[PATCH] devfs: Remove devfs_mk_dir() function from the kernel tree
[linux-2.6.git] / arch / um / drivers / ubd_kern.c
1 /*
2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 /* 2001-09-28...2002-04-17
7  * Partition stuff by James_McMechan@hotmail.com
8  * old style ubd by setting UBD_SHIFT to 0
9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
10  * partitions have changed in 2.5
11  * 2003-01-29 more tinkering for 2.5.59-1
12  * This should now address the sysfs problems and has
13  * the symlink for devfs to allow for booting with
14  * the common /dev/ubd/discX/... names rather than
15  * only /dev/ubdN/discN this version also has lots of
16  * clean ups preparing for ubd-many.
17  * James McMechan
18  */
19
20 #define MAJOR_NR UBD_MAJOR
21 #define UBD_SHIFT 4
22
23 #include "linux/config.h"
24 #include "linux/module.h"
25 #include "linux/blkdev.h"
26 #include "linux/hdreg.h"
27 #include "linux/init.h"
28 #include "linux/devfs_fs_kernel.h"
29 #include "linux/cdrom.h"
30 #include "linux/proc_fs.h"
31 #include "linux/ctype.h"
32 #include "linux/capability.h"
33 #include "linux/mm.h"
34 #include "linux/vmalloc.h"
35 #include "linux/blkpg.h"
36 #include "linux/genhd.h"
37 #include "linux/spinlock.h"
38 #include "linux/platform_device.h"
39 #include "asm/segment.h"
40 #include "asm/uaccess.h"
41 #include "asm/irq.h"
42 #include "asm/types.h"
43 #include "asm/tlbflush.h"
44 #include "user_util.h"
45 #include "mem_user.h"
46 #include "kern_util.h"
47 #include "kern.h"
48 #include "mconsole_kern.h"
49 #include "init.h"
50 #include "irq_user.h"
51 #include "irq_kern.h"
52 #include "ubd_user.h"
53 #include "os.h"
54 #include "mem.h"
55 #include "mem_kern.h"
56 #include "cow.h"
57
58 enum ubd_req { UBD_READ, UBD_WRITE };
59
60 struct io_thread_req {
61         enum ubd_req op;
62         int fds[2];
63         unsigned long offsets[2];
64         unsigned long long offset;
65         unsigned long length;
66         char *buffer;
67         int sectorsize;
68         unsigned long sector_mask;
69         unsigned long long cow_offset;
70         unsigned long bitmap_words[2];
71         int error;
72 };
73
74 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
75                          char **backing_file_out, int *bitmap_offset_out,
76                          unsigned long *bitmap_len_out, int *data_offset_out,
77                          int *create_cow_out);
78 extern int create_cow_file(char *cow_file, char *backing_file,
79                            struct openflags flags, int sectorsize,
80                            int alignment, int *bitmap_offset_out,
81                            unsigned long *bitmap_len_out,
82                            int *data_offset_out);
83 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
84 extern void do_io(struct io_thread_req *req);
85
86 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
87 {
88         __u64 n;
89         int bits, off;
90
91         bits = sizeof(data[0]) * 8;
92         n = bit / bits;
93         off = bit % bits;
94         return((data[n] & (1 << off)) != 0);
95 }
96
97 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
98 {
99         __u64 n;
100         int bits, off;
101
102         bits = sizeof(data[0]) * 8;
103         n = bit / bits;
104         off = bit % bits;
105         data[n] |= (1 << off);
106 }
107 /*End stuff from ubd_user.h*/
108
109 #define DRIVER_NAME "uml-blkdev"
110
111 static DEFINE_SPINLOCK(ubd_io_lock);
112 static DEFINE_SPINLOCK(ubd_lock);
113
114 static void (*do_ubd)(void);
115
116 static int ubd_open(struct inode * inode, struct file * filp);
117 static int ubd_release(struct inode * inode, struct file * file);
118 static int ubd_ioctl(struct inode * inode, struct file * file,
119                      unsigned int cmd, unsigned long arg);
120 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
121
122 #define MAX_DEV (8)
123
124 static struct block_device_operations ubd_blops = {
125         .owner          = THIS_MODULE,
126         .open           = ubd_open,
127         .release        = ubd_release,
128         .ioctl          = ubd_ioctl,
129         .getgeo         = ubd_getgeo,
130 };
131
132 /* Protected by the queue_lock */
133 static request_queue_t *ubd_queue;
134
135 /* Protected by ubd_lock */
136 static int fake_major = MAJOR_NR;
137
138 static struct gendisk *ubd_gendisk[MAX_DEV];
139 static struct gendisk *fake_gendisk[MAX_DEV];
140
141 #ifdef CONFIG_BLK_DEV_UBD_SYNC
142 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
143                                          .cl = 1 })
144 #else
145 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
146                                          .cl = 1 })
147 #endif
148
149 /* Not protected - changed only in ubd_setup_common and then only to
150  * to enable O_SYNC.
