
2964                 pd = pkt_devs[idx];
2965                 if (pd && (pd->pkt_dev == pkt_dev))
2966                         break;
2967         }
2968         if (idx == MAX_WRITERS) {
2969                 DPRINTK(DRIVER_NAME": dev not setup\n");
2970                 ret = -ENXIO;
2971                 goto out;
2972         }
2973
2974         if (pd->refcnt > 0) {
2975                 ret = -EBUSY;
2976                 goto out;
2977         }
2978         if (!IS_ERR(pd->cdrw.thread))
2979                 kthread_stop(pd->cdrw.thread);
2980
2981         pkt_devs[idx] = NULL;
2982
2983         pkt_debugfs_dev_remove(pd);
2984         pkt_sysfs_dev_remove(pd);
2985
2986         blkdev_put(pd->bdev, FMODE_READ | FMODE_NDELAY);
2987
2988         remove_proc_entry(pd->name, pkt_proc);
2989         DPRINTK(DRIVER_NAME": writer %s unmapped\n", pd->name);
2990
2991         del_gendisk(pd->disk);
2992         blk_cleanup_queue(pd->disk->queue);
2993         put_disk(pd->disk);
2994
2995         mempool_destroy(pd->rb_pool);
2996         kfree(pd);
2997
2998         /* This is safe: open() is still holding a reference. */
2999         module_put(THIS_MODULE);
3000
3001 out:
3002         mutex_unlock(&ctl_mutex);
3003         return ret;
3004 }
3005
3006 static void pkt_get_status(struct pkt_ctrl_command *ctrl_cmd)
3007 {
3008         struct pktcdvd_device *pd;
3009
3010         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
3011
3012         pd = pkt_find_dev_from_minor(ctrl_cmd->dev_index);
3013         if (pd) {
3014                 ctrl_cmd->dev = new_encode_dev(pd->bdev->bd_dev);
3015                 ctrl_cmd->pkt_dev = new_encode_dev(pd->pkt_dev);
3016         } else {
3017                 ctrl_cmd->dev = 0;
3018                 ctrl_cmd->pkt_dev = 0;
3019         }
3020         ctrl_cmd->num_devices = MAX_WRITERS;
3021
3022         mutex_unlock(&ctl_mutex);
3023 }
3024
3025 static int pkt_ctl_ioctl(struct inode *inode, struct file *file, unsigned int cmd, unsigned long arg)
3026 {
3027         void __user *argp = (void __user *)arg;
3028         struct pkt_ctrl_command ctrl_cmd;
3029         int ret = 0;
3030         dev_t pkt_dev = 0;
3031
3032         if (cmd != PACKET_CTRL_CMD)
3033                 return -ENOTTY;
3034
3035         if (copy_from_user(&ctrl_cmd, argp, sizeof(struct pkt_ctrl_command)))
3036                 return -EFAULT;
3037
3038         switch (ctrl_cmd.command) {
3039         case PKT_CTRL_CMD_SETUP:
3040                 if (!capable(CAP_SYS_ADMIN))
3041                         return -EPERM;
3042                 ret = pkt_setup_dev(new_decode_dev(ctrl_cmd.dev), &pkt_dev);
3043                 ctrl_cmd.pkt_dev = new_encode_dev(pkt_dev);
3044                 break;
3045         case PKT_CTRL_CMD_TEARDOWN:
3046                 if (!capable(CAP_SYS_ADMIN))
3047                         return -EPERM;
3048                 ret = pkt_remove_dev(new_decode_dev(ctrl_cmd.pkt_dev));
3049                 break;
3050         case PKT_CTRL_CMD_STATUS:
3051                 pkt_get_status(&ctrl_cmd);
3052                 break;
3053         default:
3054                 return -ENOTTY;
3055         }
3056
3057         if (copy_to_user(argp, &ctrl_cmd, sizeof(struct pkt_ctrl_command)))
3058                 return -EFAULT;
3059         return ret;
3060 }
3061
3062
3063 static const struct file_operations pkt_ctl_fops = {
3064         .ioctl   = pkt_ctl_ioctl,
3065         .owner   = THIS_MODULE,
3066 };
3067
3068 static struct miscdevice pkt_misc = {
3069         .minor          = MISC_DYNAMIC_MINOR,
3070         .name           = DRIVER_NAME,
3071         .name           = "pktcdvd/control",
3072         .fops           = &pkt_ctl_fops
3073 };
3074
3075 static int __init pkt_init(void)
3076 {
3077         int ret;
3078
3079         mutex_init(&ctl_mutex);
3080
3081         psd_pool = mempool_create_kmalloc_pool(PSD_POOL_SIZE,
3082                                         sizeof(struct packet_stacked_data));
3083         if (!psd_pool)
3084                 return -ENOMEM;
3085
3086         ret = register_blkdev(pktdev_major, DRIVER_NAME);
3087         if (ret < 0) {
3088                 printk(DRIVER_NAME": Unable to register block device\n");
3089                 goto out2;
3090         }
3091         if (!pktdev_major)
3092                 pktdev_major = ret;
3093
3094         ret = pkt_sysfs_init();
3095         if (ret)
3096                 goto out;
3097
3098         pkt_debugfs_init();
3099
3100         ret = misc_register(&pkt_misc);
3101         if (ret) {
3102                 printk(DRIVER_NAME": Unable to register misc device\n");
3103                 goto out_misc;
3104         }
3105
3106         pkt_proc = proc_mkdir("driver/"DRIVER_NAME, NULL);
3107
3108         return 0;
3109
3110 out_misc:
3111         pkt_debugfs_cleanup();
3112         pkt_sysfs_cleanup();
3113 out:
3114         unregister_blkdev(pktdev_major, DRIVER_NAME);
3115 out2:
3116         mempool_destroy(psd_pool);
3117         return ret;
3118 }
3119
3120 static void __exit pkt_exit(void)
3121 {
3122         remove_proc_entry("driver/"DRIVER_NAME, NULL);
3123         misc_deregister(&pkt_misc);
3124
3125         pkt_debugfs_cleanup();
3126         pkt_sysfs_cleanup();
3127
3128         unregister_blkdev(pktdev_major, DRIVER_NAME);
3129         mempool_destroy(psd_pool);
3130 }
3131
3132 MODULE_DESCRIPTION("Packet writing layer for CD/DVD drives");
3133 MODULE_AUTHOR("Jens Axboe <axboe@suse.de>");
3134 MODULE_LICENSE("GPL");
3135
3136 module_init(pkt_init);
3137 module_exit(pkt_exit);