Merge branch 'upstream-linus' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
[linux-2.6] / arch / um / drivers / ubd_kern.c
1 /*
2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 /* 2001-09-28...2002-04-17
7  * Partition stuff by James_McMechan@hotmail.com
8  * old style ubd by setting UBD_SHIFT to 0
9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
10  * partitions have changed in 2.5
11  * 2003-01-29 more tinkering for 2.5.59-1
12  * This should now address the sysfs problems and has
13  * the symlink for devfs to allow for booting with
14  * the common /dev/ubd/discX/... names rather than
15  * only /dev/ubdN/discN this version also has lots of
16  * clean ups preparing for ubd-many.
17  * James McMechan
18  */
19
20 #define MAJOR_NR UBD_MAJOR
21 #define UBD_SHIFT 4
22
23 #include "linux/config.h"
24 #include "linux/module.h"
25 #include "linux/blkdev.h"
26 #include "linux/hdreg.h"
27 #include "linux/init.h"
28 #include "linux/cdrom.h"
29 #include "linux/proc_fs.h"
30 #include "linux/ctype.h"
31 #include "linux/capability.h"
32 #include "linux/mm.h"
33 #include "linux/vmalloc.h"
34 #include "linux/blkpg.h"
35 #include "linux/genhd.h"
36 #include "linux/spinlock.h"
37 #include "linux/platform_device.h"
38 #include "asm/segment.h"
39 #include "asm/uaccess.h"
40 #include "asm/irq.h"
41 #include "asm/types.h"
42 #include "asm/tlbflush.h"
43 #include "user_util.h"
44 #include "mem_user.h"
45 #include "kern_util.h"
46 #include "kern.h"
47 #include "mconsole_kern.h"
48 #include "init.h"
49 #include "irq_user.h"
50 #include "irq_kern.h"
51 #include "ubd_user.h"
52 #include "os.h"
53 #include "mem.h"
54 #include "mem_kern.h"
55 #include "cow.h"
56
57 enum ubd_req { UBD_READ, UBD_WRITE };
58
59 struct io_thread_req {
60         enum ubd_req op;
61         int fds[2];
62         unsigned long offsets[2];
63         unsigned long long offset;
64         unsigned long length;
65         char *buffer;
66         int sectorsize;
67         unsigned long sector_mask;
68         unsigned long long cow_offset;
69         unsigned long bitmap_words[2];
70         int error;
71 };
72
73 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
74                          char **backing_file_out, int *bitmap_offset_out,
75                          unsigned long *bitmap_len_out, int *data_offset_out,
76                          int *create_cow_out);
77 extern int create_cow_file(char *cow_file, char *backing_file,
78                            struct openflags flags, int sectorsize,
79                            int alignment, int *bitmap_offset_out,
80                            unsigned long *bitmap_len_out,
81                            int *data_offset_out);
82 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
83 extern void do_io(struct io_thread_req *req);
84
85 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
86 {
87         __u64 n;
88         int bits, off;
89
90         bits = sizeof(data[0]) * 8;
91         n = bit / bits;
92         off = bit % bits;
93         return((data[n] & (1 << off)) != 0);
94 }
95
96 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
97 {
98         __u64 n;
99         int bits, off;
100
101         bits = sizeof(data[0]) * 8;
102         n = bit / bits;
103         off = bit % bits;
104         data[n] |= (1 << off);
105 }
106 /*End stuff from ubd_user.h*/
107
108 #define DRIVER_NAME "uml-blkdev"
109
110 static DEFINE_SPINLOCK(ubd_io_lock);
111 static DEFINE_SPINLOCK(ubd_lock);
112
113 static void (*do_ubd)(void);
114
115 static int ubd_open(struct inode * inode, struct file * filp);
116 static int ubd_release(struct inode * inode, struct file * file);
117 static int ubd_ioctl(struct inode * inode, struct file * file,
118                      unsigned int cmd, unsigned long arg);
119 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
120
121 #define MAX_DEV (8)
122
123 static struct block_device_operations ubd_blops = {
124         .owner          = THIS_MODULE,
125         .open           = ubd_open,
126         .release        = ubd_release,
127         .ioctl          = ubd_ioctl,
128         .getgeo         = ubd_getgeo,
129 };
130
131 /* Protected by the queue_lock */
132 static request_queue_t *ubd_queue;
133
134 /* Protected by ubd_lock */
135 static int fake_major = MAJOR_NR;
136
137 static struct gendisk *ubd_gendisk[MAX_DEV];
138 static struct gendisk *fake_gendisk[MAX_DEV];
139
140 #ifdef CONFIG_BLK_DEV_UBD_SYNC
141 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
142                                          .cl = 1 })
143 #else
144 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
145                                          .cl = 1 })
146 #endif
147
148 /* Not protected - changed only in ubd_setup_common and then only to
149  * to enable O_SYNC.