151  */
152 static struct openflags global_openflags = OPEN_FLAGS;
153
154 struct cow {
155         /* This is the backing file, actually */
156         char *file;
157         int fd;
158         unsigned long *bitmap;
159         unsigned long bitmap_len;
160         int bitmap_offset;
161         int data_offset;
162 };
163
164 struct ubd {
165         char *file;
166         int count;
167         int fd;
168         __u64 size;
169         struct openflags boot_openflags;
170         struct openflags openflags;
171         int shared;
172         int no_cow;
173         struct cow cow;
174         struct platform_device pdev;
175 };
176
177 #define DEFAULT_COW { \
178         .file =                 NULL, \
179         .fd =                   -1, \
180         .bitmap =               NULL, \
181         .bitmap_offset =        0, \
182         .data_offset =          0, \
183 }
184
185 #define DEFAULT_UBD { \
186         .file =                 NULL, \
187         .count =                0, \
188         .fd =                   -1, \
189         .size =                 -1, \
190         .boot_openflags =       OPEN_FLAGS, \
191         .openflags =            OPEN_FLAGS, \
192         .no_cow =               0, \
193         .shared =               0, \
194         .cow =                  DEFAULT_COW, \
195 }
196
197 struct ubd ubd_dev[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
198
199 static int ubd0_init(void)
200 {
201         struct ubd *dev = &ubd_dev[0];
202
203         if(dev->file == NULL)
204                 dev->file = "root_fs";
205         return(0);
206 }
207
208 __initcall(ubd0_init);
209
210 /* Only changed by fake_ide_setup which is a setup */
211 static int fake_ide = 0;
212 static struct proc_dir_entry *proc_ide_root = NULL;
213 static struct proc_dir_entry *proc_ide = NULL;
214
215 static void make_proc_ide(void)
216 {
217         proc_ide_root = proc_mkdir("ide", NULL);
218         proc_ide = proc_mkdir("ide0", proc_ide_root);
219 }
220
221 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
222                                int *eof, void *data)
223 {
224         int len;
225
226         strcpy(page, "disk\n");
227         len = strlen("disk\n");
228         len -= off;
229         if (len < count){
230                 *eof = 1;
231                 if (len <= 0) return 0;
232         }
233         else len = count;
234         *start = page + off;
235         return len;
236 }
237
238 static void make_ide_entries(char *dev_name)
239 {
240         struct proc_dir_entry *dir, *ent;
241         char name[64];
242
243         if(proc_ide_root == NULL) make_proc_ide();
244
245         dir = proc_mkdir(dev_name, proc_ide);
246         if(!dir) return;
247
248         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
249         if(!ent) return;
250         ent->nlink = 1;
251         ent->data = NULL;
252         ent->read_proc = proc_ide_read_media;
253         ent->write_proc = NULL;
254         sprintf(name,"ide0/%s", dev_name);
255         proc_symlink(dev_name, proc_ide_root, name);
256 }
257
258 static int fake_ide_setup(char *str)
259 {
260         fake_ide = 1;
261         return(1);
262 }
263
264 __setup("fake_ide", fake_ide_setup);
265
266 __uml_help(fake_ide_setup,
267 "fake_ide\n"
268 "    Create ide0 entries that map onto ubd devices.\n\n"
269 );
270
271 static int parse_unit(char **ptr)
272 {
273         char *str = *ptr, *end;
274         int n = -1;
275
276         if(isdigit(*str)) {
277                 n = simple_strtoul(str, &end, 0);
278                 if(end == str)
279                         return(-1);
280                 *ptr = end;
281         }
282         else if (('a' <= *str) && (*str <= 'h')) {
283                 n = *str - 'a';
284                 str++;
285                 *ptr = str;
286         }
287         return(n);
288 }
289
290 static int ubd_setup_common(char *str, int *index_out)
291 {
292         struct ubd *dev;
293         struct openflags flags = global_openflags;
294         char *backing_file;
295         int n, err, i;
296
297         if(index_out) *index_out = -1;
298         n = *str;
299         if(n == '='){
300                 char *end;
301                 int major;
302
303                 str++;
304                 if(!strcmp(str, "sync")){
305                         global_openflags = of_sync(global_openflags);
306                         return(0);
307                 }
308                 major = simple_strtoul(str, &end, 0);
309                 if((*end != '\0') || (end == str)){
310                         printk(KERN_ERR
311                                "ubd_setup : didn't parse major number\n");
312                         return(1);
313                 }
314
315                 err = 1;
316                 spin_lock(&ubd_lock);
317                 if(fake_major != MAJOR_NR){
318                         printk(KERN_ERR "Can't assign a fake major twice\n");
319                         goto out1;
320                 }
321
322                 fake_major = major;
323
324                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
325                        major);
326                 err = 0;
327         out1:
328                 spin_unlock(&ubd_lock);
329                 return(err);
330         }
331
332         n = parse_unit(&str);
333         if(n < 0){
334                 printk(KERN_ERR "ubd_setup : couldn't parse unit number "
335                        "'%s'\n", str);
336                 return(1);
337         }
338         if(n >= MAX_DEV){
339                 printk(KERN_ERR "ubd_setup : index %d out of range "
340                        "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
341                 return(1);
342         }
343
344         err = 1;
345         spin_lock(&ubd_lock);
346
347         dev = &ubd_dev[n];
348         if(dev->file != NULL){
349                 printk(KERN_ERR "ubd_setup : device already configured\n");
350                 goto out;
351         }
352
353         if (index_out)
354                 *index_out = n;
355
356         for (i = 0; i < sizeof("rscd="); i++) {
