uml: remove user_util.h
[safe/jmp/linux-2.6] / arch / um / drivers / ubd_kern.c
1 /*
2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 /* 2001-09-28...2002-04-17
7  * Partition stuff by James_McMechan@hotmail.com
8  * old style ubd by setting UBD_SHIFT to 0
9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
10  * partitions have changed in 2.5
11  * 2003-01-29 more tinkering for 2.5.59-1
12  * This should now address the sysfs problems and has
13  * the symlink for devfs to allow for booting with
14  * the common /dev/ubd/discX/... names rather than
15  * only /dev/ubdN/discN this version also has lots of
16  * clean ups preparing for ubd-many.
17  * James McMechan
18  */
19
20 #define MAJOR_NR UBD_MAJOR
21 #define UBD_SHIFT 4
22
23 #include "linux/module.h"
24 #include "linux/blkdev.h"
25 #include "linux/hdreg.h"
26 #include "linux/init.h"
27 #include "linux/cdrom.h"
28 #include "linux/proc_fs.h"
29 #include "linux/ctype.h"
30 #include "linux/capability.h"
31 #include "linux/mm.h"
32 #include "linux/vmalloc.h"
33 #include "linux/blkpg.h"
34 #include "linux/genhd.h"
35 #include "linux/spinlock.h"
36 #include "linux/platform_device.h"
37 #include "asm/segment.h"
38 #include "asm/uaccess.h"
39 #include "asm/irq.h"
40 #include "asm/types.h"
41 #include "asm/tlbflush.h"
42 #include "mem_user.h"
43 #include "kern_util.h"
44 #include "kern.h"
45 #include "mconsole_kern.h"
46 #include "init.h"
47 #include "irq_user.h"
48 #include "irq_kern.h"
49 #include "ubd_user.h"
50 #include "os.h"
51 #include "mem.h"
52 #include "mem_kern.h"
53 #include "cow.h"
54
55 enum ubd_req { UBD_READ, UBD_WRITE };
56
57 struct io_thread_req {
58         struct request *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_MUTEX(ubd_lock);
110
111 static int ubd_open(struct inode * inode, struct file * filp);
112 static int ubd_release(struct inode * inode, struct file * file);
113 static int ubd_ioctl(struct inode * inode, struct file * file,
114                      unsigned int cmd, unsigned long arg);
115 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
116
117 #define MAX_DEV (16)
118
119 static struct block_device_operations ubd_blops = {
120         .owner          = THIS_MODULE,
121         .open           = ubd_open,
122         .release        = ubd_release,
123         .ioctl          = ubd_ioctl,
124         .getgeo         = ubd_getgeo,
125 };
126
127 /* Protected by ubd_lock */
128 static int fake_major = MAJOR_NR;
129 static struct gendisk *ubd_gendisk[MAX_DEV];
130 static struct gendisk *fake_gendisk[MAX_DEV];
131
132 #ifdef CONFIG_BLK_DEV_UBD_SYNC
133 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
134                                          .cl = 1 })
135 #else
136 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
137                                          .cl = 1 })
138 #endif
139 static struct openflags global_openflags = OPEN_FLAGS;
140
141 struct cow {
142         /* backing file name */
143         char *file;
144         /* backing file fd */
145         int fd;
146         unsigned long *bitmap;
147         unsigned long bitmap_len;
148         int bitmap_offset;
149         int data_offset;
150 };
151
152 struct ubd {
153         /* name (and fd, below) of the file opened for writing, either the
154          * backing or the cow file. */
155         char *file;
156         int count;
157         int fd;
158         __u64 size;
159         struct openflags boot_openflags;
160         struct openflags openflags;
161         unsigned shared:1;
162         unsigned no_cow:1;
163         struct cow cow;
164         struct platform_device pdev;
165         struct request_queue *queue;
166         spinlock_t lock;
167         int active;
168 };
169
170 #define DEFAULT_COW { \
171         .file =                 NULL, \
172         .fd =                   -1, \
173         .bitmap =               NULL, \
174         .bitmap_offset =        0, \
175         .data_offset =          0, \
176 }
177
178 #define DEFAULT_UBD { \
179         .file =                 NULL, \
180         .count =                0, \
181         .fd =                   -1, \
182         .size =                 -1, \
183         .boot_openflags =       OPEN_FLAGS, \
184         .openflags =            OPEN_FLAGS, \
185         .no_cow =               0, \
186         .shared =               0, \
187         .cow =                  DEFAULT_COW, \
188         .lock =                 SPIN_LOCK_UNLOCKED,     \
189         .active =               0, \
190 }
191
192 /* Protected by ubd_lock */
193 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
194
195 /* Only changed by fake_ide_setup which is a setup */
196 static int fake_ide = 0;
197 static struct proc_dir_entry *proc_ide_root = NULL;
198 static struct proc_dir_entry *proc_ide = NULL;
199
200 static void make_proc_ide(void)
201 {
202         proc_ide_root = proc_mkdir("ide", NULL);
203         proc_ide = proc_mkdir("ide0", proc_ide_root);
204 }
205
206 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
207                                int *eof, void *data)
208 {
209         int len;
210
211         strcpy(page, "disk\n");
212         len = strlen("disk\n");
213         len -= off;
214         if (len < count){
215                 *eof = 1;
216                 if (len <= 0) return 0;
