cciss: Fix scsi tape io with more than 255 scatter gather elements
[linux-2.6.git] / drivers / block / pktcdvd.c
1 /*
2  * Copyright (C) 2000 Jens Axboe <axboe@suse.de>
3  * Copyright (C) 2001-2004 Peter Osterlund <petero2@telia.com>
4  * Copyright (C) 2006 Thomas Maier <balagi@justmail.de>
5  *
6  * May be copied or modified under the terms of the GNU General Public
7  * License.  See linux/COPYING for more information.
8  *
9  * Packet writing layer for ATAPI and SCSI CD-RW, DVD+RW, DVD-RW and
10  * DVD-RAM devices.
11  *
12  * Theory of operation:
13  *
14  * At the lowest level, there is the standard driver for the CD/DVD device,
15  * typically ide-cd.c or sr.c. This driver can handle read and write requests,
16  * but it doesn't know anything about the special restrictions that apply to
17  * packet writing. One restriction is that write requests must be aligned to
18  * packet boundaries on the physical media, and the size of a write request
19  * must be equal to the packet size. Another restriction is that a
20  * GPCMD_FLUSH_CACHE command has to be issued to the drive before a read
21  * command, if the previous command was a write.
22  *
23  * The purpose of the packet writing driver is to hide these restrictions from
24  * higher layers, such as file systems, and present a block device that can be
25  * randomly read and written using 2kB-sized blocks.
26  *
27  * The lowest layer in the packet writing driver is the packet I/O scheduler.
28  * Its data is defined by the struct packet_iosched and includes two bio
29  * queues with pending read and write requests. These queues are processed
30  * by the pkt_iosched_process_queue() function. The write requests in this
31  * queue are already properly aligned and sized. This layer is responsible for
32  * issuing the flush cache commands and scheduling the I/O in a good order.
33  *
34  * The next layer transforms unaligned write requests to aligned writes. This
35  * transformation requires reading missing pieces of data from the underlying
36  * block device, assembling the pieces to full packets and queuing them to the
37  * packet I/O scheduler.
38  *
39  * At the top layer there is a custom make_request_fn function that forwards
40  * read requests directly to the iosched queue and puts write requests in the
41  * unaligned write queue. A kernel thread performs the necessary read
42  * gathering to convert the unaligned writes to aligned writes and then feeds
43  * them to the packet I/O scheduler.
44  *
45  *************************************************************************/
46
47 #include <linux/pktcdvd.h>
48 #include <linux/module.h>
49 #include <linux/types.h>
50 #include <linux/kernel.h>
51 #include <linux/compat.h>
52 #include <linux/kthread.h>
53 #include <linux/errno.h>
54 #include <linux/spinlock.h>
55 #include <linux/file.h>
56 #include <linux/proc_fs.h>
57 #include <linux/seq_file.h>
58 #include <linux/miscdevice.h>
59 #include <linux/freezer.h>
60 #include <linux/mutex.h>
61 #include <linux/slab.h>
62 #include <scsi/scsi_cmnd.h>
63 #include <scsi/scsi_ioctl.h>
64 #include <scsi/scsi.h>
65 #include <linux/debugfs.h>
66 #include <linux/device.h>
67
68 #include <asm/uaccess.h>
69
70 #define DRIVER_NAME     "pktcdvd"
71
72 #if PACKET_DEBUG
73 #define DPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
74 #else
75 #define DPRINTK(fmt, args...)
76 #endif
77
78 #if PACKET_DEBUG > 1
79 #define VPRINTK(fmt, args...) printk(KERN_NOTICE fmt, ##args)
80 #else
81 #define VPRINTK(fmt, args...)
82 #endif
83
84 #define MAX_SPEED 0xffff
85
86 #define ZONE(sector, pd) (((sector) + (pd)->offset) & ~((pd)->settings.size - 1))
87
88 static DEFINE_MUTEX(pktcdvd_mutex);
89 static struct pktcdvd_device *pkt_devs[MAX_WRITERS];
90 static struct proc_dir_entry *pkt_proc;
91 static int pktdev_major;
92 static int write_congestion_on  = PKT_WRITE_CONGESTION_ON;
93 static int write_congestion_off = PKT_WRITE_CONGESTION_OFF;
94 static struct mutex ctl_mutex;  /* Serialize open/close/setup/teardown */
95 static mempool_t *psd_pool;
96
97 static struct class     *class_pktcdvd = NULL;    /* /sys/class/pktcdvd */
98 static struct dentry    *pkt_debugfs_root = NULL; /* /sys/kernel/debug/pktcdvd */
99
100 /* forward declaration */
101 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev);
102 static int pkt_remove_dev(dev_t pkt_dev);
103 static int pkt_seq_show(struct seq_file *m, void *p);
104
105
106
107 /*
108  * create and register a pktcdvd kernel object.
109  */
110 static struct pktcdvd_kobj* pkt_kobj_create(struct pktcdvd_device *pd,
111                                         const char* name,
112                                         struct kobject* parent,
113                                         struct kobj_type* ktype)
114 {
115         struct pktcdvd_kobj *p;
116         int error;
117
118         p = kzalloc(sizeof(*p), GFP_KERNEL);
119         if (!p)
120                 return NULL;
121         p->pd = pd;
122         error = kobject_init_and_add(&p->kobj, ktype, parent, "%s", name);
123         if (error) {
124                 kobject_put(&p->kobj);
125                 return NULL;
126         }
127         kobject_uevent(&p->kobj, KOBJ_ADD);
128         return p;
129 }
130 /*
131  * remove a pktcdvd kernel object.
132  */
133 static void pkt_kobj_remove(struct pktcdvd_kobj *p)
134 {
135         if (p)
136                 kobject_put(&p->kobj);
137 }
138 /*
139  * default release function for pktcdvd kernel objects.
140  */
141 static void pkt_kobj_release(struct kobject *kobj)
142 {
143         kfree(to_pktcdvdkobj(kobj));
144 }
145
146
147 /**********************************************************
148  *
149  * sysfs interface for pktcdvd
150  * by (C) 2006  Thomas Maier <balagi@justmail.de>
151  *
152  **********************************************************/
153
154 #define DEF_ATTR(_obj,_name,_mode) \
155         static struct attribute _obj = { .name = _name, .mode = _mode }
156
157 /**********************************************************
158   /sys/class/pktcdvd/pktcdvd[0-7]/
159                      stat/reset
160                      stat/packets_started
161                      stat/packets_finished
162                      stat/kb_written
163                      stat/kb_read
164                      stat/kb_read_gather
165                      write_queue/size
166                      write_queue/congestion_off
167                      write_queue/congestion_on
168  **********************************************************/
169
170 DEF_ATTR(kobj_pkt_attr_st1, "reset", 0200);
171 DEF_ATTR(kobj_pkt_attr_st2, "packets_started", 0444);
172 DEF_ATTR(kobj_pkt_attr_st3, "packets_finished", 0444);
173 DEF_ATTR(kobj_pkt_attr_st4, "kb_written", 0444);
174 DEF_ATTR(kobj_pkt_attr_st5, "kb_read", 0444);
175 DEF_ATTR(kobj_pkt_attr_st6, "kb_read_gather", 0444);
176
177 static struct attribute *kobj_pkt_attrs_stat[] = {
178         &kobj_pkt_attr_st1,
179         &kobj_pkt_attr_st2,
180         &kobj_pkt_attr_st3,
181         &kobj_pkt_attr_st4,
182         &kobj_pkt_attr_st5,
183         &kobj_pkt_attr_st6,
184         NULL
185 };
186
187 DEF_ATTR(kobj_pkt_attr_wq1, "size", 0444);
188 DEF_ATTR(kobj_pkt_attr_wq2, "congestion_off", 0644);
189 DEF_ATTR(kobj_pkt_attr_wq3, "congestion_on",  0644);
190
191 static struct attribute *kobj_pkt_attrs_wqueue[] = {
192         &kobj_pkt_attr_wq1,
193         &kobj_pkt_attr_wq2,
194         &kobj_pkt_attr_wq3,
195         NULL
196 };
197
198 static ssize_t kobj_pkt_show(struct kobject *kobj,
199                         struct attribute *attr, char *data)
200 {
201         struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
202         int n = 0;
203         int v;
204         if (strcmp(attr->name, "packets_started") == 0) {
205                 n = sprintf(data, "%lu\n", pd->stats.pkt_started);
206
207         } else if (strcmp(attr->name, "packets_finished") == 0) {
208                 n = sprintf(data, "%lu\n", pd->stats.pkt_ended);
209
210         } else if (strcmp(attr->name, "kb_written") == 0) {
211                 n = sprintf(data, "%lu\n", pd->stats.secs_w >> 1);
212
213         } else if (strcmp(attr->name, "kb_read") == 0) {
214                 n = sprintf(data, "%lu\n", pd->stats.secs_r >> 1);
215
216         } else if (strcmp(attr->name, "kb_read_gather") == 0) {
217                 n = sprintf(data, "%lu\n", pd->stats.secs_rg >> 1);
218
219         } else if (strcmp(attr->name, "size") == 0) {
220                 spin_lock(&pd->lock);
221                 v = pd->bio_queue_size;
222                 spin_unlock(&pd->lock);
223                 n = sprintf(data, "%d\n", v);
224
225         } else if (strcmp(attr->name, "congestion_off") == 0) {
226                 spin_lock(&pd->lock);
227                 v = pd->write_congestion_off;
228                 spin_unlock(&pd->lock);
229                 n = sprintf(data, "%d\n", v);
230
231         } else if (strcmp(attr->name, "congestion_on") == 0) {
232                 spin_lock(&pd->lock);
233                 v = pd->write_congestion_on;
234                 spin_unlock(&pd->lock);
235                 n = sprintf(data, "%d\n", v);
236         }
237         return n;
238 }
239
240 static void init_write_congestion_marks(int* lo, int* hi)
241 {
242         if (*hi > 0) {
243                 *hi = max(*hi, 500);
244                 *hi = min(*hi, 1000000);
245                 if (*lo <= 0)
246                         *lo = *hi - 100;
247                 else {
248                         *lo = min(*lo, *hi - 100);
249                         *lo = max(*lo, 100);
250                 }
251         } else {
252                 *hi = -1;
253                 *lo = -1;
254         }
255 }
256
257 static ssize_t kobj_pkt_store(struct kobject *kobj,
258                         struct attribute *attr,
259                         const char *data, size_t len)
260 {
261         struct pktcdvd_device *pd = to_pktcdvdkobj(kobj)->pd;
262         int val;
263
264         if (strcmp(attr->name, "reset") == 0 && len > 0) {
265                 pd->stats.pkt_started = 0;
266                 pd->stats.pkt_ended = 0;
267                 pd->stats.secs_w = 0;
268                 pd->stats.secs_rg = 0;
269                 pd->stats.secs_r = 0;
270
271         } else if (strcmp(attr->name, "congestion_off") == 0
272                    && sscanf(data, "%d", &val) == 1) {
273                 spin_lock(&pd->lock);
274                 pd->write_congestion_off = val;
275                 init_write_congestion_marks(&pd->write_congestion_off,
276                                         &pd->write_congestion_on);
277                 spin_unlock(&pd->lock);
278
279         } else if (strcmp(attr->name, "congestion_on") == 0
280                    && sscanf(data, "%d", &val) == 1) {
281                 spin_lock(&pd->lock);
282                 pd->write_congestion_on = val;
283                 init_write_congestion_marks(&pd->write_congestion_off,
284                                         &pd->write_congestion_on);
285                 spin_unlock(&pd->lock);
286         }
287         return len;
288 }
289
290 static const struct sysfs_ops kobj_pkt_ops = {
291         .show = kobj_pkt_show,
292         .store = kobj_pkt_store
293 };
294 static struct kobj_type kobj_pkt_type_stat = {
295         .release = pkt_kobj_release,
296         .sysfs_ops = &kobj_pkt_ops,
297         .default_attrs = kobj_pkt_attrs_stat
298 };
299 static struct kobj_type kobj_pkt_type_wqueue = {
300         .release = pkt_kobj_release,
301         .sysfs_ops = &kobj_pkt_ops,
302         .default_attrs = kobj_pkt_attrs_wqueue
303 };
304
305 static void pkt_sysfs_dev_new(struct pktcdvd_device *pd)
306 {
307         if (class_pktcdvd) {
308                 pd->dev = device_create(class_pktcdvd, NULL, MKDEV(0, 0), NULL,
309                                         "%s", pd->name);
310                 if (IS_ERR(pd->dev))
311                         pd->dev = NULL;
312         }
313         if (pd->dev) {
314                 pd->kobj_stat = pkt_kobj_create(pd, "stat",
315                                         &pd->dev->kobj,
316                                         &kobj_pkt_type_stat);
317                 pd->kobj_wqueue = pkt_kobj_create(pd, "write_queue",
318                                         &pd->dev->kobj,
319                                         &kobj_pkt_type_wqueue);
320         }
321 }
322
323 static void pkt_sysfs_dev_remove(struct pktcdvd_device *pd)
324 {
325         pkt_kobj_remove(pd->kobj_stat);
326         pkt_kobj_remove(pd->kobj_wqueue);
327         if (class_pktcdvd)
328                 device_unregister(pd->dev);
329 }
330
331
332 /********************************************************************
333   /sys/class/pktcdvd/
334                      add            map block device
335                      remove         unmap packet dev
336                      device_map     show mappings
337  *******************************************************************/
338
339 static void class_pktcdvd_release(struct class *cls)
340 {
341         kfree(cls);
342 }
343 static ssize_t class_pktcdvd_show_map(struct class *c,
344                                         struct class_attribute *attr,
345                                         char *data)
346 {
347         int n = 0;
348         int idx;
349         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
350         for (idx = 0; idx < MAX_WRITERS; idx++) {
351                 struct pktcdvd_device *pd = pkt_devs[idx];
352                 if (!pd)
353                         continue;
354                 n += sprintf(data+n, "%s %u:%u %u:%u\n",
355                         pd->name,
356                         MAJOR(pd->pkt_dev), MINOR(pd->pkt_dev),
357                         MAJOR(pd->bdev->bd_dev),
358                         MINOR(pd->bdev->bd_dev));
359         }
360         mutex_unlock(&ctl_mutex);
361         return n;
362 }
363
364 static ssize_t class_pktcdvd_store_add(struct class *c,
365                                         struct class_attribute *attr,
366                                         const char *buf,
367                                         size_t count)
368 {
369         unsigned int major, minor;
370
371         if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
372                 /* pkt_setup_dev() expects caller to hold reference to self */
373                 if (!try_module_get(THIS_MODULE))
374                         return -ENODEV;
375
376                 pkt_setup_dev(MKDEV(major, minor), NULL);
377
378                 module_put(THIS_MODULE);
379
380                 return count;
381         }
382
383         return -EINVAL;
384 }
385
386 static ssize_t class_pktcdvd_store_remove(struct class *c,
387                                           struct class_attribute *attr,
388                                           const char *buf,
389                                         size_t count)
390 {
391         unsigned int major, minor;
392         if (sscanf(buf, "%u:%u", &major, &minor) == 2) {
393                 pkt_remove_dev(MKDEV(major, minor));
394                 return count;
395         }
396         return -EINVAL;
397 }
398
399 static struct class_attribute class_pktcdvd_attrs[] = {
400  __ATTR(add,            0200, NULL, class_pktcdvd_store_add),
401  __ATTR(remove,         0200, NULL, class_pktcdvd_store_remove),
402  __ATTR(device_map,     0444, class_pktcdvd_show_map, NULL),
403  __ATTR_NULL
404 };
405
406
407 static int pkt_sysfs_init(void)
408 {
409         int ret = 0;
410
411         /*
412          * create control files in sysfs
413          * /sys/class/pktcdvd/...