357                 switch (*str) {
358                 case 'r':
359                         flags.w = 0;
360                         break;
361                 case 's':
362                         flags.s = 1;
363                         break;
364                 case 'd':
365                         dev->no_cow = 1;
366                         break;
367                 case 'c':
368                         dev->shared = 1;
369                         break;
370                 case '=':
371                         str++;
372                         goto break_loop;
373                 default:
374                         printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r, s, c, or d)\n");
375                         goto out;
376                 }
377                 str++;
378         }
379
380         if (*str == '=')
381                 printk(KERN_ERR "ubd_setup : Too many flags specified\n");
382         else
383                 printk(KERN_ERR "ubd_setup : Expected '='\n");
384         goto out;
385
386 break_loop:
387         err = 0;
388         backing_file = strchr(str, ',');
389
390         if (!backing_file) {
391                 backing_file = strchr(str, ':');
392         }
393
394         if(backing_file){
395                 if(dev->no_cow)
396                         printk(KERN_ERR "Can't specify both 'd' and a "
397                                "cow file\n");
398                 else {
399                         *backing_file = '\0';
400                         backing_file++;
401                 }
402         }
403         dev->file = str;
404         dev->cow.file = backing_file;
405         dev->boot_openflags = flags;
406 out:
407         spin_unlock(&ubd_lock);
408         return(err);
409 }
410
411 static int ubd_setup(char *str)
412 {
413         ubd_setup_common(str, NULL);
414         return(1);
415 }
416
417 __setup("ubd", ubd_setup);
418 __uml_help(ubd_setup,
419 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
420 "    This is used to associate a device with a file in the underlying\n"
421 "    filesystem. When specifying two filenames, the first one is the\n"
422 "    COW name and the second is the backing file name. As separator you can\n"
423 "    use either a ':' or a ',': the first one allows writing things like;\n"
424 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
425 "    while with a ',' the shell would not expand the 2nd '~'.\n"
426 "    When using only one filename, UML will detect whether to thread it like\n"
427 "    a COW file or a backing file. To override this detection, add the 'd'\n"
428 "    flag:\n"
429 "       ubd0d=BackingFile\n"
430 "    Usually, there is a filesystem in the file, but \n"
431 "    that's not required. Swap devices containing swap files can be\n"
432 "    specified like this. Also, a file which doesn't contain a\n"
433 "    filesystem can have its contents read in the virtual \n"
434 "    machine by running 'dd' on the device. <n> must be in the range\n"
435 "    0 to 7. Appending an 'r' to the number will cause that device\n"
436 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
437 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
438 );
439
440 static int udb_setup(char *str)
441 {
442         printk("udb%s specified on command line is almost certainly a ubd -> "
443                "udb TYPO\n", str);
444         return(1);
445 }
446
447 __setup("udb", udb_setup);
448 __uml_help(udb_setup,
449 "udb\n"
450 "    This option is here solely to catch ubd -> udb typos, which can be\n"
451 "    to impossible to catch visually unless you specifically look for\n"
452 "    them.  The only result of any option starting with 'udb' is an error\n"
453 "    in the boot output.\n\n"
454 );
455
456 static int fakehd_set = 0;
457 static int fakehd(char *str)
458 {
459         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
460         fakehd_set = 1;
461         return 1;
462 }
463
464 __setup("fakehd", fakehd);
465 __uml_help(fakehd,
466 "fakehd\n"
467 "    Change the ubd device name to \"hd\".\n\n"
468 );
469
470 static void do_ubd_request(request_queue_t * q);
471
472 /* Only changed by ubd_init, which is an initcall. */
473 int thread_fd = -1;
474
475 /* Changed by ubd_handler, which is serialized because interrupts only
476  * happen on CPU 0.
477  */
478 int intr_count = 0;
479
480 /* call ubd_finish if you need to serialize */
481 static void __ubd_finish(struct request *req, int error)
482 {
483         int nsect;
484
485         if(error){
486                 end_request(req, 0);
487                 return;
488         }
489         nsect = req->current_nr_sectors;
490         req->sector += nsect;
491         req->buffer += nsect << 9;
492         req->errors = 0;
493         req->nr_sectors -= nsect;
494         req->current_nr_sectors = 0;
495         end_request(req, 1);
496 }
497
498 static inline void ubd_finish(struct request *req, int error)
499 {
500         spin_lock(&ubd_io_lock);
501         __ubd_finish(req, error);
502         spin_unlock(&ubd_io_lock);
503 }
504
505 /* Called without ubd_io_lock held */
506 static void ubd_handler(void)
507 {
508         struct io_thread_req req;
509         struct request *rq = elv_next_request(ubd_queue);
510         int n;
511
512         do_ubd = NULL;
513         intr_count++;
514         n = os_read_file(thread_fd, &req, sizeof(req));
515         if(n != sizeof(req)){
516                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
517                        "err = %d\n", os_getpid(), -n);
518                 spin_lock(&ubd_io_lock);
519                 end_request(rq, 0);
520                 spin_unlock(&ubd_io_lock);
521                 return;
522         }
523
524         ubd_finish(rq, req.error);
525         reactivate_fd(thread_fd, UBD_IRQ);      
526         do_ubd_request(ubd_queue);
527 }
528
529 static irqreturn_t ubd_intr(int irq, void *dev, struct pt_regs *unused)
530 {
531         ubd_handler();
532         return(IRQ_HANDLED);
533 }
534
535 /* Only changed by ubd_init, which is an initcall. */
536 static int io_pid = -1;
537
538 void kill_io_thread(void)
539 {