217         }
218         else len = count;
219         *start = page + off;
220         return len;
221 }
222
223 static void make_ide_entries(char *dev_name)
224 {
225         struct proc_dir_entry *dir, *ent;
226         char name[64];
227
228         if(proc_ide_root == NULL) make_proc_ide();
229
230         dir = proc_mkdir(dev_name, proc_ide);
231         if(!dir) return;
232
233         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
234         if(!ent) return;
235         ent->data = NULL;
236         ent->read_proc = proc_ide_read_media;
237         ent->write_proc = NULL;
238         sprintf(name,"ide0/%s", dev_name);
239         proc_symlink(dev_name, proc_ide_root, name);
240 }
241
242 static int fake_ide_setup(char *str)
243 {
244         fake_ide = 1;
245         return(1);
246 }
247
248 __setup("fake_ide", fake_ide_setup);
249
250 __uml_help(fake_ide_setup,
251 "fake_ide\n"
252 "    Create ide0 entries that map onto ubd devices.\n\n"
253 );
254
255 static int parse_unit(char **ptr)
256 {
257         char *str = *ptr, *end;
258         int n = -1;
259
260         if(isdigit(*str)) {
261                 n = simple_strtoul(str, &end, 0);
262                 if(end == str)
263                         return(-1);
264                 *ptr = end;
265         }
266         else if (('a' <= *str) && (*str <= 'z')) {
267                 n = *str - 'a';
268                 str++;
269                 *ptr = str;
270         }
271         return(n);
272 }
273
274 /* If *index_out == -1 at exit, the passed option was a general one;
275  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
276  * should not be freed on exit.
277  */
278 static int ubd_setup_common(char *str, int *index_out, char **error_out)
279 {
280         struct ubd *ubd_dev;
281         struct openflags flags = global_openflags;
282         char *backing_file;
283         int n, err = 0, i;
284
285         if(index_out) *index_out = -1;
286         n = *str;
287         if(n == '='){
288                 char *end;
289                 int major;
290
291                 str++;
292                 if(!strcmp(str, "sync")){
293                         global_openflags = of_sync(global_openflags);
294                         goto out1;
295                 }
296
297                 err = -EINVAL;
298                 major = simple_strtoul(str, &end, 0);
299                 if((*end != '\0') || (end == str)){
300                         *error_out = "Didn't parse major number";
301                         goto out1;
302                 }
303
304                 mutex_lock(&ubd_lock);
305                 if(fake_major != MAJOR_NR){
306                         *error_out = "Can't assign a fake major twice";
307                         goto out1;
308                 }
309
310                 fake_major = major;
311
312                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
313                        major);
314                 err = 0;
315         out1:
316                 mutex_unlock(&ubd_lock);
317                 return err;
318         }
319
320         n = parse_unit(&str);
321         if(n < 0){
322                 *error_out = "Couldn't parse device number";
323                 return -EINVAL;
324         }
325         if(n >= MAX_DEV){
326                 *error_out = "Device number out of range";
327                 return 1;
328         }
329
330         err = -EBUSY;
331         mutex_lock(&ubd_lock);
332
333         ubd_dev = &ubd_devs[n];
334         if(ubd_dev->file != NULL){
335                 *error_out = "Device is already configured";
336                 goto out;
337         }
338
339         if (index_out)
340                 *index_out = n;
341
342         err = -EINVAL;
343         for (i = 0; i < sizeof("rscd="); i++) {
344                 switch (*str) {
345                 case 'r':
346                         flags.w = 0;
347                         break;
348                 case 's':
349                         flags.s = 1;
350                         break;
351                 case 'd':
352                         ubd_dev->no_cow = 1;
353                         break;
354                 case 'c':
355                         ubd_dev->shared = 1;
356                         break;
357                 case '=':
358                         str++;
359                         goto break_loop;
360                 default:
361                         *error_out = "Expected '=' or flag letter "
362                                 "(r, s, c, or d)";
363                         goto out;
364                 }
365                 str++;
366         }
367
368         if (*str == '=')
369                 *error_out = "Too many flags specified";
370         else
371                 *error_out = "Missing '='";
372         goto out;
373
374 break_loop:
375         backing_file = strchr(str, ',');
376
377         if (backing_file == NULL)
378                 backing_file = strchr(str, ':');
379
380         if(backing_file != NULL){
381                 if(ubd_dev->no_cow){
382                         *error_out = "Can't specify both 'd' and a cow file";
383                         goto out;
384                 }
385                 else {
386                         *backing_file = '\0';
