Merge to Fedora kernel-2.6.18-1.2224_FC5 patched with stable patch-2.6.18.1-vs2.0...
[linux-2.6.git] / 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 (8)
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         /* This is the backing file, actually */
154         char *file;
155         int fd;
156         unsigned long *bitmap;
157         unsigned long bitmap_len;
158         int bitmap_offset;
159         int data_offset;
160 };
161
162 struct ubd {
163         char *file;
164         int count;
165         int fd;
166         __u64 size;
167         struct openflags boot_openflags;
168         struct openflags openflags;
169         int shared;
170         int no_cow;
171         struct cow cow;
172         struct platform_device pdev;
173 };
174
175 #define DEFAULT_COW { \
176         .file =                 NULL, \
177         .fd =                   -1, \
178         .bitmap =               NULL, \
179         .bitmap_offset =        0, \
180         .data_offset =          0, \
181 }
182
183 #define DEFAULT_UBD { \
184         .file =                 NULL, \
185         .count =                0, \
186         .fd =                   -1, \
187         .size =                 -1, \
188         .boot_openflags =       OPEN_FLAGS, \
189         .openflags =            OPEN_FLAGS, \
190         .no_cow =               0, \
191         .shared =               0, \
192         .cow =                  DEFAULT_COW, \
193 }
194
195 struct ubd ubd_dev[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
196
197 static int ubd0_init(void)
198 {
199         struct ubd *dev = &ubd_dev[0];
200
201         if(dev->file == NULL)
202                 dev->file = "root_fs";
203         return(0);
204 }
205
206 __initcall(ubd0_init);
207
208 /* Only changed by fake_ide_setup which is a setup */
209 static int fake_ide = 0;
210 static struct proc_dir_entry *proc_ide_root = NULL;
211 static struct proc_dir_entry *proc_ide = NULL;
212
213 static void make_proc_ide(void)
214 {
215         proc_ide_root = proc_mkdir("ide", NULL);
216         proc_ide = proc_mkdir("ide0", proc_ide_root);
217 }
218
219 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
220                                int *eof, void *data)
221 {
222         int len;
223
224         strcpy(page, "disk\n");
225         len = strlen("disk\n");
226         len -= off;
227         if (len < count){
228                 *eof = 1;
229                 if (len <= 0) return 0;
230         }
231         else len = count;
232         *start = page + off;
233         return len;
234 }
235
236 static void make_ide_entries(char *dev_name)
237 {
238         struct proc_dir_entry *dir, *ent;
239         char name[64];
240
241         if(proc_ide_root == NULL) make_proc_ide();
242
243         dir = proc_mkdir(dev_name, proc_ide);
244         if(!dir) return;
245
246         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
247         if(!ent) return;
248         ent->nlink = 1;
249         ent->data = NULL;
250         ent->read_proc = proc_ide_read_media;
251         ent->write_proc = NULL;
252         sprintf(name,"ide0/%s", dev_name);
253         proc_symlink(dev_name, proc_ide_root, name);
254 }
255
256 static int fake_ide_setup(char *str)
257 {
258         fake_ide = 1;
259         return(1);
260 }
261
262 __setup("fake_ide", fake_ide_setup);
263
264 __uml_help(fake_ide_setup,
265 "fake_ide\n"
266 "    Create ide0 entries that map onto ubd devices.\n\n"
267 );
268
269 static int parse_unit(char **ptr)
270 {
271         char *str = *ptr, *end;
272         int n = -1;
273
274         if(isdigit(*str)) {
275                 n = simple_strtoul(str, &end, 0);
276                 if(end == str)
277                         return(-1);
278                 *ptr = end;
279         }
280         else if (('a' <= *str) && (*str <= 'h')) {
281                 n = *str - 'a';
282                 str++;
283                 *ptr = str;
284         }
285         return(n);
286 }
287
288 static int ubd_setup_common(char *str, int *index_out)
289 {
290         struct ubd *dev;
291         struct openflags flags = global_openflags;
292         char *backing_file;
293         int n, err, i;
294
295         if(index_out) *index_out = -1;
296         n = *str;
297         if(n == '='){
298                 char *end;
299                 int major;
300
301                 str++;
302                 if(!strcmp(str, "sync")){
303                         global_openflags = of_sync(global_openflags);
304                         return(0);
305                 }
306                 major = simple_strtoul(str, &end, 0);
307                 if((*end != '\0') || (end == str)){
308                         printk(KERN_ERR
309                                "ubd_setup : didn't parse major number\n");
310                         return(1);
311                 }
312
313                 err = 1;
314                 spin_lock(&ubd_lock);
315                 if(fake_major != MAJOR_NR){
316                         printk(KERN_ERR "Can't assign a fake major twice\n");
317                         goto out1;
318                 }
319
320                 fake_major = major;
321
322                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
323                        major);
