Merge branch 'master' of /home/sam/kernel/linux-2.6/
[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         int err;
631
632         disk = alloc_disk(1 << UBD_SHIFT);
633         if(disk == NULL)
634                 return(-ENOMEM);
635
636         disk->major = major;
637         disk->first_minor = unit << UBD_SHIFT;
638         disk->fops = &ubd_blops;
639         set_capacity(disk, size / 512);
640         if(major == MAJOR_NR)
641                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
642         else
643                 sprintf(disk->disk_name, "ubd_fake%d", unit);
644
645         /* sysfs register (not for ide fake devices) */
646         if (major == MAJOR_NR) {
647                 ubd_dev[unit].pdev.id   = unit;
648                 ubd_dev[unit].pdev.name = DRIVER_NAME;
649                 platform_device_register(&ubd_dev[unit].pdev);
650                 disk->driverfs_dev = &ubd_dev[unit].pdev.dev;
651         }
652
653         disk->private_data = &ubd_dev[unit];
654         disk->queue = ubd_queue;
655         add_disk(disk);
656
657         *disk_out = disk;
658         return 0;
659 }
660
661 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
662
663 static int ubd_add(int n)
664 {
665         struct ubd *dev = &ubd_dev[n];
666         int err;
667
668         err = -ENODEV;
669         if(dev->file == NULL)
670                 goto out;
671
672         if (ubd_open_dev(dev))
673                 goto out;
674
675         err = ubd_file_size(dev, &dev->size);
676         if(err < 0)
677                 goto out_close;
678
679         dev->size = ROUND_BLOCK(dev->size);
680
681         err = ubd_new_disk(MAJOR_NR, dev->size, n, &ubd_gendisk[n]);
682         if(err)
683                 goto out_close;
684
685         if(fake_major != MAJOR_NR)
686                 ubd_new_disk(fake_major, dev->size, n,
687                              &fake_gendisk[n]);
688
689         /* perhaps this should also be under the "if (fake_major)" above */
690         /* using the fake_disk->disk_name and also the fakehd_set name */
691         if (fake_ide)
692                 make_ide_entries(ubd_gendisk[n]->disk_name);
693
694         err = 0;
695 out_close:
696         ubd_close(dev);
697 out:
698         return err;
699 }
700
701 static int ubd_config(char *str)
702 {
703         int n, err;
704
705         str = kstrdup(str, GFP_KERNEL);
706         if(str == NULL){
707                 printk(KERN_ERR "ubd_config failed to strdup string\n");
708                 return(1);
709         }
710         err = ubd_setup_common(str, &n);
711         if(err){
712                 kfree(str);
713                 return(-1);
714         }
715         if(n == -1) return(0);
716
717         spin_lock(&ubd_lock);
718         err = ubd_add(n);
719         if(err)
720                 ubd_dev[n].file = NULL;
721         spin_unlock(&ubd_lock);
722
723         return(err);
724 }
725
726 static int ubd_get_config(char *name, char *str, int size, char **error_out)
727 {
728         struct ubd *dev;
729         int n, len = 0;
730
731         n = parse_unit(&name);
732         if((n >= MAX_DEV) || (n < 0)){
733                 *error_out = "ubd_get_config : device number out of range";
734                 return(-1);
735         }
736
737         dev = &ubd_dev[n];
738         spin_lock(&ubd_lock);
739
740         if(dev->file == NULL){
741                 CONFIG_CHUNK(str, size, len, "", 1);
742                 goto out;
743         }
744
745         CONFIG_CHUNK(str, size, len, dev->file, 0);
746
747         if(dev->cow.file != NULL){
748                 CONFIG_CHUNK(str, size, len, ",", 0);
749                 CONFIG_CHUNK(str, size, len, dev->cow.file, 1);
750         }
751         else CONFIG_CHUNK(str, size, len, "", 1);
752
753  out:
754         spin_unlock(&ubd_lock);
755         return(len);
756 }
757
758 static int ubd_id(char **str, int *start_out, int *end_out)
759 {
760         int n;
761
762         n = parse_unit(str);
763         *start_out = 0;
764         *end_out = MAX_DEV - 1;
765         return n;
766 }
767
768 static int ubd_remove(int n)
769 {
770         struct ubd *dev;
771         int err = -ENODEV;
772
773         spin_lock(&ubd_lock);
774
775         if(ubd_gendisk[n] == NULL)
776                 goto out;
777
778         dev = &ubd_dev[n];
779
780         if(dev->file == NULL)
781                 goto out;
782
783         /* you cannot remove a open disk */
784         err = -EBUSY;
785         if(dev->count > 0)
786                 goto out;
787
788         del_gendisk(ubd_gendisk[n]);
789         put_disk(ubd_gendisk[n]);
790         ubd_gendisk[n] = NULL;
791
792         if(fake_gendisk[n] != NULL){
793                 del_gendisk(fake_gendisk[n]);
794                 put_disk(fake_gendisk[n]);
795                 fake_gendisk[n] = NULL;
796         }
797
798         platform_device_unregister(&dev->pdev);
799         *dev = ((struct ubd) DEFAULT_UBD);
800         err = 0;
801 out:
802         spin_unlock(&ubd_lock);
803         return err;
804 }
805
806 static struct mc_device ubd_mc = {
807         .name           = "ubd",
808         .config         = ubd_config,
809         .get_config     = ubd_get_config,
810         .id             = ubd_id,
811         .remove         = ubd_remove,
812 };
813
814 static int ubd_mc_init(void)
815 {
816         mconsole_register_dev(&ubd_mc);
817         return 0;
818 }
819
820 __initcall(ubd_mc_init);
821
822 static struct platform_driver ubd_driver = {
823         .driver = {
824                 .name  = DRIVER_NAME,
825         },
826 };
827
828 int ubd_init(void)
829 {
830         int i;
831
832         if (register_blkdev(MAJOR_NR, "ubd"))
833                 return -1;
834
835         ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
836         if (!ubd_queue) {
837                 unregister_blkdev(MAJOR_NR, "ubd");
838                 return -1;
839         }
840                 
841         if (fake_major != MAJOR_NR) {
842                 char name[sizeof("ubd_nnn\0")];
843
844                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
845                 if (register_blkdev(fake_major, "ubd"))
846                         return -1;
847         }
848         platform_driver_register(&ubd_driver);
849         for (i = 0; i < MAX_DEV; i++)
850                 ubd_add(i);
851         return 0;
852 }
853
854 late_initcall(ubd_init);
855
856 int ubd_driver_init(void){
857         unsigned long stack;
858         int err;
859
860         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
861         if(global_openflags.s){
862                 printk(KERN_INFO "ubd: Synchronous mode\n");
863                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
864                  * enough. So use anyway the io thread. */
865         }
866         stack = alloc_stack(0, 0);
867         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
868                                  &thread_fd);
869         if(io_pid < 0){
870                 printk(KERN_ERR
871                        "ubd : Failed to start I/O thread (errno = %d) - "
872                        "falling back to synchronous I/O\n", -io_pid);
873                 io_pid = -1;
874                 return(0);
875         }
876         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
877                              IRQF_DISABLED, "ubd", ubd_dev);
878         if(err != 0)
879                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
880         return 0;
881 }
882
883 device_initcall(ubd_driver_init);
884
885 static int ubd_open(struct inode *inode, struct file *filp)
886 {
887         struct gendisk *disk = inode->i_bdev->bd_disk;
888         struct ubd *dev = disk->private_data;
889         int err = 0;
890
891         if(dev->count == 0){
892                 err = ubd_open_dev(dev);
893                 if(err){
894                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
895                                disk->disk_name, dev->file, -err);
896                         goto out;
897                 }
898         }
899         dev->count++;
900         set_disk_ro(disk, !dev->openflags.w);
901
902         /* This should no more be needed. And it didn't work anyway to exclude
903          * read-write remounting of filesystems.*/
904         /*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
905                 if(--dev->count == 0) ubd_close(dev);
906                 err = -EROFS;
907         }*/
908  out:
909         return(err);
910 }
911
912 static int ubd_release(struct inode * inode, struct file * file)
913 {
914         struct gendisk *disk = inode->i_bdev->bd_disk;
915         struct ubd *dev = disk->private_data;
916
917         if(--dev->count == 0)
918                 ubd_close(dev);
919         return(0);
920 }
921
922 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
923                           __u64 *cow_offset, unsigned long *bitmap,
924                           __u64 bitmap_offset, unsigned long *bitmap_words,
925                           __u64 bitmap_len)
926 {
927         __u64 sector = io_offset >> 9;
928         int i, update_bitmap = 0;
929
930         for(i = 0; i < length >> 9; i++){
931                 if(cow_mask != NULL)
932                         ubd_set_bit(i, (unsigned char *) cow_mask);
933                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
934                         continue;
935
936                 update_bitmap = 1;
937                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
938         }
939
940         if(!update_bitmap)
941                 return;
942
943         *cow_offset = sector / (sizeof(unsigned long) * 8);
944
945         /* This takes care of the case where we're exactly at the end of the
946          * device, and *cow_offset + 1 is off the end.  So, just back it up
947          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
948          * for the original diagnosis.
