[PATCH] uml ubd driver: document some struct fields
[safe/jmp/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/module.h"
24 #include "linux/blkdev.h"
25 #include "linux/hdreg.h"
26 #include "linux/init.h"
27 #include "linux/cdrom.h"
28 #include "linux/proc_fs.h"
29 #include "linux/ctype.h"
30 #include "linux/capability.h"
31 #include "linux/mm.h"
32 #include "linux/vmalloc.h"
33 #include "linux/blkpg.h"
34 #include "linux/genhd.h"
35 #include "linux/spinlock.h"
36 #include "linux/platform_device.h"
37 #include "asm/segment.h"
38 #include "asm/uaccess.h"
39 #include "asm/irq.h"
40 #include "asm/types.h"
41 #include "asm/tlbflush.h"
42 #include "user_util.h"
43 #include "mem_user.h"
44 #include "kern_util.h"
45 #include "kern.h"
46 #include "mconsole_kern.h"
47 #include "init.h"
48 #include "irq_user.h"
49 #include "irq_kern.h"
50 #include "ubd_user.h"
51 #include "os.h"
52 #include "mem.h"
53 #include "mem_kern.h"
54 #include "cow.h"
55
56 enum ubd_req { UBD_READ, UBD_WRITE };
57
58 struct io_thread_req {
59         enum ubd_req op;
60         int fds[2];
61         unsigned long offsets[2];
62         unsigned long long offset;
63         unsigned long length;
64         char *buffer;
65         int sectorsize;
66         unsigned long sector_mask;
67         unsigned long long cow_offset;
68         unsigned long bitmap_words[2];
69         int error;
70 };
71
72 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
73                          char **backing_file_out, int *bitmap_offset_out,
74                          unsigned long *bitmap_len_out, int *data_offset_out,
75                          int *create_cow_out);
76 extern int create_cow_file(char *cow_file, char *backing_file,
77                            struct openflags flags, int sectorsize,
78                            int alignment, int *bitmap_offset_out,
79                            unsigned long *bitmap_len_out,
80                            int *data_offset_out);
81 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
82 extern void do_io(struct io_thread_req *req);
83
84 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
85 {
86         __u64 n;
87         int bits, off;
88
89         bits = sizeof(data[0]) * 8;
90         n = bit / bits;
91         off = bit % bits;
92         return((data[n] & (1 << off)) != 0);
93 }
94
95 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
96 {
97         __u64 n;
98         int bits, off;
99
100         bits = sizeof(data[0]) * 8;
101         n = bit / bits;
102         off = bit % bits;
103         data[n] |= (1 << off);
104 }
105 /*End stuff from ubd_user.h*/
106
107 #define DRIVER_NAME "uml-blkdev"
108
109 static DEFINE_SPINLOCK(ubd_io_lock);
110 static DEFINE_SPINLOCK(ubd_lock);
111
112 static void (*do_ubd)(void);
113
114 static int ubd_open(struct inode * inode, struct file * filp);
115 static int ubd_release(struct inode * inode, struct file * file);
116 static int ubd_ioctl(struct inode * inode, struct file * file,
117                      unsigned int cmd, unsigned long arg);
118 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
119
120 #define MAX_DEV (16)
121
122 static struct block_device_operations ubd_blops = {
123         .owner          = THIS_MODULE,
124         .open           = ubd_open,
125         .release        = ubd_release,
126         .ioctl          = ubd_ioctl,
127         .getgeo         = ubd_getgeo,
128 };
129
130 /* Protected by the queue_lock */
131 static request_queue_t *ubd_queue;
132
133 /* Protected by ubd_lock */
134 static int fake_major = MAJOR_NR;
135
136 static struct gendisk *ubd_gendisk[MAX_DEV];
137 static struct gendisk *fake_gendisk[MAX_DEV];
138
139 #ifdef CONFIG_BLK_DEV_UBD_SYNC
140 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
141                                          .cl = 1 })
142 #else
143 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
144                                          .cl = 1 })
145 #endif
146
147 /* Not protected - changed only in ubd_setup_common and then only to
148  * to enable O_SYNC.