324                 err = 0;
325         out1:
326                 spin_unlock(&ubd_lock);
327                 return(err);
328         }
329
330         n = parse_unit(&str);
331         if(n < 0){
332                 printk(KERN_ERR "ubd_setup : couldn't parse unit number "
333                        "'%s'\n", str);
334                 return(1);
335         }
336         if(n >= MAX_DEV){
337                 printk(KERN_ERR "ubd_setup : index %d out of range "
338                        "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
339                 return(1);
340         }
341
342         err = 1;
343         spin_lock(&ubd_lock);
344
345         dev = &ubd_dev[n];
346         if(dev->file != NULL){
347                 printk(KERN_ERR "ubd_setup : device already configured\n");
348                 goto out;
349         }
350
351         if (index_out)
352                 *index_out = n;
353
354         for (i = 0; i < sizeof("rscd="); i++) {
355                 switch (*str) {
356                 case 'r':
357                         flags.w = 0;
358                         break;
359                 case 's':
360                         flags.s = 1;
361                         break;
362                 case 'd':
363                         dev->no_cow = 1;
364                         break;
365                 case 'c':
366                         dev->shared = 1;
367                         break;
368                 case '=':
369                         str++;
370                         goto break_loop;
371                 default:
372                         printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r, s, c, or d)\n");
373                         goto out;
374                 }
375                 str++;
376         }
377
378         if (*str == '=')
379                 printk(KERN_ERR "ubd_setup : Too many flags specified\n");
380         else
381                 printk(KERN_ERR "ubd_setup : Expected '='\n");
382         goto out;
383
384 break_loop:
385         err = 0;
386         backing_file = strchr(str, ',');
387
388         if (!backing_file) {
389                 backing_file = strchr(str, ':');
390         }
391
392         if(backing_file){
393                 if(dev->no_cow)
394                         printk(KERN_ERR "Can't specify both 'd' and a "
395                                "cow file\n");
396                 else {
397                         *backing_file = '\0';
398                         backing_file++;
399                 }
400         }
401         dev->file = str;
402         dev->cow.file = backing_file;
403         dev->boot_openflags = flags;
404 out:
405         spin_unlock(&ubd_lock);
406         return(err);
407 }
408
409 static int ubd_setup(char *str)
410 {
411         ubd_setup_common(str, NULL);
412         return(1);
413 }
414
415 __setup("ubd", ubd_setup);
416 __uml_help(ubd_setup,
417 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
418 "    This is used to associate a device with a file in the underlying\n"
419 "    filesystem. When specifying two filenames, the first one is the\n"
420 "    COW name and the second is the backing file name. As separator you can\n"
421 "    use either a ':' or a ',': the first one allows writing things like;\n"
422 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
423 "    while with a ',' the shell would not expand the 2nd '~'.\n"
424 "    When using only one filename, UML will detect whether to thread it like\n"
425 "    a COW file or a backing file. To override this detection, add the 'd'\n"
426 "    flag:\n"
427 "       ubd0d=BackingFile\n"
428 "    Usually, there is a filesystem in the file, but \n"
429 "    that's not required. Swap devices containing swap files can be\n"
430 "    specified like this. Also, a file which doesn't contain a\n"
431 "    filesystem can have its contents read in the virtual \n"
432 "    machine by running 'dd' on the device. <n> must be in the range\n"
433 "    0 to 7. Appending an 'r' to the number will cause that device\n"
434 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
435 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
436 );
437
438 static int udb_setup(char *str)
439 {
440         printk("udb%s specified on command line is almost certainly a ubd -> "
441                "udb TYPO\n", str);
442         return(1);
443 }
444
445 __setup("udb", udb_setup);
446 __uml_help(udb_setup,
447 "udb\n"
448 "    This option is here solely to catch ubd -> udb typos, which can be\n"
449 "    to impossible to catch visually unless you specifically look for\n"
450 "    them.  The only result of any option starting with 'udb' is an error\n"
451 "    in the boot output.\n\n"
452 );
453
454 static int fakehd_set = 0;
455 static int fakehd(char *str)
456 {
457         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
458         fakehd_set = 1;
459         return 1;
460 }
461
462 __setup("fakehd", fakehd);
463 __uml_help(fakehd,
464 "fakehd\n"
465 "    Change the ubd device name to \"hd\".\n\n"
466 );
467
468 static void do_ubd_request(request_queue_t * q);
469
470 /* Only changed by ubd_init, which is an initcall. */
471 int thread_fd = -1;
472
473 /* Changed by ubd_handler, which is serialized because interrupts only
474  * happen on CPU 0.