414          */
415         class_pktcdvd = kzalloc(sizeof(*class_pktcdvd), GFP_KERNEL);
416         if (!class_pktcdvd)
417                 return -ENOMEM;
418         class_pktcdvd->name = DRIVER_NAME;
419         class_pktcdvd->owner = THIS_MODULE;
420         class_pktcdvd->class_release = class_pktcdvd_release;
421         class_pktcdvd->class_attrs = class_pktcdvd_attrs;
422         ret = class_register(class_pktcdvd);
423         if (ret) {
424                 kfree(class_pktcdvd);
425                 class_pktcdvd = NULL;
426                 printk(DRIVER_NAME": failed to create class pktcdvd\n");
427                 return ret;
428         }
429         return 0;
430 }
431
432 static void pkt_sysfs_cleanup(void)
433 {
434         if (class_pktcdvd)
435                 class_destroy(class_pktcdvd);
436         class_pktcdvd = NULL;
437 }
438
439 /********************************************************************
440   entries in debugfs
441
442   /sys/kernel/debug/pktcdvd[0-7]/
443                         info
444
445  *******************************************************************/
446
447 static int pkt_debugfs_seq_show(struct seq_file *m, void *p)
448 {
449         return pkt_seq_show(m, p);
450 }
451
452 static int pkt_debugfs_fops_open(struct inode *inode, struct file *file)
453 {
454         return single_open(file, pkt_debugfs_seq_show, inode->i_private);
455 }
456
457 static const struct file_operations debug_fops = {
458         .open           = pkt_debugfs_fops_open,
459         .read           = seq_read,
460         .llseek         = seq_lseek,
461         .release        = single_release,
462         .owner          = THIS_MODULE,
463 };
464
465 static void pkt_debugfs_dev_new(struct pktcdvd_device *pd)
466 {
467         if (!pkt_debugfs_root)
468                 return;
469         pd->dfs_f_info = NULL;
470         pd->dfs_d_root = debugfs_create_dir(pd->name, pkt_debugfs_root);
471         if (IS_ERR(pd->dfs_d_root)) {
472                 pd->dfs_d_root = NULL;
473                 return;
474         }
475         pd->dfs_f_info = debugfs_create_file("info", S_IRUGO,
476                                 pd->dfs_d_root, pd, &debug_fops);
477         if (IS_ERR(pd->dfs_f_info)) {
478                 pd->dfs_f_info = NULL;
479                 return;
480         }
481 }
482
483 static void pkt_debugfs_dev_remove(struct pktcdvd_device *pd)
484 {
485         if (!pkt_debugfs_root)
486                 return;
487         if (pd->dfs_f_info)
488                 debugfs_remove(pd->dfs_f_info);
489         pd->dfs_f_info = NULL;
490         if (pd->dfs_d_root)
491                 debugfs_remove(pd->dfs_d_root);
492         pd->dfs_d_root = NULL;
493 }
494
495 static void pkt_debugfs_init(void)
496 {
497         pkt_debugfs_root = debugfs_create_dir(DRIVER_NAME, NULL);
498         if (IS_ERR(pkt_debugfs_root)) {
499                 pkt_debugfs_root = NULL;
500                 return;
501         }
502 }
503
504 static void pkt_debugfs_cleanup(void)
505 {
506         if (!pkt_debugfs_root)
507                 return;
508         debugfs_remove(pkt_debugfs_root);
509         pkt_debugfs_root = NULL;
510 }
511
512 /* ----------------------------------------------------------*/
513
514
515 static void pkt_bio_finished(struct pktcdvd_device *pd)
516 {
517         BUG_ON(atomic_read(&pd->cdrw.pending_bios) <= 0);
518         if (atomic_dec_and_test(&pd->cdrw.pending_bios)) {
519                 VPRINTK(DRIVER_NAME": queue empty\n");
520                 atomic_set(&pd->iosched.attention, 1);
521                 wake_up(&pd->wqueue);
522         }
523 }
524
525 static void pkt_bio_destructor(struct bio *bio)
526 {
527         kfree(bio->bi_io_vec);
528         kfree(bio);
529 }
530
531 static struct bio *pkt_bio_alloc(int nr_iovecs)
532 {
533         struct bio_vec *bvl = NULL;
534         struct bio *bio;
535
536         bio = kmalloc(sizeof(struct bio), GFP_KERNEL);
537         if (!bio)
538                 goto no_bio;
539         bio_init(bio);
540
541         bvl = kcalloc(nr_iovecs, sizeof(struct bio_vec), GFP_KERNEL);
542         if (!bvl)
543                 goto no_bvl;
544
545         bio->bi_max_vecs = nr_iovecs;
546         bio->bi_io_vec = bvl;
547         bio->bi_destructor = pkt_bio_destructor;
548
549         return bio;
550
551  no_bvl:
552         kfree(bio);
553  no_bio:
554         return NULL;
555 }
556
557 /*
558  * Allocate a packet_data struct
559  */
560 static struct packet_data *pkt_alloc_packet_data(int frames)
561 {
562         int i;
563         struct packet_data *pkt;
564
565         pkt = kzalloc(sizeof(struct packet_data), GFP_KERNEL);
566         if (!pkt)
567                 goto no_pkt;
568
569         pkt->frames = frames;
570         pkt->w_bio = pkt_bio_alloc(frames);
571         if (!pkt->w_bio)
572                 goto no_bio;
573
574         for (i = 0; i < frames / FRAMES_PER_PAGE; i++) {
575                 pkt->pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO);
576                 if (!pkt->pages[i])
577                         goto no_page;
578         }
579
580         spin_lock_init(&pkt->lock);
581         bio_list_init(&pkt->orig_bios);
582
583         for (i = 0; i < frames; i++) {
584                 struct bio *bio = pkt_bio_alloc(1);
585                 if (!bio)
586                         goto no_rd_bio;
587                 pkt->r_bios[i] = bio;
588         }
589
590         return pkt;
591
592 no_rd_bio:
593         for (i = 0; i < frames; i++) {
594                 struct bio *bio = pkt->r_bios[i];
595                 if (bio)
596                         bio_put(bio);
597         }
598
599 no_page:
600         for (i = 0; i < frames / FRAMES_PER_PAGE; i++)
601                 if (pkt->pages[i])
602                         __free_page(pkt->pages[i]);
603         bio_put(pkt->w_bio);
604 no_bio:
605         kfree(pkt);
606 no_pkt:
607         return NULL;
608 }
609
610 /*
611  * Free a packet_data struct
612  */
613 static void pkt_free_packet_data(struct packet_data *pkt)
614 {
615         int i;
616
617         for (i = 0; i < pkt->frames; i++) {
618                 struct bio *bio = pkt->r_bios[i];
619                 if (bio)
620                         bio_put(bio);
621         }
622         for (i = 0; i < pkt->frames / FRAMES_PER_PAGE; i++)
623                 __free_page(pkt->pages[i]);
624         bio_put(pkt->w_bio);
625         kfree(pkt);
626 }
627
628 static void pkt_shrink_pktlist(struct pktcdvd_device *pd)
629 {
630         struct packet_data *pkt, *next;
631
632         BUG_ON(!list_empty(&pd->cdrw.pkt_active_list));
633
634         list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_free_list, list) {
635                 pkt_free_packet_data(pkt);
636         }
637         INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
638 }
639
640 static int pkt_grow_pktlist(struct pktcdvd_device *pd, int nr_packets)
641 {
642         struct packet_data *pkt;
643
644         BUG_ON(!list_empty(&pd->cdrw.pkt_free_list));
645
646         while (nr_packets > 0) {
647                 pkt = pkt_alloc_packet_data(pd->settings.size >> 2);
648                 if (!pkt) {
649                         pkt_shrink_pktlist(pd);
650                         return 0;
651                 }
652                 pkt->id = nr_packets;
653                 pkt->pd = pd;
654                 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
655                 nr_packets--;
656         }
657         return 1;
658 }
659
660 static inline struct pkt_rb_node *pkt_rbtree_next(struct pkt_rb_node *node)
661 {
662         struct rb_node *n = rb_next(&node->rb_node);
663         if (!n)
664                 return NULL;
665         return rb_entry(n, struct pkt_rb_node, rb_node);
666 }
667
668 static void pkt_rbtree_erase(struct pktcdvd_device *pd, struct pkt_rb_node *node)
669 {
670         rb_erase(&node->rb_node, &pd->bio_queue);
671         mempool_free(node, pd->rb_pool);
672         pd->bio_queue_size--;
673         BUG_ON(pd->bio_queue_size < 0);
674 }
675
676 /*
677  * Find the first node in the pd->bio_queue rb tree with a starting sector >= s.
678  */
679 static struct pkt_rb_node *pkt_rbtree_find(struct pktcdvd_device *pd, sector_t s)
680 {
681         struct rb_node *n = pd->bio_queue.rb_node;
682         struct rb_node *next;
683         struct pkt_rb_node *tmp;
684
685         if (!n) {
686                 BUG_ON(pd->bio_queue_size > 0);
687                 return NULL;
688         }
689
690         for (;;) {
691                 tmp = rb_entry(n, struct pkt_rb_node, rb_node);
692                 if (s <= tmp->bio->bi_sector)
693                         next = n->rb_left;
694                 else
695                         next = n->rb_right;
696                 if (!next)
697                         break;
698                 n = next;
699         }
700
701         if (s > tmp->bio->bi_sector) {
702                 tmp = pkt_rbtree_next(tmp);
703                 if (!tmp)
704                         return NULL;
705         }
706         BUG_ON(s > tmp->bio->bi_sector);
707         return tmp;
708 }
709
710 /*
711  * Insert a node into the pd->bio_queue rb tree.
712  */
713 static void pkt_rbtree_insert(struct pktcdvd_device *pd, struct pkt_rb_node *node)
714 {
715         struct rb_node **p = &pd->bio_queue.rb_node;
716         struct rb_node *parent = NULL;
717         sector_t s = node->bio->bi_sector;
718         struct pkt_rb_node *tmp;
719
720         while (*p) {
721                 parent = *p;
722                 tmp = rb_entry(parent, struct pkt_rb_node, rb_node);
723                 if (s < tmp->bio->bi_sector)
724                         p = &(*p)->rb_left;
725                 else
726                         p = &(*p)->rb_right;
727         }
728         rb_link_node(&node->rb_node, parent, p);
729         rb_insert_color(&node->rb_node, &pd->bio_queue);
730         pd->bio_queue_size++;
731 }
732
733 /*
734  * Send a packet_command to the underlying block device and
735  * wait for completion.
736  */
737 static int pkt_generic_packet(struct pktcdvd_device *pd, struct packet_command *cgc)
738 {
739         struct request_queue *q = bdev_get_queue(pd->bdev);
740         struct request *rq;
741         int ret = 0;
742
743         rq = blk_get_request(q, (cgc->data_direction == CGC_DATA_WRITE) ?
744                              WRITE : READ, __GFP_WAIT);
745
746         if (cgc->buflen) {
747                 if (blk_rq_map_kern(q, rq, cgc->buffer, cgc->buflen, __GFP_WAIT))
748                         goto out;
749         }
750
751         rq->cmd_len = COMMAND_SIZE(cgc->cmd[0]);
752         memcpy(rq->cmd, cgc->cmd, CDROM_PACKET_SIZE);
753
754         rq->timeout = 60*HZ;
755         rq->cmd_type = REQ_TYPE_BLOCK_PC;
756         if (cgc->quiet)
757                 rq->cmd_flags |= REQ_QUIET;
758
759         blk_execute_rq(rq->q, pd->bdev->bd_disk, rq, 0);
760         if (rq->errors)
761                 ret = -EIO;
762 out:
763         blk_put_request(rq);
764         return ret;
765 }
766
767 /*
768  * A generic sense dump / resolve mechanism should be implemented across
769  * all ATAPI + SCSI devices.
770  */
771 static void pkt_dump_sense(struct packet_command *cgc)
772 {
773         static char *info[9] = { "No sense", "Recovered error", "Not ready",
774                                  "Medium error", "Hardware error", "Illegal request",
775                                  "Unit attention", "Data protect", "Blank check" };
776         int i;
777         struct request_sense *sense = cgc->sense;
778
779         printk(DRIVER_NAME":");
780         for (i = 0; i < CDROM_PACKET_SIZE; i++)
781                 printk(" %02x", cgc->cmd[i]);
782         printk(" - ");
783
784         if (sense == NULL) {
785                 printk("no sense\n");
786                 return;
787         }
788
789         printk("sense %02x.%02x.%02x", sense->sense_key, sense->asc, sense->ascq);
790
791         if (sense->sense_key > 8) {
792                 printk(" (INVALID)\n");
793                 return;
794         }
795
796         printk(" (%s)\n", info[sense->sense_key]);
797 }
798
799 /*
800  * flush the drive cache to media
801  */
802 static int pkt_flush_cache(struct pktcdvd_device *pd)
803 {
804         struct packet_command cgc;
805
806         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
807         cgc.cmd[0] = GPCMD_FLUSH_CACHE;
808         cgc.quiet = 1;
809
810         /*
811          * the IMMED bit -- we default to not setting it, although that
812          * would allow a much faster close, this is safer
813          */
814 #if 0
815         cgc.cmd[1] = 1 << 1;
816 #endif
817         return pkt_generic_packet(pd, &cgc);
818 }
819
820 /*
821  * speed is given as the normal factor, e.g. 4 for 4x
822  */
823 static noinline_for_stack int pkt_set_speed(struct pktcdvd_device *pd,
824                                 unsigned write_speed, unsigned read_speed)
825 {
826         struct packet_command cgc;
827         struct request_sense sense;
828         int ret;
829
830         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
831         cgc.sense = &sense;
832         cgc.cmd[0] = GPCMD_SET_SPEED;
833         cgc.cmd[2] = (read_speed >> 8) & 0xff;
834         cgc.cmd[3] = read_speed & 0xff;
835         cgc.cmd[4] = (write_speed >> 8) & 0xff;
836         cgc.cmd[5] = write_speed & 0xff;
837
838         if ((ret = pkt_generic_packet(pd, &cgc)))
839                 pkt_dump_sense(&cgc);
840
841         return ret;
842 }
843
844 /*
845  * Queue a bio for processing by the low-level CD device. Must be called
846  * from process context.