149  */
150 static struct openflags global_openflags = OPEN_FLAGS;
151
152 struct cow {
153         /* backing file name */
154         char *file;
155         /* backing file fd */
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         /* name (and fd, below) of the file opened for writing, either the
165          * backing or the cow file. */
166         char *file;
167         int count;
168         int fd;
169         __u64 size;
170         struct openflags boot_openflags;
171         struct openflags openflags;
172         int shared;
173         int no_cow;
174         struct cow cow;
175         struct platform_device pdev;
176 };
177
178 #define DEFAULT_COW { \
179         .file =                 NULL, \
180         .fd =                   -1, \
181         .bitmap =               NULL, \
182         .bitmap_offset =        0, \
183         .data_offset =          0, \
184 }
185
186 #define DEFAULT_UBD { \
187         .file =                 NULL, \
188         .count =                0, \
189         .fd =                   -1, \
190         .size =                 -1, \
191         .boot_openflags =       OPEN_FLAGS, \
192         .openflags =            OPEN_FLAGS, \
193         .no_cow =               0, \
194         .shared =               0, \
195         .cow =                  DEFAULT_COW, \
196 }
197
198 struct ubd ubd_dev[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
199
200 static int ubd0_init(void)
201 {
202         struct ubd *dev = &ubd_dev[0];
203
204         if(dev->file == NULL)
205                 dev->file = "root_fs";
206         return(0);
207 }
208
209 __initcall(ubd0_init);
210
211 /* Only changed by fake_ide_setup which is a setup */
212 static int fake_ide = 0;
213 static struct proc_dir_entry *proc_ide_root = NULL;
214 static struct proc_dir_entry *proc_ide = NULL;
215
216 static void make_proc_ide(void)
217 {
218         proc_ide_root = proc_mkdir("ide", NULL);
219         proc_ide = proc_mkdir("ide0", proc_ide_root);
220 }
221
222 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
223                                int *eof, void *data)
224 {
225         int len;
226
227         strcpy(page, "disk\n");
228         len = strlen("disk\n");
229         len -= off;
230         if (len < count){
231                 *eof = 1;
232                 if (len <= 0) return 0;
233         }
234         else len = count;
235         *start = page + off;
236         return len;
237 }
238
239 static void make_ide_entries(char *dev_name)
240 {
241         struct proc_dir_entry *dir, *ent;
242         char name[64];
243
244         if(proc_ide_root == NULL) make_proc_ide();
245
246         dir = proc_mkdir(dev_name, proc_ide);
247         if(!dir) return;
248
249         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
250         if(!ent) return;
251         ent->nlink = 1;
252         ent->data = NULL;
253         ent->read_proc = proc_ide_read_media;
254         ent->write_proc = NULL;
255         sprintf(name,"ide0/%s", dev_name);
256         proc_symlink(dev_name, proc_ide_root, name);
257 }
258
259 static int fake_ide_setup(char *str)
260 {
261         fake_ide = 1;
262         return(1);
263 }
264
265 __setup("fake_ide", fake_ide_setup);
266
267 __uml_help(fake_ide_setup,
268 "fake_ide\n"
269 "    Create ide0 entries that map onto ubd devices.\n\n"
270 );
271
272 static int parse_unit(char **ptr)
273 {
274         char *str = *ptr, *end;
275         int n = -1;
276
277         if(isdigit(*str)) {
278                 n = simple_strtoul(str, &end, 0);
279                 if(end == str)
280                         return(-1);
281                 *ptr = end;
282         }
283         else if (('a' <= *str) && (*str <= 'z')) {
284                 n = *str - 'a';
285                 str++;
286                 *ptr = str;
287         }
288         return(n);
289 }
290
291 static int ubd_setup_common(char *str, int *index_out)
292 {
293         struct ubd *dev;
294         struct openflags flags = global_openflags;
295         char *backing_file;
296         int n, err, i;
297
298         if(index_out) *index_out = -1;
299         n = *str;
300         if(n == '='){
301                 char *end;
302                 int major;
303
304                 str++;
305                 if(!strcmp(str, "sync")){
306                         global_openflags = of_sync(global_openflags);
307                         return(0);
308                 }
309                 major = simple_strtoul(str, &end, 0);
310                 if((*end != '\0') || (end == str)){
311                         printk(KERN_ERR
312                                "ubd_setup : didn't parse major number\n");
313                         return(1);
314                 }
315
316                 err = 1;
317                 spin_lock(&ubd_lock);
318                 if(fake_major != MAJOR_NR){
319                         printk(KERN_ERR "Can't assign a fake major twice\n");
320                         goto out1;
321                 }
322
323                 fake_major = major;
324
325                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
326                        major);
327                 err = 0;
328         out1:
329                 spin_unlock(&ubd_lock);
330                 return(err);
331         }
332
333         n = parse_unit(&str);
334         if(n < 0){
335                 printk(KERN_ERR "ubd_setup : couldn't parse unit number "
336                        "'%s'\n", str);
337                 return(1);
338         }
339         if(n >= MAX_DEV){
340                 printk(KERN_ERR "ubd_setup : index %d out of range "
341                        "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
342                 return(1);
343         }
344
345         err = 1;
346         spin_lock(&ubd_lock);
347
348         dev = &ubd_dev[n];
349         if(dev->file != NULL){
350                 printk(KERN_ERR "ubd_setup : device already configured\n");
351                 goto out;
352         }
353