475  */
476 int intr_count = 0;
477
478 /* call ubd_finish if you need to serialize */
479 static void __ubd_finish(struct request *req, int error)
480 {
481         int nsect;
482
483         if(error){
484                 end_request(req, 0);
485                 return;
486         }
487         nsect = req->current_nr_sectors;
488         req->sector += nsect;
489         req->buffer += nsect << 9;
490         req->errors = 0;
491         req->nr_sectors -= nsect;
492         req->current_nr_sectors = 0;
493         end_request(req, 1);
494 }
495
496 static inline void ubd_finish(struct request *req, int error)
497 {
498         spin_lock(&ubd_io_lock);
499         __ubd_finish(req, error);
500         spin_unlock(&ubd_io_lock);
501 }
502
503 /* Called without ubd_io_lock held */
504 static void ubd_handler(void)
505 {
506         struct io_thread_req req;
507         struct request *rq = elv_next_request(ubd_queue);
508         int n;
509
510         do_ubd = NULL;
511         intr_count++;
512         n = os_read_file(thread_fd, &req, sizeof(req));
513         if(n != sizeof(req)){
514                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
515                        "err = %d\n", os_getpid(), -n);
516                 spin_lock(&ubd_io_lock);
517                 end_request(rq, 0);
518                 spin_unlock(&ubd_io_lock);
519                 return;
520         }
521
522         ubd_finish(rq, req.error);
523         reactivate_fd(thread_fd, UBD_IRQ);      
524         do_ubd_request(ubd_queue);
525 }
526
527 static irqreturn_t ubd_intr(int irq, void *dev, struct pt_regs *unused)
528 {
529         ubd_handler();
530         return(IRQ_HANDLED);
531 }
532
533 /* Only changed by ubd_init, which is an initcall. */
534 static int io_pid = -1;
535
536 void kill_io_thread(void)
537 {
538         if(io_pid != -1)
539                 os_kill_process(io_pid, 1);
540 }
541
542 __uml_exitcall(kill_io_thread);
543
544 static int ubd_file_size(struct ubd *dev, __u64 *size_out)
545 {
546         char *file;
547
548         file = dev->cow.file ? dev->cow.file : dev->file;
549         return(os_file_size(file, size_out));
550 }
551
552 static void ubd_close(struct ubd *dev)
553 {
554         os_close_file(dev->fd);
555         if(dev->cow.file == NULL)
556                 return;
557
558         os_close_file(dev->cow.fd);
559         vfree(dev->cow.bitmap);
560         dev->cow.bitmap = NULL;
561 }
562
563 static int ubd_open_dev(struct ubd *dev)
564 {
565         struct openflags flags;
566         char **back_ptr;
567         int err, create_cow, *create_ptr;
568
569         dev->openflags = dev->boot_openflags;
570         create_cow = 0;
571         create_ptr = (dev->cow.file != NULL) ? &create_cow : NULL;
572         back_ptr = dev->no_cow ? NULL : &dev->cow.file;
573         dev->fd = open_ubd_file(dev->file, &dev->openflags, dev->shared,
574                                 back_ptr, &dev->cow.bitmap_offset,
575                                 &dev->cow.bitmap_len, &dev->cow.data_offset,
576                                 create_ptr);
577
578         if((dev->fd == -ENOENT) && create_cow){
579                 dev->fd = create_cow_file(dev->file, dev->cow.file,
580                                           dev->openflags, 1 << 9, PAGE_SIZE,
581                                           &dev->cow.bitmap_offset,
582                                           &dev->cow.bitmap_len,
583                                           &dev->cow.data_offset);
584                 if(dev->fd >= 0){
585                         printk(KERN_INFO "Creating \"%s\" as COW file for "
586                                "\"%s\"\n", dev->file, dev->cow.file);
587                 }
588         }
589
590         if(dev->fd < 0){
591                 printk("Failed to open '%s', errno = %d\n", dev->file,
592                        -dev->fd);
593                 return(dev->fd);
594         }
595
596         if(dev->cow.file != NULL){
597                 err = -ENOMEM;
598                 dev->cow.bitmap = (void *) vmalloc(dev->cow.bitmap_len);
599                 if(dev->cow.bitmap == NULL){
600                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
601                         goto error;
602                 }
603                 flush_tlb_kernel_vm();
604
605                 err = read_cow_bitmap(dev->fd, dev->cow.bitmap,
606                                       dev->cow.bitmap_offset,
607                                       dev->cow.bitmap_len);
608                 if(err < 0)
609                         goto error;
610
611                 flags = dev->openflags;
612                 flags.w = 0;
613                 err = open_ubd_file(dev->cow.file, &flags, dev->shared, NULL,
614                                     NULL, NULL, NULL, NULL);
615                 if(err < 0) goto error;