387                         backing_file++;
388                 }
389         }
390         err = 0;
391         ubd_dev->file = str;
392         ubd_dev->cow.file = backing_file;
393         ubd_dev->boot_openflags = flags;
394 out:
395         mutex_unlock(&ubd_lock);
396         return err;
397 }
398
399 static int ubd_setup(char *str)
400 {
401         char *error;
402         int err;
403
404         err = ubd_setup_common(str, NULL, &error);
405         if(err)
406                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
407                        "%s\n", str, error);
408         return 1;
409 }
410
411 __setup("ubd", ubd_setup);
412 __uml_help(ubd_setup,
413 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
414 "    This is used to associate a device with a file in the underlying\n"
415 "    filesystem. When specifying two filenames, the first one is the\n"
416 "    COW name and the second is the backing file name. As separator you can\n"
417 "    use either a ':' or a ',': the first one allows writing things like;\n"
418 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
419 "    while with a ',' the shell would not expand the 2nd '~'.\n"
420 "    When using only one filename, UML will detect whether to treat it like\n"
421 "    a COW file or a backing file. To override this detection, add the 'd'\n"
422 "    flag:\n"
423 "       ubd0d=BackingFile\n"
424 "    Usually, there is a filesystem in the file, but \n"
425 "    that's not required. Swap devices containing swap files can be\n"
426 "    specified like this. Also, a file which doesn't contain a\n"
427 "    filesystem can have its contents read in the virtual \n"
428 "    machine by running 'dd' on the device. <n> must be in the range\n"
429 "    0 to 7. Appending an 'r' to the number will cause that device\n"
430 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
431 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
432 );
433
434 static int udb_setup(char *str)
435 {
436         printk("udb%s specified on command line is almost certainly a ubd -> "
437                "udb TYPO\n", str);
438         return(1);
439 }
440
441 __setup("udb", udb_setup);
442 __uml_help(udb_setup,
443 "udb\n"
444 "    This option is here solely to catch ubd -> udb typos, which can be\n"
445 "    to impossible to catch visually unless you specifically look for\n"
446 "    them.  The only result of any option starting with 'udb' is an error\n"
447 "    in the boot output.\n\n"
448 );
449
450 static int fakehd_set = 0;
451 static int fakehd(char *str)
452 {
453         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
454         fakehd_set = 1;
455         return 1;
456 }
457
458 __setup("fakehd", fakehd);
459 __uml_help(fakehd,
460 "fakehd\n"
461 "    Change the ubd device name to \"hd\".\n\n"
462 );
463
464 static void do_ubd_request(request_queue_t * q);
465
466 /* Only changed by ubd_init, which is an initcall. */
467 int thread_fd = -1;
468
469 /* call ubd_finish if you need to serialize */
470 static void __ubd_finish(struct request *req, int error)
471 {
472         int nsect;
473
474         if(error){
475                 end_request(req, 0);
476                 return;
477         }
478         nsect = req->current_nr_sectors;
479         req->sector += nsect;
480         req->buffer += nsect << 9;
481         req->errors = 0;
482         req->nr_sectors -= nsect;
483         req->current_nr_sectors = 0;
484         end_request(req, 1);
485 }
486
487 /* Callable only from interrupt context - otherwise you need to do
488  * spin_lock_irq()/spin_lock_irqsave() */
489 static inline void ubd_finish(struct request *req, int error)
490 {
491         struct ubd *dev = req->rq_disk->private_data;
492
493         spin_lock(&dev->lock);
494         __ubd_finish(req, error);
495         spin_unlock(&dev->lock);
496 }
497
498 /* XXX - move this inside ubd_intr. */
499 /* Called without dev->lock held, and only in interrupt context. */
500 static void ubd_handler(void)
501 {
502         struct io_thread_req req;
503         struct request *rq;
504         struct ubd *dev;
505         int n;
506
507         n = os_read_file(thread_fd, &req, sizeof(req));
508         if(n != sizeof(req)){
509                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
510                        "err = %d\n", os_getpid(), -n);
511                 return;
512         }
513
514         rq = req.req;
515         dev = rq->rq_disk->private_data;
516         dev->active = 0;
517
518         ubd_finish(rq, req.error);
519         reactivate_fd(thread_fd, UBD_IRQ);
520         spin_lock(&dev->lock);
521         do_ubd_request(dev->queue);
522         spin_unlock(&dev->lock);
523 }
524
525 static irqreturn_t ubd_intr(int irq, void *dev)
526 {
527         ubd_handler();
528         return(IRQ_HANDLED);
529 }
530
531 /* Only changed by ubd_init, which is an initcall. */
532 static int io_pid = -1;
533
534 void kill_io_thread(void)
535 {
536         if(io_pid != -1)
537                 os_kill_process(io_pid, 1);
538 }
539
540 __uml_exitcall(kill_io_thread);
541
542 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
543 {
544         char *file;
545
546         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
547         return(os_file_size(file, size_out));
548 }