354         if (index_out)
355                 *index_out = n;
356
357         for (i = 0; i < sizeof("rscd="); i++) {
358                 switch (*str) {
359                 case 'r':
360                         flags.w = 0;
361                         break;
362                 case 's':
363                         flags.s = 1;
364                         break;
365                 case 'd':
366                         dev->no_cow = 1;
367                         break;
368                 case 'c':
369                         dev->shared = 1;
370                         break;
371                 case '=':
372                         str++;
373                         goto break_loop;
374                 default:
375                         printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r, s, c, or d)\n");
376                         goto out;
377                 }
378                 str++;
379         }
380
381         if (*str == '=')
382                 printk(KERN_ERR "ubd_setup : Too many flags specified\n");
383         else
384                 printk(KERN_ERR "ubd_setup : Expected '='\n");
385         goto out;
386
387 break_loop:
388         err = 0;
389         backing_file = strchr(str, ',');
390
391         if (!backing_file) {
392                 backing_file = strchr(str, ':');
393         }
394
395         if(backing_file){
396                 if(dev->no_cow)
397                         printk(KERN_ERR "Can't specify both 'd' and a "
398                                "cow file\n");
399                 else {
400                         *backing_file = '\0';
401                         backing_file++;
402                 }
403         }
404         dev->file = str;
405         dev->cow.file = backing_file;
406         dev->boot_openflags = flags;
407 out:
408         spin_unlock(&ubd_lock);
409         return(err);
410 }
411
412 static int ubd_setup(char *str)
413 {
414         ubd_setup_common(str, NULL);
415         return(1);
416 }
417
418 __setup("ubd", ubd_setup);
419 __uml_help(ubd_setup,
420 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
421 "    This is used to associate a device with a file in the underlying\n"
422 "    filesystem. When specifying two filenames, the first one is the\n"
423 "    COW name and the second is the backing file name. As separator you can\n"
424 "    use either a ':' or a ',': the first one allows writing things like;\n"
425 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
426 "    while with a ',' the shell would not expand the 2nd '~'.\n"
427 "    When using only one filename, UML will detect whether to thread it like\n"
428 "    a COW file or a backing file. To override this detection, add the 'd'\n"
429 "    flag:\n"
430 "       ubd0d=BackingFile\n"
431 "    Usually, there is a filesystem in the file, but \n"
432 "    that's not required. Swap devices containing swap files can be\n"
433 "    specified like this. Also, a file which doesn't contain a\n"
434 "    filesystem can have its contents read in the virtual \n"
435 "    machine by running 'dd' on the device. <n> must be in the range\n"
436 "    0 to 7. Appending an 'r' to the number will cause that device\n"
437 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
438 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
439 );
440
441 static int udb_setup(char *str)
442 {
443         printk("udb%s specified on command line is almost certainly a ubd -> "
444                "udb TYPO\n", str);
445         return(1);
446 }
447
448 __setup("udb", udb_setup);
449 __uml_help(udb_setup,
450 "udb\n"
451 "    This option is here solely to catch ubd -> udb typos, which can be\n"
452 "    to impossible to catch visually unless you specifically look for\n"
453 "    them.  The only result of any option starting with 'udb' is an error\n"
454 "    in the boot output.\n\n"
455 );
456
457 static int fakehd_set = 0;
458 static int fakehd(char *str)
459 {
460         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
461         fakehd_set = 1;
462         return 1;
463 }
464
465 __setup("fakehd", fakehd);
466 __uml_help(fakehd,
467 "fakehd\n"
468 "    Change the ubd device name to \"hd\".\n\n"
469 );
470
471 static void do_ubd_request(request_queue_t * q);
472
473 /* Only changed by ubd_init, which is an initcall. */
474 int thread_fd = -1;
475
476 /* Changed by ubd_handler, which is serialized because interrupts only
477  * happen on CPU 0.
478  */
479 int intr_count = 0;
480
481 /* call ubd_finish if you need to serialize */
482 static void __ubd_finish(struct request *req, int error)
483 {
484         int nsect;
485
486         if(error){
487                 end_request(req, 0);
488                 return;
489         }
490         nsect = req->current_nr_sectors;
491         req->sector += nsect;
492         req->buffer += nsect << 9;
493         req->errors = 0;
494         req->nr_sectors -= nsect;
495         req->current_nr_sectors = 0;
496         end_request(req, 1);
497 }
498
499 static inline void ubd_finish(struct request *req, int error)
500 {
501         spin_lock(&ubd_io_lock);
502         __ubd_finish(req, error);
503         spin_unlock(&ubd_io_lock);
504 }
505
506 /* Called without ubd_io_lock held */
507 static void ubd_handler(void)
508 {
509         struct io_thread_req req;
510         struct request *rq = elv_next_request(ubd_queue);
511         int n;
512
513         do_ubd = NULL;
514         intr_count++;
515         n = os_read_file(thread_fd, &req, sizeof(req));
516         if(n != sizeof(req)){
517                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
518                        "err = %d\n", os_getpid(), -n);
519                 spin_lock(&ubd_io_lock);
520                 end_request(rq, 0);
521                 spin_unlock(&ubd_io_lock);
522                 return;
523         }
524
525         ubd_finish(rq, req.error);
526         reactivate_fd(thread_fd, UBD_IRQ);      
527         do_ubd_request(ubd_queue);
528 }
529
530 static irqreturn_t ubd_intr(int irq, void *dev)