847  */
848 static void pkt_queue_bio(struct pktcdvd_device *pd, struct bio *bio)
849 {
850         spin_lock(&pd->iosched.lock);
851         if (bio_data_dir(bio) == READ)
852                 bio_list_add(&pd->iosched.read_queue, bio);
853         else
854                 bio_list_add(&pd->iosched.write_queue, bio);
855         spin_unlock(&pd->iosched.lock);
856
857         atomic_set(&pd->iosched.attention, 1);
858         wake_up(&pd->wqueue);
859 }
860
861 /*
862  * Process the queued read/write requests. This function handles special
863  * requirements for CDRW drives:
864  * - A cache flush command must be inserted before a read request if the
865  *   previous request was a write.
866  * - Switching between reading and writing is slow, so don't do it more often
867  *   than necessary.
868  * - Optimize for throughput at the expense of latency. This means that streaming
869  *   writes will never be interrupted by a read, but if the drive has to seek
870  *   before the next write, switch to reading instead if there are any pending
871  *   read requests.
872  * - Set the read speed according to current usage pattern. When only reading
873  *   from the device, it's best to use the highest possible read speed, but
874  *   when switching often between reading and writing, it's better to have the
875  *   same read and write speeds.
876  */
877 static void pkt_iosched_process_queue(struct pktcdvd_device *pd)
878 {
879
880         if (atomic_read(&pd->iosched.attention) == 0)
881                 return;
882         atomic_set(&pd->iosched.attention, 0);
883
884         for (;;) {
885                 struct bio *bio;
886                 int reads_queued, writes_queued;
887
888                 spin_lock(&pd->iosched.lock);
889                 reads_queued = !bio_list_empty(&pd->iosched.read_queue);
890                 writes_queued = !bio_list_empty(&pd->iosched.write_queue);
891                 spin_unlock(&pd->iosched.lock);
892
893                 if (!reads_queued && !writes_queued)
894                         break;
895
896                 if (pd->iosched.writing) {
897                         int need_write_seek = 1;
898                         spin_lock(&pd->iosched.lock);
899                         bio = bio_list_peek(&pd->iosched.write_queue);
900                         spin_unlock(&pd->iosched.lock);
901                         if (bio && (bio->bi_sector == pd->iosched.last_write))
902                                 need_write_seek = 0;
903                         if (need_write_seek && reads_queued) {
904                                 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
905                                         VPRINTK(DRIVER_NAME": write, waiting\n");
906                                         break;
907                                 }
908                                 pkt_flush_cache(pd);
909                                 pd->iosched.writing = 0;
910                         }
911                 } else {
912                         if (!reads_queued && writes_queued) {
913                                 if (atomic_read(&pd->cdrw.pending_bios) > 0) {
914                                         VPRINTK(DRIVER_NAME": read, waiting\n");
915                                         break;
916                                 }
917                                 pd->iosched.writing = 1;
918                         }
919                 }
920
921                 spin_lock(&pd->iosched.lock);
922                 if (pd->iosched.writing)
923                         bio = bio_list_pop(&pd->iosched.write_queue);
924                 else
925                         bio = bio_list_pop(&pd->iosched.read_queue);
926                 spin_unlock(&pd->iosched.lock);
927
928                 if (!bio)
929                         continue;
930
931                 if (bio_data_dir(bio) == READ)
932                         pd->iosched.successive_reads += bio->bi_size >> 10;
933                 else {
934                         pd->iosched.successive_reads = 0;
935                         pd->iosched.last_write = bio->bi_sector + bio_sectors(bio);
936                 }
937                 if (pd->iosched.successive_reads >= HI_SPEED_SWITCH) {
938                         if (pd->read_speed == pd->write_speed) {
939                                 pd->read_speed = MAX_SPEED;
940                                 pkt_set_speed(pd, pd->write_speed, pd->read_speed);
941                         }
942                 } else {
943                         if (pd->read_speed != pd->write_speed) {
944                                 pd->read_speed = pd->write_speed;
945                                 pkt_set_speed(pd, pd->write_speed, pd->read_speed);
946                         }
947                 }
948
949                 atomic_inc(&pd->cdrw.pending_bios);
950                 generic_make_request(bio);
951         }
952 }
953
954 /*
955  * Special care is needed if the underlying block device has a small
956  * max_phys_segments value.
957  */
958 static int pkt_set_segment_merging(struct pktcdvd_device *pd, struct request_queue *q)
959 {
960         if ((pd->settings.size << 9) / CD_FRAMESIZE
961             <= queue_max_segments(q)) {
962                 /*
963                  * The cdrom device can handle one segment/frame
964                  */
965                 clear_bit(PACKET_MERGE_SEGS, &pd->flags);
966                 return 0;
967         } else if ((pd->settings.size << 9) / PAGE_SIZE
968                    <= queue_max_segments(q)) {
969                 /*
970                  * We can handle this case at the expense of some extra memory
971                  * copies during write operations
972                  */
973                 set_bit(PACKET_MERGE_SEGS, &pd->flags);
974                 return 0;
975         } else {
976                 printk(DRIVER_NAME": cdrom max_phys_segments too small\n");
977                 return -EIO;
978         }
979 }
980
981 /*
982  * Copy CD_FRAMESIZE bytes from src_bio into a destination page
983  */
984 static void pkt_copy_bio_data(struct bio *src_bio, int seg, int offs, struct page *dst_page, int dst_offs)
985 {
986         unsigned int copy_size = CD_FRAMESIZE;
987
988         while (copy_size > 0) {
989                 struct bio_vec *src_bvl = bio_iovec_idx(src_bio, seg);
990                 void *vfrom = kmap_atomic(src_bvl->bv_page, KM_USER0) +
991                         src_bvl->bv_offset + offs;
992                 void *vto = page_address(dst_page) + dst_offs;
993                 int len = min_t(int, copy_size, src_bvl->bv_len - offs);
994
995                 BUG_ON(len < 0);
996                 memcpy(vto, vfrom, len);
997                 kunmap_atomic(vfrom, KM_USER0);
998
999                 seg++;
1000                 offs = 0;
1001                 dst_offs += len;
1002                 copy_size -= len;
1003         }
1004 }
1005
1006 /*
1007  * Copy all data for this packet to pkt->pages[], so that
1008  * a) The number of required segments for the write bio is minimized, which
1009  *    is necessary for some scsi controllers.
1010  * b) The data can be used as cache to avoid read requests if we receive a
1011  *    new write request for the same zone.
1012  */
1013 static void pkt_make_local_copy(struct packet_data *pkt, struct bio_vec *bvec)
1014 {
1015         int f, p, offs;
1016
1017         /* Copy all data to pkt->pages[] */
1018         p = 0;
1019         offs = 0;
1020         for (f = 0; f < pkt->frames; f++) {
1021                 if (bvec[f].bv_page != pkt->pages[p]) {
1022                         void *vfrom = kmap_atomic(bvec[f].bv_page, KM_USER0) + bvec[f].bv_offset;
1023                         void *vto = page_address(pkt->pages[p]) + offs;
1024                         memcpy(vto, vfrom, CD_FRAMESIZE);
1025                         kunmap_atomic(vfrom, KM_USER0);
1026                         bvec[f].bv_page = pkt->pages[p];
1027                         bvec[f].bv_offset = offs;
1028                 } else {
1029                         BUG_ON(bvec[f].bv_offset != offs);
1030                 }
1031                 offs += CD_FRAMESIZE;
1032                 if (offs >= PAGE_SIZE) {
1033                         offs = 0;
1034                         p++;
1035                 }
1036         }
1037 }
1038
1039 static void pkt_end_io_read(struct bio *bio, int err)
1040 {
1041         struct packet_data *pkt = bio->bi_private;
1042         struct pktcdvd_device *pd = pkt->pd;
1043         BUG_ON(!pd);
1044
1045         VPRINTK("pkt_end_io_read: bio=%p sec0=%llx sec=%llx err=%d\n", bio,
1046                 (unsigned long long)pkt->sector, (unsigned long long)bio->bi_sector, err);
1047
1048         if (err)
1049                 atomic_inc(&pkt->io_errors);
1050         if (atomic_dec_and_test(&pkt->io_wait)) {
1051                 atomic_inc(&pkt->run_sm);
1052                 wake_up(&pd->wqueue);
1053         }
1054         pkt_bio_finished(pd);
1055 }
1056
1057 static void pkt_end_io_packet_write(struct bio *bio, int err)
1058 {
1059         struct packet_data *pkt = bio->bi_private;
1060         struct pktcdvd_device *pd = pkt->pd;
1061         BUG_ON(!pd);
1062
1063         VPRINTK("pkt_end_io_packet_write: id=%d, err=%d\n", pkt->id, err);
1064
1065         pd->stats.pkt_ended++;
1066
1067         pkt_bio_finished(pd);
1068         atomic_dec(&pkt->io_wait);
1069         atomic_inc(&pkt->run_sm);
1070         wake_up(&pd->wqueue);
1071 }
1072
1073 /*
1074  * Schedule reads for the holes in a packet
1075  */
1076 static void pkt_gather_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1077 {
1078         int frames_read = 0;
1079         struct bio *bio;
1080         int f;
1081         char written[PACKET_MAX_SIZE];
1082
1083         BUG_ON(bio_list_empty(&pkt->orig_bios));
1084
1085         atomic_set(&pkt->io_wait, 0);
1086         atomic_set(&pkt->io_errors, 0);
1087
1088         /*
1089          * Figure out which frames we need to read before we can write.
1090          */
1091         memset(written, 0, sizeof(written));
1092         spin_lock(&pkt->lock);
1093         bio_list_for_each(bio, &pkt->orig_bios) {
1094                 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1095                 int num_frames = bio->bi_size / CD_FRAMESIZE;
1096                 pd->stats.secs_w += num_frames * (CD_FRAMESIZE >> 9);
1097                 BUG_ON(first_frame < 0);
1098                 BUG_ON(first_frame + num_frames > pkt->frames);
1099                 for (f = first_frame; f < first_frame + num_frames; f++)
1100                         written[f] = 1;
1101         }
1102         spin_unlock(&pkt->lock);
1103
1104         if (pkt->cache_valid) {
1105                 VPRINTK("pkt_gather_data: zone %llx cached\n",
1106                         (unsigned long long)pkt->sector);
1107                 goto out_account;
1108         }
1109
1110         /*
1111          * Schedule reads for missing parts of the packet.
1112          */
1113         for (f = 0; f < pkt->frames; f++) {
1114                 struct bio_vec *vec;
1115
1116                 int p, offset;
1117                 if (written[f])
1118                         continue;
1119                 bio = pkt->r_bios[f];
1120                 vec = bio->bi_io_vec;
1121                 bio_init(bio);
1122                 bio->bi_max_vecs = 1;
1123                 bio->bi_sector = pkt->sector + f * (CD_FRAMESIZE >> 9);
1124                 bio->bi_bdev = pd->bdev;
1125                 bio->bi_end_io = pkt_end_io_read;
1126                 bio->bi_private = pkt;
1127                 bio->bi_io_vec = vec;
1128                 bio->bi_destructor = pkt_bio_destructor;
1129
1130                 p = (f * CD_FRAMESIZE) / PAGE_SIZE;
1131                 offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1132                 VPRINTK("pkt_gather_data: Adding frame %d, page:%p offs:%d\n",
1133                         f, pkt->pages[p], offset);
1134                 if (!bio_add_page(bio, pkt->pages[p], CD_FRAMESIZE, offset))
1135                         BUG();
1136
1137                 atomic_inc(&pkt->io_wait);
1138                 bio->bi_rw = READ;
1139                 pkt_queue_bio(pd, bio);
1140                 frames_read++;
1141         }
1142
1143 out_account:
1144         VPRINTK("pkt_gather_data: need %d frames for zone %llx\n",
1145                 frames_read, (unsigned long long)pkt->sector);
1146         pd->stats.pkt_started++;
1147         pd->stats.secs_rg += frames_read * (CD_FRAMESIZE >> 9);
1148 }
1149
1150 /*
1151  * Find a packet matching zone, or the least recently used packet if
1152  * there is no match.
1153  */
1154 static struct packet_data *pkt_get_packet_data(struct pktcdvd_device *pd, int zone)
1155 {
1156         struct packet_data *pkt;
1157
1158         list_for_each_entry(pkt, &pd->cdrw.pkt_free_list, list) {
1159                 if (pkt->sector == zone || pkt->list.next == &pd->cdrw.pkt_free_list) {
1160                         list_del_init(&pkt->list);
1161                         if (pkt->sector != zone)
1162                                 pkt->cache_valid = 0;
1163                         return pkt;
1164                 }
1165         }
1166         BUG();
1167         return NULL;
1168 }
1169
1170 static void pkt_put_packet_data(struct pktcdvd_device *pd, struct packet_data *pkt)
1171 {
1172         if (pkt->cache_valid) {
1173                 list_add(&pkt->list, &pd->cdrw.pkt_free_list);
1174         } else {
1175                 list_add_tail(&pkt->list, &pd->cdrw.pkt_free_list);
1176         }
1177 }
1178
1179 /*
1180  * recover a failed write, query for relocation if possible
1181  *
1182  * returns 1 if recovery is possible, or 0 if not
1183  *
1184  */
1185 static int pkt_start_recovery(struct packet_data *pkt)
1186 {
1187         /*
1188          * FIXME. We need help from the file system to implement
1189          * recovery handling.
1190          */
1191         return 0;
1192 #if 0
1193         struct request *rq = pkt->rq;
1194         struct pktcdvd_device *pd = rq->rq_disk->private_data;
1195         struct block_device *pkt_bdev;
1196         struct super_block *sb = NULL;
1197         unsigned long old_block, new_block;
1198         sector_t new_sector;
1199
1200         pkt_bdev = bdget(kdev_t_to_nr(pd->pkt_dev));
1201         if (pkt_bdev) {
1202                 sb = get_super(pkt_bdev);
1203                 bdput(pkt_bdev);
1204         }
1205
1206         if (!sb)
1207                 return 0;
1208
1209         if (!sb->s_op->relocate_blocks)
1210                 goto out;
1211
1212         old_block = pkt->sector / (CD_FRAMESIZE >> 9);
1213         if (sb->s_op->relocate_blocks(sb, old_block, &new_block))
1214                 goto out;
1215
1216         new_sector = new_block * (CD_FRAMESIZE >> 9);
1217         pkt->sector = new_sector;
1218
1219         pkt->bio->bi_sector = new_sector;
1220         pkt->bio->bi_next = NULL;
1221         pkt->bio->bi_flags = 1 << BIO_UPTODATE;
1222         pkt->bio->bi_idx = 0;
1223
1224         BUG_ON(pkt->bio->bi_rw != REQ_WRITE);
1225         BUG_ON(pkt->bio->bi_vcnt != pkt->frames);
1226         BUG_ON(pkt->bio->bi_size != pkt->frames * CD_FRAMESIZE);
1227         BUG_ON(pkt->bio->bi_end_io != pkt_end_io_packet_write);
1228         BUG_ON(pkt->bio->bi_private != pkt);
1229
1230         drop_super(sb);
1231         return 1;
1232
1233 out:
1234         drop_super(sb);
1235         return 0;
1236 #endif
1237 }
1238
1239 static inline void pkt_set_state(struct packet_data *pkt, enum packet_data_state state)
1240 {
1241 #if PACKET_DEBUG > 1
1242         static const char *state_name[] = {
1243                 "IDLE", "WAITING", "READ_WAIT", "WRITE_WAIT", "RECOVERY", "FINISHED"
1244         };
1245         enum packet_data_state old_state = pkt->state;
1246         VPRINTK("pkt %2d : s=%6llx %s -> %s\n", pkt->id, (unsigned long long)pkt->sector,
1247                 state_name[old_state], state_name[state]);
1248 #endif
1249         pkt->state = state;
1250 }
1251
1252 /*
1253  * Scan the work queue to see if we can start a new packet.