540         if(io_pid != -1)
541                 os_kill_process(io_pid, 1);
542 }
543
544 __uml_exitcall(kill_io_thread);
545
546 static int ubd_file_size(struct ubd *dev, __u64 *size_out)
547 {
548         char *file;
549
550         file = dev->cow.file ? dev->cow.file : dev->file;
551         return(os_file_size(file, size_out));
552 }
553
554 static void ubd_close(struct ubd *dev)
555 {
556         os_close_file(dev->fd);
557         if(dev->cow.file == NULL)
558                 return;
559
560         os_close_file(dev->cow.fd);
561         vfree(dev->cow.bitmap);
562         dev->cow.bitmap = NULL;
563 }
564
565 static int ubd_open_dev(struct ubd *dev)
566 {
567         struct openflags flags;
568         char **back_ptr;
569         int err, create_cow, *create_ptr;
570
571         dev->openflags = dev->boot_openflags;
572         create_cow = 0;
573         create_ptr = (dev->cow.file != NULL) ? &create_cow : NULL;
574         back_ptr = dev->no_cow ? NULL : &dev->cow.file;
575         dev->fd = open_ubd_file(dev->file, &dev->openflags, dev->shared,
576                                 back_ptr, &dev->cow.bitmap_offset,
577                                 &dev->cow.bitmap_len, &dev->cow.data_offset,
578                                 create_ptr);
579
580         if((dev->fd == -ENOENT) && create_cow){
581                 dev->fd = create_cow_file(dev->file, dev->cow.file,
582                                           dev->openflags, 1 << 9, PAGE_SIZE,
583                                           &dev->cow.bitmap_offset,
584                                           &dev->cow.bitmap_len,
585                                           &dev->cow.data_offset);
586                 if(dev->fd >= 0){
587                         printk(KERN_INFO "Creating \"%s\" as COW file for "
588                                "\"%s\"\n", dev->file, dev->cow.file);
589                 }
590         }
591
592         if(dev->fd < 0){
593                 printk("Failed to open '%s', errno = %d\n", dev->file,
594                        -dev->fd);
595                 return(dev->fd);
596         }
597
598         if(dev->cow.file != NULL){
599                 err = -ENOMEM;
600                 dev->cow.bitmap = (void *) vmalloc(dev->cow.bitmap_len);
601                 if(dev->cow.bitmap == NULL){
602                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
603                         goto error;
604                 }
605                 flush_tlb_kernel_vm();
606
607                 err = read_cow_bitmap(dev->fd, dev->cow.bitmap,
608                                       dev->cow.bitmap_offset,
609                                       dev->cow.bitmap_len);
610                 if(err < 0)
611                         goto error;
612
613                 flags = dev->openflags;
614                 flags.w = 0;
615                 err = open_ubd_file(dev->cow.file, &flags, dev->shared, NULL,
616                                     NULL, NULL, NULL, NULL);
617                 if(err < 0) goto error;
618                 dev->cow.fd = err;
619         }
620         return(0);
621  error:
622         os_close_file(dev->fd);
623         return(err);
624 }
625
626 static int ubd_new_disk(int major, u64 size, int unit,
627                         struct gendisk **disk_out)
628                         
629 {
630         struct gendisk *disk;
631         char from[sizeof("ubd/nnnnn\0")], to[sizeof("discnnnnn/disc\0")];
632         int err;
633
634         disk = alloc_disk(1 << UBD_SHIFT);
635         if(disk == NULL)
636                 return(-ENOMEM);
637
638         disk->major = major;
639         disk->first_minor = unit << UBD_SHIFT;
640         disk->fops = &ubd_blops;
641         set_capacity(disk, size / 512);
642         if(major == MAJOR_NR){
643                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
644                 sprintf(disk->devfs_name, "ubd/disc%d", unit);
645                 sprintf(from, "ubd/%d", unit);
646                 sprintf(to, "disc%d/disc", unit);
647                 err = devfs_mk_symlink(from, to);
648                 if(err)
649                         printk("ubd_new_disk failed to make link from %s to "
650                                "%s, error = %d\n", from, to, err);
651         }
652         else {
653                 sprintf(disk->disk_name, "ubd_fake%d", unit);
654                 sprintf(disk->devfs_name, "ubd_fake/disc%d", unit);
655         }
656
657         /* sysfs register (not for ide fake devices) */
658         if (major == MAJOR_NR) {
659                 ubd_dev[unit].pdev.id   = unit;
660                 ubd_dev[unit].pdev.name = DRIVER_NAME;
661                 platform_device_register(&ubd_dev[unit].pdev);
662                 disk->driverfs_dev = &ubd_dev[unit].pdev.dev;
663         }
664
665         disk->private_data = &ubd_dev[unit];
666         disk->queue = ubd_queue;
667         add_disk(disk);
668
669         *disk_out = disk;
670         return 0;
671 }
672
673 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
674
675 static int ubd_add(int n)
676 {
677         struct ubd *dev = &ubd_dev[n];
678         int err;
679
680         err = -ENODEV;
681         if(dev->file == NULL)
682                 goto out;
683
684         if (ubd_open_dev(dev))
685                 goto out;
686
687         err = ubd_file_size(dev, &dev->size);
688         if(err < 0)
689                 goto out_close;
690
691         dev->size = ROUND_BLOCK(dev->size);
692
693         err = ubd_new_disk(MAJOR_NR, dev->size, n, &ubd_gendisk[n]);
694         if(err)
695                 goto out_close;
696
697         if(fake_major != MAJOR_NR)
698                 ubd_new_disk(fake_major, dev->size, n,
699                              &fake_gendisk[n]);
700
701         /* perhaps this should also be under the "if (fake_major)" above */
702         /* using the fake_disk->disk_name and also the fakehd_set name */
703         if (fake_ide)
704                 make_ide_entries(ubd_gendisk[n]->disk_name);
705
706         err = 0;
707 out_close:
708         ubd_close(dev);
709 out:
710         return err;
711 }
712