531 {
532         ubd_handler();
533         return(IRQ_HANDLED);
534 }
535
536 /* Only changed by ubd_init, which is an initcall. */
537 static int io_pid = -1;
538
539 void kill_io_thread(void)
540 {
541         if(io_pid != -1)
542                 os_kill_process(io_pid, 1);
543 }
544
545 __uml_exitcall(kill_io_thread);
546
547 static int ubd_file_size(struct ubd *dev, __u64 *size_out)
548 {
549         char *file;
550
551         file = dev->cow.file ? dev->cow.file : dev->file;
552         return(os_file_size(file, size_out));
553 }
554
555 static void ubd_close(struct ubd *dev)
556 {
557         os_close_file(dev->fd);
558         if(dev->cow.file == NULL)
559                 return;
560
561         os_close_file(dev->cow.fd);
562         vfree(dev->cow.bitmap);
563         dev->cow.bitmap = NULL;
564 }
565
566 static int ubd_open_dev(struct ubd *dev)
567 {
568         struct openflags flags;
569         char **back_ptr;
570         int err, create_cow, *create_ptr;
571
572         dev->openflags = dev->boot_openflags;
573         create_cow = 0;
574         create_ptr = (dev->cow.file != NULL) ? &create_cow : NULL;
575         back_ptr = dev->no_cow ? NULL : &dev->cow.file;
576         dev->fd = open_ubd_file(dev->file, &dev->openflags, dev->shared,
577                                 back_ptr, &dev->cow.bitmap_offset,
578                                 &dev->cow.bitmap_len, &dev->cow.data_offset,
579                                 create_ptr);
580
581         if((dev->fd == -ENOENT) && create_cow){
582                 dev->fd = create_cow_file(dev->file, dev->cow.file,
583                                           dev->openflags, 1 << 9, PAGE_SIZE,
584                                           &dev->cow.bitmap_offset,
585                                           &dev->cow.bitmap_len,
586                                           &dev->cow.data_offset);
587                 if(dev->fd >= 0){
588                         printk(KERN_INFO "Creating \"%s\" as COW file for "
589                                "\"%s\"\n", dev->file, dev->cow.file);
590                 }
591         }
592
593         if(dev->fd < 0){
594                 printk("Failed to open '%s', errno = %d\n", dev->file,
595                        -dev->fd);
596                 return(dev->fd);
597         }
598
599         if(dev->cow.file != NULL){
600                 err = -ENOMEM;
601                 dev->cow.bitmap = (void *) vmalloc(dev->cow.bitmap_len);
602                 if(dev->cow.bitmap == NULL){
603                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
604                         goto error;
605                 }
606                 flush_tlb_kernel_vm();
607
608                 err = read_cow_bitmap(dev->fd, dev->cow.bitmap,
609                                       dev->cow.bitmap_offset,
610                                       dev->cow.bitmap_len);
611                 if(err < 0)
612                         goto error;
613
614                 flags = dev->openflags;
615                 flags.w = 0;
616                 err = open_ubd_file(dev->cow.file, &flags, dev->shared, NULL,
617                                     NULL, NULL, NULL, NULL);
618                 if(err < 0) goto error;
619                 dev->cow.fd = err;
620         }
621         return(0);
622  error:
623         os_close_file(dev->fd);
624         return(err);
625 }
626
627 static int ubd_new_disk(int major, u64 size, int unit,
628                         struct gendisk **disk_out)
629                         
630 {
631         struct gendisk *disk;
632
633         disk = alloc_disk(1 << UBD_SHIFT);
634         if(disk == NULL)
635                 return(-ENOMEM);
636
637         disk->major = major;
638         disk->first_minor = unit << UBD_SHIFT;
639         disk->fops = &ubd_blops;
640         set_capacity(disk, size / 512);
641         if(major == MAJOR_NR)
642                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
643         else
644                 sprintf(disk->disk_name, "ubd_fake%d", unit);
645
646         /* sysfs register (not for ide fake devices) */
647         if (major == MAJOR_NR) {
648                 ubd_dev[unit].pdev.id   = unit;
649                 ubd_dev[unit].pdev.name = DRIVER_NAME;
650                 platform_device_register(&ubd_dev[unit].pdev);
651                 disk->driverfs_dev = &ubd_dev[unit].pdev.dev;
652         }
653
654         disk->private_data = &ubd_dev[unit];
655         disk->queue = ubd_queue;
656         add_disk(disk);
657
658         *disk_out = disk;
659         return 0;
660 }
661
662 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
663
664 static int ubd_add(int n)
665 {
666         struct ubd *dev = &ubd_dev[n];
667         int err;
668
669         err = -ENODEV;
670         if(dev->file == NULL)
671                 goto out;
672
673         err = ubd_file_size(dev, &dev->size);
674         if(err < 0)
675                 goto out;
676
677         dev->size = ROUND_BLOCK(dev->size);
678
679         err = ubd_new_disk(MAJOR_NR, dev->size, n, &ubd_gendisk[n]);
680         if(err)
681                 goto out;
682
683         if(fake_major != MAJOR_NR)
684                 ubd_new_disk(fake_major, dev->size, n,
685                              &fake_gendisk[n]);
686
687         /* perhaps this should also be under the "if (fake_major)" above */
688         /* using the fake_disk->disk_name and also the fakehd_set name */
689         if (fake_ide)
690                 make_ide_entries(ubd_gendisk[n]->disk_name);
691
692         err = 0;
693 out:
694         return err;
695 }
696
697 static int ubd_config(char *str)
698 {
699         int n, err;
700
701         str = kstrdup(str, GFP_KERNEL);
702         if(str == NULL){
703                 printk(KERN_ERR "ubd_config failed to strdup string\n");
704                 return(1);
705         }
706         err = ubd_setup_common(str, &n);
707         if(err){
708                 kfree(str);
709                 return(-1);