1254  * returns non-zero if any work was done.
1255  */
1256 static int pkt_handle_queue(struct pktcdvd_device *pd)
1257 {
1258         struct packet_data *pkt, *p;
1259         struct bio *bio = NULL;
1260         sector_t zone = 0; /* Suppress gcc warning */
1261         struct pkt_rb_node *node, *first_node;
1262         struct rb_node *n;
1263         int wakeup;
1264
1265         VPRINTK("handle_queue\n");
1266
1267         atomic_set(&pd->scan_queue, 0);
1268
1269         if (list_empty(&pd->cdrw.pkt_free_list)) {
1270                 VPRINTK("handle_queue: no pkt\n");
1271                 return 0;
1272         }
1273
1274         /*
1275          * Try to find a zone we are not already working on.
1276          */
1277         spin_lock(&pd->lock);
1278         first_node = pkt_rbtree_find(pd, pd->current_sector);
1279         if (!first_node) {
1280                 n = rb_first(&pd->bio_queue);
1281                 if (n)
1282                         first_node = rb_entry(n, struct pkt_rb_node, rb_node);
1283         }
1284         node = first_node;
1285         while (node) {
1286                 bio = node->bio;
1287                 zone = ZONE(bio->bi_sector, pd);
1288                 list_for_each_entry(p, &pd->cdrw.pkt_active_list, list) {
1289                         if (p->sector == zone) {
1290                                 bio = NULL;
1291                                 goto try_next_bio;
1292                         }
1293                 }
1294                 break;
1295 try_next_bio:
1296                 node = pkt_rbtree_next(node);
1297                 if (!node) {
1298                         n = rb_first(&pd->bio_queue);
1299                         if (n)
1300                                 node = rb_entry(n, struct pkt_rb_node, rb_node);
1301                 }
1302                 if (node == first_node)
1303                         node = NULL;
1304         }
1305         spin_unlock(&pd->lock);
1306         if (!bio) {
1307                 VPRINTK("handle_queue: no bio\n");
1308                 return 0;
1309         }
1310
1311         pkt = pkt_get_packet_data(pd, zone);
1312
1313         pd->current_sector = zone + pd->settings.size;
1314         pkt->sector = zone;
1315         BUG_ON(pkt->frames != pd->settings.size >> 2);
1316         pkt->write_size = 0;
1317
1318         /*
1319          * Scan work queue for bios in the same zone and link them
1320          * to this packet.
1321          */
1322         spin_lock(&pd->lock);
1323         VPRINTK("pkt_handle_queue: looking for zone %llx\n", (unsigned long long)zone);
1324         while ((node = pkt_rbtree_find(pd, zone)) != NULL) {
1325                 bio = node->bio;
1326                 VPRINTK("pkt_handle_queue: found zone=%llx\n",
1327                         (unsigned long long)ZONE(bio->bi_sector, pd));
1328                 if (ZONE(bio->bi_sector, pd) != zone)
1329                         break;
1330                 pkt_rbtree_erase(pd, node);
1331                 spin_lock(&pkt->lock);
1332                 bio_list_add(&pkt->orig_bios, bio);
1333                 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
1334                 spin_unlock(&pkt->lock);
1335         }
1336         /* check write congestion marks, and if bio_queue_size is
1337            below, wake up any waiters */
1338         wakeup = (pd->write_congestion_on > 0
1339                         && pd->bio_queue_size <= pd->write_congestion_off);
1340         spin_unlock(&pd->lock);
1341         if (wakeup) {
1342                 clear_bdi_congested(&pd->disk->queue->backing_dev_info,
1343                                         BLK_RW_ASYNC);
1344         }
1345
1346         pkt->sleep_time = max(PACKET_WAIT_TIME, 1);
1347         pkt_set_state(pkt, PACKET_WAITING_STATE);
1348         atomic_set(&pkt->run_sm, 1);
1349
1350         spin_lock(&pd->cdrw.active_list_lock);
1351         list_add(&pkt->list, &pd->cdrw.pkt_active_list);
1352         spin_unlock(&pd->cdrw.active_list_lock);
1353
1354         return 1;
1355 }
1356
1357 /*
1358  * Assemble a bio to write one packet and queue the bio for processing
1359  * by the underlying block device.
1360  */
1361 static void pkt_start_write(struct pktcdvd_device *pd, struct packet_data *pkt)
1362 {
1363         struct bio *bio;
1364         int f;
1365         int frames_write;
1366         struct bio_vec *bvec = pkt->w_bio->bi_io_vec;
1367
1368         for (f = 0; f < pkt->frames; f++) {
1369                 bvec[f].bv_page = pkt->pages[(f * CD_FRAMESIZE) / PAGE_SIZE];
1370                 bvec[f].bv_offset = (f * CD_FRAMESIZE) % PAGE_SIZE;
1371         }
1372
1373         /*
1374          * Fill-in bvec with data from orig_bios.
1375          */
1376         frames_write = 0;
1377         spin_lock(&pkt->lock);
1378         bio_list_for_each(bio, &pkt->orig_bios) {
1379                 int segment = bio->bi_idx;
1380                 int src_offs = 0;
1381                 int first_frame = (bio->bi_sector - pkt->sector) / (CD_FRAMESIZE >> 9);
1382                 int num_frames = bio->bi_size / CD_FRAMESIZE;
1383                 BUG_ON(first_frame < 0);
1384                 BUG_ON(first_frame + num_frames > pkt->frames);
1385                 for (f = first_frame; f < first_frame + num_frames; f++) {
1386                         struct bio_vec *src_bvl = bio_iovec_idx(bio, segment);
1387
1388                         while (src_offs >= src_bvl->bv_len) {
1389                                 src_offs -= src_bvl->bv_len;
1390                                 segment++;
1391                                 BUG_ON(segment >= bio->bi_vcnt);
1392                                 src_bvl = bio_iovec_idx(bio, segment);
1393                         }
1394
1395                         if (src_bvl->bv_len - src_offs >= CD_FRAMESIZE) {
1396                                 bvec[f].bv_page = src_bvl->bv_page;
1397                                 bvec[f].bv_offset = src_bvl->bv_offset + src_offs;
1398                         } else {
1399                                 pkt_copy_bio_data(bio, segment, src_offs,
1400                                                   bvec[f].bv_page, bvec[f].bv_offset);
1401                         }
1402                         src_offs += CD_FRAMESIZE;
1403                         frames_write++;
1404                 }
1405         }
1406         pkt_set_state(pkt, PACKET_WRITE_WAIT_STATE);
1407         spin_unlock(&pkt->lock);
1408
1409         VPRINTK("pkt_start_write: Writing %d frames for zone %llx\n",
1410                 frames_write, (unsigned long long)pkt->sector);
1411         BUG_ON(frames_write != pkt->write_size);
1412
1413         if (test_bit(PACKET_MERGE_SEGS, &pd->flags) || (pkt->write_size < pkt->frames)) {
1414                 pkt_make_local_copy(pkt, bvec);
1415                 pkt->cache_valid = 1;
1416         } else {
1417                 pkt->cache_valid = 0;
1418         }
1419
1420         /* Start the write request */
1421         bio_init(pkt->w_bio);
1422         pkt->w_bio->bi_max_vecs = PACKET_MAX_SIZE;
1423         pkt->w_bio->bi_sector = pkt->sector;
1424         pkt->w_bio->bi_bdev = pd->bdev;
1425         pkt->w_bio->bi_end_io = pkt_end_io_packet_write;
1426         pkt->w_bio->bi_private = pkt;
1427         pkt->w_bio->bi_io_vec = bvec;
1428         pkt->w_bio->bi_destructor = pkt_bio_destructor;
1429         for (f = 0; f < pkt->frames; f++)
1430                 if (!bio_add_page(pkt->w_bio, bvec[f].bv_page, CD_FRAMESIZE, bvec[f].bv_offset))
1431                         BUG();
1432         VPRINTK(DRIVER_NAME": vcnt=%d\n", pkt->w_bio->bi_vcnt);
1433
1434         atomic_set(&pkt->io_wait, 1);
1435         pkt->w_bio->bi_rw = WRITE;
1436         pkt_queue_bio(pd, pkt->w_bio);
1437 }
1438
1439 static void pkt_finish_packet(struct packet_data *pkt, int uptodate)
1440 {
1441         struct bio *bio;
1442
1443         if (!uptodate)
1444                 pkt->cache_valid = 0;
1445
1446         /* Finish all bios corresponding to this packet */
1447         while ((bio = bio_list_pop(&pkt->orig_bios)))
1448                 bio_endio(bio, uptodate ? 0 : -EIO);
1449 }
1450
1451 static void pkt_run_state_machine(struct pktcdvd_device *pd, struct packet_data *pkt)
1452 {
1453         int uptodate;
1454
1455         VPRINTK("run_state_machine: pkt %d\n", pkt->id);
1456
1457         for (;;) {
1458                 switch (pkt->state) {
1459                 case PACKET_WAITING_STATE:
1460                         if ((pkt->write_size < pkt->frames) && (pkt->sleep_time > 0))
1461                                 return;
1462
1463                         pkt->sleep_time = 0;
1464                         pkt_gather_data(pd, pkt);
1465                         pkt_set_state(pkt, PACKET_READ_WAIT_STATE);
1466                         break;
1467
1468                 case PACKET_READ_WAIT_STATE:
1469                         if (atomic_read(&pkt->io_wait) > 0)
1470                                 return;
1471
1472                         if (atomic_read(&pkt->io_errors) > 0) {
1473                                 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1474                         } else {
1475                                 pkt_start_write(pd, pkt);
1476                         }
1477                         break;
1478
1479                 case PACKET_WRITE_WAIT_STATE:
1480                         if (atomic_read(&pkt->io_wait) > 0)
1481                                 return;
1482
1483                         if (test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags)) {
1484                                 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1485                         } else {
1486                                 pkt_set_state(pkt, PACKET_RECOVERY_STATE);
1487                         }
1488                         break;
1489
1490                 case PACKET_RECOVERY_STATE:
1491                         if (pkt_start_recovery(pkt)) {
1492                                 pkt_start_write(pd, pkt);
1493                         } else {
1494                                 VPRINTK("No recovery possible\n");
1495                                 pkt_set_state(pkt, PACKET_FINISHED_STATE);
1496                         }
1497                         break;
1498
1499                 case PACKET_FINISHED_STATE:
1500                         uptodate = test_bit(BIO_UPTODATE, &pkt->w_bio->bi_flags);
1501                         pkt_finish_packet(pkt, uptodate);
1502                         return;
1503
1504                 default:
1505                         BUG();
1506                         break;
1507                 }
1508         }
1509 }
1510
1511 static void pkt_handle_packets(struct pktcdvd_device *pd)
1512 {
1513         struct packet_data *pkt, *next;
1514
1515         VPRINTK("pkt_handle_packets\n");
1516
1517         /*
1518          * Run state machine for active packets
1519          */
1520         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1521                 if (atomic_read(&pkt->run_sm) > 0) {
1522                         atomic_set(&pkt->run_sm, 0);
1523                         pkt_run_state_machine(pd, pkt);
1524                 }
1525         }
1526
1527         /*
1528          * Move no longer active packets to the free list
1529          */
1530         spin_lock(&pd->cdrw.active_list_lock);
1531         list_for_each_entry_safe(pkt, next, &pd->cdrw.pkt_active_list, list) {
1532                 if (pkt->state == PACKET_FINISHED_STATE) {
1533                         list_del(&pkt->list);
1534                         pkt_put_packet_data(pd, pkt);
1535                         pkt_set_state(pkt, PACKET_IDLE_STATE);
1536                         atomic_set(&pd->scan_queue, 1);
1537                 }
1538         }
1539         spin_unlock(&pd->cdrw.active_list_lock);
1540 }
1541
1542 static void pkt_count_states(struct pktcdvd_device *pd, int *states)
1543 {
1544         struct packet_data *pkt;
1545         int i;
1546
1547         for (i = 0; i < PACKET_NUM_STATES; i++)
1548                 states[i] = 0;
1549
1550         spin_lock(&pd->cdrw.active_list_lock);
1551         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1552                 states[pkt->state]++;
1553         }
1554         spin_unlock(&pd->cdrw.active_list_lock);
1555 }
1556
1557 /*
1558  * kcdrwd is woken up when writes have been queued for one of our
1559  * registered devices
1560  */
1561 static int kcdrwd(void *foobar)
1562 {
1563         struct pktcdvd_device *pd = foobar;
1564         struct packet_data *pkt;
1565         long min_sleep_time, residue;
1566
1567         set_user_nice(current, -20);
1568         set_freezable();
1569
1570         for (;;) {
1571                 DECLARE_WAITQUEUE(wait, current);
1572
1573                 /*
1574                  * Wait until there is something to do
1575                  */
1576                 add_wait_queue(&pd->wqueue, &wait);
1577                 for (;;) {
1578                         set_current_state(TASK_INTERRUPTIBLE);
1579
1580                         /* Check if we need to run pkt_handle_queue */
1581                         if (atomic_read(&pd->scan_queue) > 0)
1582                                 goto work_to_do;
1583
1584                         /* Check if we need to run the state machine for some packet */
1585                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1586                                 if (atomic_read(&pkt->run_sm) > 0)
1587                                         goto work_to_do;
1588                         }
1589
1590                         /* Check if we need to process the iosched queues */
1591                         if (atomic_read(&pd->iosched.attention) != 0)
1592                                 goto work_to_do;
1593
1594                         /* Otherwise, go to sleep */
1595                         if (PACKET_DEBUG > 1) {
1596                                 int states[PACKET_NUM_STATES];
1597                                 pkt_count_states(pd, states);
1598                                 VPRINTK("kcdrwd: i:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
1599                                         states[0], states[1], states[2], states[3],
1600                                         states[4], states[5]);
1601                         }
1602
1603                         min_sleep_time = MAX_SCHEDULE_TIMEOUT;
1604                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1605                                 if (pkt->sleep_time && pkt->sleep_time < min_sleep_time)
1606                                         min_sleep_time = pkt->sleep_time;
1607                         }
1608
1609                         VPRINTK("kcdrwd: sleeping\n");
1610                         residue = schedule_timeout(min_sleep_time);
1611                         VPRINTK("kcdrwd: wake up\n");
1612
1613                         /* make swsusp happy with our thread */
1614                         try_to_freeze();
1615
1616                         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
1617                                 if (!pkt->sleep_time)
1618                                         continue;
1619                                 pkt->sleep_time -= min_sleep_time - residue;
1620                                 if (pkt->sleep_time <= 0) {
1621                                         pkt->sleep_time = 0;
1622                                         atomic_inc(&pkt->run_sm);
1623                                 }
1624                         }
1625
1626                         if (kthread_should_stop())
1627                                 break;
1628                 }
1629 work_to_do:
1630                 set_current_state(TASK_RUNNING);
1631                 remove_wait_queue(&pd->wqueue, &wait);
1632
1633                 if (kthread_should_stop())
1634                         break;
1635
1636                 /*
1637                  * if pkt_handle_queue returns true, we can queue
1638                  * another request.