713 static int ubd_config(char *str)
714 {
715         int n, err;
716
717         str = kstrdup(str, GFP_KERNEL);
718         if(str == NULL){
719                 printk(KERN_ERR "ubd_config failed to strdup string\n");
720                 return(1);
721         }
722         err = ubd_setup_common(str, &n);
723         if(err){
724                 kfree(str);
725                 return(-1);
726         }
727         if(n == -1) return(0);
728
729         spin_lock(&ubd_lock);
730         err = ubd_add(n);
731         if(err)
732                 ubd_dev[n].file = NULL;
733         spin_unlock(&ubd_lock);
734
735         return(err);
736 }
737
738 static int ubd_get_config(char *name, char *str, int size, char **error_out)
739 {
740         struct ubd *dev;
741         int n, len = 0;
742
743         n = parse_unit(&name);
744         if((n >= MAX_DEV) || (n < 0)){
745                 *error_out = "ubd_get_config : device number out of range";
746                 return(-1);
747         }
748
749         dev = &ubd_dev[n];
750         spin_lock(&ubd_lock);
751
752         if(dev->file == NULL){
753                 CONFIG_CHUNK(str, size, len, "", 1);
754                 goto out;
755         }
756
757         CONFIG_CHUNK(str, size, len, dev->file, 0);
758
759         if(dev->cow.file != NULL){
760                 CONFIG_CHUNK(str, size, len, ",", 0);
761                 CONFIG_CHUNK(str, size, len, dev->cow.file, 1);
762         }
763         else CONFIG_CHUNK(str, size, len, "", 1);
764
765  out:
766         spin_unlock(&ubd_lock);
767         return(len);
768 }
769
770 static int ubd_id(char **str, int *start_out, int *end_out)
771 {
772         int n;
773
774         n = parse_unit(str);
775         *start_out = 0;
776         *end_out = MAX_DEV - 1;
777         return n;
778 }
779
780 static int ubd_remove(int n)
781 {
782         struct ubd *dev;
783         int err = -ENODEV;
784
785         spin_lock(&ubd_lock);
786
787         if(ubd_gendisk[n] == NULL)
788                 goto out;
789
790         dev = &ubd_dev[n];
791
792         if(dev->file == NULL)
793                 goto out;
794
795         /* you cannot remove a open disk */
796         err = -EBUSY;
797         if(dev->count > 0)
798                 goto out;
799
800         del_gendisk(ubd_gendisk[n]);
801         put_disk(ubd_gendisk[n]);
802         ubd_gendisk[n] = NULL;
803
804         if(fake_gendisk[n] != NULL){
805                 del_gendisk(fake_gendisk[n]);
806                 put_disk(fake_gendisk[n]);
807                 fake_gendisk[n] = NULL;
808         }
809
810         platform_device_unregister(&dev->pdev);
811         *dev = ((struct ubd) DEFAULT_UBD);
812         err = 0;
813 out:
814         spin_unlock(&ubd_lock);
815         return err;
816 }
817
818 static struct mc_device ubd_mc = {
819         .name           = "ubd",
820         .config         = ubd_config,
821         .get_config     = ubd_get_config,
822         .id             = ubd_id,
823         .remove         = ubd_remove,
824 };
825
826 static int ubd_mc_init(void)
827 {
828         mconsole_register_dev(&ubd_mc);
829         return 0;
830 }
831
832 __initcall(ubd_mc_init);
833
834 static struct platform_driver ubd_driver = {
835         .driver = {
836                 .name  = DRIVER_NAME,
837         },
838 };
839
840 int ubd_init(void)
841 {
842         int i;
843
844         if (register_blkdev(MAJOR_NR, "ubd"))
845                 return -1;
846
847         ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
848         if (!ubd_queue) {
849                 unregister_blkdev(MAJOR_NR, "ubd");
850                 return -1;
851         }
852                 
853         if (fake_major != MAJOR_NR) {
854                 char name[sizeof("ubd_nnn\0")];
855
856                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
857                 if (register_blkdev(fake_major, "ubd"))
858                         return -1;
859         }
860         platform_driver_register(&ubd_driver);
861         for (i = 0; i < MAX_DEV; i++)
862                 ubd_add(i);
863         return 0;
864 }
865
866 late_initcall(ubd_init);
867
868 int ubd_driver_init(void){
869         unsigned long stack;
870         int err;
871
872         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
873         if(global_openflags.s){
874                 printk(KERN_INFO "ubd: Synchronous mode\n");
875                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
876                  * enough. So use anyway the io thread. */
877         }
878         stack = alloc_stack(0, 0);
879         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
880                                  &thread_fd);
881         if(io_pid < 0){
882                 printk(KERN_ERR
883                        "ubd : Failed to start I/O thread (errno = %d) - "
884                        "falling back to synchronous I/O\n", -io_pid);
885                 io_pid = -1;
886                 return(0);
887         }
888         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
889                              SA_INTERRUPT, "ubd", ubd_dev);
890         if(err != 0)
891                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
892         return 0;
893 }
894
895 device_initcall(ubd_driver_init);
896
897 static int ubd_open(struct inode *inode, struct file *filp)
898 {
899         struct gendisk *disk = inode->i_bdev->bd_disk;
900         struct ubd *dev = disk->private_data;
901         int err = 0;
902
903         if(dev->count == 0){
904                 err = ubd_open_dev(dev);
905                 if(err){
906                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
907                                disk->disk_name, dev->file, -err);
908                         goto out;
909                 }
910         }
911         dev->count++;
912         set_disk_ro(disk, !dev->openflags.w);
913
914         /* This should no more be needed. And it didn't work anyway to exclude
915          * read-write remounting of filesystems.*/
916         /*if((filp->f_mode & FMODE_WRITE) && !dev->openflags.w){