549
550 static void ubd_close_dev(struct ubd *ubd_dev)
551 {
552         os_close_file(ubd_dev->fd);
553         if(ubd_dev->cow.file == NULL)
554                 return;
555
556         os_close_file(ubd_dev->cow.fd);
557         vfree(ubd_dev->cow.bitmap);
558         ubd_dev->cow.bitmap = NULL;
559 }
560
561 static int ubd_open_dev(struct ubd *ubd_dev)
562 {
563         struct openflags flags;
564         char **back_ptr;
565         int err, create_cow, *create_ptr;
566         int fd;
567
568         ubd_dev->openflags = ubd_dev->boot_openflags;
569         create_cow = 0;
570         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
571         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
572
573         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
574                                 back_ptr, &ubd_dev->cow.bitmap_offset,
575                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
576                                 create_ptr);
577
578         if((fd == -ENOENT) && create_cow){
579                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
580                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
581                                           &ubd_dev->cow.bitmap_offset,
582                                           &ubd_dev->cow.bitmap_len,
583                                           &ubd_dev->cow.data_offset);
584                 if(fd >= 0){
585                         printk(KERN_INFO "Creating \"%s\" as COW file for "
586                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
587                 }
588         }
589
590         if(fd < 0){
591                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
592                        -fd);
593                 return fd;
594         }
595         ubd_dev->fd = fd;
596
597         if(ubd_dev->cow.file != NULL){
598                 err = -ENOMEM;
599                 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
600                 if(ubd_dev->cow.bitmap == NULL){
601                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
602                         goto error;
603                 }
604                 flush_tlb_kernel_vm();
605
606                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
607                                       ubd_dev->cow.bitmap_offset,
608                                       ubd_dev->cow.bitmap_len);
609                 if(err < 0)
610                         goto error;
611
612                 flags = ubd_dev->openflags;
613                 flags.w = 0;
614                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
615                                     NULL, NULL, NULL, NULL);
616                 if(err < 0) goto error;
617                 ubd_dev->cow.fd = err;
618         }
619         return(0);
620  error:
621         os_close_file(ubd_dev->fd);
622         return(err);
623 }
624
625 static int ubd_disk_register(int major, u64 size, int unit,
626                              struct gendisk **disk_out)
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_devs[unit].pdev.id   = unit;
646                 ubd_devs[unit].pdev.name = DRIVER_NAME;
647                 platform_device_register(&ubd_devs[unit].pdev);
648                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
649         }
650
651         disk->private_data = &ubd_devs[unit];
652         disk->queue = ubd_devs[unit].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, char **error_out)
662 {
663         struct ubd *ubd_dev = &ubd_devs[n];
664         int err = 0;
665
666         if(ubd_dev->file == NULL)
667                 goto out;
668
669         err = ubd_file_size(ubd_dev, &ubd_dev->size);
670         if(err < 0){
671                 *error_out = "Couldn't determine size of device's file";
672                 goto out;
673         }
674
675         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
676
677         err = -ENOMEM;
678         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
679         if (ubd_dev->queue == NULL) {
680                 *error_out = "Failed to initialize device queue";
681                 goto out;
682         }
683         ubd_dev->queue->queuedata = ubd_dev;
684
685         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
686         if(err){
687                 *error_out = "Failed to register device";
688                 goto out_cleanup;
689         }
690
691         if(fake_major != MAJOR_NR)
692                 ubd_disk_register(fake_major, ubd_dev->size, n,
693                                   &fake_gendisk[n]);
694
695         /* perhaps this should also be under the "if (fake_major)" above */
696         /* using the fake_disk->disk_name and also the fakehd_set name */
697         if (fake_ide)
698                 make_ide_entries(ubd_gendisk[n]->disk_name);
699
700         err = 0;
701 out:
702         return err;
703
704 out_cleanup:
705         blk_cleanup_queue(ubd_dev->queue);
706         goto out;
707 }
708
709 static int ubd_config(char *str, char **error_out)
710 {
711         int n, ret;
712
713         /* This string is possibly broken up and stored, so it's only
714          * freed if ubd_setup_common fails, or if only general options
715          * were set.