150  */
151 static struct openflags global_openflags = OPEN_FLAGS;
152
153 struct cow {
154         /* This is the backing file, actually */
155         char *file;
156         int fd;
157         unsigned long *bitmap;
158         unsigned long bitmap_len;
159         int bitmap_offset;
160         int data_offset;
161 };
162
163 struct ubd {
164         char *file;
165         int count;
166         int fd;
167         __u64 size;
168         struct openflags boot_openflags;
169         struct openflags openflags;
170         int shared;
171         int no_cow;
172         struct cow cow;
173         struct platform_device pdev;
174 };
175
176 #define DEFAULT_COW { \
177         .file =                 NULL, \
178         .fd =                   -1, \
179         .bitmap =               NULL, \
180         .bitmap_offset =        0, \
181         .data_offset =          0, \
182 }
183
184 #define DEFAULT_UBD { \
185         .file =                 NULL, \
186         .count =                0, \
187         .fd =                   -1, \
188         .size =                 -1, \
189         .boot_openflags =       OPEN_FLAGS, \
190         .openflags =            OPEN_FLAGS, \
191         .no_cow =               0, \
192         .shared =               0, \
193         .cow =                  DEFAULT_COW, \
194 }
195
196 struct ubd ubd_dev[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
197
198 static int ubd0_init(void)
199 {
200         struct ubd *dev = &ubd_dev[0];
201
202         if(dev->file == NULL)
203                 dev->file = "root_fs";
204         return(0);
205 }
206
207 __initcall(ubd0_init);
208
209 /* Only changed by fake_ide_setup which is a setup */
210 static int fake_ide = 0;
211 static struct proc_dir_entry *proc_ide_root = NULL;
212 static struct proc_dir_entry *proc_ide = NULL;
213
214 static void make_proc_ide(void)
215 {
216         proc_ide_root = proc_mkdir("ide", NULL);
217         proc_ide = proc_mkdir("ide0", proc_ide_root);
218 }
219
220 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
221                                int *eof, void *data)
222 {
223         int len;
224
225         strcpy(page, "disk\n");
226         len = strlen("disk\n");
227         len -= off;
228         if (len < count){
229                 *eof = 1;
230                 if (len <= 0) return 0;
231         }
232         else len = count;
233         *start = page + off;
234         return len;
235 }
236
237 static void make_ide_entries(char *dev_name)
238 {
239         struct proc_dir_entry *dir, *ent;
240         char name[64];
241
242         if(proc_ide_root == NULL) make_proc_ide();
243
244         dir = proc_mkdir(dev_name, proc_ide);
245         if(!dir) return;
246
247         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
248         if(!ent) return;
249         ent->nlink = 1;
250         ent->data = NULL;
251         ent->read_proc = proc_ide_read_media;
252         ent->write_proc = NULL;
253         sprintf(name,"ide0/%s", dev_name);
254         proc_symlink(dev_name, proc_ide_root, name);
255 }
256
257 static int fake_ide_setup(char *str)
258 {
259         fake_ide = 1;
260         return(1);
261 }
262
263 __setup("fake_ide", fake_ide_setup);
264
265 __uml_help(fake_ide_setup,
266 "fake_ide\n"
267 "    Create ide0 entries that map onto ubd devices.\n\n"
268 );
269
270 static int parse_unit(char **ptr)
271 {
272         char *str = *ptr, *end;
273         int n = -1;
274
275         if(isdigit(*str)) {
276                 n = simple_strtoul(str, &end, 0);
277                 if(end == str)
278                         return(-1);
279                 *ptr = end;
280         }
281         else if (('a' <= *str) && (*str <= 'h')) {
282                 n = *str - 'a';
283                 str++;
284                 *ptr = str;
285         }
286         return(n);
287 }
288
289 static int ubd_setup_common(char *str, int *index_out)
290 {
291         struct ubd *dev;
292         struct openflags flags = global_openflags;
293         char *backing_file;
294         int n, err, i;
295
296         if(index_out) *index_out = -1;
297         n = *str;
298         if(n == '='){
299                 char *end;
300                 int major;
301
302                 str++;
303                 if(!strcmp(str, "sync")){
304                         global_openflags = of_sync(global_openflags);
305                         return(0);
306                 }
307                 major = simple_strtoul(str, &end, 0);
308                 if((*end != '\0') || (end == str)){
309                         printk(KERN_ERR
310                                "ubd_setup : didn't parse major number\n");
311                         return(1);
312                 }
313
314                 err = 1;
315                 spin_lock(&ubd_lock);
316                 if(fake_major != MAJOR_NR){
317                         printk(KERN_ERR "Can't assign a fake major twice\n");
318                         goto out1;
319                 }
320
321                 fake_major = major;
322
323                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
324                        major);
325                 err = 0;
326         out1:
327                 spin_unlock(&ubd_lock);
328                 return(err);
329         }
330
331         n = parse_unit(&str);
332         if(n < 0){
333                 printk(KERN_ERR "ubd_setup : couldn't parse unit number "
334                        "'%s'\n", str);
335                 return(1);
336         }
337         if(n >= MAX_DEV){
338                 printk(KERN_ERR "ubd_setup : index %d out of range "
339                        "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
340                 return(1);
341         }
342
343         err = 1;
344         spin_lock(&ubd_lock);
345
346         dev = &ubd_dev[n];
347         if(dev->file != NULL){
348                 printk(KERN_ERR "ubd_setup : device already configured\n");
349                 goto out;
350         }
351
352         if (index_out)
353                 *index_out = n;
354
355         for (i = 0; i < sizeof("rscd="); i++) {
356                 switch (*str) {
357                 case 'r':
358                         flags.w = 0;
359                         break;
360                 case 's':
361                         flags.s = 1;
362                         break;
363                 case 'd':
364                         dev->no_cow = 1;
365                         break;
366                 case 'c':
367                         dev->shared = 1;
368                         break;
369                 case '=':
370                         str++;
371                         goto break_loop;
372                 default:
373                         printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r, s, c, or d)\n");
374                         goto out;
375                 }
376                 str++;
377         }
378
379         if (*str == '=')
380                 printk(KERN_ERR "ubd_setup : Too many flags specified\n");
381         else
382                 printk(KERN_ERR "ubd_setup : Expected '='\n");
383         goto out;
384
385 break_loop:
386         err = 0;
387         backing_file = strchr(str, ',');
388
389         if (!backing_file) {
390                 backing_file = strchr(str, ':');
391         }
392
393         if(backing_file){
394                 if(dev->no_cow)
395                         printk(KERN_ERR "Can't specify both 'd' and a "
396                                "cow file\n");
397                 else {
398                         *backing_file = '\0';
399                         backing_file++;
400                 }
401         }
402         dev->file = str;
403         dev->cow.file = backing_file;
404         dev->boot_openflags = flags;
405 out:
406         spin_unlock(&ubd_lock);
407         return(err);
408 }
409
410 static int ubd_setup(char *str)
411 {
412         ubd_setup_common(str, NULL);
413         return(1);
414 }
415
416 __setup("ubd", ubd_setup);
417 __uml_help(ubd_setup,
418 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
419 "    This is used to associate a device with a file in the underlying\n"
420 "    filesystem. When specifying two filenames, the first one is the\n"
421 "    COW name and the second is the backing file name. As separator you can\n"
422 "    use either a ':' or a ',': the first one allows writing things like;\n"
423 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
424 "    while with a ',' the shell would not expand the 2nd '~'.\n"
425 "    When using only one filename, UML will detect whether to thread it like\n"
426 "    a COW file or a backing file. To override this detection, add the 'd'\n"
427 "    flag:\n"
428 "       ubd0d=BackingFile\n"
429 "    Usually, there is a filesystem in the file, but \n"
430 "    that's not required. Swap devices containing swap files can be\n"
431 "    specified like this. Also, a file which doesn't contain a\n"
432 "    filesystem can have its contents read in the virtual \n"
433 "    machine by running 'dd' on the device. <n> must be in the range\n"
434 "    0 to 7. Appending an 'r' to the number will cause that device\n"
435 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
436 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
437 );
438
439 static int udb_setup(char *str)
440 {
441         printk("udb%s specified on command line is almost certainly a ubd -> "
442                "udb TYPO\n", str);
443         return(1);
444 }
445
446 __setup("udb", udb_setup);
447 __uml_help(udb_setup,
448 "udb\n"
449 "    This option is here solely to catch ubd -> udb typos, which can be\n"
450 "    to impossible to catch visually unless you specifically look for\n"
451 "    them.  The only result of any option starting with 'udb' is an error\n"
452 "    in the boot output.\n\n"
453 );
454
455 static int fakehd_set = 0;
456 static int fakehd(char *str)
457 {
458         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
459         fakehd_set = 1;
460         return 1;
461 }
462
463 __setup("fakehd", fakehd);
464 __uml_help(fakehd,
465 "fakehd\n"
466 "    Change the ubd device name to \"hd\".\n\n"
467 );
468
469 static void do_ubd_request(request_queue_t * q);
470
471 /* Only changed by ubd_init, which is an initcall. */
472 int thread_fd = -1;
473
474 /* Changed by ubd_handler, which is serialized because interrupts only
475  * happen on CPU 0.