949          */
950         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
951                            sizeof(unsigned long) - 1))
952                 (*cow_offset)--;
953
954         bitmap_words[0] = bitmap[*cow_offset];
955         bitmap_words[1] = bitmap[*cow_offset + 1];
956
957         *cow_offset *= sizeof(unsigned long);
958         *cow_offset += bitmap_offset;
959 }
960
961 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
962                        __u64 bitmap_offset, __u64 bitmap_len)
963 {
964         __u64 sector = req->offset >> 9;
965         int i;
966
967         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
968                 panic("Operation too long");
969
970         if(req->op == UBD_READ) {
971                 for(i = 0; i < req->length >> 9; i++){
972                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
973                                 ubd_set_bit(i, (unsigned char *)
974                                             &req->sector_mask);
975                 }
976         }
977         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
978                            &req->cow_offset, bitmap, bitmap_offset,
979                            req->bitmap_words, bitmap_len);
980 }
981
982 /* Called with ubd_io_lock held */
983 static int prepare_request(struct request *req, struct io_thread_req *io_req)
984 {
985         struct gendisk *disk = req->rq_disk;
986         struct ubd *dev = disk->private_data;
987         __u64 offset;
988         int len;
989
990         if(req->rq_status == RQ_INACTIVE) return(1);
991
992         /* This should be impossible now */
993         if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
994                 printk("Write attempted on readonly ubd device %s\n",
995                        disk->disk_name);
996                 end_request(req, 0);
997                 return(1);
998         }
999
1000         offset = ((__u64) req->sector) << 9;
1001         len = req->current_nr_sectors << 9;
1002
1003         io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
1004         io_req->fds[1] = dev->fd;
1005         io_req->cow_offset = -1;
1006         io_req->offset = offset;
1007         io_req->length = len;
1008         io_req->error = 0;
1009         io_req->sector_mask = 0;
1010
1011         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1012         io_req->offsets[0] = 0;
1013         io_req->offsets[1] = dev->cow.data_offset;
1014         io_req->buffer = req->buffer;
1015         io_req->sectorsize = 1 << 9;
1016
1017         if(dev->cow.file != NULL)
1018                 cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset,
1019                            dev->cow.bitmap_len);
1020
1021         return(0);
1022 }
1023
1024 /* Called with ubd_io_lock held */
1025 static void do_ubd_request(request_queue_t *q)
1026 {
1027         struct io_thread_req io_req;
1028         struct request *req;
1029         int err, n;
1030
1031         if(thread_fd == -1){
1032                 while((req = elv_next_request(q)) != NULL){
1033                         err = prepare_request(req, &io_req);
1034                         if(!err){
1035                                 do_io(&io_req);
1036                                 __ubd_finish(req, io_req.error);
1037                         }
1038                 }
1039         }
1040         else {
1041                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1042                         return;
1043                 err = prepare_request(req, &io_req);
1044                 if(!err){
1045                         do_ubd = ubd_handler;
1046                         n = os_write_file(thread_fd, (char *) &io_req,
1047                                          sizeof(io_req));
1048                         if(n != sizeof(io_req))
1049                                 printk("write to io thread failed, "
1050                                        "errno = %d\n", -n);
1051                 }
1052         }
1053 }
1054
1055 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1056 {
1057         struct ubd *dev = bdev->bd_disk->private_data;
1058
1059         geo->heads = 128;
1060         geo->sectors = 32;
1061         geo->cylinders = dev->size / (128 * 32 * 512);
1062         return 0;
1063 }
1064
1065 static int ubd_ioctl(struct inode * inode, struct file * file,
1066                      unsigned int cmd, unsigned long arg)
1067 {
1068         struct ubd *dev = inode->i_bdev->bd_disk->private_data;
1069         struct hd_driveid ubd_id = {
1070                 .cyls           = 0,
1071                 .heads          = 128,
1072                 .sectors        = 32,
1073         };
1074
1075         switch (cmd) {
1076                 struct cdrom_volctrl volume;
1077         case HDIO_GET_IDENTITY:
1078                 ubd_id.cyls = dev->size / (128 * 32 * 512);
1079                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1080                                  sizeof(ubd_id)))
1081                         return(-EFAULT);
1082                 return(0);
1083                 
1084         case CDROMVOLREAD:
1085                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1086                         return(-EFAULT);
1087                 volume.channel0 = 255;
1088                 volume.channel1 = 255;
1089                 volume.channel2 = 255;
1090                 volume.channel3 = 255;
1091                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1092                         return(-EFAULT);
1093                 return(0);
1094         }
