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