[PATCH] uml ubd driver: use bitfields where possible
[safe/jmp/linux-2.6] / arch / um / drivers / ubd_kern.c
1 /*
2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 /* 2001-09-28...2002-04-17
7  * Partition stuff by James_McMechan@hotmail.com
8  * old style ubd by setting UBD_SHIFT to 0
9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
10  * partitions have changed in 2.5
11  * 2003-01-29 more tinkering for 2.5.59-1
12  * This should now address the sysfs problems and has
13  * the symlink for devfs to allow for booting with
14  * the common /dev/ubd/discX/... names rather than
15  * only /dev/ubdN/discN this version also has lots of
16  * clean ups preparing for ubd-many.
17  * James McMechan
18  */
19
20 #define MAJOR_NR UBD_MAJOR
21 #define UBD_SHIFT 4
22
23 #include "linux/module.h"
24 #include "linux/blkdev.h"
25 #include "linux/hdreg.h"
26 #include "linux/init.h"
27 #include "linux/cdrom.h"
28 #include "linux/proc_fs.h"
29 #include "linux/ctype.h"
30 #include "linux/capability.h"
31 #include "linux/mm.h"
32 #include "linux/vmalloc.h"
33 #include "linux/blkpg.h"
34 #include "linux/genhd.h"
35 #include "linux/spinlock.h"
36 #include "linux/platform_device.h"
37 #include "asm/segment.h"
38 #include "asm/uaccess.h"
39 #include "asm/irq.h"
40 #include "asm/types.h"
41 #include "asm/tlbflush.h"
42 #include "user_util.h"
43 #include "mem_user.h"
44 #include "kern_util.h"
45 #include "kern.h"
46 #include "mconsole_kern.h"
47 #include "init.h"
48 #include "irq_user.h"
49 #include "irq_kern.h"
50 #include "ubd_user.h"
51 #include "os.h"
52 #include "mem.h"
53 #include "mem_kern.h"
54 #include "cow.h"
55
56 enum ubd_req { UBD_READ, UBD_WRITE };
57
58 struct io_thread_req {
59         enum ubd_req op;
60         int fds[2];
61         unsigned long offsets[2];
62         unsigned long long offset;
63         unsigned long length;
64         char *buffer;
65         int sectorsize;
66         unsigned long sector_mask;
67         unsigned long long cow_offset;
68         unsigned long bitmap_words[2];
69         int error;
70 };
71
72 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
73                          char **backing_file_out, int *bitmap_offset_out,
74                          unsigned long *bitmap_len_out, int *data_offset_out,
75                          int *create_cow_out);
76 extern int create_cow_file(char *cow_file, char *backing_file,
77                            struct openflags flags, int sectorsize,
78                            int alignment, int *bitmap_offset_out,
79                            unsigned long *bitmap_len_out,
80                            int *data_offset_out);
81 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
82 extern void do_io(struct io_thread_req *req);
83
84 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
85 {
86         __u64 n;
87         int bits, off;
88
89         bits = sizeof(data[0]) * 8;
90         n = bit / bits;
91         off = bit % bits;
92         return((data[n] & (1 << off)) != 0);
93 }
94
95 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
96 {
97         __u64 n;
98         int bits, off;
99
100         bits = sizeof(data[0]) * 8;
101         n = bit / bits;
102         off = bit % bits;
103         data[n] |= (1 << off);
104 }
105 /*End stuff from ubd_user.h*/
106
107 #define DRIVER_NAME "uml-blkdev"
108
109 /* Can be taken in interrupt context, and is passed to the block layer to lock
110  * the request queue. Kernel side code knows that. */
111 static DEFINE_SPINLOCK(ubd_io_lock);
112
113 static DEFINE_MUTEX(ubd_lock);
114
115 /* XXX - this made sense in 2.4 days, now it's only used as a boolean, and
116  * probably it doesn't make sense even for that. */
117 static int do_ubd;
118
119 static int ubd_open(struct inode * inode, struct file * filp);
120 static int ubd_release(struct inode * inode, struct file * file);
121 static int ubd_ioctl(struct inode * inode, struct file * file,
122                      unsigned int cmd, unsigned long arg);
123 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
124
125 #define MAX_DEV (16)
126
127 static struct block_device_operations ubd_blops = {
128         .owner          = THIS_MODULE,
129         .open           = ubd_open,
130         .release        = ubd_release,
131         .ioctl          = ubd_ioctl,
132         .getgeo         = ubd_getgeo,
133 };
134
135 /* Protected by the queue_lock */
136 static request_queue_t *ubd_queue;
137
138 /* Protected by ubd_lock */
139 static int fake_major = MAJOR_NR;
140
141 static struct gendisk *ubd_gendisk[MAX_DEV];
142 static struct gendisk *fake_gendisk[MAX_DEV];
143
144 #ifdef CONFIG_BLK_DEV_UBD_SYNC
145 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
146                                          .cl = 1 })
147 #else
148 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
149                                          .cl = 1 })
150 #endif
151
152 /* Not protected - changed only in ubd_setup_common and then only to
153  * to enable O_SYNC.