917                 if(--dev->count == 0) ubd_close(dev);
918                 err = -EROFS;
919         }*/
920  out:
921         return(err);
922 }
923
924 static int ubd_release(struct inode * inode, struct file * file)
925 {
926         struct gendisk *disk = inode->i_bdev->bd_disk;
927         struct ubd *dev = disk->private_data;
928
929         if(--dev->count == 0)
930                 ubd_close(dev);
931         return(0);
932 }
933
934 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
935                           __u64 *cow_offset, unsigned long *bitmap,
936                           __u64 bitmap_offset, unsigned long *bitmap_words,
937                           __u64 bitmap_len)
938 {
939         __u64 sector = io_offset >> 9;
940         int i, update_bitmap = 0;
941
942         for(i = 0; i < length >> 9; i++){
943                 if(cow_mask != NULL)
944                         ubd_set_bit(i, (unsigned char *) cow_mask);
945                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
946                         continue;
947
948                 update_bitmap = 1;
949                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
950         }
951
952         if(!update_bitmap)
953                 return;
954
955         *cow_offset = sector / (sizeof(unsigned long) * 8);
956
957         /* This takes care of the case where we're exactly at the end of the
958          * device, and *cow_offset + 1 is off the end.  So, just back it up
959          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
960          * for the original diagnosis.
961          */
962         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
963                            sizeof(unsigned long) - 1))
964                 (*cow_offset)--;
965
966         bitmap_words[0] = bitmap[*cow_offset];
967         bitmap_words[1] = bitmap[*cow_offset + 1];
968
969         *cow_offset *= sizeof(unsigned long);
970         *cow_offset += bitmap_offset;
971 }
972
973 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
974                        __u64 bitmap_offset, __u64 bitmap_len)
975 {
976         __u64 sector = req->offset >> 9;
977         int i;
978
979         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
980                 panic("Operation too long");
981
982         if(req->op == UBD_READ) {
983                 for(i = 0; i < req->length >> 9; i++){
984                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
985                                 ubd_set_bit(i, (unsigned char *)
986                                             &req->sector_mask);
987                 }
988         }
989         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
990                            &req->cow_offset, bitmap, bitmap_offset,
991                            req->bitmap_words, bitmap_len);
992 }
993
994 /* Called with ubd_io_lock held */
995 static int prepare_request(struct request *req, struct io_thread_req *io_req)
996 {
997         struct gendisk *disk = req->rq_disk;
998         struct ubd *dev = disk->private_data;
999         __u64 offset;
1000         int len;
1001
1002         if(req->rq_status == RQ_INACTIVE) return(1);
1003
1004         /* This should be impossible now */
1005         if((rq_data_dir(req) == WRITE) && !dev->openflags.w){
1006                 printk("Write attempted on readonly ubd device %s\n",
1007                        disk->disk_name);
1008                 end_request(req, 0);
1009                 return(1);
1010         }
1011
1012         offset = ((__u64) req->sector) << 9;
1013         len = req->current_nr_sectors << 9;
1014
1015         io_req->fds[0] = (dev->cow.file != NULL) ? dev->cow.fd : dev->fd;
1016         io_req->fds[1] = dev->fd;
1017         io_req->cow_offset = -1;
1018         io_req->offset = offset;
1019         io_req->length = len;
1020         io_req->error = 0;
1021         io_req->sector_mask = 0;
1022
1023         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1024         io_req->offsets[0] = 0;
1025         io_req->offsets[1] = dev->cow.data_offset;
1026         io_req->buffer = req->buffer;
1027         io_req->sectorsize = 1 << 9;
1028
1029         if(dev->cow.file != NULL)
1030                 cowify_req(io_req, dev->cow.bitmap, dev->cow.bitmap_offset,
1031                            dev->cow.bitmap_len);
1032
1033         return(0);
1034 }
1035
1036 /* Called with ubd_io_lock held */
1037 static void do_ubd_request(request_queue_t *q)
1038 {
1039         struct io_thread_req io_req;
1040         struct request *req;
1041         int err, n;
1042
1043         if(thread_fd == -1){
1044                 while((req = elv_next_request(q)) != NULL){
1045                         err = prepare_request(req, &io_req);
1046                         if(!err){
1047                                 do_io(&io_req);
1048                                 __ubd_finish(req, io_req.error);
1049                         }
1050                 }
1051         }
1052         else {
1053                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1054                         return;
1055                 err = prepare_request(req, &io_req);
1056                 if(!err){
1057                         do_ubd = ubd_handler;
1058                         n = os_write_file(thread_fd, (char *) &io_req,
1059                                          sizeof(io_req));
1060                         if(n != sizeof(io_req))
1061                                 printk("write to io thread failed, "
1062                                        "errno = %d\n", -n);
1063                 }
1064         }
1065 }
1066
1067 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1068 {
1069         struct ubd *dev = bdev->bd_disk->private_data;
1070
1071         geo->heads = 128;
1072         geo->sectors = 32;
1073         geo->cylinders = dev->size / (128 * 32 * 512);
1074         return 0;
1075 }
1076
1077 static int ubd_ioctl(struct inode * inode, struct file * file,
1078                      unsigned int cmd, unsigned long arg)
1079 {
1080         struct ubd *dev = inode->i_bdev->bd_disk->private_data;
1081         struct hd_driveid ubd_id = {