716          */
717         str = kstrdup(str, GFP_KERNEL);
718         if (str == NULL) {
719                 *error_out = "Failed to allocate memory";
720                 return -ENOMEM;
721         }
722
723         ret = ubd_setup_common(str, &n, error_out);
724         if (ret)
725                 goto err_free;
726
727         if (n == -1) {
728                 ret = 0;
729                 goto err_free;
730         }
731
732         mutex_lock(&ubd_lock);
733         ret = ubd_add(n, error_out);
734         if (ret)
735                 ubd_devs[n].file = NULL;
736         mutex_unlock(&ubd_lock);
737
738 out:
739         return ret;
740
741 err_free:
742         kfree(str);
743         goto out;
744 }
745
746 static int ubd_get_config(char *name, char *str, int size, char **error_out)
747 {
748         struct ubd *ubd_dev;
749         int n, len = 0;
750
751         n = parse_unit(&name);
752         if((n >= MAX_DEV) || (n < 0)){
753                 *error_out = "ubd_get_config : device number out of range";
754                 return(-1);
755         }
756
757         ubd_dev = &ubd_devs[n];
758         mutex_lock(&ubd_lock);
759
760         if(ubd_dev->file == NULL){
761                 CONFIG_CHUNK(str, size, len, "", 1);
762                 goto out;
763         }
764
765         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
766
767         if(ubd_dev->cow.file != NULL){
768                 CONFIG_CHUNK(str, size, len, ",", 0);
769                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
770         }
771         else CONFIG_CHUNK(str, size, len, "", 1);
772
773  out:
774         mutex_unlock(&ubd_lock);
775         return(len);
776 }
777
778 static int ubd_id(char **str, int *start_out, int *end_out)
779 {
780         int n;
781
782         n = parse_unit(str);
783         *start_out = 0;
784         *end_out = MAX_DEV - 1;
785         return n;
786 }
787
788 static int ubd_remove(int n, char **error_out)
789 {
790         struct gendisk *disk;
791         struct ubd *ubd_dev;
792         int err = -ENODEV;
793
794         mutex_lock(&ubd_lock);
795
796         ubd_dev = &ubd_devs[n];
797
798         if(ubd_dev->file == NULL)
799                 goto out;
800
801         /* you cannot remove a open disk */
802         err = -EBUSY;
803         if(ubd_dev->count > 0)
804                 goto out;
805
806         disk = ubd_gendisk[n];
807         ubd_gendisk[n] = NULL;
808         if(disk != NULL){
809                 del_gendisk(disk);
810                 put_disk(disk);
811         }
812
813         if(fake_gendisk[n] != NULL){
814                 del_gendisk(fake_gendisk[n]);
815                 put_disk(fake_gendisk[n]);
816                 fake_gendisk[n] = NULL;
817         }
818
819         blk_cleanup_queue(ubd_dev->queue);
820         platform_device_unregister(&ubd_dev->pdev);
821         *ubd_dev = ((struct ubd) DEFAULT_UBD);
822         err = 0;
823 out:
824         mutex_unlock(&ubd_lock);
825         return err;
826 }
827
828 /* All these are called by mconsole in process context and without
829  * ubd-specific locks.  The structure itself is const except for .list.
830  */
831 static struct mc_device ubd_mc = {
832         .list           = LIST_HEAD_INIT(ubd_mc.list),
833         .name           = "ubd",
834         .config         = ubd_config,
835         .get_config     = ubd_get_config,
836         .id             = ubd_id,
837         .remove         = ubd_remove,
838 };
839
840 static int __init ubd_mc_init(void)
841 {
842         mconsole_register_dev(&ubd_mc);
843         return 0;
844 }
845
846 __initcall(ubd_mc_init);
847
848 static int __init ubd0_init(void)
849 {
850         struct ubd *ubd_dev = &ubd_devs[0];
851
852         mutex_lock(&ubd_lock);
853         if(ubd_dev->file == NULL)
854                 ubd_dev->file = "root_fs";
855         mutex_unlock(&ubd_lock);
856
857         return(0);
858 }
859
860 __initcall(ubd0_init);
861
862 /* Used in ubd_init, which is an initcall */
863 static struct platform_driver ubd_driver = {
864         .driver = {
865                 .name  = DRIVER_NAME,
866         },
867 };
868
869 static int __init ubd_init(void)
870 {
871         char *error;
872         int i, err;
873
874         if (register_blkdev(MAJOR_NR, "ubd"))
875                 return -1;
876
877         if (fake_major != MAJOR_NR) {
878                 char name[sizeof("ubd_nnn\0")];
879
880                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
881                 if (register_blkdev(fake_major, "ubd"))
882                         return -1;
883         }
884         platform_driver_register(&ubd_driver);
885         mutex_lock(&ubd_lock);
886         for (i = 0; i < MAX_DEV; i++){
887                 err = ubd_add(i, &error);
888                 if(err)
889                         printk(KERN_ERR "Failed to initialize ubd device %d :"
890                                "%s\n", i, error);
891         }
892         mutex_unlock(&ubd_lock);
893         return 0;
894 }
895
896 late_initcall(ubd_init);
897
898 static int __init ubd_driver_init(void){
899         unsigned long stack;
900         int err;
901
902         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
903         if(global_openflags.s){
904                 printk(KERN_INFO "ubd: Synchronous mode\n");
905                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
906                  * enough. So use anyway the io thread. */
907         }
908         stack = alloc_stack(0, 0);
909         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
910                                  &thread_fd);
911         if(io_pid < 0){
912                 printk(KERN_ERR
913                        "ubd : Failed to start I/O thread (errno = %d) - "
914                        "falling back to synchronous I/O\n", -io_pid);
915                 io_pid = -1;
916                 return(0);
917         }
918         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
919                              IRQF_DISABLED, "ubd", ubd_devs);
920         if(err != 0)
921                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
922         return 0;
923 }
924
925 device_initcall(ubd_driver_init);
926
927 static int ubd_open(struct inode *inode, struct file *filp)
928 {
929         struct gendisk *disk = inode->i_bdev->bd_disk;
930         struct ubd *ubd_dev = disk->private_data;