1639                  */
1640                 while (pkt_handle_queue(pd))
1641                         ;
1642
1643                 /*
1644                  * Handle packet state machine
1645                  */
1646                 pkt_handle_packets(pd);
1647
1648                 /*
1649                  * Handle iosched queues
1650                  */
1651                 pkt_iosched_process_queue(pd);
1652         }
1653
1654         return 0;
1655 }
1656
1657 static void pkt_print_settings(struct pktcdvd_device *pd)
1658 {
1659         printk(DRIVER_NAME": %s packets, ", pd->settings.fp ? "Fixed" : "Variable");
1660         printk("%u blocks, ", pd->settings.size >> 2);
1661         printk("Mode-%c disc\n", pd->settings.block_mode == 8 ? '1' : '2');
1662 }
1663
1664 static int pkt_mode_sense(struct pktcdvd_device *pd, struct packet_command *cgc, int page_code, int page_control)
1665 {
1666         memset(cgc->cmd, 0, sizeof(cgc->cmd));
1667
1668         cgc->cmd[0] = GPCMD_MODE_SENSE_10;
1669         cgc->cmd[2] = page_code | (page_control << 6);
1670         cgc->cmd[7] = cgc->buflen >> 8;
1671         cgc->cmd[8] = cgc->buflen & 0xff;
1672         cgc->data_direction = CGC_DATA_READ;
1673         return pkt_generic_packet(pd, cgc);
1674 }
1675
1676 static int pkt_mode_select(struct pktcdvd_device *pd, struct packet_command *cgc)
1677 {
1678         memset(cgc->cmd, 0, sizeof(cgc->cmd));
1679         memset(cgc->buffer, 0, 2);
1680         cgc->cmd[0] = GPCMD_MODE_SELECT_10;
1681         cgc->cmd[1] = 0x10;             /* PF */
1682         cgc->cmd[7] = cgc->buflen >> 8;
1683         cgc->cmd[8] = cgc->buflen & 0xff;
1684         cgc->data_direction = CGC_DATA_WRITE;
1685         return pkt_generic_packet(pd, cgc);
1686 }
1687
1688 static int pkt_get_disc_info(struct pktcdvd_device *pd, disc_information *di)
1689 {
1690         struct packet_command cgc;
1691         int ret;
1692
1693         /* set up command and get the disc info */
1694         init_cdrom_command(&cgc, di, sizeof(*di), CGC_DATA_READ);
1695         cgc.cmd[0] = GPCMD_READ_DISC_INFO;
1696         cgc.cmd[8] = cgc.buflen = 2;
1697         cgc.quiet = 1;
1698
1699         if ((ret = pkt_generic_packet(pd, &cgc)))
1700                 return ret;
1701
1702         /* not all drives have the same disc_info length, so requeue
1703          * packet with the length the drive tells us it can supply
1704          */
1705         cgc.buflen = be16_to_cpu(di->disc_information_length) +
1706                      sizeof(di->disc_information_length);
1707
1708         if (cgc.buflen > sizeof(disc_information))
1709                 cgc.buflen = sizeof(disc_information);
1710
1711         cgc.cmd[8] = cgc.buflen;
1712         return pkt_generic_packet(pd, &cgc);
1713 }
1714
1715 static int pkt_get_track_info(struct pktcdvd_device *pd, __u16 track, __u8 type, track_information *ti)
1716 {
1717         struct packet_command cgc;
1718         int ret;
1719
1720         init_cdrom_command(&cgc, ti, 8, CGC_DATA_READ);
1721         cgc.cmd[0] = GPCMD_READ_TRACK_RZONE_INFO;
1722         cgc.cmd[1] = type & 3;
1723         cgc.cmd[4] = (track & 0xff00) >> 8;
1724         cgc.cmd[5] = track & 0xff;
1725         cgc.cmd[8] = 8;
1726         cgc.quiet = 1;
1727
1728         if ((ret = pkt_generic_packet(pd, &cgc)))
1729                 return ret;
1730
1731         cgc.buflen = be16_to_cpu(ti->track_information_length) +
1732                      sizeof(ti->track_information_length);
1733
1734         if (cgc.buflen > sizeof(track_information))
1735                 cgc.buflen = sizeof(track_information);
1736
1737         cgc.cmd[8] = cgc.buflen;
1738         return pkt_generic_packet(pd, &cgc);
1739 }
1740
1741 static noinline_for_stack int pkt_get_last_written(struct pktcdvd_device *pd,
1742                                                 long *last_written)
1743 {
1744         disc_information di;
1745         track_information ti;
1746         __u32 last_track;
1747         int ret = -1;
1748
1749         if ((ret = pkt_get_disc_info(pd, &di)))
1750                 return ret;
1751
1752         last_track = (di.last_track_msb << 8) | di.last_track_lsb;
1753         if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1754                 return ret;
1755
1756         /* if this track is blank, try the previous. */
1757         if (ti.blank) {
1758                 last_track--;
1759                 if ((ret = pkt_get_track_info(pd, last_track, 1, &ti)))
1760                         return ret;
1761         }
1762
1763         /* if last recorded field is valid, return it. */
1764         if (ti.lra_v) {
1765                 *last_written = be32_to_cpu(ti.last_rec_address);
1766         } else {
1767                 /* make it up instead */
1768                 *last_written = be32_to_cpu(ti.track_start) +
1769                                 be32_to_cpu(ti.track_size);
1770                 if (ti.free_blocks)
1771                         *last_written -= (be32_to_cpu(ti.free_blocks) + 7);
1772         }
1773         return 0;
1774 }
1775
1776 /*
1777  * write mode select package based on pd->settings
1778  */
1779 static noinline_for_stack int pkt_set_write_settings(struct pktcdvd_device *pd)
1780 {
1781         struct packet_command cgc;
1782         struct request_sense sense;
1783         write_param_page *wp;
1784         char buffer[128];
1785         int ret, size;
1786
1787         /* doesn't apply to DVD+RW or DVD-RAM */
1788         if ((pd->mmc3_profile == 0x1a) || (pd->mmc3_profile == 0x12))
1789                 return 0;
1790
1791         memset(buffer, 0, sizeof(buffer));
1792         init_cdrom_command(&cgc, buffer, sizeof(*wp), CGC_DATA_READ);
1793         cgc.sense = &sense;
1794         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1795                 pkt_dump_sense(&cgc);
1796                 return ret;
1797         }
1798
1799         size = 2 + ((buffer[0] << 8) | (buffer[1] & 0xff));
1800         pd->mode_offset = (buffer[6] << 8) | (buffer[7] & 0xff);
1801         if (size > sizeof(buffer))
1802                 size = sizeof(buffer);
1803
1804         /*
1805          * now get it all
1806          */
1807         init_cdrom_command(&cgc, buffer, size, CGC_DATA_READ);
1808         cgc.sense = &sense;
1809         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WRITE_PARMS_PAGE, 0))) {
1810                 pkt_dump_sense(&cgc);
1811                 return ret;
1812         }
1813
1814         /*
1815          * write page is offset header + block descriptor length
1816          */
1817         wp = (write_param_page *) &buffer[sizeof(struct mode_page_header) + pd->mode_offset];
1818
1819         wp->fp = pd->settings.fp;
1820         wp->track_mode = pd->settings.track_mode;
1821         wp->write_type = pd->settings.write_type;
1822         wp->data_block_type = pd->settings.block_mode;
1823
1824         wp->multi_session = 0;
1825
1826 #ifdef PACKET_USE_LS
1827         wp->link_size = 7;
1828         wp->ls_v = 1;
1829 #endif
1830
1831         if (wp->data_block_type == PACKET_BLOCK_MODE1) {
1832                 wp->session_format = 0;
1833                 wp->subhdr2 = 0x20;
1834         } else if (wp->data_block_type == PACKET_BLOCK_MODE2) {
1835                 wp->session_format = 0x20;
1836                 wp->subhdr2 = 8;
1837 #if 0
1838                 wp->mcn[0] = 0x80;
1839                 memcpy(&wp->mcn[1], PACKET_MCN, sizeof(wp->mcn) - 1);
1840 #endif
1841         } else {
1842                 /*
1843                  * paranoia
1844                  */
1845                 printk(DRIVER_NAME": write mode wrong %d\n", wp->data_block_type);
1846                 return 1;
1847         }
1848         wp->packet_size = cpu_to_be32(pd->settings.size >> 2);
1849
1850         cgc.buflen = cgc.cmd[8] = size;
1851         if ((ret = pkt_mode_select(pd, &cgc))) {
1852                 pkt_dump_sense(&cgc);
1853                 return ret;
1854         }
1855
1856         pkt_print_settings(pd);
1857         return 0;
1858 }
1859
1860 /*
1861  * 1 -- we can write to this track, 0 -- we can't
1862  */
1863 static int pkt_writable_track(struct pktcdvd_device *pd, track_information *ti)
1864 {
1865         switch (pd->mmc3_profile) {
1866                 case 0x1a: /* DVD+RW */
1867                 case 0x12: /* DVD-RAM */
1868                         /* The track is always writable on DVD+RW/DVD-RAM */
1869                         return 1;
1870                 default:
1871                         break;
1872         }
1873
1874         if (!ti->packet || !ti->fp)
1875                 return 0;
1876
1877         /*
1878          * "good" settings as per Mt Fuji.
1879          */
1880         if (ti->rt == 0 && ti->blank == 0)
1881                 return 1;
1882
1883         if (ti->rt == 0 && ti->blank == 1)
1884                 return 1;
1885
1886         if (ti->rt == 1 && ti->blank == 0)
1887                 return 1;
1888
1889         printk(DRIVER_NAME": bad state %d-%d-%d\n", ti->rt, ti->blank, ti->packet);
1890         return 0;
1891 }
1892
1893 /*
1894  * 1 -- we can write to this disc, 0 -- we can't
1895  */
1896 static int pkt_writable_disc(struct pktcdvd_device *pd, disc_information *di)
1897 {
1898         switch (pd->mmc3_profile) {
1899                 case 0x0a: /* CD-RW */
1900                 case 0xffff: /* MMC3 not supported */
1901                         break;
1902                 case 0x1a: /* DVD+RW */
1903                 case 0x13: /* DVD-RW */
1904                 case 0x12: /* DVD-RAM */
1905                         return 1;
1906                 default:
1907                         VPRINTK(DRIVER_NAME": Wrong disc profile (%x)\n", pd->mmc3_profile);
1908                         return 0;
1909         }
1910
1911         /*
1912          * for disc type 0xff we should probably reserve a new track.
1913          * but i'm not sure, should we leave this to user apps? probably.
1914          */
1915         if (di->disc_type == 0xff) {
1916                 printk(DRIVER_NAME": Unknown disc. No track?\n");
1917                 return 0;
1918         }
1919
1920         if (di->disc_type != 0x20 && di->disc_type != 0) {
1921                 printk(DRIVER_NAME": Wrong disc type (%x)\n", di->disc_type);
1922                 return 0;
1923         }
1924
1925         if (di->erasable == 0) {
1926                 printk(DRIVER_NAME": Disc not erasable\n");
1927                 return 0;
1928         }
1929
1930         if (di->border_status == PACKET_SESSION_RESERVED) {
1931                 printk(DRIVER_NAME": Can't write to last track (reserved)\n");
1932                 return 0;
1933         }
1934
1935         return 1;
1936 }
1937
1938 static noinline_for_stack int pkt_probe_settings(struct pktcdvd_device *pd)
1939 {
1940         struct packet_command cgc;
1941         unsigned char buf[12];
1942         disc_information di;
1943         track_information ti;
1944         int ret, track;
1945
1946         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
1947         cgc.cmd[0] = GPCMD_GET_CONFIGURATION;
1948         cgc.cmd[8] = 8;
1949         ret = pkt_generic_packet(pd, &cgc);
1950         pd->mmc3_profile = ret ? 0xffff : buf[6] << 8 | buf[7];
1951
1952         memset(&di, 0, sizeof(disc_information));
1953         memset(&ti, 0, sizeof(track_information));
1954
1955         if ((ret = pkt_get_disc_info(pd, &di))) {
1956                 printk("failed get_disc\n");
1957                 return ret;
1958         }
1959
1960         if (!pkt_writable_disc(pd, &di))
1961                 return -EROFS;
1962
1963         pd->type = di.erasable ? PACKET_CDRW : PACKET_CDR;
1964
1965         track = 1; /* (di.last_track_msb << 8) | di.last_track_lsb; */
1966         if ((ret = pkt_get_track_info(pd, track, 1, &ti))) {
1967                 printk(DRIVER_NAME": failed get_track\n");
1968                 return ret;
1969         }
1970
1971         if (!pkt_writable_track(pd, &ti)) {
1972                 printk(DRIVER_NAME": can't write to this track\n");
1973                 return -EROFS;
1974         }
1975
1976         /*
1977          * we keep packet size in 512 byte units, makes it easier to
1978          * deal with request calculations.
1979          */
1980         pd->settings.size = be32_to_cpu(ti.fixed_packet_size) << 2;
1981         if (pd->settings.size == 0) {
1982                 printk(DRIVER_NAME": detected zero packet size!\n");
1983                 return -ENXIO;
1984         }
1985         if (pd->settings.size > PACKET_MAX_SECTORS) {
1986                 printk(DRIVER_NAME": packet size is too big\n");
1987                 return -EROFS;
1988         }
1989         pd->settings.fp = ti.fp;
1990         pd->offset = (be32_to_cpu(ti.track_start) << 2) & (pd->settings.size - 1);
1991
1992         if (ti.nwa_v) {
1993                 pd->nwa = be32_to_cpu(ti.next_writable);
1994                 set_bit(PACKET_NWA_VALID, &pd->flags);
1995         }
1996
1997         /*
1998          * in theory we could use lra on -RW media as well and just zero
1999          * blocks that haven't been written yet, but in practice that
2000          * is just a no-go. we'll use that for -R, naturally.