154  */
155 static struct openflags global_openflags = OPEN_FLAGS;
156
157 struct cow {
158         /* backing file name */
159         char *file;
160         /* backing file fd */
161         int fd;
162         unsigned long *bitmap;
163         unsigned long bitmap_len;
164         int bitmap_offset;
165         int data_offset;
166 };
167
168 struct ubd {
169         /* name (and fd, below) of the file opened for writing, either the
170          * backing or the cow file. */
171         char *file;
172         int count;
173         int fd;
174         __u64 size;
175         struct openflags boot_openflags;
176         struct openflags openflags;
177         unsigned shared:1;
178         unsigned no_cow:1;
179         struct cow cow;
180         struct platform_device pdev;
181 };
182
183 #define DEFAULT_COW { \
184         .file =                 NULL, \
185         .fd =                   -1, \
186         .bitmap =               NULL, \
187         .bitmap_offset =        0, \
188         .data_offset =          0, \
189 }
190
191 #define DEFAULT_UBD { \
192         .file =                 NULL, \
193         .count =                0, \
194         .fd =                   -1, \
195         .size =                 -1, \
196         .boot_openflags =       OPEN_FLAGS, \
197         .openflags =            OPEN_FLAGS, \
198         .no_cow =               0, \
199         .shared =               0, \
200         .cow =                  DEFAULT_COW, \
201 }
202
203 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
204
205 static int ubd0_init(void)
206 {
207         struct ubd *ubd_dev = &ubd_devs[0];
208
209         if(ubd_dev->file == NULL)
210                 ubd_dev->file = "root_fs";
211         return(0);
212 }
213
214 __initcall(ubd0_init);
215
216 /* Only changed by fake_ide_setup which is a setup */
217 static int fake_ide = 0;
218 static struct proc_dir_entry *proc_ide_root = NULL;
219 static struct proc_dir_entry *proc_ide = NULL;
220
221 static void make_proc_ide(void)
222 {
223         proc_ide_root = proc_mkdir("ide", NULL);
224         proc_ide = proc_mkdir("ide0", proc_ide_root);
225 }
226
227 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
228                                int *eof, void *data)
229 {
230         int len;
231
232         strcpy(page, "disk\n");
233         len = strlen("disk\n");
234         len -= off;
235         if (len < count){
236                 *eof = 1;
237                 if (len <= 0) return 0;
238         }
239         else len = count;
240         *start = page + off;
241         return len;
242 }
243
244 static void make_ide_entries(char *dev_name)
245 {
246         struct proc_dir_entry *dir, *ent;
247         char name[64];
248
249         if(proc_ide_root == NULL) make_proc_ide();
250
251         dir = proc_mkdir(dev_name, proc_ide);
252         if(!dir) return;
253
254         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
255         if(!ent) return;
256         ent->nlink = 1;
257         ent->data = NULL;
258         ent->read_proc = proc_ide_read_media;
259         ent->write_proc = NULL;
260         sprintf(name,"ide0/%s", dev_name);
261         proc_symlink(dev_name, proc_ide_root, name);
262 }
263
264 static int fake_ide_setup(char *str)
265 {
266         fake_ide = 1;
267         return(1);
268 }
269
270 __setup("fake_ide", fake_ide_setup);
271
272 __uml_help(fake_ide_setup,
273 "fake_ide\n"
274 "    Create ide0 entries that map onto ubd devices.\n\n"
275 );
276
277 static int parse_unit(char **ptr)
278 {
279         char *str = *ptr, *end;
280         int n = -1;
281
282         if(isdigit(*str)) {
283                 n = simple_strtoul(str, &end, 0);
284                 if(end == str)
285                         return(-1);
286                 *ptr = end;
287         }
288         else if (('a' <= *str) && (*str <= 'z')) {
289                 n = *str - 'a';
290                 str++;
291                 *ptr = str;
292         }
293         return(n);
294 }
295
296 static int ubd_setup_common(char *str, int *index_out)
297 {
298         struct ubd *ubd_dev;
299         struct openflags flags = global_openflags;
300         char *backing_file;
301         int n, err, i;
302
303         if(index_out) *index_out = -1;
304         n = *str;
305         if(n == '='){
306                 char *end;
307                 int major;
308
309                 str++;
310                 if(!strcmp(str, "sync")){
311                         global_openflags = of_sync(global_openflags);
312                         return(0);
313                 }