476  */
477 int intr_count = 0;
478
479 /* call ubd_finish if you need to serialize */
480 static void __ubd_finish(struct request *req, int error)
481 {
482         int nsect;
483
484         if(error){
485                 end_request(req, 0);
486                 return;
487         }
488         nsect = req->current_nr_sectors;
489         req->sector += nsect;
490         req->buffer += nsect << 9;
491         req->errors = 0;
492         req->nr_sectors -= nsect;
493         req->current_nr_sectors = 0;
494         end_request(req, 1);
495 }
496
497 static inline void ubd_finish(struct request *req, int error)
498 {
499         spin_lock(&ubd_io_lock);
500         __ubd_finish(req, error);
501         spin_unlock(&ubd_io_lock);
502 }
503
504 /* Called without ubd_io_lock held */
505 static void ubd_handler(void)
506 {
507         struct io_thread_req req;
508         struct request *rq = elv_next_request(ubd_queue);
509         int n;
510
511         do_ubd = NULL;
512         intr_count++;
513         n = os_read_file(thread_fd, &req, sizeof(req));
514         if(n != sizeof(req)){
515                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
516                        "err = %d\n", os_getpid(), -n);
517                 spin_lock(&ubd_io_lock);
518                 end_request(rq, 0);
519                 spin_unlock(&ubd_io_lock);
520                 return;
521         }
522
523         ubd_finish(rq, req.error);
524         reactivate_fd(thread_fd, UBD_IRQ);      
525         do_ubd_request(ubd_queue);
526 }
527
528 static irqreturn_t ubd_intr(int irq, void *dev, struct pt_regs *unused)
529 {
530         ubd_handler();
531         return(IRQ_HANDLED);
532 }
533
534 /* Only changed by ubd_init, which is an initcall. */
535 static int io_pid = -1;
536
537 void kill_io_thread(void)
538 {
539         if(io_pid != -1)
540                 os_kill_process(io_pid, 1);
541 }
542
543 __uml_exitcall(kill_io_thread);
544
545 static int ubd_file_size(struct ubd *dev, __u64 *size_out)
546 {
547         char *file;
548
549         file = dev->cow.file ? dev->cow.file : dev->file;
550         return(os_file_size(file, size_out));
551 }
552
553 static void ubd_close(struct ubd *dev)
554 {
555         os_close_file(dev->fd);
556         if(dev->cow.file == NULL)
557                 return;
558
559         os_close_file(dev->cow.fd);
560         vfree(dev->cow.bitmap);
561         dev->cow.bitmap = NULL;
562 }
563
564 static int ubd_open_dev(struct ubd *dev)
565 {
566         struct openflags flags;
567         char **back_ptr;
568         int err, create_cow, *create_ptr;
569
570         dev->openflags = dev->boot_openflags;
571         create_cow = 0;
572         create_ptr = (dev->cow.file != NULL) ? &create_cow : NULL;
573         back_ptr = dev->no_cow ? NULL : &dev->cow.file;
574         dev->fd = open_ubd_file(dev->file, &dev->openflags, dev->shared,
575                                 back_ptr, &dev->cow.bitmap_offset,
576                                 &dev->cow.bitmap_len, &dev->cow.data_offset,
577                                 create_ptr);
578
579         if((dev->fd == -ENOENT) && create_cow){
580                 dev->fd = create_cow_file(dev->file, dev->cow.file,
581                                           dev->openflags, 1 << 9, PAGE_SIZE,
582                                           &dev->cow.bitmap_offset,
583                                           &dev->cow.bitmap_len,
584                                           &dev->cow.data_offset);
585                 if(dev->fd >= 0){
586                         printk(KERN_INFO "Creating \"%s\" as COW file for "
587                                "\"%s\"\n", dev->file, dev->cow.file);
588                 }
589         }
590
591         if(dev->fd < 0){
592                 printk("Failed to open '%s', errno = %d\n", dev->file,
593                        -dev->fd);
594                 return(dev->fd);
595         }
596
597         if(dev->cow.file != NULL){
598                 err = -ENOMEM;
599                 dev->cow.bitmap = (void *) vmalloc(dev->cow.bitmap_len);
600                 if(dev->cow.bitmap == NULL){
601                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
602                         goto error;
603                 }
604                 flush_tlb_kernel_vm();
605
606                 err = read_cow_bitmap(dev->fd, dev->cow.bitmap,
607                                       dev->cow.bitmap_offset,
608                                       dev->cow.bitmap_len);
609                 if(err < 0)
610                         goto error;
611
612                 flags = dev->openflags;
613                 flags.w = 0;
614                 err = open_ubd_file(dev->cow.file, &flags, dev->shared, NULL,
615                                     NULL, NULL, NULL, NULL);
616                 if(err < 0) goto error;
617                 dev->cow.fd = err;
618         }
619         return(0);
620  error:
621         os_close_file(dev->fd);
622         return(err);
623 }
624
625 static int ubd_new_disk(int major, u64 size, int unit,
626                         struct gendisk **disk_out)
627                         
628 {
629         struct gendisk *disk;
630
631         disk = alloc_disk(1 << UBD_SHIFT);
632         if(disk == NULL)
633                 return(-ENOMEM);
634
635         disk->major = major;
636         disk->first_minor = unit << UBD_SHIFT;
637         disk->fops = &ubd_blops;
638         set_capacity(disk, size / 512);
639         if(major == MAJOR_NR)
640                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
641         else
642                 sprintf(disk->disk_name, "ubd_fake%d", unit);
643
644         /* sysfs register (not for ide fake devices) */
645         if (major == MAJOR_NR) {
646                 ubd_dev[unit].pdev.id   = unit;
647                 ubd_dev[unit].pdev.name = DRIVER_NAME;
648                 platform_device_register(&ubd_dev[unit].pdev);
649                 disk->driverfs_dev = &ubd_dev[unit].pdev.dev;
650         }
651
652         disk->private_data = &ubd_dev[unit];
653         disk->queue = ubd_queue;
654         add_disk(disk);
655
656         *disk_out = disk;
657         return 0;
658 }
659
660 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
661
662 static int ubd_add(int n)
663 {
664         struct ubd *dev = &ubd_dev[n];