2001          */
2002         if (ti.lra_v) {
2003                 pd->lra = be32_to_cpu(ti.last_rec_address);
2004                 set_bit(PACKET_LRA_VALID, &pd->flags);
2005         } else {
2006                 pd->lra = 0xffffffff;
2007                 set_bit(PACKET_LRA_VALID, &pd->flags);
2008         }
2009
2010         /*
2011          * fine for now
2012          */
2013         pd->settings.link_loss = 7;
2014         pd->settings.write_type = 0;    /* packet */
2015         pd->settings.track_mode = ti.track_mode;
2016
2017         /*
2018          * mode1 or mode2 disc
2019          */
2020         switch (ti.data_mode) {
2021                 case PACKET_MODE1:
2022                         pd->settings.block_mode = PACKET_BLOCK_MODE1;
2023                         break;
2024                 case PACKET_MODE2:
2025                         pd->settings.block_mode = PACKET_BLOCK_MODE2;
2026                         break;
2027                 default:
2028                         printk(DRIVER_NAME": unknown data mode\n");
2029                         return -EROFS;
2030         }
2031         return 0;
2032 }
2033
2034 /*
2035  * enable/disable write caching on drive
2036  */
2037 static noinline_for_stack int pkt_write_caching(struct pktcdvd_device *pd,
2038                                                 int set)
2039 {
2040         struct packet_command cgc;
2041         struct request_sense sense;
2042         unsigned char buf[64];
2043         int ret;
2044
2045         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_READ);
2046         cgc.sense = &sense;
2047         cgc.buflen = pd->mode_offset + 12;
2048
2049         /*
2050          * caching mode page might not be there, so quiet this command
2051          */
2052         cgc.quiet = 1;
2053
2054         if ((ret = pkt_mode_sense(pd, &cgc, GPMODE_WCACHING_PAGE, 0)))
2055                 return ret;
2056
2057         buf[pd->mode_offset + 10] |= (!!set << 2);
2058
2059         cgc.buflen = cgc.cmd[8] = 2 + ((buf[0] << 8) | (buf[1] & 0xff));
2060         ret = pkt_mode_select(pd, &cgc);
2061         if (ret) {
2062                 printk(DRIVER_NAME": write caching control failed\n");
2063                 pkt_dump_sense(&cgc);
2064         } else if (!ret && set)
2065                 printk(DRIVER_NAME": enabled write caching on %s\n", pd->name);
2066         return ret;
2067 }
2068
2069 static int pkt_lock_door(struct pktcdvd_device *pd, int lockflag)
2070 {
2071         struct packet_command cgc;
2072
2073         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2074         cgc.cmd[0] = GPCMD_PREVENT_ALLOW_MEDIUM_REMOVAL;
2075         cgc.cmd[4] = lockflag ? 1 : 0;
2076         return pkt_generic_packet(pd, &cgc);
2077 }
2078
2079 /*
2080  * Returns drive maximum write speed
2081  */
2082 static noinline_for_stack int pkt_get_max_speed(struct pktcdvd_device *pd,
2083                                                 unsigned *write_speed)
2084 {
2085         struct packet_command cgc;
2086         struct request_sense sense;
2087         unsigned char buf[256+18];
2088         unsigned char *cap_buf;
2089         int ret, offset;
2090
2091         cap_buf = &buf[sizeof(struct mode_page_header) + pd->mode_offset];
2092         init_cdrom_command(&cgc, buf, sizeof(buf), CGC_DATA_UNKNOWN);
2093         cgc.sense = &sense;
2094
2095         ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2096         if (ret) {
2097                 cgc.buflen = pd->mode_offset + cap_buf[1] + 2 +
2098                              sizeof(struct mode_page_header);
2099                 ret = pkt_mode_sense(pd, &cgc, GPMODE_CAPABILITIES_PAGE, 0);
2100                 if (ret) {
2101                         pkt_dump_sense(&cgc);
2102                         return ret;
2103                 }
2104         }
2105
2106         offset = 20;                        /* Obsoleted field, used by older drives */
2107         if (cap_buf[1] >= 28)
2108                 offset = 28;                /* Current write speed selected */
2109         if (cap_buf[1] >= 30) {
2110                 /* If the drive reports at least one "Logical Unit Write
2111                  * Speed Performance Descriptor Block", use the information
2112                  * in the first block. (contains the highest speed)
2113                  */
2114                 int num_spdb = (cap_buf[30] << 8) + cap_buf[31];
2115                 if (num_spdb > 0)
2116                         offset = 34;
2117         }
2118
2119         *write_speed = (cap_buf[offset] << 8) | cap_buf[offset + 1];
2120         return 0;
2121 }
2122
2123 /* These tables from cdrecord - I don't have orange book */
2124 /* standard speed CD-RW (1-4x) */
2125 static char clv_to_speed[16] = {
2126         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2127            0, 2, 4, 6, 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2128 };
2129 /* high speed CD-RW (-10x) */
2130 static char hs_clv_to_speed[16] = {
2131         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2132            0, 2, 4, 6, 10, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
2133 };
2134 /* ultra high speed CD-RW */
2135 static char us_clv_to_speed[16] = {
2136         /* 0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 */
2137            0, 2, 4, 8, 0, 0,16, 0,24,32,40,48, 0, 0, 0, 0
2138 };
2139
2140 /*
2141  * reads the maximum media speed from ATIP
2142  */
2143 static noinline_for_stack int pkt_media_speed(struct pktcdvd_device *pd,
2144                                                 unsigned *speed)
2145 {
2146         struct packet_command cgc;
2147         struct request_sense sense;
2148         unsigned char buf[64];
2149         unsigned int size, st, sp;
2150         int ret;
2151
2152         init_cdrom_command(&cgc, buf, 2, CGC_DATA_READ);
2153         cgc.sense = &sense;
2154         cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2155         cgc.cmd[1] = 2;
2156         cgc.cmd[2] = 4; /* READ ATIP */
2157         cgc.cmd[8] = 2;
2158         ret = pkt_generic_packet(pd, &cgc);
2159         if (ret) {
2160                 pkt_dump_sense(&cgc);
2161                 return ret;
2162         }
2163         size = ((unsigned int) buf[0]<<8) + buf[1] + 2;
2164         if (size > sizeof(buf))
2165                 size = sizeof(buf);
2166
2167         init_cdrom_command(&cgc, buf, size, CGC_DATA_READ);
2168         cgc.sense = &sense;
2169         cgc.cmd[0] = GPCMD_READ_TOC_PMA_ATIP;
2170         cgc.cmd[1] = 2;
2171         cgc.cmd[2] = 4;
2172         cgc.cmd[8] = size;
2173         ret = pkt_generic_packet(pd, &cgc);
2174         if (ret) {
2175                 pkt_dump_sense(&cgc);
2176                 return ret;
2177         }
2178
2179         if (!(buf[6] & 0x40)) {
2180                 printk(DRIVER_NAME": Disc type is not CD-RW\n");
2181                 return 1;
2182         }
2183         if (!(buf[6] & 0x4)) {
2184                 printk(DRIVER_NAME": A1 values on media are not valid, maybe not CDRW?\n");
2185                 return 1;
2186         }
2187
2188         st = (buf[6] >> 3) & 0x7; /* disc sub-type */
2189
2190         sp = buf[16] & 0xf; /* max speed from ATIP A1 field */
2191
2192         /* Info from cdrecord */
2193         switch (st) {
2194                 case 0: /* standard speed */
2195                         *speed = clv_to_speed[sp];
2196                         break;
2197                 case 1: /* high speed */
2198                         *speed = hs_clv_to_speed[sp];
2199                         break;
2200                 case 2: /* ultra high speed */
2201                         *speed = us_clv_to_speed[sp];
2202                         break;
2203                 default:
2204                         printk(DRIVER_NAME": Unknown disc sub-type %d\n",st);
2205                         return 1;
2206         }
2207         if (*speed) {
2208                 printk(DRIVER_NAME": Max. media speed: %d\n",*speed);
2209                 return 0;
2210         } else {
2211                 printk(DRIVER_NAME": Unknown speed %d for sub-type %d\n",sp,st);
2212                 return 1;
2213         }
2214 }
2215
2216 static noinline_for_stack int pkt_perform_opc(struct pktcdvd_device *pd)
2217 {
2218         struct packet_command cgc;
2219         struct request_sense sense;
2220         int ret;
2221
2222         VPRINTK(DRIVER_NAME": Performing OPC\n");
2223
2224         init_cdrom_command(&cgc, NULL, 0, CGC_DATA_NONE);
2225         cgc.sense = &sense;
2226         cgc.timeout = 60*HZ;
2227         cgc.cmd[0] = GPCMD_SEND_OPC;
2228         cgc.cmd[1] = 1;
2229         if ((ret = pkt_generic_packet(pd, &cgc)))
2230                 pkt_dump_sense(&cgc);
2231         return ret;
2232 }
2233
2234 static int pkt_open_write(struct pktcdvd_device *pd)
2235 {
2236         int ret;
2237         unsigned int write_speed, media_write_speed, read_speed;
2238
2239         if ((ret = pkt_probe_settings(pd))) {
2240                 VPRINTK(DRIVER_NAME": %s failed probe\n", pd->name);
2241                 return ret;
2242         }
2243
2244         if ((ret = pkt_set_write_settings(pd))) {
2245                 DPRINTK(DRIVER_NAME": %s failed saving write settings\n", pd->name);
2246                 return -EIO;
2247         }
2248
2249         pkt_write_caching(pd, USE_WCACHING);
2250
2251         if ((ret = pkt_get_max_speed(pd, &write_speed)))
2252                 write_speed = 16 * 177;
2253         switch (pd->mmc3_profile) {
2254                 case 0x13: /* DVD-RW */
2255                 case 0x1a: /* DVD+RW */
2256                 case 0x12: /* DVD-RAM */
2257                         DPRINTK(DRIVER_NAME": write speed %ukB/s\n", write_speed);
2258                         break;
2259                 default:
2260                         if ((ret = pkt_media_speed(pd, &media_write_speed)))
2261                                 media_write_speed = 16;
2262                         write_speed = min(write_speed, media_write_speed * 177);
2263                         DPRINTK(DRIVER_NAME": write speed %ux\n", write_speed / 176);
2264                         break;
2265         }
2266         read_speed = write_speed;
2267
2268         if ((ret = pkt_set_speed(pd, write_speed, read_speed))) {
2269                 DPRINTK(DRIVER_NAME": %s couldn't set write speed\n", pd->name);
2270                 return -EIO;
2271         }
2272         pd->write_speed = write_speed;
2273         pd->read_speed = read_speed;
2274
2275         if ((ret = pkt_perform_opc(pd))) {
2276                 DPRINTK(DRIVER_NAME": %s Optimum Power Calibration failed\n", pd->name);
2277         }
2278
2279         return 0;
2280 }
2281
2282 /*
2283  * called at open time.
2284  */
2285 static int pkt_open_dev(struct pktcdvd_device *pd, fmode_t write)
2286 {
2287         int ret;
2288         long lba;
2289         struct request_queue *q;
2290
2291         /*
2292          * We need to re-open the cdrom device without O_NONBLOCK to be able
2293          * to read/write from/to it. It is already opened in O_NONBLOCK mode
2294          * so bdget() can't fail.
2295          */
2296         bdget(pd->bdev->bd_dev);
2297         if ((ret = blkdev_get(pd->bdev, FMODE_READ | FMODE_EXCL, pd)))
2298                 goto out;
2299
2300         if ((ret = pkt_get_last_written(pd, &lba))) {
2301                 printk(DRIVER_NAME": pkt_get_last_written failed\n");
2302                 goto out_putdev;
2303         }
2304
2305         set_capacity(pd->disk, lba << 2);
2306         set_capacity(pd->bdev->bd_disk, lba << 2);
2307         bd_set_size(pd->bdev, (loff_t)lba << 11);
2308
2309         q = bdev_get_queue(pd->bdev);
2310         if (write) {
2311                 if ((ret = pkt_open_write(pd)))
2312                         goto out_putdev;
2313                 /*
2314                  * Some CDRW drives can not handle writes larger than one packet,
2315                  * even if the size is a multiple of the packet size.
2316                  */
2317                 spin_lock_irq(q->queue_lock);
2318                 blk_queue_max_hw_sectors(q, pd->settings.size);
2319                 spin_unlock_irq(q->queue_lock);
2320                 set_bit(PACKET_WRITABLE, &pd->flags);
2321         } else {
2322                 pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2323                 clear_bit(PACKET_WRITABLE, &pd->flags);
2324         }
2325
2326         if ((ret = pkt_set_segment_merging(pd, q)))
2327                 goto out_putdev;
2328
2329         if (write) {
2330                 if (!pkt_grow_pktlist(pd, CONFIG_CDROM_PKTCDVD_BUFFERS)) {
2331                         printk(DRIVER_NAME": not enough memory for buffers\n");
2332                         ret = -ENOMEM;
2333                         goto out_putdev;
2334                 }
2335                 printk(DRIVER_NAME": %lukB available on disc\n", lba << 1);
2336         }
2337
2338         return 0;
2339
2340 out_putdev:
2341         blkdev_put(pd->bdev, FMODE_READ | FMODE_EXCL);
2342 out:
2343         return ret;
2344 }
2345
2346 /*
2347  * called when the device is closed. makes sure that the device flushes
2348  * the internal cache before we close.