314                 major = simple_strtoul(str, &end, 0);
315                 if((*end != '\0') || (end == str)){
316                         printk(KERN_ERR
317                                "ubd_setup : didn't parse major number\n");
318                         return(1);
319                 }
320
321                 err = 1;
322                 mutex_lock(&ubd_lock);
323                 if(fake_major != MAJOR_NR){
324                         printk(KERN_ERR "Can't assign a fake major twice\n");
325                         goto out1;
326                 }
327
328                 fake_major = major;
329
330                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
331                        major);
332                 err = 0;
333         out1:
334                 mutex_unlock(&ubd_lock);
335                 return(err);
336         }
337
338         n = parse_unit(&str);
339         if(n < 0){
340                 printk(KERN_ERR "ubd_setup : couldn't parse unit number "
341                        "'%s'\n", str);
342                 return(1);
343         }
344         if(n >= MAX_DEV){
345                 printk(KERN_ERR "ubd_setup : index %d out of range "
346                        "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
347                 return(1);
348         }
349
350         err = 1;
351         mutex_lock(&ubd_lock);
352
353         ubd_dev = &ubd_devs[n];
354         if(ubd_dev->file != NULL){
355                 printk(KERN_ERR "ubd_setup : device already configured\n");
356                 goto out;
357         }
358
359         if (index_out)
360                 *index_out = n;
361
362         for (i = 0; i < sizeof("rscd="); i++) {
363                 switch (*str) {
364                 case 'r':
365                         flags.w = 0;
366                         break;
367                 case 's':
368                         flags.s = 1;
369                         break;
370                 case 'd':
371                         ubd_dev->no_cow = 1;
372                         break;
373                 case 'c':
374                         ubd_dev->shared = 1;
375                         break;
376                 case '=':
377                         str++;
378                         goto break_loop;
379                 default:
380                         printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r, s, c, or d)\n");
381                         goto out;
382                 }
383                 str++;
384         }
385
386         if (*str == '=')
387                 printk(KERN_ERR "ubd_setup : Too many flags specified\n");
388         else
389                 printk(KERN_ERR "ubd_setup : Expected '='\n");
390         goto out;
391
392 break_loop:
393         err = 0;
394         backing_file = strchr(str, ',');
395
396         if (!backing_file) {
397                 backing_file = strchr(str, ':');
398         }
399
400         if(backing_file){
401                 if(ubd_dev->no_cow)
402                         printk(KERN_ERR "Can't specify both 'd' and a "
403                                "cow file\n");
404                 else {
405                         *backing_file = '\0';
406                         backing_file++;
407                 }
408         }
409         ubd_dev->file = str;
410         ubd_dev->cow.file = backing_file;
411         ubd_dev->boot_openflags = flags;
412 out:
413         mutex_unlock(&ubd_lock);
414         return(err);
415 }
416
417 static int ubd_setup(char *str)
418 {
419         ubd_setup_common(str, NULL);
420         return(1);
421 }
422
423 __setup("ubd", ubd_setup);
424 __uml_help(ubd_setup,
425 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
426 "    This is used to associate a device with a file in the underlying\n"
427 "    filesystem. When specifying two filenames, the first one is the\n"
428 "    COW name and the second is the backing file name. As separator you can\n"
429 "    use either a ':' or a ',': the first one allows writing things like;\n"
430 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
431 "    while with a ',' the shell would not expand the 2nd '~'.\n"
432 "    When using only one filename, UML will detect whether to thread it like\n"
433 "    a COW file or a backing file. To override this detection, add the 'd'\n"
434 "    flag:\n"
435 "       ubd0d=BackingFile\n"
436 "    Usually, there is a filesystem in the file, but \n"
437 "    that's not required. Swap devices containing swap files can be\n"
438 "    specified like this. Also, a file which doesn't contain a\n"