616                 dev->cow.fd = err;
617         }
618         return(0);
619  error:
620         os_close_file(dev->fd);
621         return(err);
622 }
623
624 static int ubd_new_disk(int major, u64 size, int unit,
625                         struct gendisk **disk_out)
626                         
627 {
628         struct gendisk *disk;
629
630         disk = alloc_disk(1 << UBD_SHIFT);
631         if(disk == NULL)
632                 return(-ENOMEM);
633
634         disk->major = major;
635         disk->first_minor = unit << UBD_SHIFT;
636         disk->fops = &ubd_blops;
637         set_capacity(disk, size / 512);
638         if(major == MAJOR_NR)
639                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
640         else
641                 sprintf(disk->disk_name, "ubd_fake%d", unit);
642
643         /* sysfs register (not for ide fake devices) */
644         if (major == MAJOR_NR) {
645                 ubd_dev[unit].pdev.id   = unit;
646                 ubd_dev[unit].pdev.name = DRIVER_NAME;
647                 platform_device_register(&ubd_dev[unit].pdev);
648                 disk->driverfs_dev = &ubd_dev[unit].pdev.dev;
649         }
650
651         disk->private_data = &ubd_dev[unit];
652         disk->queue = ubd_queue;
653         add_disk(disk);
654
655         *disk_out = disk;
656         return 0;
657 }
658
659 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
660
661 static int ubd_add(int n)
662 {
663         struct ubd *dev = &ubd_dev[n];
664         int err;
665
666         err = -ENODEV;
667         if(dev->file == NULL)
668                 goto out;
669
670         if (ubd_open_dev(dev))
671                 goto out;
672
673         err = ubd_file_size(dev, &dev->size);
674         if(err < 0)
675                 goto out_close;
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_close;
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_close:
694         ubd_close(dev);
695 out:
696         return err;
697 }
698
699 static int ubd_config(char *str)
700 {
701         int n, err;
702
703         str = kstrdup(str, GFP_KERNEL);
704         if(str == NULL){
705                 printk(KERN_ERR "ubd_config failed to strdup string\n");
706                 return(1);
707         }
708         err = ubd_setup_common(str, &n);
709         if(err){
710                 kfree(str);
711                 return(-1);
712         }
713         if(n == -1) return(0);
714
715         spin_lock(&ubd_lock);
716         err = ubd_add(n);
717         if(err)
718                 ubd_dev[n].file = NULL;
719         spin_unlock(&ubd_lock);
720
721         return(err);
722 }
723
724 static int ubd_get_config(char *name, char *str, int size, char **error_out)
725 {
726         struct ubd *dev;
727         int n, len = 0;
728
729         n = parse_unit(&name);
730         if((n >= MAX_DEV) || (n < 0)){
731                 *error_out = "ubd_get_config : device number out of range";
732                 return(-1);
733         }
734
735         dev = &ubd_dev[n];
736         spin_lock(&ubd_lock);
737
738         if(dev->file == NULL){
739                 CONFIG_CHUNK(str, size, len, "", 1);
740                 goto out;
741         }
742
743         CONFIG_CHUNK(str, size, len, dev->file, 0);
744
745         if(dev->cow.file != NULL){
746                 CONFIG_CHUNK(str, size, len, ",", 0);
747                 CONFIG_CHUNK(str, size, len, dev->cow.file, 1);
748         }
749         else CONFIG_CHUNK(str, size, len, "", 1);
750
751  out:
752         spin_unlock(&ubd_lock);
753         return(len);
754 }
755
756 static int ubd_id(char **str, int *start_out, int *end_out)
757 {
758         int n;
759
760         n = parse_unit(str);
761         *start_out = 0;
762         *end_out = MAX_DEV - 1;
763         return n;
764 }
765
766 static int ubd_remove(int n)
767 {
768         struct ubd *dev;
769         int err = -ENODEV;
770
771         spin_lock(&ubd_lock);
772
773         if(ubd_gendisk[n] == NULL)
774                 goto out;
775
776         dev = &ubd_dev[n];
777
778         if(dev->file == NULL)
779                 goto out;
780
781         /* you cannot remove a open disk */
782         err = -EBUSY;
783         if(dev->count > 0)
784                 goto out;
785
786         del_gendisk(ubd_gendisk[n]);
787         put_disk(ubd_gendisk[n]);
788         ubd_gendisk[n] = NULL;
789
790         if(fake_gendisk[n] != NULL){
791                 del_gendisk(fake_gendisk[n]);
792                 put_disk(fake_gendisk[n]);
793                 fake_gendisk[n] = NULL;
794         }
795
796         platform_device_unregister(&dev->pdev);
797         *dev = ((struct ubd) DEFAULT_UBD);
798         err = 0;
799 out:
800         spin_unlock(&ubd_lock);
801         return err;
802 }
803
804 static struct mc_device ubd_mc = {
805         .name           = "ubd",