665         int err;
666
667         err = -ENODEV;
668         if(dev->file == NULL)
669                 goto out;
670
671         err = ubd_file_size(dev, &dev->size);
672         if(err < 0)
673                 goto out;
674
675         dev->size = ROUND_BLOCK(dev->size);
676
677         err = ubd_new_disk(MAJOR_NR, dev->size, n, &ubd_gendisk[n]);
678         if(err)
679                 goto out;
680
681         if(fake_major != MAJOR_NR)
682                 ubd_new_disk(fake_major, dev->size, n,
683                              &fake_gendisk[n]);
684
685         /* perhaps this should also be under the "if (fake_major)" above */
686         /* using the fake_disk->disk_name and also the fakehd_set name */
687         if (fake_ide)
688                 make_ide_entries(ubd_gendisk[n]->disk_name);
689
690         err = 0;
691 out:
692         return err;
693 }
694
695 static int ubd_config(char *str)
696 {
697         int n, err;
698
699         str = kstrdup(str, GFP_KERNEL);
700         if(str == NULL){
701                 printk(KERN_ERR "ubd_config failed to strdup string\n");
702                 return(1);
703         }
704         err = ubd_setup_common(str, &n);
705         if(err){
706                 kfree(str);
707                 return(-1);
708         }
709         if(n == -1) return(0);
710
711         spin_lock(&ubd_lock);
712         err = ubd_add(n);
713         if(err)
714                 ubd_dev[n].file = NULL;
715         spin_unlock(&ubd_lock);
716
717         return(err);
718 }
719
720 static int ubd_get_config(char *name, char *str, int size, char **error_out)
721 {
722         struct ubd *dev;
723         int n, len = 0;
724
725         n = parse_unit(&name);
726         if((n >= MAX_DEV) || (n < 0)){
727                 *error_out = "ubd_get_config : device number out of range";
728                 return(-1);
729         }
730
731         dev = &ubd_dev[n];
732         spin_lock(&ubd_lock);
733
734         if(dev->file == NULL){
735                 CONFIG_CHUNK(str, size, len, "", 1);
736                 goto out;
737         }
738
739         CONFIG_CHUNK(str, size, len, dev->file, 0);
740
741         if(dev->cow.file != NULL){
742                 CONFIG_CHUNK(str, size, len, ",", 0);
743                 CONFIG_CHUNK(str, size, len, dev->cow.file, 1);
744         }
745         else CONFIG_CHUNK(str, size, len, "", 1);
746
747  out:
748         spin_unlock(&ubd_lock);
749         return(len);
750 }
751
752 static int ubd_id(char **str, int *start_out, int *end_out)
753 {
754         int n;
755
756         n = parse_unit(str);
757         *start_out = 0;
758         *end_out = MAX_DEV - 1;
759         return n;
760 }
761
762 static int ubd_remove(int n)
763 {
764         struct ubd *dev;
765         int err = -ENODEV;
766
767         spin_lock(&ubd_lock);
768
769         if(ubd_gendisk[n] == NULL)
770                 goto out;
771
772         dev = &ubd_dev[n];
773
774         if(dev->file == NULL)
775                 goto out;
776
777         /* you cannot remove a open disk */
778         err = -EBUSY;
779         if(dev->count > 0)
780                 goto out;
781
782         del_gendisk(ubd_gendisk[n]);
783         put_disk(ubd_gendisk[n]);
784         ubd_gendisk[n] = NULL;
785
786         if(fake_gendisk[n] != NULL){
787                 del_gendisk(fake_gendisk[n]);
788                 put_disk(fake_gendisk[n]);
789                 fake_gendisk[n] = NULL;
790         }
791
792         platform_device_unregister(&dev->pdev);
793         *dev = ((struct ubd) DEFAULT_UBD);
794         err = 0;
795 out:
796         spin_unlock(&ubd_lock);
797         return err;
798 }
799
800 static struct mc_device ubd_mc = {
801         .name           = "ubd",
802         .config         = ubd_config,
803         .get_config     = ubd_get_config,
804         .id             = ubd_id,
805         .remove         = ubd_remove,
806 };
807
808 static int ubd_mc_init(void)
809 {
810         mconsole_register_dev(&ubd_mc);
811         return 0;
812 }
813
814 __initcall(ubd_mc_init);
815
816 static struct platform_driver ubd_driver = {
817         .driver = {
818                 .name  = DRIVER_NAME,
819         },
820 };
821
822 int ubd_init(void)
823 {
824         int i;
825
826         if (register_blkdev(MAJOR_NR, "ubd"))
827                 return -1;
828
829         ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
830         if (!ubd_queue) {
831                 unregister_blkdev(MAJOR_NR, "ubd");
832                 return -1;
833         }
834                 
835         if (fake_major != MAJOR_NR) {
836                 char name[sizeof("ubd_nnn\0")];
837
838                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
839                 if (register_blkdev(fake_major, "ubd"))
840                         return -1;
841         }
842         platform_driver_register(&ubd_driver);
843         for (i = 0; i < MAX_DEV; i++)
844                 ubd_add(i);
845         return 0;
846 }
847
848 late_initcall(ubd_init);
849
850 int ubd_driver_init(void){
851         unsigned long stack;
852         int err;
853
854         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
855         if(global_openflags.s){
856                 printk(KERN_INFO "ubd: Synchronous mode\n");
857                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
858                  * enough. So use anyway the io thread. */
859         }
860         stack = alloc_stack(0, 0);
861         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
862                                  &thread_fd);
863         if(io_pid < 0){
864                 printk(KERN_ERR
865                        "ubd : Failed to start I/O thread (errno = %d) - "
866                        "falling back to synchronous I/O\n", -io_pid);
867                 io_pid = -1;
868                 return(0);
869         }
870         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
871                              IRQF_DISABLED, "ubd", ubd_dev);
872         if(err != 0)
873                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
874         return 0;
875 }
876
877 device_initcall(ubd_driver_init);
878
879 static int ubd_open(struct inode *inode, struct file *filp)
880 {
881         struct gendisk *disk = inode->i_bdev->bd_disk;