439 "    filesystem can have its contents read in the virtual \n"
440 "    machine by running 'dd' on the device. <n> must be in the range\n"
441 "    0 to 7. Appending an 'r' to the number will cause that device\n"
442 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
443 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
444 );
445
446 static int udb_setup(char *str)
447 {
448         printk("udb%s specified on command line is almost certainly a ubd -> "
449                "udb TYPO\n", str);
450         return(1);
451 }
452
453 __setup("udb", udb_setup);
454 __uml_help(udb_setup,
455 "udb\n"
456 "    This option is here solely to catch ubd -> udb typos, which can be\n"
457 "    to impossible to catch visually unless you specifically look for\n"
458 "    them.  The only result of any option starting with 'udb' is an error\n"
459 "    in the boot output.\n\n"
460 );
461
462 static int fakehd_set = 0;
463 static int fakehd(char *str)
464 {
465         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
466         fakehd_set = 1;
467         return 1;
468 }
469
470 __setup("fakehd", fakehd);
471 __uml_help(fakehd,
472 "fakehd\n"
473 "    Change the ubd device name to \"hd\".\n\n"
474 );
475
476 static void do_ubd_request(request_queue_t * q);
477
478 /* Only changed by ubd_init, which is an initcall. */
479 int thread_fd = -1;
480
481 /* Changed by ubd_handler, which is serialized because interrupts only
482  * happen on CPU 0.
483  */
484 int intr_count = 0;
485
486 /* call ubd_finish if you need to serialize */
487 static void __ubd_finish(struct request *req, int error)
488 {
489         int nsect;
490
491         if(error){
492                 end_request(req, 0);
493                 return;
494         }
495         nsect = req->current_nr_sectors;
496         req->sector += nsect;
497         req->buffer += nsect << 9;
498         req->errors = 0;
499         req->nr_sectors -= nsect;
500         req->current_nr_sectors = 0;
501         end_request(req, 1);
502 }
503
504 /* Callable only from interrupt context - otherwise you need to do
505  * spin_lock_irq()/spin_lock_irqsave() */
506 static inline void ubd_finish(struct request *req, int error)
507 {
508         spin_lock(&ubd_io_lock);
509         __ubd_finish(req, error);
510         spin_unlock(&ubd_io_lock);
511 }
512
513 /* XXX - move this inside ubd_intr. */
514 /* Called without ubd_io_lock held, and only in interrupt context. */
515 static void ubd_handler(void)
516 {
517         struct io_thread_req req;
518         struct request *rq = elv_next_request(ubd_queue);
519         int n;
520
521         do_ubd = 0;
522         intr_count++;
523         n = os_read_file(thread_fd, &req, sizeof(req));
524         if(n != sizeof(req)){
525                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
526                        "err = %d\n", os_getpid(), -n);
527                 spin_lock(&ubd_io_lock);
528                 end_request(rq, 0);
529                 spin_unlock(&ubd_io_lock);
530                 return;
531         }
532
533         ubd_finish(rq, req.error);
534         reactivate_fd(thread_fd, UBD_IRQ);      
535         spin_lock(&ubd_io_lock);
536         do_ubd_request(ubd_queue);
537         spin_unlock(&ubd_io_lock);
538 }
539
540 static irqreturn_t ubd_intr(int irq, void *dev)
541 {
542         ubd_handler();
543         return(IRQ_HANDLED);
544 }
545
546 /* Only changed by ubd_init, which is an initcall. */
547 static int io_pid = -1;
548
549 void kill_io_thread(void)
550 {
551         if(io_pid != -1)
552                 os_kill_process(io_pid, 1);
553 }
554
555 __uml_exitcall(kill_io_thread);
556
557 static int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
558 {
559         char *file;
560
561         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
562         return(os_file_size(file, size_out));
563 }
564
565 static void ubd_close_dev(struct ubd *ubd_dev)
566 {
567         os_close_file(ubd_dev->fd);
568         if(ubd_dev->cow.file == NULL)
569                 return;
570
571         os_close_file(ubd_dev->cow.fd);
572         vfree(ubd_dev->cow.bitmap);
573         ubd_dev->cow.bitmap = NULL;
574 }
575
576 static int ubd_open_dev(struct ubd *ubd_dev)
577 {
578         struct openflags flags;
579         char **back_ptr;
580         int err, create_cow, *create_ptr;
581
582         ubd_dev->openflags = ubd_dev->boot_openflags;
583         create_cow = 0;
584         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
585         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