806         .config         = ubd_config,
807         .get_config     = ubd_get_config,
808         .id             = ubd_id,
809         .remove         = ubd_remove,
810 };
811
812 static int ubd_mc_init(void)
813 {
814         mconsole_register_dev(&ubd_mc);
815         return 0;
816 }
817
818 __initcall(ubd_mc_init);
819
820 static struct platform_driver ubd_driver = {
821         .driver = {
822                 .name  = DRIVER_NAME,
823         },
824 };
825
826 int ubd_init(void)
827 {
828         int i;
829
830         if (register_blkdev(MAJOR_NR, "ubd"))
831                 return -1;
832
833         ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
834         if (!ubd_queue) {
835                 unregister_blkdev(MAJOR_NR, "ubd");
836                 return -1;
837         }
838                 
839         if (fake_major != MAJOR_NR) {
840                 char name[sizeof("ubd_nnn\0")];
841
842                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
843                 if (register_blkdev(fake_major, "ubd"))
844                         return -1;
845         }
846         platform_driver_register(&ubd_driver);
847         for (i = 0; i < MAX_DEV; i++)
848                 ubd_add(i);
849         return 0;
850 }
851
852 late_initcall(ubd_init);
853
854 int ubd_driver_init(void){
855         unsigned long stack;
856         int err;
857
858         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
859         if(global_openflags.s){
860                 printk(KERN_INFO "ubd: Synchronous mode\n");
861                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
862                  * enough. So use anyway the io thread. */
863         }
864         stack = alloc_stack(0, 0);
865         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
866                                  &thread_fd);
867         if(io_pid < 0){
868                 printk(KERN_ERR
869                        "ubd : Failed to start I/O thread (errno = %d) - "
870                        "falling back to synchronous I/O\n", -io_pid);
871                 io_pid = -1;
872                 return(0);
873         }
874         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
875                              IRQF_DISABLED, "ubd", ubd_dev);
876         if(err != 0)
877                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
878         return 0;
879 }
880
881 device_initcall(ubd_driver_init);
882
883 static int ubd_open(struct inode *inode, struct file *filp)
884 {
885         struct gendisk *disk = inode->i_bdev->bd_disk;
886         struct ubd *dev = disk->private_data;
887         int err = 0;
888
889         if(dev->count == 0){
890                 err = ubd_open_dev(dev);
891                 if(err){
892                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
893                                disk->disk_name, dev->file, -err);
894                         goto out;
895                 }
896         }
897         dev->count++;
898         set_disk_ro(disk, !dev->openflags.w);
899
900         /* This should no more be needed. And it didn't work anyway to exclude
901          * read-write remounting of filesystems.*/
902         /*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
903                 if(--dev->count == 0) ubd_close(dev);
904                 err = -EROFS;
905         }*/
906  out:
907         return(err);
908 }
909
910 static int ubd_release(struct inode * inode, struct file * file)
911 {
912         struct gendisk *disk = inode->i_bdev->bd_disk;
913         struct ubd *dev = disk->private_data;
914
915         if(--dev->count == 0)
916                 ubd_close(dev);
917         return(0);
918 }
919
920 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
921                           __u64 *cow_offset, unsigned long *bitmap,
922                           __u64 bitmap_offset, unsigned long *bitmap_words,
923                           __u64 bitmap_len)
924 {
925         __u64 sector = io_offset >> 9;
926         int i, update_bitmap = 0;
927
928         for(i = 0; i < length >> 9; i++){
929                 if(cow_mask != NULL)
930                         ubd_set_bit(i, (unsigned char *) cow_mask);
931                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
932                         continue;
933
934                 update_bitmap = 1;
935                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
936         }
937
938         if(!update_bitmap)
939                 return;
940
941         *cow_offset = sector / (sizeof(unsigned long) * 8);
942
943         /* This takes care of the case where we're exactly at the end of the
944          * device, and *cow_offset + 1 is off the end.  So, just back it up
945          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
946          * for the original diagnosis.