1095         return(-EINVAL);
1096 }
1097
1098 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1099 {
1100         struct uml_stat buf1, buf2;
1101         int err;
1102
1103         if(from_cmdline == NULL)
1104                 return 0;
1105         if(!strcmp(from_cmdline, from_cow))
1106                 return 0;
1107
1108         err = os_stat_file(from_cmdline, &buf1);
1109         if(err < 0){
1110                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1111                 return 0;
1112         }
1113         err = os_stat_file(from_cow, &buf2);
1114         if(err < 0){
1115                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1116                 return 1;
1117         }
1118         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1119                 return 0;
1120
1121         printk("Backing file mismatch - \"%s\" requested,\n"
1122                "\"%s\" specified in COW header of \"%s\"\n",
1123                from_cmdline, from_cow, cow);
1124         return 1;
1125 }
1126
1127 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1128 {
1129         unsigned long modtime;
1130         unsigned long long actual;
1131         int err;
1132
1133         err = os_file_modtime(file, &modtime);
1134         if(err < 0){
1135                 printk("Failed to get modification time of backing file "
1136                        "\"%s\", err = %d\n", file, -err);
1137                 return(err);
1138         }
1139
1140         err = os_file_size(file, &actual);
1141         if(err < 0){
1142                 printk("Failed to get size of backing file \"%s\", "
1143                        "err = %d\n", file, -err);
1144                 return(err);
1145         }
1146
1147         if(actual != size){
1148                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1149                  * the typecast.*/
1150                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1151                        "file\n", (unsigned long long) size, actual);
1152                 return(-EINVAL);
1153         }
1154         if(modtime != mtime){
1155                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1156                        "file\n", mtime, modtime);
1157                 return(-EINVAL);
1158         }
1159         return(0);
1160 }
1161
1162 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1163 {
1164         int err;
1165
1166         err = os_seek_file(fd, offset);
1167         if(err < 0)
1168                 return(err);
1169
1170         err = os_read_file(fd, buf, len);
1171         if(err < 0)
1172                 return(err);
1173
1174         return(0);
1175 }
1176
1177 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1178                   char **backing_file_out, int *bitmap_offset_out,
1179                   unsigned long *bitmap_len_out, int *data_offset_out,
1180                   int *create_cow_out)
1181 {
1182         time_t mtime;
1183         unsigned long long size;
1184         __u32 version, align;
1185         char *backing_file;
1186         int fd, err, sectorsize, asked_switch, mode = 0644;
1187
1188         fd = os_open_file(file, *openflags, mode);
1189         if (fd < 0) {
1190                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1191                         *create_cow_out = 1;
1192                 if (!openflags->w ||
1193                    ((fd != -EROFS) && (fd != -EACCES)))
1194                         return fd;
1195                 openflags->w = 0;
1196                 fd = os_open_file(file, *openflags, mode);
1197                 if (fd < 0)
1198                         return fd;
1199         }
1200
1201         if(shared)
1202                 printk("Not locking \"%s\" on the host\n", file);
1203         else {
1204                 err = os_lock_file(fd, openflags->w);
1205                 if(err < 0){
1206                         printk("Failed to lock '%s', err = %d\n", file, -err);
1207                         goto out_close;
1208                 }
1209         }
1210
1211         /* Successful return case! */
1212         if(backing_file_out == NULL)
1213                 return(fd);
1214
1215         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1216                               &size, &sectorsize, &align, bitmap_offset_out);
1217         if(err && (*backing_file_out != NULL)){
1218                 printk("Failed to read COW header from COW file \"%s\", "
1219                        "errno = %d\n", file, -err);
1220                 goto out_close;
1221         }
1222         if(err)
1223                 return(fd);
1224
1225         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1226
1227         /* Allow switching only if no mismatch. */
1228         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1229                 printk("Switching backing file to '%s'\n", *backing_file_out);
1230                 err = write_cow_header(file, fd, *backing_file_out,
1231                                        sectorsize, align, &size);
1232                 if (err) {
1233                         printk("Switch failed, errno = %d\n", -err);
1234                         goto out_close;
1235                 }
1236         } else {
1237                 *backing_file_out = backing_file;
1238                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1239                 if (err)
1240                         goto out_close;
1241         }
1242
1243         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1244                   bitmap_len_out, data_offset_out);
1245