882         struct ubd *dev = disk->private_data;
883         int err = 0;
884
885         if(dev->count == 0){
886                 err = ubd_open_dev(dev);
887                 if(err){
888                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
889                                disk->disk_name, dev->file, -err);
890                         goto out;
891                 }
892         }
893         dev->count++;
894         set_disk_ro(disk, !dev->openflags.w);
895
896         /* This should no more be needed. And it didn't work anyway to exclude
897          * read-write remounting of filesystems.*/
898         /*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
899                 if(--dev->count == 0) ubd_close(dev);
900                 err = -EROFS;
901         }*/
902  out:
903         return(err);
904 }
905
906 static int ubd_release(struct inode * inode, struct file * file)
907 {
908         struct gendisk *disk = inode->i_bdev->bd_disk;
909         struct ubd *dev = disk->private_data;
910
911         if(--dev->count == 0)
912                 ubd_close(dev);
913         return(0);
914 }
915
916 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
917                           __u64 *cow_offset, unsigned long *bitmap,
918                           __u64 bitmap_offset, unsigned long *bitmap_words,
919                           __u64 bitmap_len)
920 {
921         __u64 sector = io_offset >> 9;
922         int i, update_bitmap = 0;
923
924         for(i = 0; i < length >> 9; i++){
925                 if(cow_mask != NULL)
926                         ubd_set_bit(i, (unsigned char *) cow_mask);
927                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
928                         continue;
929
930                 update_bitmap = 1;
931                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
932         }
933
934         if(!update_bitmap)
935                 return;
936
937         *cow_offset = sector / (sizeof(unsigned long) * 8);
938
939         /* This takes care of the case where we're exactly at the end of the
940          * device, and *cow_offset + 1 is off the end.  So, just back it up
941          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
942          * for the original diagnosis.
943          */
944         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
945                            sizeof(unsigned long) - 1))
946                 (*cow_offset)--;
947
948         bitmap_words[0] = bitmap[*cow_offset];
949         bitmap_words[1] = bitmap[*cow_offset + 1];
950
951         *cow_offset *= sizeof(unsigned long);
952         *cow_offset += bitmap_offset;
953 }
954
955 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
956                        __u64 bitmap_offset, __u64 bitmap_len)
957 {
958         __u64 sector = req->offset >> 9;
959         int i;
960
961         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
962                 panic("Operation too long");
963
964         if(req->op == UBD_READ) {
965                 for(i = 0; i < req->length >> 9; i++){
966                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
967                                 ubd_set_bit(i, (unsigned char *)
968                                             &req->sector_mask);
969                 }
970         }
971         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
972                            &req->cow_offset, bitmap, bitmap_offset,
973                            req->bitmap_words, bitmap_len);
974 }
975
976 /* Called with ubd_io_lock held */
977 static int prepare_request(struct request *req, struct io_thread_req *io_req)
978 {
979         struct gendisk *disk = req->rq_disk;
980         struct ubd *dev = disk->private_data;
981         __u64 offset;
982         int len;
983
984         /* This should be impossible now */
985         if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
986                 printk("Write attempted on readonly ubd device %s\n",
987                        disk->disk_name);
988                 end_request(req, 0);
989                 return(1);
990         }
991
992         offset = ((__u64) req->sector) << 9;
993         len = req->current_nr_sectors << 9;
994
995         io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
996         io_req->fds[1] = dev->fd;
997         io_req->cow_offset = -1;
998         io_req->offset = offset;
999         io_req->length = len;
1000         io_req->error = 0;
1001         io_req->sector_mask = 0;
1002
1003         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1004         io_req->offsets[0] = 0;
1005         io_req->offsets[1] = dev->cow.data_offset;
1006         io_req->buffer = req->buffer;
1007         io_req->sectorsize = 1 << 9;
1008
1009         if(dev->cow.file != NULL)
1010                 cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset,
1011                            dev->cow.bitmap_len);
1012
1013         return(0);
1014 }
1015
1016 /* Called with ubd_io_lock held */
1017 static void do_ubd_request(request_queue_t *q)
1018 {
1019         struct io_thread_req io_req;
1020         struct request *req;
1021         int err, n;
1022
1023         if(thread_fd == -1){
1024                 while((req = elv_next_request(q)) != NULL){
1025                         err = prepare_request(req, &io_req);
1026                         if(!err){
1027                                 do_io(&io_req);
1028                                 __ubd_finish(req, io_req.error);
1029                         }
1030                 }
1031         }
1032         else {
1033                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1034                         return;
1035                 err = prepare_request(req, &io_req);
1036                 if(!err){
1037                         do_ubd = ubd_handler;
1038                         n = os_write_file(thread_fd, (char *) &io_req,
1039                                          sizeof(io_req));
1040                         if(n != sizeof(io_req))
1041                                 printk("write to io thread failed, "
1042                                        "errno = %d\n", -n);
1043                 }
1044         }
1045 }
1046
1047 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1048 {
1049         struct ubd *dev = bdev->bd_disk->private_data;