947          */
948         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
949                            sizeof(unsigned long) - 1))
950                 (*cow_offset)--;
951
952         bitmap_words[0] = bitmap[*cow_offset];
953         bitmap_words[1] = bitmap[*cow_offset + 1];
954
955         *cow_offset *= sizeof(unsigned long);
956         *cow_offset += bitmap_offset;
957 }
958
959 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
960                        __u64 bitmap_offset, __u64 bitmap_len)
961 {
962         __u64 sector = req->offset >> 9;
963         int i;
964
965         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
966                 panic("Operation too long");
967
968         if(req->op == UBD_READ) {
969                 for(i = 0; i < req->length >> 9; i++){
970                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
971                                 ubd_set_bit(i, (unsigned char *)
972                                             &req->sector_mask);
973                 }
974         }
975         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
976                            &req->cow_offset, bitmap, bitmap_offset,
977                            req->bitmap_words, bitmap_len);
978 }
979
980 /* Called with ubd_io_lock held */
981 static int prepare_request(struct request *req, struct io_thread_req *io_req)
982 {
983         struct gendisk *disk = req->rq_disk;
984         struct ubd *dev = disk->private_data;
985         __u64 offset;
986         int len;
987
988         if(req->rq_status == RQ_INACTIVE) return(1);
989
990         /* This should be impossible now */
991         if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
992                 printk("Write attempted on readonly ubd device %s\n",
993                        disk->disk_name);
994                 end_request(req, 0);
995                 return(1);
996         }
997
998         offset = ((__u64) req->sector) << 9;
999         len = req->current_nr_sectors << 9;
1000
1001         io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
1002         io_req->fds[1] = dev->fd;
1003         io_req->cow_offset = -1;
1004         io_req->offset = offset;
1005         io_req->length = len;
1006         io_req->error = 0;
1007         io_req->sector_mask = 0;
1008
1009         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1010         io_req->offsets[0] = 0;
1011         io_req->offsets[1] = dev->cow.data_offset;
1012         io_req->buffer = req->buffer;
1013         io_req->sectorsize = 1 << 9;
1014
1015         if(dev->cow.file != NULL)
1016                 cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset,
1017                            dev->cow.bitmap_len);
1018
1019         return(0);
1020 }
1021
1022 /* Called with ubd_io_lock held */
1023 static void do_ubd_request(request_queue_t *q)
1024 {
1025         struct io_thread_req io_req;
1026         struct request *req;
1027         int err, n;
1028
1029         if(thread_fd == -1){
1030                 while((req = elv_next_request(q)) != NULL){
1031                         err = prepare_request(req, &io_req);
1032                         if(!err){
1033                                 do_io(&io_req);
1034                                 __ubd_finish(req, io_req.error);
1035                         }
1036                 }
1037         }
1038         else {
1039                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1040                         return;
1041                 err = prepare_request(req, &io_req);
1042                 if(!err){
1043                         do_ubd = ubd_handler;
1044                         n = os_write_file(thread_fd, (char *) &io_req,
1045                                          sizeof(io_req));
1046                         if(n != sizeof(io_req))
1047                                 printk("write to io thread failed, "
1048                                        "errno = %d\n", -n);
1049                 }
1050         }
1051 }
1052
1053 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1054 {
1055         struct ubd *dev = bdev->bd_disk->private_data;
1056
1057         geo->heads = 128;
1058         geo->sectors = 32;
1059         geo->cylinders = dev->size / (128 * 32 * 512);
1060         return 0;
1061 }
1062
1063 static int ubd_ioctl(struct inode * inode, struct file * file,
1064                      unsigned int cmd, unsigned long arg)
1065 {
1066         struct ubd *dev = inode->i_bdev->bd_disk->private_data;
1067         struct hd_driveid ubd_id = {
1068                 .cyls           = 0,
1069                 .heads          = 128,
1070                 .sectors        = 32,
1071         };
1072
1073         switch (cmd) {
1074                 struct cdrom_volctrl volume;
1075         case HDIO_GET_IDENTITY:
1076                 ubd_id.cyls = dev->size / (128 * 32 * 512);
1077                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1078                                  sizeof(ubd_id)))
1079                         return(-EFAULT);
1080                 return(0);
1081                 
1082         case CDROMVOLREAD:
1083                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1084                         return(-EFAULT);
1085                 volume.channel0 = 255;
1086                 volume.channel1 = 255;
1087                 volume.channel2 = 255;
1088                 volume.channel3 = 255;
1089                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1090                         return(-EFAULT);
1091                 return(0);
1092         }