1246         return fd;
1247  out_close:
1248         os_close_file(fd);
1249         return err;
1250 }
1251
1252 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1253                     int sectorsize, int alignment, int *bitmap_offset_out,
1254                     unsigned long *bitmap_len_out, int *data_offset_out)
1255 {
1256         int err, fd;
1257
1258         flags.c = 1;
1259         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1260         if(fd < 0){
1261                 err = fd;
1262                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1263                        -err);
1264                 goto out;
1265         }
1266
1267         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1268                             bitmap_offset_out, bitmap_len_out,
1269                             data_offset_out);
1270         if(!err)
1271                 return(fd);
1272         os_close_file(fd);
1273  out:
1274         return(err);
1275 }
1276
1277 static int update_bitmap(struct io_thread_req *req)
1278 {
1279         int n;
1280
1281         if(req->cow_offset == -1)
1282                 return(0);
1283
1284         n = os_seek_file(req->fds[1], req->cow_offset);
1285         if(n < 0){
1286                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1287                 return(1);
1288         }
1289
1290         n = os_write_file(req->fds[1], &req->bitmap_words,
1291                           sizeof(req->bitmap_words));
1292         if(n != sizeof(req->bitmap_words)){
1293                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1294                        req->fds[1]);
1295                 return(1);
1296         }
1297
1298         return(0);
1299 }
1300
1301 void do_io(struct io_thread_req *req)
1302 {
1303         char *buf;
1304         unsigned long len;
1305         int n, nsectors, start, end, bit;
1306         int err;
1307         __u64 off;
1308
1309         nsectors = req->length / req->sectorsize;
1310         start = 0;
1311         do {
1312                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1313                 end = start;
1314                 while((end < nsectors) &&
1315                       (ubd_test_bit(end, (unsigned char *)
1316                                     &req->sector_mask) == bit))
1317                         end++;
1318
1319                 off = req->offset + req->offsets[bit] +
1320                         start * req->sectorsize;
1321                 len = (end - start) * req->sectorsize;
1322                 buf = &req->buffer[start * req->sectorsize];
1323
1324                 err = os_seek_file(req->fds[bit], off);
1325                 if(err < 0){
1326                         printk("do_io - lseek failed : err = %d\n", -err);
1327                         req->error = 1;
1328                         return;
1329                 }
1330                 if(req->op == UBD_READ){
1331                         n = 0;
1332                         do {
1333                                 buf = &buf[n];
1334                                 len -= n;
1335                                 n = os_read_file(req->fds[bit], buf, len);
1336                                 if (n < 0) {
1337                                         printk("do_io - read failed, err = %d "
1338                                                "fd = %d\n", -n, req->fds[bit]);
1339                                         req->error = 1;
1340                                         return;
1341                                 }
1342                         } while((n < len) && (n != 0));
1343                         if (n < len) memset(&buf[n], 0, len - n);
1344                 } else {
1345                         n = os_write_file(req->fds[bit], buf, len);
1346                         if(n != len){
1347                                 printk("do_io - write failed err = %d "
1348                                        "fd = %d\n", -n, req->fds[bit]);
1349                                 req->error = 1;
1350                                 return;
1351                         }
1352                 }
1353
1354                 start = end;
1355         } while(start < nsectors);
1356
1357         req->error = update_bitmap(req);
1358 }
1359
1360 /* Changed in start_io_thread, which is serialized by being called only
1361  * from ubd_init, which is an initcall.
1362  */
1363 int kernel_fd = -1;
1364
1365 /* Only changed by the io thread */
1366 int io_count = 0;
1367
1368 int io_thread(void *arg)
1369 {
1370         struct io_thread_req req;
1371         int n;
1372
1373         ignore_sigwinch_sig();
1374         while(1){
1375                 n = os_read_file(kernel_fd, &req, sizeof(req));
1376                 if(n != sizeof(req)){
1377                         if(n < 0)
1378                                 printk("io_thread - read failed, fd = %d, "
1379                                        "err = %d\n", kernel_fd, -n);
1380                         else {
1381                                 printk("io_thread - short read, fd = %d, "
1382                                        "length = %d\n", kernel_fd, n);
1383                         }
1384                         continue;
1385                 }
1386                 io_count++;
1387                 do_io(&req);
1388                 n = os_write_file(kernel_fd, &req, sizeof(req));
1389                 if(n != sizeof(req))
1390                         printk("io_thread - write failed, fd = %d, err = %d\n",
1391                                kernel_fd, -n);
1392         }
1393
1394         return 0;
1395 }