710         }
711         if(n == -1) return(0);
712
713         spin_lock(&ubd_lock);
714         err = ubd_add(n);
715         if(err)
716                 ubd_dev[n].file = NULL;
717         spin_unlock(&ubd_lock);
718
719         return(err);
720 }
721
722 static int ubd_get_config(char *name, char *str, int size, char **error_out)
723 {
724         struct ubd *dev;
725         int n, len = 0;
726
727         n = parse_unit(&name);
728         if((n >= MAX_DEV) || (n < 0)){
729                 *error_out = "ubd_get_config : device number out of range";
730                 return(-1);
731         }
732
733         dev = &ubd_dev[n];
734         spin_lock(&ubd_lock);
735
736         if(dev->file == NULL){
737                 CONFIG_CHUNK(str, size, len, "", 1);
738                 goto out;
739         }
740
741         CONFIG_CHUNK(str, size, len, dev->file, 0);
742
743         if(dev->cow.file != NULL){
744                 CONFIG_CHUNK(str, size, len, ",", 0);
745                 CONFIG_CHUNK(str, size, len, dev->cow.file, 1);
746         }
747         else CONFIG_CHUNK(str, size, len, "", 1);
748
749  out:
750         spin_unlock(&ubd_lock);
751         return(len);
752 }
753
754 static int ubd_id(char **str, int *start_out, int *end_out)
755 {
756         int n;
757
758         n = parse_unit(str);
759         *start_out = 0;
760         *end_out = MAX_DEV - 1;
761         return n;
762 }
763
764 static int ubd_remove(int n)
765 {
766         struct ubd *dev;
767         int err = -ENODEV;
768
769         spin_lock(&ubd_lock);
770
771         if(ubd_gendisk[n] == NULL)
772                 goto out;
773
774         dev = &ubd_dev[n];
775
776         if(dev->file == NULL)
777                 goto out;
778
779         /* you cannot remove a open disk */
780         err = -EBUSY;
781         if(dev->count > 0)
782                 goto out;
783
784         del_gendisk(ubd_gendisk[n]);
785         put_disk(ubd_gendisk[n]);
786         ubd_gendisk[n] = NULL;
787
788         if(fake_gendisk[n] != NULL){
789                 del_gendisk(fake_gendisk[n]);
790                 put_disk(fake_gendisk[n]);
791                 fake_gendisk[n] = NULL;
792         }
793
794         platform_device_unregister(&dev->pdev);
795         *dev = ((struct ubd) DEFAULT_UBD);
796         err = 0;
797 out:
798         spin_unlock(&ubd_lock);
799         return err;
800 }
801
802 static struct mc_device ubd_mc = {
803         .name           = "ubd",
804         .config         = ubd_config,
805         .get_config     = ubd_get_config,
806         .id             = ubd_id,
807         .remove         = ubd_remove,
808 };
809
810 static int ubd_mc_init(void)
811 {
812         mconsole_register_dev(&ubd_mc);
813         return 0;
814 }
815
816 __initcall(ubd_mc_init);
817
818 static struct platform_driver ubd_driver = {
819         .driver = {
820                 .name  = DRIVER_NAME,
821         },
822 };
823
824 int ubd_init(void)
825 {
826         int i;
827
828         if (register_blkdev(MAJOR_NR, "ubd"))
829                 return -1;
830
831         ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
832         if (!ubd_queue) {
833                 unregister_blkdev(MAJOR_NR, "ubd");
834                 return -1;
835         }
836                 
837         if (fake_major != MAJOR_NR) {
838                 char name[sizeof("ubd_nnn\0")];
839
840                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
841                 if (register_blkdev(fake_major, "ubd"))
842                         return -1;
843         }
844         platform_driver_register(&ubd_driver);
845         for (i = 0; i < MAX_DEV; i++)
846                 ubd_add(i);
847         return 0;
848 }
849
850 late_initcall(ubd_init);
851
852 int ubd_driver_init(void){
853         unsigned long stack;
854         int err;
855
856         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
857         if(global_openflags.s){
858                 printk(KERN_INFO "ubd: Synchronous mode\n");
859                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
860                  * enough. So use anyway the io thread. */
861         }
862         stack = alloc_stack(0, 0);
863         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
864                                  &thread_fd);
865         if(io_pid < 0){
866                 printk(KERN_ERR
867                        "ubd : Failed to start I/O thread (errno = %d) - "
868                        "falling back to synchronous I/O\n", -io_pid);
869                 io_pid = -1;
870                 return(0);
871         }
872         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
873                              IRQF_DISABLED, "ubd", ubd_dev);
874         if(err != 0)
875                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
876         return 0;
877 }
878
879 device_initcall(ubd_driver_init);
880
881 static int ubd_open(struct inode *inode, struct file *filp)
882 {
883         struct gendisk *disk = inode->i_bdev->bd_disk;
884         struct ubd *dev = disk->private_data;
885         int err = 0;
886
887         if(dev->count == 0){
888                 err = ubd_open_dev(dev);
889                 if(err){
890                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
891                                disk->disk_name, dev->file, -err);
892                         goto out;
893                 }
894         }
895         dev->count++;
896         set_disk_ro(disk, !dev->openflags.w);
897
898         /* This should no more be needed. And it didn't work anyway to exclude
899          * read-write remounting of filesystems.*/
900         /*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
901                 if(--dev->count == 0) ubd_close(dev);
902                 err = -EROFS;
903         }*/
904  out:
905         return(err);
906 }
907
908 static int ubd_release(struct inode * inode, struct file * file)
909 {
910         struct gendisk *disk = inode->i_bdev->bd_disk;