2349  */
2350 static void pkt_release_dev(struct pktcdvd_device *pd, int flush)
2351 {
2352         if (flush && pkt_flush_cache(pd))
2353                 DPRINTK(DRIVER_NAME": %s not flushing cache\n", pd->name);
2354
2355         pkt_lock_door(pd, 0);
2356
2357         pkt_set_speed(pd, MAX_SPEED, MAX_SPEED);
2358         blkdev_put(pd->bdev, FMODE_READ | FMODE_EXCL);
2359
2360         pkt_shrink_pktlist(pd);
2361 }
2362
2363 static struct pktcdvd_device *pkt_find_dev_from_minor(unsigned int dev_minor)
2364 {
2365         if (dev_minor >= MAX_WRITERS)
2366                 return NULL;
2367         return pkt_devs[dev_minor];
2368 }
2369
2370 static int pkt_open(struct block_device *bdev, fmode_t mode)
2371 {
2372         struct pktcdvd_device *pd = NULL;
2373         int ret;
2374
2375         VPRINTK(DRIVER_NAME": entering open\n");
2376
2377         mutex_lock(&pktcdvd_mutex);
2378         mutex_lock(&ctl_mutex);
2379         pd = pkt_find_dev_from_minor(MINOR(bdev->bd_dev));
2380         if (!pd) {
2381                 ret = -ENODEV;
2382                 goto out;
2383         }
2384         BUG_ON(pd->refcnt < 0);
2385
2386         pd->refcnt++;
2387         if (pd->refcnt > 1) {
2388                 if ((mode & FMODE_WRITE) &&
2389                     !test_bit(PACKET_WRITABLE, &pd->flags)) {
2390                         ret = -EBUSY;
2391                         goto out_dec;
2392                 }
2393         } else {
2394                 ret = pkt_open_dev(pd, mode & FMODE_WRITE);
2395                 if (ret)
2396                         goto out_dec;
2397                 /*
2398                  * needed here as well, since ext2 (among others) may change
2399                  * the blocksize at mount time
2400                  */
2401                 set_blocksize(bdev, CD_FRAMESIZE);
2402         }
2403
2404         mutex_unlock(&ctl_mutex);
2405         mutex_unlock(&pktcdvd_mutex);
2406         return 0;
2407
2408 out_dec:
2409         pd->refcnt--;
2410 out:
2411         VPRINTK(DRIVER_NAME": failed open (%d)\n", ret);
2412         mutex_unlock(&ctl_mutex);
2413         mutex_unlock(&pktcdvd_mutex);
2414         return ret;
2415 }
2416
2417 static int pkt_close(struct gendisk *disk, fmode_t mode)
2418 {
2419         struct pktcdvd_device *pd = disk->private_data;
2420         int ret = 0;
2421
2422         mutex_lock(&pktcdvd_mutex);
2423         mutex_lock(&ctl_mutex);
2424         pd->refcnt--;
2425         BUG_ON(pd->refcnt < 0);
2426         if (pd->refcnt == 0) {
2427                 int flush = test_bit(PACKET_WRITABLE, &pd->flags);
2428                 pkt_release_dev(pd, flush);
2429         }
2430         mutex_unlock(&ctl_mutex);
2431         mutex_unlock(&pktcdvd_mutex);
2432         return ret;
2433 }
2434
2435
2436 static void pkt_end_io_read_cloned(struct bio *bio, int err)
2437 {
2438         struct packet_stacked_data *psd = bio->bi_private;
2439         struct pktcdvd_device *pd = psd->pd;
2440
2441         bio_put(bio);
2442         bio_endio(psd->bio, err);
2443         mempool_free(psd, psd_pool);
2444         pkt_bio_finished(pd);
2445 }
2446
2447 static void pkt_make_request(struct request_queue *q, struct bio *bio)
2448 {
2449         struct pktcdvd_device *pd;
2450         char b[BDEVNAME_SIZE];
2451         sector_t zone;
2452         struct packet_data *pkt;
2453         int was_empty, blocked_bio;
2454         struct pkt_rb_node *node;
2455
2456         pd = q->queuedata;
2457         if (!pd) {
2458                 printk(DRIVER_NAME": %s incorrect request queue\n", bdevname(bio->bi_bdev, b));
2459                 goto end_io;
2460         }
2461
2462         /*
2463          * Clone READ bios so we can have our own bi_end_io callback.
2464          */
2465         if (bio_data_dir(bio) == READ) {
2466                 struct bio *cloned_bio = bio_clone(bio, GFP_NOIO);
2467                 struct packet_stacked_data *psd = mempool_alloc(psd_pool, GFP_NOIO);
2468
2469                 psd->pd = pd;
2470                 psd->bio = bio;
2471                 cloned_bio->bi_bdev = pd->bdev;
2472                 cloned_bio->bi_private = psd;
2473                 cloned_bio->bi_end_io = pkt_end_io_read_cloned;
2474                 pd->stats.secs_r += bio->bi_size >> 9;
2475                 pkt_queue_bio(pd, cloned_bio);
2476                 return;
2477         }
2478
2479         if (!test_bit(PACKET_WRITABLE, &pd->flags)) {
2480                 printk(DRIVER_NAME": WRITE for ro device %s (%llu)\n",
2481                         pd->name, (unsigned long long)bio->bi_sector);
2482                 goto end_io;
2483         }
2484
2485         if (!bio->bi_size || (bio->bi_size % CD_FRAMESIZE)) {
2486                 printk(DRIVER_NAME": wrong bio size\n");
2487                 goto end_io;
2488         }
2489
2490         blk_queue_bounce(q, &bio);
2491
2492         zone = ZONE(bio->bi_sector, pd);
2493         VPRINTK("pkt_make_request: start = %6llx stop = %6llx\n",
2494                 (unsigned long long)bio->bi_sector,
2495                 (unsigned long long)(bio->bi_sector + bio_sectors(bio)));
2496
2497         /* Check if we have to split the bio */
2498         {
2499                 struct bio_pair *bp;
2500                 sector_t last_zone;
2501                 int first_sectors;
2502
2503                 last_zone = ZONE(bio->bi_sector + bio_sectors(bio) - 1, pd);
2504                 if (last_zone != zone) {
2505                         BUG_ON(last_zone != zone + pd->settings.size);
2506                         first_sectors = last_zone - bio->bi_sector;
2507                         bp = bio_split(bio, first_sectors);
2508                         BUG_ON(!bp);
2509                         pkt_make_request(q, &bp->bio1);
2510                         pkt_make_request(q, &bp->bio2);
2511                         bio_pair_release(bp);
2512                         return;
2513                 }
2514         }
2515
2516         /*
2517          * If we find a matching packet in state WAITING or READ_WAIT, we can
2518          * just append this bio to that packet.
2519          */
2520         spin_lock(&pd->cdrw.active_list_lock);
2521         blocked_bio = 0;
2522         list_for_each_entry(pkt, &pd->cdrw.pkt_active_list, list) {
2523                 if (pkt->sector == zone) {
2524                         spin_lock(&pkt->lock);
2525                         if ((pkt->state == PACKET_WAITING_STATE) ||
2526                             (pkt->state == PACKET_READ_WAIT_STATE)) {
2527                                 bio_list_add(&pkt->orig_bios, bio);
2528                                 pkt->write_size += bio->bi_size / CD_FRAMESIZE;
2529                                 if ((pkt->write_size >= pkt->frames) &&
2530                                     (pkt->state == PACKET_WAITING_STATE)) {
2531                                         atomic_inc(&pkt->run_sm);
2532                                         wake_up(&pd->wqueue);
2533                                 }
2534                                 spin_unlock(&pkt->lock);
2535                                 spin_unlock(&pd->cdrw.active_list_lock);
2536                                 return;
2537                         } else {
2538                                 blocked_bio = 1;
2539                         }
2540                         spin_unlock(&pkt->lock);
2541                 }
2542         }
2543         spin_unlock(&pd->cdrw.active_list_lock);
2544
2545         /*
2546          * Test if there is enough room left in the bio work queue
2547          * (queue size >= congestion on mark).
2548          * If not, wait till the work queue size is below the congestion off mark.
2549          */
2550         spin_lock(&pd->lock);
2551         if (pd->write_congestion_on > 0
2552             && pd->bio_queue_size >= pd->write_congestion_on) {
2553                 set_bdi_congested(&q->backing_dev_info, BLK_RW_ASYNC);
2554                 do {
2555                         spin_unlock(&pd->lock);
2556                         congestion_wait(BLK_RW_ASYNC, HZ);
2557                         spin_lock(&pd->lock);
2558                 } while(pd->bio_queue_size > pd->write_congestion_off);
2559         }
2560         spin_unlock(&pd->lock);
2561
2562         /*
2563          * No matching packet found. Store the bio in the work queue.
2564          */
2565         node = mempool_alloc(pd->rb_pool, GFP_NOIO);
2566         node->bio = bio;
2567         spin_lock(&pd->lock);
2568         BUG_ON(pd->bio_queue_size < 0);
2569         was_empty = (pd->bio_queue_size == 0);
2570         pkt_rbtree_insert(pd, node);
2571         spin_unlock(&pd->lock);
2572
2573         /*
2574          * Wake up the worker thread.
2575          */
2576         atomic_set(&pd->scan_queue, 1);
2577         if (was_empty) {
2578                 /* This wake_up is required for correct operation */
2579                 wake_up(&pd->wqueue);
2580         } else if (!list_empty(&pd->cdrw.pkt_free_list) && !blocked_bio) {
2581                 /*
2582                  * This wake up is not required for correct operation,
2583                  * but improves performance in some cases.
2584                  */
2585                 wake_up(&pd->wqueue);
2586         }
2587         return;
2588 end_io:
2589         bio_io_error(bio);
2590 }
2591
2592
2593
2594 static int pkt_merge_bvec(struct request_queue *q, struct bvec_merge_data *bmd,
2595                           struct bio_vec *bvec)
2596 {
2597         struct pktcdvd_device *pd = q->queuedata;
2598         sector_t zone = ZONE(bmd->bi_sector, pd);
2599         int used = ((bmd->bi_sector - zone) << 9) + bmd->bi_size;
2600         int remaining = (pd->settings.size << 9) - used;
2601         int remaining2;
2602
2603         /*
2604          * A bio <= PAGE_SIZE must be allowed. If it crosses a packet
2605          * boundary, pkt_make_request() will split the bio.
2606          */
2607         remaining2 = PAGE_SIZE - bmd->bi_size;
2608         remaining = max(remaining, remaining2);
2609
2610         BUG_ON(remaining < 0);
2611         return remaining;
2612 }
2613
2614 static void pkt_init_queue(struct pktcdvd_device *pd)
2615 {
2616         struct request_queue *q = pd->disk->queue;
2617
2618         blk_queue_make_request(q, pkt_make_request);
2619         blk_queue_logical_block_size(q, CD_FRAMESIZE);
2620         blk_queue_max_hw_sectors(q, PACKET_MAX_SECTORS);
2621         blk_queue_merge_bvec(q, pkt_merge_bvec);
2622         q->queuedata = pd;
2623 }
2624
2625 static int pkt_seq_show(struct seq_file *m, void *p)
2626 {
2627         struct pktcdvd_device *pd = m->private;
2628         char *msg;
2629         char bdev_buf[BDEVNAME_SIZE];
2630         int states[PACKET_NUM_STATES];
2631
2632         seq_printf(m, "Writer %s mapped to %s:\n", pd->name,
2633                    bdevname(pd->bdev, bdev_buf));
2634
2635         seq_printf(m, "\nSettings:\n");
2636         seq_printf(m, "\tpacket size:\t\t%dkB\n", pd->settings.size / 2);
2637
2638         if (pd->settings.write_type == 0)
2639                 msg = "Packet";
2640         else
2641                 msg = "Unknown";
2642         seq_printf(m, "\twrite type:\t\t%s\n", msg);
2643
2644         seq_printf(m, "\tpacket type:\t\t%s\n", pd->settings.fp ? "Fixed" : "Variable");
2645         seq_printf(m, "\tlink loss:\t\t%d\n", pd->settings.link_loss);
2646
2647         seq_printf(m, "\ttrack mode:\t\t%d\n", pd->settings.track_mode);
2648
2649         if (pd->settings.block_mode == PACKET_BLOCK_MODE1)
2650                 msg = "Mode 1";
2651         else if (pd->settings.block_mode == PACKET_BLOCK_MODE2)
2652                 msg = "Mode 2";
2653         else
2654                 msg = "Unknown";
2655         seq_printf(m, "\tblock mode:\t\t%s\n", msg);
2656
2657         seq_printf(m, "\nStatistics:\n");
2658         seq_printf(m, "\tpackets started:\t%lu\n", pd->stats.pkt_started);
2659         seq_printf(m, "\tpackets ended:\t\t%lu\n", pd->stats.pkt_ended);
2660         seq_printf(m, "\twritten:\t\t%lukB\n", pd->stats.secs_w >> 1);
2661         seq_printf(m, "\tread gather:\t\t%lukB\n", pd->stats.secs_rg >> 1);
2662         seq_printf(m, "\tread:\t\t\t%lukB\n", pd->stats.secs_r >> 1);
2663
2664         seq_printf(m, "\nMisc:\n");
2665         seq_printf(m, "\treference count:\t%d\n", pd->refcnt);
2666         seq_printf(m, "\tflags:\t\t\t0x%lx\n", pd->flags);
2667         seq_printf(m, "\tread speed:\t\t%ukB/s\n", pd->read_speed);
2668         seq_printf(m, "\twrite speed:\t\t%ukB/s\n", pd->write_speed);
2669         seq_printf(m, "\tstart offset:\t\t%lu\n", pd->offset);
2670         seq_printf(m, "\tmode page offset:\t%u\n", pd->mode_offset);
2671
2672         seq_printf(m, "\nQueue state:\n");
2673         seq_printf(m, "\tbios queued:\t\t%d\n", pd->bio_queue_size);
2674         seq_printf(m, "\tbios pending:\t\t%d\n", atomic_read(&pd->cdrw.pending_bios));
2675         seq_printf(m, "\tcurrent sector:\t\t0x%llx\n", (unsigned long long)pd->current_sector);
2676
2677         pkt_count_states(pd, states);
2678         seq_printf(m, "\tstate:\t\t\ti:%d ow:%d rw:%d ww:%d rec:%d fin:%d\n",
2679                    states[0], states[1], states[2], states[3], states[4], states[5]);
2680
2681         seq_printf(m, "\twrite congestion marks:\toff=%d on=%d\n",
2682                         pd->write_congestion_off,
2683                         pd->write_congestion_on);
2684         return 0;
2685 }
2686
2687 static int pkt_seq_open(struct inode *inode, struct file *file)
2688 {
2689         return single_open(file, pkt_seq_show, PDE(inode)->data);
2690 }
2691
2692 static const struct file_operations pkt_proc_fops = {
2693         .open   = pkt_seq_open,
2694         .read   = seq_read,
2695         .llseek = seq_lseek,
2696         .release = single_release
2697 };
2698
2699 static int pkt_new_dev(struct pktcdvd_device *pd, dev_t dev)
2700 {
2701         int i;
2702         int ret = 0;
2703         char b[BDEVNAME_SIZE];
2704         struct block_device *bdev;
2705
2706         if (pd->pkt_dev == dev) {
2707                 printk(DRIVER_NAME": Recursive setup not allowed\n");
2708                 return -EBUSY;
2709         }
2710         for (i = 0; i < MAX_WRITERS; i++) {
2711                 struct pktcdvd_device *pd2 = pkt_devs[i];
2712                 if (!pd2)
2713                         continue;
2714                 if (pd2->bdev->bd_dev == dev) {
2715                         printk(DRIVER_NAME": %s already setup\n", bdevname(pd2->bdev, b));
2716                         return -EBUSY;
2717                 }
2718                 if (pd2->pkt_dev == dev) {
2719                         printk(DRIVER_NAME": Can't chain pktcdvd devices\n");
2720                         return -EBUSY;
2721                 }
2722         }
2723
2724         bdev = bdget(dev);
2725         if (!bdev)
2726                 return -ENOMEM;
2727         ret = blkdev_get(bdev, FMODE_READ | FMODE_NDELAY, NULL);
2728         if (ret)
2729                 return ret;
2730
2731         /* This is safe, since we have a reference from open(). */
2732         __module_get(THIS_MODULE);
2733
2734         pd->bdev = bdev;
2735         set_blocksize(bdev, CD_FRAMESIZE);
2736
2737         pkt_init_queue(pd);
2738
2739         atomic_set(&pd->cdrw.pending_bios, 0);
2740         pd->cdrw.thread = kthread_run(kcdrwd, pd, "%s", pd->name);
2741         if (IS_ERR(pd->cdrw.thread)) {
2742                 printk(DRIVER_NAME": can't start kernel thread\n");
2743                 ret = -ENOMEM;
2744                 goto out_mem;
2745         }
2746
2747         proc_create_data(pd->name, 0, pkt_proc, &pkt_proc_fops, pd);
2748         DPRINTK(DRIVER_NAME": writer %s mapped to %s\n", pd->name, bdevname(bdev, b));
2749         return 0;
2750
2751 out_mem:
2752         blkdev_put(bdev, FMODE_READ | FMODE_NDELAY);
2753         /* This is safe: open() is still holding a reference. */
2754         module_put(THIS_MODULE);
2755         return ret;
2756 }
2757
2758 static int pkt_ioctl(struct block_device *bdev, fmode_t mode, unsigned int cmd, unsigned long arg)
2759 {
2760         struct pktcdvd_device *pd = bdev->bd_disk->private_data;
2761         int ret;
2762
2763         VPRINTK("pkt_ioctl: cmd %x, dev %d:%d\n", cmd,
2764                 MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev));
2765
2766         mutex_lock(&pktcdvd_mutex);
2767         switch (cmd) {
2768         case CDROMEJECT:
2769                 /*
2770                  * The door gets locked when the device is opened, so we
2771                  * have to unlock it or else the eject command fails.