1050
1051         geo->heads = 128;
1052         geo->sectors = 32;
1053         geo->cylinders = dev->size / (128 * 32 * 512);
1054         return 0;
1055 }
1056
1057 static int ubd_ioctl(struct inode * inode, struct file * file,
1058                      unsigned int cmd, unsigned long arg)
1059 {
1060         struct ubd *dev = inode->i_bdev->bd_disk->private_data;
1061         struct hd_driveid ubd_id = {
1062                 .cyls           = 0,
1063                 .heads          = 128,
1064                 .sectors        = 32,
1065         };
1066
1067         switch (cmd) {
1068                 struct cdrom_volctrl volume;
1069         case HDIO_GET_IDENTITY:
1070                 ubd_id.cyls = dev->size / (128 * 32 * 512);
1071                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1072                                  sizeof(ubd_id)))
1073                         return(-EFAULT);
1074                 return(0);
1075                 
1076         case CDROMVOLREAD:
1077                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1078                         return(-EFAULT);
1079                 volume.channel0 = 255;
1080                 volume.channel1 = 255;
1081                 volume.channel2 = 255;
1082                 volume.channel3 = 255;
1083                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1084                         return(-EFAULT);
1085                 return(0);
1086         }
1087         return(-EINVAL);
1088 }
1089
1090 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1091 {
1092         struct uml_stat buf1, buf2;
1093         int err;
1094
1095         if(from_cmdline == NULL)
1096                 return 0;
1097         if(!strcmp(from_cmdline, from_cow))
1098                 return 0;
1099
1100         err = os_stat_file(from_cmdline, &buf1);
1101         if(err < 0){
1102                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1103                 return 0;
1104         }
1105         err = os_stat_file(from_cow, &buf2);
1106         if(err < 0){
1107                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1108                 return 1;
1109         }
1110         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1111                 return 0;
1112
1113         printk("Backing file mismatch - \"%s\" requested,\n"
1114                "\"%s\" specified in COW header of \"%s\"\n",
1115                from_cmdline, from_cow, cow);
1116         return 1;
1117 }
1118
1119 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1120 {
1121         unsigned long modtime;
1122         unsigned long long actual;
1123         int err;
1124
1125         err = os_file_modtime(file, &modtime);
1126         if(err < 0){
1127                 printk("Failed to get modification time of backing file "
1128                        "\"%s\", err = %d\n", file, -err);
1129                 return(err);
1130         }
1131
1132         err = os_file_size(file, &actual);
1133         if(err < 0){
1134                 printk("Failed to get size of backing file \"%s\", "
1135                        "err = %d\n", file, -err);
1136                 return(err);
1137         }
1138
1139         if(actual != size){
1140                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1141                  * the typecast.*/
1142                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1143                        "file\n", (unsigned long long) size, actual);
1144                 return(-EINVAL);
1145         }
1146         if(modtime != mtime){
1147                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1148                        "file\n", mtime, modtime);
1149                 return(-EINVAL);
1150         }
1151         return(0);
1152 }
1153
1154 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1155 {
1156         int err;
1157
1158         err = os_seek_file(fd, offset);
1159         if(err < 0)
1160                 return(err);
1161
1162         err = os_read_file(fd, buf, len);
1163         if(err < 0)
1164                 return(err);
1165
1166         return(0);
1167 }
1168
1169 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1170                   char **backing_file_out, int *bitmap_offset_out,
1171                   unsigned long *bitmap_len_out, int *data_offset_out,
1172                   int *create_cow_out)
1173 {
1174         time_t mtime;
1175         unsigned long long size;
1176         __u32 version, align;
1177         char *backing_file;
1178         int fd, err, sectorsize, asked_switch, mode = 0644;
1179
1180         fd = os_open_file(file, *openflags, mode);
1181         if (fd < 0) {
1182                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1183                         *create_cow_out = 1;
1184                 if (!openflags->w ||
1185                    ((fd != -EROFS) && (fd != -EACCES)))
1186                         return fd;
1187                 openflags->w = 0;
1188                 fd = os_open_file(file, *openflags, mode);
1189                 if (fd < 0)
1190                         return fd;
1191         }
1192
1193         if(shared)
1194                 printk("Not locking \"%s\" on the host\n", file);
1195         else {
1196                 err = os_lock_file(fd, openflags->w);
1197                 if(err < 0){
1198                         printk("Failed to lock '%s', err = %d\n", file, -err);
1199                         goto out_close;
1200                 }
1201         }
1202
1203         /* Successful return case! */
1204         if(backing_file_out == NULL)
1205                 return(fd);
1206
1207         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1208                               &size, &sectorsize, &align, bitmap_offset_out);
1209         if(err && (*backing_file_out != NULL)){
1210                 printk("Failed to read COW header from COW file \"%s\", "
1211                        "errno = %d\n", file, -err);
1212                 goto out_close;
1213         }
1214         if(err)
1215                 return(fd);
1216
1217         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1218
1219         /* Allow switching only if no mismatch. */
1220         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1221                 printk("Switching backing file to '%s'\n", *backing_file_out);