911         struct ubd *dev = disk->private_data;
912
913         if(--dev->count == 0)
914                 ubd_close(dev);
915         return(0);
916 }
917
918 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
919                           __u64 *cow_offset, unsigned long *bitmap,
920                           __u64 bitmap_offset, unsigned long *bitmap_words,
921                           __u64 bitmap_len)
922 {
923         __u64 sector = io_offset >> 9;
924         int i, update_bitmap = 0;
925
926         for(i = 0; i < length >> 9; i++){
927                 if(cow_mask != NULL)
928                         ubd_set_bit(i, (unsigned char *) cow_mask);
929                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
930                         continue;
931
932                 update_bitmap = 1;
933                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
934         }
935
936         if(!update_bitmap)
937                 return;
938
939         *cow_offset = sector / (sizeof(unsigned long) * 8);
940
941         /* This takes care of the case where we're exactly at the end of the
942          * device, and *cow_offset + 1 is off the end.  So, just back it up
943          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
944          * for the original diagnosis.
945          */
946         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
947                            sizeof(unsigned long) - 1))
948                 (*cow_offset)--;
949
950         bitmap_words[0] = bitmap[*cow_offset];
951         bitmap_words[1] = bitmap[*cow_offset + 1];
952
953         *cow_offset *= sizeof(unsigned long);
954         *cow_offset += bitmap_offset;
955 }
956
957 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
958                        __u64 bitmap_offset, __u64 bitmap_len)
959 {
960         __u64 sector = req->offset >> 9;
961         int i;
962
963         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
964                 panic("Operation too long");
965
966         if(req->op == UBD_READ) {
967                 for(i = 0; i < req->length >> 9; i++){
968                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
969                                 ubd_set_bit(i, (unsigned char *)
970                                             &req->sector_mask);
971                 }
972         }
973         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
974                            &req->cow_offset, bitmap, bitmap_offset,
975                            req->bitmap_words, bitmap_len);
976 }
977
978 /* Called with ubd_io_lock held */
979 static int prepare_request(struct request *req, struct io_thread_req *io_req)
980 {
981         struct gendisk *disk = req->rq_disk;
982         struct ubd *dev = disk->private_data;
983         __u64 offset;
984         int len;
985
986         /* This should be impossible now */
987         if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
988                 printk("Write attempted on readonly ubd device %s\n",
989                        disk->disk_name);
990                 end_request(req, 0);
991                 return(1);
992         }
993
994         offset = ((__u64) req->sector) << 9;
995         len = req->current_nr_sectors << 9;
996
997         io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
998         io_req->fds[1] = dev->fd;
999         io_req->cow_offset = -1;
1000         io_req->offset = offset;
1001         io_req->length = len;
1002         io_req->error = 0;
1003         io_req->sector_mask = 0;
1004
1005         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1006         io_req->offsets[0] = 0;
1007         io_req->offsets[1] = dev->cow.data_offset;
1008         io_req->buffer = req->buffer;
1009         io_req->sectorsize = 1 << 9;
1010
1011         if(dev->cow.file != NULL)
1012                 cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset,
1013                            dev->cow.bitmap_len);
1014
1015         return(0);
1016 }
1017
1018 /* Called with ubd_io_lock held */
1019 static void do_ubd_request(request_queue_t *q)
1020 {
1021         struct io_thread_req io_req;
1022         struct request *req;
1023         int err, n;
1024
1025         if(thread_fd == -1){
1026                 while((req = elv_next_request(q)) != NULL){
1027                         err = prepare_request(req, &io_req);
1028                         if(!err){
1029                                 do_io(&io_req);
1030                                 __ubd_finish(req, io_req.error);
1031                         }
1032                 }
1033         }
1034         else {
1035                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1036                         return;
1037                 err = prepare_request(req, &io_req);
1038                 if(!err){
1039                         do_ubd = ubd_handler;
1040                         n = os_write_file(thread_fd, (char *) &io_req,
1041                                          sizeof(io_req));
1042                         if(n != sizeof(io_req))
1043                                 printk("write to io thread failed, "
1044                                        "errno = %d\n", -n);
1045                 }
1046         }
1047 }
1048
1049 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1050 {
1051         struct ubd *dev = bdev->bd_disk->private_data;
1052
1053         geo->heads = 128;
1054         geo->sectors = 32;
1055         geo->cylinders = dev->size / (128 * 32 * 512);
1056         return 0;
1057 }
1058
1059 static int ubd_ioctl(struct inode * inode, struct file * file,
1060                      unsigned int cmd, unsigned long arg)
1061 {
1062         struct ubd *dev = inode->i_bdev->bd_disk->private_data;
1063         struct hd_driveid ubd_id = {
1064                 .cyls           = 0,
1065                 .heads          = 128,
1066                 .sectors        = 32,
1067         };
1068
1069         switch (cmd) {
1070                 struct cdrom_volctrl volume;