1082                 .cyls           = 0,
1083                 .heads          = 128,
1084                 .sectors        = 32,
1085         };
1086
1087         switch (cmd) {
1088                 struct cdrom_volctrl volume;
1089         case HDIO_GET_IDENTITY:
1090                 ubd_id.cyls = dev->size / (128 * 32 * 512);
1091                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1092                                  sizeof(ubd_id)))
1093                         return(-EFAULT);
1094                 return(0);
1095                 
1096         case CDROMVOLREAD:
1097                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1098                         return(-EFAULT);
1099                 volume.channel0 = 255;
1100                 volume.channel1 = 255;
1101                 volume.channel2 = 255;
1102                 volume.channel3 = 255;
1103                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1104                         return(-EFAULT);
1105                 return(0);
1106         }
1107         return(-EINVAL);
1108 }
1109
1110 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1111 {
1112         struct uml_stat buf1, buf2;
1113         int err;
1114
1115         if(from_cmdline == NULL)
1116                 return 0;
1117         if(!strcmp(from_cmdline, from_cow))
1118                 return 0;
1119
1120         err = os_stat_file(from_cmdline, &buf1);
1121         if(err < 0){
1122                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1123                 return 0;
1124         }
1125         err = os_stat_file(from_cow, &buf2);
1126         if(err < 0){
1127                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1128                 return 1;
1129         }
1130         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1131                 return 0;
1132
1133         printk("Backing file mismatch - \"%s\" requested,\n"
1134                "\"%s\" specified in COW header of \"%s\"\n",
1135                from_cmdline, from_cow, cow);
1136         return 1;
1137 }
1138
1139 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1140 {
1141         unsigned long modtime;
1142         unsigned long long actual;
1143         int err;
1144
1145         err = os_file_modtime(file, &modtime);
1146         if(err < 0){
1147                 printk("Failed to get modification time of backing file "
1148                        "\"%s\", err = %d\n", file, -err);
1149                 return(err);
1150         }
1151
1152         err = os_file_size(file, &actual);
1153         if(err < 0){
1154                 printk("Failed to get size of backing file \"%s\", "
1155                        "err = %d\n", file, -err);
1156                 return(err);
1157         }
1158
1159         if(actual != size){
1160                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1161                  * the typecast.*/
1162                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1163                        "file\n", (unsigned long long) size, actual);
1164                 return(-EINVAL);
1165         }
1166         if(modtime != mtime){
1167                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1168                        "file\n", mtime, modtime);
1169                 return(-EINVAL);
1170         }
1171         return(0);
1172 }
1173
1174 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1175 {
1176         int err;
1177
1178         err = os_seek_file(fd, offset);
1179         if(err < 0)
1180                 return(err);
1181
1182         err = os_read_file(fd, buf, len);
1183         if(err < 0)
1184                 return(err);
1185
1186         return(0);
1187 }
1188
1189 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1190                   char **backing_file_out, int *bitmap_offset_out,
1191                   unsigned long *bitmap_len_out, int *data_offset_out,
1192                   int *create_cow_out)
1193 {
1194         time_t mtime;
1195         unsigned long long size;
1196         __u32 version, align;
1197         char *backing_file;
1198         int fd, err, sectorsize, asked_switch, mode = 0644;
1199
1200         fd = os_open_file(file, *openflags, mode);
1201         if (fd < 0) {
1202                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1203                         *create_cow_out = 1;
1204                 if (!openflags->w ||
1205                    ((fd != -EROFS) && (fd != -EACCES)))
1206                         return fd;
1207                 openflags->w = 0;
1208                 fd = os_open_file(file, *openflags, mode);
1209                 if (fd < 0)
1210                         return fd;
1211         }
1212
1213         if(shared)
1214                 printk("Not locking \"%s\" on the host\n", file);
1215         else {
1216                 err = os_lock_file(fd, openflags->w);
1217                 if(err < 0){
1218                         printk("Failed to lock '%s', err = %d\n", file, -err);
1219                         goto out_close;
1220                 }
1221         }
1222
1223         /* Succesful return case! */
1224         if(backing_file_out == NULL)
1225                 return(fd);
1226
1227         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1228                               &size, &sectorsize, &align, bitmap_offset_out);
1229         if(err && (*backing_file_out != NULL)){
1230                 printk("Failed to read COW header from COW file \"%s\", "
1231                        "errno = %d\n", file, -err);
1232                 goto out_close;
1233         }
1234         if(err)
1235                 return(fd);
1236
1237         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1238
1239         /* Allow switching only if no mismatch. */
1240         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1241                 printk("Switching backing file to '%s'\n", *backing_file_out);
1242                 err = write_cow_header(file, fd, *backing_file_out,
1243                                        sectorsize, align, &size);
1244                 if (err) {