1222                 err = write_cow_header(file, fd, *backing_file_out,
1223                                        sectorsize, align, &size);
1224                 if (err) {
1225                         printk("Switch failed, errno = %d\n", -err);
1226                         goto out_close;
1227                 }
1228         } else {
1229                 *backing_file_out = backing_file;
1230                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1231                 if (err)
1232                         goto out_close;
1233         }
1234
1235         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1236                   bitmap_len_out, data_offset_out);
1237
1238         return fd;
1239  out_close:
1240         os_close_file(fd);
1241         return err;
1242 }
1243
1244 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1245                     int sectorsize, int alignment, int *bitmap_offset_out,
1246                     unsigned long *bitmap_len_out, int *data_offset_out)
1247 {
1248         int err, fd;
1249
1250         flags.c = 1;
1251         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1252         if(fd < 0){
1253                 err = fd;
1254                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1255                        -err);
1256                 goto out;
1257         }
1258
1259         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1260                             bitmap_offset_out, bitmap_len_out,
1261                             data_offset_out);
1262         if(!err)
1263                 return(fd);
1264         os_close_file(fd);
1265  out:
1266         return(err);
1267 }
1268
1269 static int update_bitmap(struct io_thread_req *req)
1270 {
1271         int n;
1272
1273         if(req->cow_offset == -1)
1274                 return(0);
1275
1276         n = os_seek_file(req->fds[1], req->cow_offset);
1277         if(n < 0){
1278                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1279                 return(1);
1280         }
1281
1282         n = os_write_file(req->fds[1], &req->bitmap_words,
1283                           sizeof(req->bitmap_words));
1284         if(n != sizeof(req->bitmap_words)){
1285                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1286                        req->fds[1]);
1287                 return(1);
1288         }
1289
1290         return(0);
1291 }
1292
1293 void do_io(struct io_thread_req *req)
1294 {
1295         char *buf;
1296         unsigned long len;
1297         int n, nsectors, start, end, bit;
1298         int err;
1299         __u64 off;
1300
1301         nsectors = req->length / req->sectorsize;
1302         start = 0;
1303         do {
1304                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1305                 end = start;
1306                 while((end < nsectors) &&
1307                       (ubd_test_bit(end, (unsigned char *)
1308                                     &req->sector_mask) == bit))
1309                         end++;
1310
1311                 off = req->offset + req->offsets[bit] +
1312                         start * req->sectorsize;
1313                 len = (end - start) * req->sectorsize;
1314                 buf = &req->buffer[start * req->sectorsize];
1315
1316                 err = os_seek_file(req->fds[bit], off);
1317                 if(err < 0){
1318                         printk("do_io - lseek failed : err = %d\n", -err);
1319                         req->error = 1;
1320                         return;
1321                 }
1322                 if(req->op == UBD_READ){
1323                         n = 0;
1324                         do {
1325                                 buf = &buf[n];
1326                                 len -= n;
1327                                 n = os_read_file(req->fds[bit], buf, len);
1328                                 if (n < 0) {
1329                                         printk("do_io - read failed, err = %d "
1330                                                "fd = %d\n", -n, req->fds[bit]);
1331                                         req->error = 1;
1332                                         return;
1333                                 }
1334                         } while((n < len) && (n != 0));
1335                         if (n < len) memset(&buf[n], 0, len - n);
1336                 } else {
1337                         n = os_write_file(req->fds[bit], buf, len);
1338                         if(n != len){
1339                                 printk("do_io - write failed err = %d "
1340                                        "fd = %d\n", -n, req->fds[bit]);
1341                                 req->error = 1;
1342                                 return;
1343                         }
1344                 }
1345
1346                 start = end;
1347         } while(start < nsectors);
1348
1349         req->error = update_bitmap(req);
1350 }
1351
1352 /* Changed in start_io_thread, which is serialized by being called only
1353  * from ubd_init, which is an initcall.
1354  */
1355 int kernel_fd = -1;
1356
1357 /* Only changed by the io thread */
1358 int io_count = 0;
1359
1360 int io_thread(void *arg)
1361 {
1362         struct io_thread_req req;
1363         int n;
1364
1365         ignore_sigwinch_sig();
1366         while(1){
1367                 n = os_read_file(kernel_fd, &req, sizeof(req));
1368                 if(n != sizeof(req)){
1369                         if(n < 0)
1370                                 printk("io_thread - read failed, fd = %d, "
1371                                        "err = %d\n", kernel_fd, -n);
1372                         else {
1373                                 printk("io_thread - short read, fd = %d, "
1374                                        "length = %d\n", kernel_fd, n);
1375                         }
1376                         continue;
1377                 }
1378                 io_count++;
1379                 do_io(&req);
1380                 n = os_write_file(kernel_fd, &req, sizeof(req));
1381                 if(n != sizeof(req))
1382                         printk("io_thread - write failed, fd = %d, err = %d\n",
1383                                kernel_fd, -n);
1384         }
1385
1386         return 0;
1387 }