1071         case HDIO_GET_IDENTITY:
1072                 ubd_id.cyls = dev->size / (128 * 32 * 512);
1073                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1074                                  sizeof(ubd_id)))
1075                         return(-EFAULT);
1076                 return(0);
1077                 
1078         case CDROMVOLREAD:
1079                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1080                         return(-EFAULT);
1081                 volume.channel0 = 255;
1082                 volume.channel1 = 255;
1083                 volume.channel2 = 255;
1084                 volume.channel3 = 255;
1085                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1086                         return(-EFAULT);
1087                 return(0);
1088         }
1089         return(-EINVAL);
1090 }
1091
1092 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1093 {
1094         struct uml_stat buf1, buf2;
1095         int err;
1096
1097         if(from_cmdline == NULL)
1098                 return 0;
1099         if(!strcmp(from_cmdline, from_cow))
1100                 return 0;
1101
1102         err = os_stat_file(from_cmdline, &buf1);
1103         if(err < 0){
1104                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1105                 return 0;
1106         }
1107         err = os_stat_file(from_cow, &buf2);
1108         if(err < 0){
1109                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1110                 return 1;
1111         }
1112         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1113                 return 0;
1114
1115         printk("Backing file mismatch - \"%s\" requested,\n"
1116                "\"%s\" specified in COW header of \"%s\"\n",
1117                from_cmdline, from_cow, cow);
1118         return 1;
1119 }
1120
1121 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1122 {
1123         unsigned long modtime;
1124         unsigned long long actual;
1125         int err;
1126
1127         err = os_file_modtime(file, &modtime);
1128         if(err < 0){
1129                 printk("Failed to get modification time of backing file "
1130                        "\"%s\", err = %d\n", file, -err);
1131                 return(err);
1132         }
1133
1134         err = os_file_size(file, &actual);
1135         if(err < 0){
1136                 printk("Failed to get size of backing file \"%s\", "
1137                        "err = %d\n", file, -err);
1138                 return(err);
1139         }
1140
1141         if(actual != size){
1142                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1143                  * the typecast.*/
1144                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1145                        "file\n", (unsigned long long) size, actual);
1146                 return(-EINVAL);
1147         }
1148         if(modtime != mtime){
1149                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1150                        "file\n", mtime, modtime);
1151                 return(-EINVAL);
1152         }
1153         return(0);
1154 }
1155
1156 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1157 {
1158         int err;
1159
1160         err = os_seek_file(fd, offset);
1161         if(err < 0)
1162                 return(err);
1163
1164         err = os_read_file(fd, buf, len);
1165         if(err < 0)
1166                 return(err);
1167
1168         return(0);
1169 }
1170
1171 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1172                   char **backing_file_out, int *bitmap_offset_out,
1173                   unsigned long *bitmap_len_out, int *data_offset_out,
1174                   int *create_cow_out)
1175 {
1176         time_t mtime;
1177         unsigned long long size;
1178         __u32 version, align;
1179         char *backing_file;
1180         int fd, err, sectorsize, asked_switch, mode = 0644;
1181
1182         fd = os_open_file(file, *openflags, mode);
1183         if (fd < 0) {
1184                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1185                         *create_cow_out = 1;
1186                 if (!openflags->w ||
1187                    ((fd != -EROFS) && (fd != -EACCES)))
1188                         return fd;
1189                 openflags->w = 0;
1190                 fd = os_open_file(file, *openflags, mode);
1191                 if (fd < 0)
1192                         return fd;
1193         }
1194
1195         if(shared)
1196                 printk("Not locking \"%s\" on the host\n", file);
1197         else {
1198                 err = os_lock_file(fd, openflags->w);
1199                 if(err < 0){
1200                         printk("Failed to lock '%s', err = %d\n", file, -err);
1201                         goto out_close;
1202                 }
1203         }
1204
1205         /* Successful return case! */
1206         if(backing_file_out == NULL)
1207                 return(fd);
1208
1209         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1210                               &size, &sectorsize, &align, bitmap_offset_out);
1211         if(err && (*backing_file_out != NULL)){
1212                 printk("Failed to read COW header from COW file \"%s\", "
1213                        "errno = %d\n", file, -err);
1214                 goto out_close;
1215         }
1216         if(err)
1217                 return(fd);
1218
1219         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1220
1221         /* Allow switching only if no mismatch. */
1222         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1223                 printk("Switching backing file to '%s'\n", *backing_file_out);
1224                 err = write_cow_header(file, fd, *backing_file_out,
1225                                        sectorsize, align, &size);
1226                 if (err) {
1227                         printk("Switch failed, errno = %d\n", -err);
1228                         goto out_close;