1245                         printk("Switch failed, errno = %d\n", -err);
1246                         goto out_close;
1247                 }
1248         } else {
1249                 *backing_file_out = backing_file;
1250                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1251                 if (err)
1252                         goto out_close;
1253         }
1254
1255         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1256                   bitmap_len_out, data_offset_out);
1257
1258         return fd;
1259  out_close:
1260         os_close_file(fd);
1261         return err;
1262 }
1263
1264 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1265                     int sectorsize, int alignment, int *bitmap_offset_out,
1266                     unsigned long *bitmap_len_out, int *data_offset_out)
1267 {
1268         int err, fd;
1269
1270         flags.c = 1;
1271         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1272         if(fd < 0){
1273                 err = fd;
1274                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1275                        -err);
1276                 goto out;
1277         }
1278
1279         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1280                             bitmap_offset_out, bitmap_len_out,
1281                             data_offset_out);
1282         if(!err)
1283                 return(fd);
1284         os_close_file(fd);
1285  out:
1286         return(err);
1287 }
1288
1289 static int update_bitmap(struct io_thread_req *req)
1290 {
1291         int n;
1292
1293         if(req->cow_offset == -1)
1294                 return(0);
1295
1296         n = os_seek_file(req->fds[1], req->cow_offset);
1297         if(n < 0){
1298                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1299                 return(1);
1300         }
1301
1302         n = os_write_file(req->fds[1], &req->bitmap_words,
1303                           sizeof(req->bitmap_words));
1304         if(n != sizeof(req->bitmap_words)){
1305                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1306                        req->fds[1]);
1307                 return(1);
1308         }
1309
1310         return(0);
1311 }
1312
1313 void do_io(struct io_thread_req *req)
1314 {
1315         char *buf;
1316         unsigned long len;
1317         int n, nsectors, start, end, bit;
1318         int err;
1319         __u64 off;
1320
1321         nsectors = req->length / req->sectorsize;
1322         start = 0;
1323         do {
1324                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1325                 end = start;
1326                 while((end < nsectors) &&
1327                       (ubd_test_bit(end, (unsigned char *)
1328                                     &req->sector_mask) == bit))
1329                         end++;
1330
1331                 off = req->offset + req->offsets[bit] +
1332                         start * req->sectorsize;
1333                 len = (end - start) * req->sectorsize;
1334                 buf = &req->buffer[start * req->sectorsize];
1335
1336                 err = os_seek_file(req->fds[bit], off);
1337                 if(err < 0){
1338                         printk("do_io - lseek failed : err = %d\n", -err);
1339                         req->error = 1;
1340                         return;
1341                 }
1342                 if(req->op == UBD_READ){
1343                         n = 0;
1344                         do {
1345                                 buf = &buf[n];
1346                                 len -= n;
1347                                 n = os_read_file(req->fds[bit], buf, len);
1348                                 if (n < 0) {
1349                                         printk("do_io - read failed, err = %d "
1350                                                "fd = %d\n", -n, req->fds[bit]);
1351                                         req->error = 1;
1352                                         return;
1353                                 }
1354                         } while((n < len) && (n != 0));
1355                         if (n < len) memset(&buf[n], 0, len - n);
1356                 } else {
1357                         n = os_write_file(req->fds[bit], buf, len);
1358                         if(n != len){
1359                                 printk("do_io - write failed err = %d "
1360                                        "fd = %d\n", -n, req->fds[bit]);
1361                                 req->error = 1;
1362                                 return;
1363                         }
1364                 }
1365
1366                 start = end;
1367         } while(start < nsectors);
1368
1369         req->error = update_bitmap(req);
1370 }
1371
1372 /* Changed in start_io_thread, which is serialized by being called only
1373  * from ubd_init, which is an initcall.
1374  */
1375 int kernel_fd = -1;
1376
1377 /* Only changed by the io thread */
1378 int io_count = 0;
1379
1380 int io_thread(void *arg)
1381 {
1382         struct io_thread_req req;
1383         int n;
1384
1385         ignore_sigwinch_sig();
1386         while(1){
1387                 n = os_read_file(kernel_fd, &req, sizeof(req));
1388                 if(n != sizeof(req)){
1389                         if(n < 0)
1390                                 printk("io_thread - read failed, fd = %d, "
1391                                        "err = %d\n", kernel_fd, -n);
1392                         else {
1393                                 printk("io_thread - short read, fd = %d, "
1394                                        "length = %d\n", kernel_fd, n);
1395                         }
1396                         continue;
1397                 }
1398                 io_count++;
1399                 do_io(&req);
1400                 n = os_write_file(kernel_fd, &req, sizeof(req));
1401                 if(n != sizeof(req))
1402                         printk("io_thread - write failed, fd = %d, err = %d\n",
1403                                kernel_fd, -n);
1404         }
1405
1406         return 0;
1407 }