931         int err = 0;
932
933         if(ubd_dev->count == 0){
934                 err = ubd_open_dev(ubd_dev);
935                 if(err){
936                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
937                                disk->disk_name, ubd_dev->file, -err);
938                         goto out;
939                 }
940         }
941         ubd_dev->count++;
942         set_disk_ro(disk, !ubd_dev->openflags.w);
943
944         /* This should no more be needed. And it didn't work anyway to exclude
945          * read-write remounting of filesystems.*/
946         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
947                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
948                 err = -EROFS;
949         }*/
950  out:
951         return(err);
952 }
953
954 static int ubd_release(struct inode * inode, struct file * file)
955 {
956         struct gendisk *disk = inode->i_bdev->bd_disk;
957         struct ubd *ubd_dev = disk->private_data;
958
959         if(--ubd_dev->count == 0)
960                 ubd_close_dev(ubd_dev);
961         return(0);
962 }
963
964 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
965                           __u64 *cow_offset, unsigned long *bitmap,
966                           __u64 bitmap_offset, unsigned long *bitmap_words,
967                           __u64 bitmap_len)
968 {
969         __u64 sector = io_offset >> 9;
970         int i, update_bitmap = 0;
971
972         for(i = 0; i < length >> 9; i++){
973                 if(cow_mask != NULL)
974                         ubd_set_bit(i, (unsigned char *) cow_mask);
975                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
976                         continue;
977
978                 update_bitmap = 1;
979                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
980         }
981
982         if(!update_bitmap)
983                 return;
984
985         *cow_offset = sector / (sizeof(unsigned long) * 8);
986
987         /* This takes care of the case where we're exactly at the end of the
988          * device, and *cow_offset + 1 is off the end.  So, just back it up
989          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
990          * for the original diagnosis.
991          */
992         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
993                            sizeof(unsigned long) - 1))
994                 (*cow_offset)--;
995
996         bitmap_words[0] = bitmap[*cow_offset];
997         bitmap_words[1] = bitmap[*cow_offset + 1];
998
999         *cow_offset *= sizeof(unsigned long);
1000         *cow_offset += bitmap_offset;
1001 }
1002
1003 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1004                        __u64 bitmap_offset, __u64 bitmap_len)
1005 {
1006         __u64 sector = req->offset >> 9;
1007         int i;
1008
1009         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1010                 panic("Operation too long");
1011
1012         if(req->op == UBD_READ) {
1013                 for(i = 0; i < req->length >> 9; i++){
1014                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1015                                 ubd_set_bit(i, (unsigned char *)
1016                                             &req->sector_mask);
1017                 }
1018         }
1019         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1020                            &req->cow_offset, bitmap, bitmap_offset,
1021                            req->bitmap_words, bitmap_len);
1022 }
1023
1024 /* Called with dev->lock held */
1025 static int prepare_request(struct request *req, struct io_thread_req *io_req)
1026 {
1027         struct gendisk *disk = req->rq_disk;
1028         struct ubd *ubd_dev = disk->private_data;
1029         __u64 offset;
1030         int len;
1031
1032         /* This should be impossible now */
1033         if((rq_data_dir(req) == WRITE) && !ubd_dev->openflags.w){
1034                 printk("Write attempted on readonly ubd device %s\n",
1035                        disk->disk_name);
1036                 end_request(req, 0);
1037                 return(1);
1038         }
1039
1040         offset = ((__u64) req->sector) << 9;
1041         len = req->current_nr_sectors << 9;
1042
1043         io_req->req = req;
1044         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd : ubd_dev->fd;
1045         io_req->fds[1] = ubd_dev->fd;
1046         io_req->cow_offset = -1;
1047         io_req->offset = offset;
1048         io_req->length = len;
1049         io_req->error = 0;
1050         io_req->sector_mask = 0;
1051
1052         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1053         io_req->offsets[0] = 0;
1054         io_req->offsets[1] = ubd_dev->cow.data_offset;
1055         io_req->buffer = req->buffer;
1056         io_req->sectorsize = 1 << 9;
1057
1058         if(ubd_dev->cow.file != NULL)
1059                 cowify_req(io_req, ubd_dev->cow.bitmap, ubd_dev->cow.bitmap_offset,
1060                            ubd_dev->cow.bitmap_len);
1061
1062         return(0);
1063 }
1064
1065 /* Called with dev->lock held */
1066 static void do_ubd_request(request_queue_t *q)
1067 {
1068         struct io_thread_req io_req;
1069         struct request *req;
1070         int err, n;
1071
1072         if(thread_fd == -1){
1073                 while((req = elv_next_request(q)) != NULL){
1074                         err = prepare_request(req, &io_req);
1075                         if(!err){
1076                                 do_io(&io_req);
1077                                 __ubd_finish(req, io_req.error);
1078                         }
1079                 }
1080         }
1081         else {
1082                 struct ubd *dev = q->queuedata;
1083                 if(dev->active || (req = elv_next_request(q)) == NULL)
1084                         return;
1085                 err = prepare_request(req, &io_req);
1086                 if(!err){
1087                         dev->active = 1;
1088                         n = os_write_file(thread_fd, (char *) &io_req,
1089                                          sizeof(io_req));
1090                         if(n != sizeof(io_req))
1091                                 printk("write to io thread failed, "
1092                                        "errno = %d\n", -n);
1093                 }
1094         }
1095 }
1096