1093         return(-EINVAL);
1094 }
1095
1096 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1097 {
1098         struct uml_stat buf1, buf2;
1099         int err;
1100
1101         if(from_cmdline == NULL)
1102                 return 0;
1103         if(!strcmp(from_cmdline, from_cow))
1104                 return 0;
1105
1106         err = os_stat_file(from_cmdline, &buf1);
1107         if(err < 0){
1108                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1109                 return 0;
1110         }
1111         err = os_stat_file(from_cow, &buf2);
1112         if(err < 0){
1113                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1114                 return 1;
1115         }
1116         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1117                 return 0;
1118
1119         printk("Backing file mismatch - \"%s\" requested,\n"
1120                "\"%s\" specified in COW header of \"%s\"\n",
1121                from_cmdline, from_cow, cow);
1122         return 1;
1123 }
1124
1125 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1126 {
1127         unsigned long modtime;
1128         unsigned long long actual;
1129         int err;
1130
1131         err = os_file_modtime(file, &modtime);
1132         if(err < 0){
1133                 printk("Failed to get modification time of backing file "
1134                        "\"%s\", err = %d\n", file, -err);
1135                 return(err);
1136         }
1137
1138         err = os_file_size(file, &actual);
1139         if(err < 0){
1140                 printk("Failed to get size of backing file \"%s\", "
1141                        "err = %d\n", file, -err);
1142                 return(err);
1143         }
1144
1145         if(actual != size){
1146                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1147                  * the typecast.*/
1148                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1149                        "file\n", (unsigned long long) size, actual);
1150                 return(-EINVAL);
1151         }
1152         if(modtime != mtime){
1153                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1154                        "file\n", mtime, modtime);
1155                 return(-EINVAL);
1156         }
1157         return(0);
1158 }
1159
1160 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1161 {
1162         int err;
1163
1164         err = os_seek_file(fd, offset);
1165         if(err < 0)
1166                 return(err);
1167
1168         err = os_read_file(fd, buf, len);
1169         if(err < 0)
1170                 return(err);
1171
1172         return(0);
1173 }
1174
1175 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1176                   char **backing_file_out, int *bitmap_offset_out,
1177                   unsigned long *bitmap_len_out, int *data_offset_out,
1178                   int *create_cow_out)
1179 {
1180         time_t mtime;
1181         unsigned long long size;
1182         __u32 version, align;
1183         char *backing_file;
1184         int fd, err, sectorsize, asked_switch, mode = 0644;
1185
1186         fd = os_open_file(file, *openflags, mode);
1187         if (fd < 0) {
1188                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1189                         *create_cow_out = 1;
1190                 if (!openflags->w ||
1191                    ((fd != -EROFS) && (fd != -EACCES)))
1192                         return fd;
1193                 openflags->w = 0;
1194                 fd = os_open_file(file, *openflags, mode);
1195                 if (fd < 0)
1196                         return fd;
1197         }
1198
1199         if(shared)
1200                 printk("Not locking \"%s\" on the host\n", file);
1201         else {
1202                 err = os_lock_file(fd, openflags->w);
1203                 if(err < 0){
1204                         printk("Failed to lock '%s', err = %d\n", file, -err);
1205                         goto out_close;
1206                 }
1207         }
1208
1209         /* Successful return case! */
1210         if(backing_file_out == NULL)
1211                 return(fd);
1212
1213         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1214                               &size, &sectorsize, &align, bitmap_offset_out);
1215         if(err && (*backing_file_out != NULL)){
1216                 printk("Failed to read COW header from COW file \"%s\", "
1217                        "errno = %d\n", file, -err);
1218                 goto out_close;
1219         }
1220         if(err)
1221                 return(fd);
1222
1223         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1224
1225         /* Allow switching only if no mismatch. */
1226         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1227                 printk("Switching backing file to '%s'\n", *backing_file_out);
1228                 err = write_cow_header(file, fd, *backing_file_out,
1229                                        sectorsize, align, &size);
1230                 if (err) {
1231                         printk("Switch failed, errno = %d\n", -err);
1232                         goto out_close;
1233                 }
1234         } else {
1235                 *backing_file_out = backing_file;
1236                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1237                 if (err)
1238                         goto out_close;
1239         }
1240
1241         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1242                   bitmap_len_out, data_offset_out);
1243