2772                  */
2773                 if (pd->refcnt == 1)
2774                         pkt_lock_door(pd, 0);
2775                 /* fallthru */
2776         /*
2777          * forward selected CDROM ioctls to CD-ROM, for UDF
2778          */
2779         case CDROMMULTISESSION:
2780         case CDROMREADTOCENTRY:
2781         case CDROM_LAST_WRITTEN:
2782         case CDROM_SEND_PACKET:
2783         case SCSI_IOCTL_SEND_COMMAND:
2784                 ret = __blkdev_driver_ioctl(pd->bdev, mode, cmd, arg);
2785                 break;
2786
2787         default:
2788                 VPRINTK(DRIVER_NAME": Unknown ioctl for %s (%x)\n", pd->name, cmd);
2789                 ret = -ENOTTY;
2790         }
2791         mutex_unlock(&pktcdvd_mutex);
2792
2793         return ret;
2794 }
2795
2796 static unsigned int pkt_check_events(struct gendisk *disk,
2797                                      unsigned int clearing)
2798 {
2799         struct pktcdvd_device *pd = disk->private_data;
2800         struct gendisk *attached_disk;
2801
2802         if (!pd)
2803                 return 0;
2804         if (!pd->bdev)
2805                 return 0;
2806         attached_disk = pd->bdev->bd_disk;
2807         if (!attached_disk || !attached_disk->fops->check_events)
2808                 return 0;
2809         return attached_disk->fops->check_events(attached_disk, clearing);
2810 }
2811
2812 static const struct block_device_operations pktcdvd_ops = {
2813         .owner =                THIS_MODULE,
2814         .open =                 pkt_open,
2815         .release =              pkt_close,
2816         .ioctl =                pkt_ioctl,
2817         .check_events =         pkt_check_events,
2818 };
2819
2820 static char *pktcdvd_devnode(struct gendisk *gd, umode_t *mode)
2821 {
2822         return kasprintf(GFP_KERNEL, "pktcdvd/%s", gd->disk_name);
2823 }
2824
2825 /*
2826  * Set up mapping from pktcdvd device to CD-ROM device.
2827  */
2828 static int pkt_setup_dev(dev_t dev, dev_t* pkt_dev)
2829 {
2830         int idx;
2831         int ret = -ENOMEM;
2832         struct pktcdvd_device *pd;
2833         struct gendisk *disk;
2834
2835         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2836
2837         for (idx = 0; idx < MAX_WRITERS; idx++)
2838                 if (!pkt_devs[idx])
2839                         break;
2840         if (idx == MAX_WRITERS) {
2841                 printk(DRIVER_NAME": max %d writers supported\n", MAX_WRITERS);
2842                 ret = -EBUSY;
2843                 goto out_mutex;
2844         }
2845
2846         pd = kzalloc(sizeof(struct pktcdvd_device), GFP_KERNEL);
2847         if (!pd)
2848                 goto out_mutex;
2849
2850         pd->rb_pool = mempool_create_kmalloc_pool(PKT_RB_POOL_SIZE,
2851                                                   sizeof(struct pkt_rb_node));
2852         if (!pd->rb_pool)
2853                 goto out_mem;
2854
2855         INIT_LIST_HEAD(&pd->cdrw.pkt_free_list);
2856         INIT_LIST_HEAD(&pd->cdrw.pkt_active_list);
2857         spin_lock_init(&pd->cdrw.active_list_lock);
2858
2859         spin_lock_init(&pd->lock);
2860         spin_lock_init(&pd->iosched.lock);
2861         bio_list_init(&pd->iosched.read_queue);
2862         bio_list_init(&pd->iosched.write_queue);
2863         sprintf(pd->name, DRIVER_NAME"%d", idx);
2864         init_waitqueue_head(&pd->wqueue);
2865         pd->bio_queue = RB_ROOT;
2866
2867         pd->write_congestion_on  = write_congestion_on;
2868         pd->write_congestion_off = write_congestion_off;
2869
2870         disk = alloc_disk(1);
2871         if (!disk)
2872                 goto out_mem;
2873         pd->disk = disk;
2874         disk->major = pktdev_major;
2875         disk->first_minor = idx;
2876         disk->fops = &pktcdvd_ops;
2877         disk->flags = GENHD_FL_REMOVABLE;
2878         strcpy(disk->disk_name, pd->name);
2879         disk->devnode = pktcdvd_devnode;
2880         disk->private_data = pd;
2881         disk->queue = blk_alloc_queue(GFP_KERNEL);
2882         if (!disk->queue)
2883                 goto out_mem2;
2884
2885         pd->pkt_dev = MKDEV(pktdev_major, idx);
2886         ret = pkt_new_dev(pd, dev);
2887         if (ret)
2888                 goto out_new_dev;
2889
2890         /* inherit events of the host device */
2891         disk->events = pd->bdev->bd_disk->events;
2892         disk->async_events = pd->bdev->bd_disk->async_events;
2893
2894         add_disk(disk);
2895
2896         pkt_sysfs_dev_new(pd);
2897         pkt_debugfs_dev_new(pd);
2898
2899         pkt_devs[idx] = pd;
2900         if (pkt_dev)
2901                 *pkt_dev = pd->pkt_dev;
2902
2903         mutex_unlock(&ctl_mutex);
2904         return 0;
2905
2906 out_new_dev:
2907         blk_cleanup_queue(disk->queue);
2908 out_mem2:
2909         put_disk(disk);
2910 out_mem:
2911         if (pd->rb_pool)
2912                 mempool_destroy(pd->rb_pool);
2913         kfree(pd);
2914 out_mutex:
2915         mutex_unlock(&ctl_mutex);
2916         printk(DRIVER_NAME": setup of pktcdvd device failed\n");
2917         return ret;
2918 }
2919
2920 /*
2921  * Tear down mapping from pktcdvd device to CD-ROM device.
2922  */
2923 static int pkt_remove_dev(dev_t pkt_dev)
2924 {
2925         struct pktcdvd_device *pd;
2926         int idx;
2927         int ret = 0;
2928
2929         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2930
2931         for (idx = 0; idx < MAX_WRITERS; idx++) {
2932                 pd = pkt_devs[idx];
2933                 if (pd && (pd->pkt_dev == pkt_dev))
2934                         break;
2935         }
2936         if (idx == MAX_WRITERS) {
2937                 DPRINTK(DRIVER_NAME": dev not setup\n");
2938                 ret = -ENXIO;
2939                 goto out;
2940         }
2941
2942         if (pd->refcnt > 0) {
2943                 ret = -EBUSY;
2944                 goto out;
2945         }
2946         if (!IS_ERR(pd->cdrw.thread))
2947                 kthread_stop(pd->cdrw.thread);
2948
2949         pkt_devs[idx] = NULL;
2950
2951         pkt_debugfs_dev_remove(pd);
2952         pkt_sysfs_dev_remove(pd);
2953
2954         blkdev_put(pd->bdev, FMODE_READ | FMODE_NDELAY);
2955
2956         remove_proc_entry(pd->name, pkt_proc);
2957         DPRINTK(DRIVER_NAME": writer %s unmapped\n", pd->name);
2958
2959         del_gendisk(pd->disk);
2960         blk_cleanup_queue(pd->disk->queue);
2961         put_disk(pd->disk);
2962
2963         mempool_destroy(pd->rb_pool);
2964         kfree(pd);
2965
2966         /* This is safe: open() is still holding a reference. */
2967         module_put(THIS_MODULE);
2968
2969 out:
2970         mutex_unlock(&ctl_mutex);
2971         return ret;
2972 }
2973
2974 static void pkt_get_status(struct pkt_ctrl_command *ctrl_cmd)
2975 {
2976         struct pktcdvd_device *pd;
2977
2978         mutex_lock_nested(&ctl_mutex, SINGLE_DEPTH_NESTING);
2979
2980         pd = pkt_find_dev_from_minor(ctrl_cmd->dev_index);
2981         if (pd) {
2982                 ctrl_cmd->dev = new_encode_dev(pd->bdev->bd_dev);
2983                 ctrl_cmd->pkt_dev = new_encode_dev(pd->pkt_dev);
2984         } else {
2985                 ctrl_cmd->dev = 0;
2986                 ctrl_cmd->pkt_dev = 0;
2987         }
2988         ctrl_cmd->num_devices = MAX_WRITERS;
2989
2990         mutex_unlock(&ctl_mutex);
2991 }
2992
2993 static long pkt_ctl_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
2994 {
2995         void __user *argp = (void __user *)arg;
2996         struct pkt_ctrl_command ctrl_cmd;
2997         int ret = 0;
2998         dev_t pkt_dev = 0;
2999
3000         if (cmd != PACKET_CTRL_CMD)
3001                 return -ENOTTY;
3002
3003         if (copy_from_user(&ctrl_cmd, argp, sizeof(struct pkt_ctrl_command)))
3004                 return -EFAULT;
3005
3006         switch (ctrl_cmd.command) {
3007         case PKT_CTRL_CMD_SETUP:
3008                 if (!capable(CAP_SYS_ADMIN))
3009                         return -EPERM;
3010                 ret = pkt_setup_dev(new_decode_dev(ctrl_cmd.dev), &pkt_dev);
3011                 ctrl_cmd.pkt_dev = new_encode_dev(pkt_dev);
3012                 break;
3013         case PKT_CTRL_CMD_TEARDOWN:
3014                 if (!capable(CAP_SYS_ADMIN))
3015                         return -EPERM;
3016                 ret = pkt_remove_dev(new_decode_dev(ctrl_cmd.pkt_dev));
3017                 break;
3018         case PKT_CTRL_CMD_STATUS:
3019                 pkt_get_status(&ctrl_cmd);
3020                 break;
3021         default:
3022                 return -ENOTTY;
3023         }
3024
3025         if (copy_to_user(argp, &ctrl_cmd, sizeof(struct pkt_ctrl_command)))
3026                 return -EFAULT;
3027         return ret;
3028 }
3029
3030 #ifdef CONFIG_COMPAT
3031 static long pkt_ctl_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3032 {
3033         return pkt_ctl_ioctl(file, cmd, (unsigned long)compat_ptr(arg));
3034 }
3035 #endif
3036
3037 static const struct file_operations pkt_ctl_fops = {
3038         .open           = nonseekable_open,
3039         .unlocked_ioctl = pkt_ctl_ioctl,
3040 #ifdef CONFIG_COMPAT
3041         .compat_ioctl   = pkt_ctl_compat_ioctl,
3042 #endif
3043         .owner          = THIS_MODULE,
3044         .llseek         = no_llseek,
3045 };
3046
3047 static struct miscdevice pkt_misc = {
3048         .minor          = MISC_DYNAMIC_MINOR,
3049         .name           = DRIVER_NAME,
3050         .nodename       = "pktcdvd/control",
3051         .fops           = &pkt_ctl_fops
3052 };
3053
3054 static int __init pkt_init(void)
3055 {
3056         int ret;
3057
3058         mutex_init(&ctl_mutex);
3059
3060         psd_pool = mempool_create_kmalloc_pool(PSD_POOL_SIZE,
3061                                         sizeof(struct packet_stacked_data));
3062         if (!psd_pool)
3063                 return -ENOMEM;
3064
3065         ret = register_blkdev(pktdev_major, DRIVER_NAME);
3066         if (ret < 0) {
3067                 printk(DRIVER_NAME": Unable to register block device\n");
3068                 goto out2;
3069         }
3070         if (!pktdev_major)
3071                 pktdev_major = ret;
3072
3073         ret = pkt_sysfs_init();
3074         if (ret)
3075                 goto out;
3076
3077         pkt_debugfs_init();
3078
3079         ret = misc_register(&pkt_misc);
3080         if (ret) {
3081                 printk(DRIVER_NAME": Unable to register misc device\n");
3082                 goto out_misc;
3083         }
3084
3085         pkt_proc = proc_mkdir("driver/"DRIVER_NAME, NULL);
3086
3087         return 0;
3088
3089 out_misc:
3090         pkt_debugfs_cleanup();
3091         pkt_sysfs_cleanup();
3092 out:
3093         unregister_blkdev(pktdev_major, DRIVER_NAME);
3094 out2:
3095         mempool_destroy(psd_pool);
3096         return ret;
3097 }
3098
3099 static void __exit pkt_exit(void)
3100 {
3101         remove_proc_entry("driver/"DRIVER_NAME, NULL);
3102         misc_deregister(&pkt_misc);
3103
3104         pkt_debugfs_cleanup();
3105         pkt_sysfs_cleanup();
3106
3107         unregister_blkdev(pktdev_major, DRIVER_NAME);
3108         mempool_destroy(psd_pool);
3109 }
3110
3111 MODULE_DESCRIPTION("Packet writing layer for CD/DVD drives");
3112 MODULE_AUTHOR("Jens Axboe <axboe@suse.de>");
3113 MODULE_LICENSE("GPL");
3114
3115 module_init(pkt_init);
3116 module_exit(pkt_exit);