1097 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1098 {
1099         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1100
1101         geo->heads = 128;
1102         geo->sectors = 32;
1103         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1104         return 0;
1105 }
1106
1107 static int ubd_ioctl(struct inode * inode, struct file * file,
1108                      unsigned int cmd, unsigned long arg)
1109 {
1110         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1111         struct hd_driveid ubd_id = {
1112                 .cyls           = 0,
1113                 .heads          = 128,
1114                 .sectors        = 32,
1115         };
1116
1117         switch (cmd) {
1118                 struct cdrom_volctrl volume;
1119         case HDIO_GET_IDENTITY:
1120                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1121                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1122                                  sizeof(ubd_id)))
1123                         return(-EFAULT);
1124                 return(0);
1125
1126         case CDROMVOLREAD:
1127                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1128                         return(-EFAULT);
1129                 volume.channel0 = 255;
1130                 volume.channel1 = 255;
1131                 volume.channel2 = 255;
1132                 volume.channel3 = 255;
1133                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1134                         return(-EFAULT);
1135                 return(0);
1136         }
1137         return(-EINVAL);
1138 }
1139
1140 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1141 {
1142         struct uml_stat buf1, buf2;
1143         int err;
1144
1145         if(from_cmdline == NULL)
1146                 return 0;
1147         if(!strcmp(from_cmdline, from_cow))
1148                 return 0;
1149
1150         err = os_stat_file(from_cmdline, &buf1);
1151         if(err < 0){
1152                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1153                 return 0;
1154         }
1155         err = os_stat_file(from_cow, &buf2);
1156         if(err < 0){
1157                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1158                 return 1;
1159         }
1160         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1161                 return 0;
1162
1163         printk("Backing file mismatch - \"%s\" requested,\n"
1164                "\"%s\" specified in COW header of \"%s\"\n",
1165                from_cmdline, from_cow, cow);
1166         return 1;
1167 }
1168
1169 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1170 {
1171         unsigned long modtime;
1172         unsigned long long actual;
1173         int err;
1174
1175         err = os_file_modtime(file, &modtime);
1176         if(err < 0){
1177                 printk("Failed to get modification time of backing file "
1178                        "\"%s\", err = %d\n", file, -err);
1179                 return(err);
1180         }
1181
1182         err = os_file_size(file, &actual);
1183         if(err < 0){
1184                 printk("Failed to get size of backing file \"%s\", "
1185                        "err = %d\n", file, -err);
1186                 return(err);
1187         }
1188
1189         if(actual != size){
1190                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1191                  * the typecast.*/
1192                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1193                        "file\n", (unsigned long long) size, actual);
1194                 return(-EINVAL);
1195         }
1196         if(modtime != mtime){
1197                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1198                        "file\n", mtime, modtime);
1199                 return(-EINVAL);
1200         }
1201         return(0);
1202 }
1203
1204 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1205 {
1206         int err;
1207
1208         err = os_seek_file(fd, offset);
1209         if(err < 0)
1210                 return(err);
1211
1212         err = os_read_file(fd, buf, len);
1213         if(err < 0)
1214                 return(err);
1215
1216         return(0);
1217 }
1218
1219 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1220                   char **backing_file_out, int *bitmap_offset_out,
1221                   unsigned long *bitmap_len_out, int *data_offset_out,
1222                   int *create_cow_out)
1223 {
1224         time_t mtime;
1225         unsigned long long size;
1226         __u32 version, align;
1227         char *backing_file;
1228         int fd, err, sectorsize, asked_switch, mode = 0644;
1229
1230         fd = os_open_file(file, *openflags, mode);
1231         if (fd < 0) {
1232                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1233                         *create_cow_out = 1;
1234                 if (!openflags->w ||
1235                    ((fd != -EROFS) && (fd != -EACCES)))
1236                         return fd;
1237                 openflags->w = 0;
1238                 fd = os_open_file(file, *openflags, mode);
1239                 if (fd < 0)
1240                         return fd;
1241         }
1242
1243         if(shared)
1244                 printk("Not locking \"%s\" on the host\n", file);
1245         else {
1246                 err = os_lock_file(fd, openflags->w);
1247                 if(err < 0){
1248                         printk("Failed to lock '%s', err = %d\n", file, -err);
1249                         goto out_close;
1250                 }
1251         }
1252
1253         /* Successful return case! */
1254         if(backing_file_out == NULL)
1255                 return(fd);
1256
1257         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1258                               &size, &sectorsize, &align, bitmap_offset_out);
1259         if(err && (*backing_file_out != NULL)){
1260                 printk("Failed to read COW header from COW file \"%s\", "
1261                        "errno = %d\n", file, -err);
1262                 goto out_close;
1263         }
1264         if(err)
1265                 return(fd);
1266
1267         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1268
1269         /* Allow switching only if no mismatch. */
1270         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1271                 printk("Switching backing file to '%s'\n", *backing_file_out);
