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