1244         return fd;
1245  out_close:
1246         os_close_file(fd);
1247         return err;
1248 }
1249
1250 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1251                     int sectorsize, int alignment, int *bitmap_offset_out,
1252                     unsigned long *bitmap_len_out, int *data_offset_out)
1253 {
1254         int err, fd;
1255
1256         flags.c = 1;
1257         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1258         if(fd < 0){
1259                 err = fd;
1260                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1261                        -err);
1262                 goto out;
1263         }
1264
1265         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1266                             bitmap_offset_out, bitmap_len_out,
1267                             data_offset_out);
1268         if(!err)
1269                 return(fd);
1270         os_close_file(fd);
1271  out:
1272         return(err);
1273 }
1274
1275 static int update_bitmap(struct io_thread_req *req)
1276 {
1277         int n;
1278
1279         if(req->cow_offset == -1)
1280                 return(0);
1281
1282         n = os_seek_file(req->fds[1], req->cow_offset);
1283         if(n < 0){
1284                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1285                 return(1);
1286         }
1287
1288         n = os_write_file(req->fds[1], &req->bitmap_words,
1289                           sizeof(req->bitmap_words));
1290         if(n != sizeof(req->bitmap_words)){
1291                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1292                        req->fds[1]);
1293                 return(1);
1294         }
1295
1296         return(0);
1297 }
1298
1299 void do_io(struct io_thread_req *req)
1300 {
1301         char *buf;
1302         unsigned long len;
1303         int n, nsectors, start, end, bit;
1304         int err;
1305         __u64 off;
1306
1307         nsectors = req->length / req->sectorsize;
1308         start = 0;
1309         do {
1310                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1311                 end = start;
1312                 while((end < nsectors) &&
1313                       (ubd_test_bit(end, (unsigned char *)
1314                                     &req->sector_mask) == bit))
1315                         end++;
1316
1317                 off = req->offset + req->offsets[bit] +
1318                         start * req->sectorsize;
1319                 len = (end - start) * req->sectorsize;
1320                 buf = &req->buffer[start * req->sectorsize];
1321
1322                 err = os_seek_file(req->fds[bit], off);
1323                 if(err < 0){
1324                         printk("do_io - lseek failed : err = %d\n", -err);
1325                         req->error = 1;
1326                         return;
1327                 }
1328                 if(req->op == UBD_READ){
1329                         n = 0;
1330                         do {
1331                                 buf = &buf[n];
1332                                 len -= n;
1333                                 n = os_read_file(req->fds[bit], buf, len);
1334                                 if (n < 0) {
1335                                         printk("do_io - read failed, err = %d "
1336                                                "fd = %d\n", -n, req->fds[bit]);
1337                                         req->error = 1;
1338                                         return;
1339                                 }
1340                         } while((n < len) && (n != 0));
1341                         if (n < len) memset(&buf[n], 0, len - n);
1342                 } else {
1343                         n = os_write_file(req->fds[bit], buf, len);
1344                         if(n != len){
1345                                 printk("do_io - write failed err = %d "
1346                                        "fd = %d\n", -n, req->fds[bit]);
1347                                 req->error = 1;
1348                                 return;
1349                         }
1350                 }
1351
1352                 start = end;
1353         } while(start < nsectors);
1354
1355         req->error = update_bitmap(req);
1356 }
1357
1358 /* Changed in start_io_thread, which is serialized by being called only
1359  * from ubd_init, which is an initcall.
1360  */
1361 int kernel_fd = -1;
1362
1363 /* Only changed by the io thread */
1364 int io_count = 0;
1365
1366 int io_thread(void *arg)
1367 {
1368         struct io_thread_req req;
1369         int n;
1370
1371         ignore_sigwinch_sig();
1372         while(1){
1373                 n = os_read_file(kernel_fd, &req, sizeof(req));
1374                 if(n != sizeof(req)){
1375                         if(n < 0)
1376                                 printk("io_thread - read failed, fd = %d, "
1377                                        "err = %d\n", kernel_fd, -n);
1378                         else {
1379                                 printk("io_thread - short read, fd = %d, "
1380                                        "length = %d\n", kernel_fd, n);
1381                         }
1382                         continue;
1383                 }
1384                 io_count++;
1385                 do_io(&req);
1386                 n = os_write_file(kernel_fd, &req, sizeof(req));
1387                 if(n != sizeof(req))
1388                         printk("io_thread - write failed, fd = %d, err = %d\n",
1389                                kernel_fd, -n);
1390         }
1391
1392         return 0;
1393 }