1272                 err = write_cow_header(file, fd, *backing_file_out,
1273                                        sectorsize, align, &size);
1274                 if (err) {
1275                         printk("Switch failed, errno = %d\n", -err);
1276                         goto out_close;
1277                 }
1278         } else {
1279                 *backing_file_out = backing_file;
1280                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1281                 if (err)
1282                         goto out_close;
1283         }
1284
1285         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1286                   bitmap_len_out, data_offset_out);
1287
1288         return fd;
1289  out_close:
1290         os_close_file(fd);
1291         return err;
1292 }
1293
1294 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1295                     int sectorsize, int alignment, int *bitmap_offset_out,
1296                     unsigned long *bitmap_len_out, int *data_offset_out)
1297 {
1298         int err, fd;
1299
1300         flags.c = 1;
1301         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1302         if(fd < 0){
1303                 err = fd;
1304                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1305                        -err);
1306                 goto out;
1307         }
1308
1309         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1310                             bitmap_offset_out, bitmap_len_out,
1311                             data_offset_out);
1312         if(!err)
1313                 return(fd);
1314         os_close_file(fd);
1315  out:
1316         return(err);
1317 }
1318
1319 static int update_bitmap(struct io_thread_req *req)
1320 {
1321         int n;
1322
1323         if(req->cow_offset == -1)
1324                 return(0);
1325
1326         n = os_seek_file(req->fds[1], req->cow_offset);
1327         if(n < 0){
1328                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1329                 return(1);
1330         }
1331
1332         n = os_write_file(req->fds[1], &req->bitmap_words,
1333                           sizeof(req->bitmap_words));
1334         if(n != sizeof(req->bitmap_words)){
1335                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1336                        req->fds[1]);
1337                 return(1);
1338         }
1339
1340         return(0);
1341 }
1342
1343 void do_io(struct io_thread_req *req)
1344 {
1345         char *buf;
1346         unsigned long len;
1347         int n, nsectors, start, end, bit;
1348         int err;
1349         __u64 off;
1350
1351         nsectors = req->length / req->sectorsize;
1352         start = 0;
1353         do {
1354                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1355                 end = start;
1356                 while((end < nsectors) &&
1357                       (ubd_test_bit(end, (unsigned char *)
1358                                     &req->sector_mask) == bit))
1359                         end++;
1360
1361                 off = req->offset + req->offsets[bit] +
1362                         start * req->sectorsize;
1363                 len = (end - start) * req->sectorsize;
1364                 buf = &req->buffer[start * req->sectorsize];
1365
1366                 err = os_seek_file(req->fds[bit], off);
1367                 if(err < 0){
1368                         printk("do_io - lseek failed : err = %d\n", -err);
1369                         req->error = 1;
1370                         return;
1371                 }
1372                 if(req->op == UBD_READ){
1373                         n = 0;
1374                         do {
1375                                 buf = &buf[n];
1376                                 len -= n;
1377                                 n = os_read_file(req->fds[bit], buf, len);
1378                                 if (n < 0) {
1379                                         printk("do_io - read failed, err = %d "
1380                                                "fd = %d\n", -n, req->fds[bit]);
1381                                         req->error = 1;
1382                                         return;
1383                                 }
1384                         } while((n < len) && (n != 0));
1385                         if (n < len) memset(&buf[n], 0, len - n);
1386                 } else {
1387                         n = os_write_file(req->fds[bit], buf, len);
1388                         if(n != len){
1389                                 printk("do_io - write failed err = %d "
1390                                        "fd = %d\n", -n, req->fds[bit]);
1391                                 req->error = 1;
1392                                 return;
1393                         }
1394                 }
1395
1396                 start = end;
1397         } while(start < nsectors);
1398
1399         req->error = update_bitmap(req);
1400 }
1401
1402 /* Changed in start_io_thread, which is serialized by being called only
1403  * from ubd_init, which is an initcall.
1404  */
1405 int kernel_fd = -1;
1406
1407 /* Only changed by the io thread. XXX: currently unused. */
1408 static int io_count = 0;
1409
1410 int io_thread(void *arg)
1411 {
1412         struct io_thread_req req;
1413         int n;
1414
1415         ignore_sigwinch_sig();
1416         while(1){
1417                 n = os_read_file(kernel_fd, &req, sizeof(req));
1418                 if(n != sizeof(req)){
1419                         if(n < 0)
1420                                 printk("io_thread - read failed, fd = %d, "
1421                                        "err = %d\n", kernel_fd, -n);
1422                         else {
1423                                 printk("io_thread - short read, fd = %d, "
1424                                        "length = %d\n", kernel_fd, n);
1425                         }
1426                         continue;
1427                 }
1428                 io_count++;
1429                 do_io(&req);
1430                 n = os_write_file(kernel_fd, &req, sizeof(req));
1431                 if(n != sizeof(req))
1432                         printk("io_thread - write failed, fd = %d, err = %d\n",
1433                                kernel_fd, -n);
1434         }
1435
1436         return 0;
1437 }