1229                 }
1230         } else {
1231                 *backing_file_out = backing_file;
1232                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1233                 if (err)
1234                         goto out_close;
1235         }
1236
1237         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1238                   bitmap_len_out, data_offset_out);
1239
1240         return fd;
1241  out_close:
1242         os_close_file(fd);
1243         return err;
1244 }
1245
1246 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1247                     int sectorsize, int alignment, int *bitmap_offset_out,
1248                     unsigned long *bitmap_len_out, int *data_offset_out)
1249 {
1250         int err, fd;
1251
1252         flags.c = 1;
1253         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1254         if(fd < 0){
1255                 err = fd;
1256                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1257                        -err);
1258                 goto out;
1259         }
1260
1261         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1262                             bitmap_offset_out, bitmap_len_out,
1263                             data_offset_out);
1264         if(!err)
1265                 return(fd);
1266         os_close_file(fd);
1267  out:
1268         return(err);
1269 }
1270
1271 static int update_bitmap(struct io_thread_req *req)
1272 {
1273         int n;
1274
1275         if(req->cow_offset == -1)
1276                 return(0);
1277
1278         n = os_seek_file(req->fds[1], req->cow_offset);
1279         if(n < 0){
1280                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1281                 return(1);
1282         }
1283
1284         n = os_write_file(req->fds[1], &req->bitmap_words,
1285                           sizeof(req->bitmap_words));
1286         if(n != sizeof(req->bitmap_words)){
1287                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1288                        req->fds[1]);
1289                 return(1);
1290         }
1291
1292         return(0);
1293 }
1294
1295 void do_io(struct io_thread_req *req)
1296 {
1297         char *buf;
1298         unsigned long len;
1299         int n, nsectors, start, end, bit;
1300         int err;
1301         __u64 off;
1302
1303         nsectors = req->length / req->sectorsize;
1304         start = 0;
1305         do {
1306                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1307                 end = start;
1308                 while((end < nsectors) &&
1309                       (ubd_test_bit(end, (unsigned char *)
1310                                     &req->sector_mask) == bit))
1311                         end++;
1312
1313                 off = req->offset + req->offsets[bit] +
1314                         start * req->sectorsize;
1315                 len = (end - start) * req->sectorsize;
1316                 buf = &req->buffer[start * req->sectorsize];
1317
1318                 err = os_seek_file(req->fds[bit], off);
1319                 if(err < 0){
1320                         printk("do_io - lseek failed : err = %d\n", -err);
1321                         req->error = 1;
1322                         return;
1323                 }
1324                 if(req->op == UBD_READ){
1325                         n = 0;
1326                         do {
1327                                 buf = &buf[n];
1328                                 len -= n;
1329                                 n = os_read_file(req->fds[bit], buf, len);
1330                                 if (n < 0) {
1331                                         printk("do_io - read failed, err = %d "
1332                                                "fd = %d\n", -n, req->fds[bit]);
1333                                         req->error = 1;
1334                                         return;
1335                                 }
1336                         } while((n < len) && (n != 0));
1337                         if (n < len) memset(&buf[n], 0, len - n);
1338                 } else {
1339                         n = os_write_file(req->fds[bit], buf, len);
1340                         if(n != len){
1341                                 printk("do_io - write failed err = %d "
1342                                        "fd = %d\n", -n, req->fds[bit]);
1343                                 req->error = 1;
1344                                 return;
1345                         }
1346                 }
1347
1348                 start = end;
1349         } while(start < nsectors);
1350
1351         req->error = update_bitmap(req);
1352 }
1353
1354 /* Changed in start_io_thread, which is serialized by being called only
1355  * from ubd_init, which is an initcall.
1356  */
1357 int kernel_fd = -1;
1358
1359 /* Only changed by the io thread */
1360 int io_count = 0;
1361
1362 int io_thread(void *arg)
1363 {
1364         struct io_thread_req req;
1365         int n;
1366
1367         ignore_sigwinch_sig();
1368         while(1){
1369                 n = os_read_file(kernel_fd, &req, sizeof(req));
1370                 if(n != sizeof(req)){
1371                         if(n < 0)
1372                                 printk("io_thread - read failed, fd = %d, "
1373                                        "err = %d\n", kernel_fd, -n);
1374                         else {
1375                                 printk("io_thread - short read, fd = %d, "
1376                                        "length = %d\n", kernel_fd, n);
1377                         }
1378                         continue;
1379                 }
1380                 io_count++;
1381                 do_io(&req);
1382                 n = os_write_file(kernel_fd, &req, sizeof(req));
1383                 if(n != sizeof(req))
1384                         printk("io_thread - write failed, fd = %d, err = %d\n",
1385                                kernel_fd, -n);
1386         }
1387
1388         return 0;
1389 }