586         ubd_dev->fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
587                                 back_ptr, &ubd_dev->cow.bitmap_offset,
588                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
589                                 create_ptr);
590
591         if((ubd_dev->fd == -ENOENT) && create_cow){
592                 ubd_dev->fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
593                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
594                                           &ubd_dev->cow.bitmap_offset,
595                                           &ubd_dev->cow.bitmap_len,
596                                           &ubd_dev->cow.data_offset);
597                 if(ubd_dev->fd >= 0){
598                         printk(KERN_INFO "Creating \"%s\" as COW file for "
599                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
600                 }
601         }
602
603         if(ubd_dev->fd < 0){
604                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
605                        -ubd_dev->fd);
606                 return(ubd_dev->fd);
607         }
608
609         if(ubd_dev->cow.file != NULL){
610                 err = -ENOMEM;
611                 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
612                 if(ubd_dev->cow.bitmap == NULL){
613                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
614                         goto error;
615                 }
616                 flush_tlb_kernel_vm();
617
618                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
619                                       ubd_dev->cow.bitmap_offset,
620                                       ubd_dev->cow.bitmap_len);
621                 if(err < 0)
622                         goto error;
623
624                 flags = ubd_dev->openflags;
625                 flags.w = 0;
626                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
627                                     NULL, NULL, NULL, NULL);
628                 if(err < 0) goto error;
629                 ubd_dev->cow.fd = err;
630         }
631         return(0);
632  error:
633         os_close_file(ubd_dev->fd);
634         return(err);
635 }
636
637 static int ubd_disk_register(int major, u64 size, int unit,
638                         struct gendisk **disk_out)
639                         
640 {
641         struct gendisk *disk;
642
643         disk = alloc_disk(1 << UBD_SHIFT);
644         if(disk == NULL)
645                 return(-ENOMEM);
646
647         disk->major = major;
648         disk->first_minor = unit << UBD_SHIFT;
649         disk->fops = &ubd_blops;
650         set_capacity(disk, size / 512);
651         if(major == MAJOR_NR)
652                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
653         else
654                 sprintf(disk->disk_name, "ubd_fake%d", unit);
655
656         /* sysfs register (not for ide fake devices) */
657         if (major == MAJOR_NR) {
658                 ubd_devs[unit].pdev.id   = unit;
659                 ubd_devs[unit].pdev.name = DRIVER_NAME;
660                 platform_device_register(&ubd_devs[unit].pdev);
661                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
662         }
663
664         disk->private_data = &ubd_devs[unit];
665         disk->queue = ubd_queue;
666         add_disk(disk);
667
668         *disk_out = disk;
669         return 0;
670 }
671
672 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
673
674 static int ubd_add(int n)
675 {
676         struct ubd *ubd_dev = &ubd_devs[n];
677         int err;
678
679         err = -ENODEV;
680         if(ubd_dev->file == NULL)
681                 goto out;
682
683         err = ubd_file_size(ubd_dev, &ubd_dev->size);
684         if(err < 0)
685                 goto out;
686
687         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
688
689         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
690         if(err)
691                 goto out;
692
693         if(fake_major != MAJOR_NR)
694                 ubd_disk_register(fake_major, ubd_dev->size, n,
695                              &fake_gendisk[n]);
696
697         /* perhaps this should also be under the "if (fake_major)" above */
698         /* using the fake_disk->disk_name and also the fakehd_set name */
699         if (fake_ide)
700                 make_ide_entries(ubd_gendisk[n]->disk_name);
701
702         err = 0;
703 out:
704         return err;
705 }
706
707 static int ubd_config(char *str)
708 {
709         int n, ret;
710
711         str = kstrdup(str, GFP_KERNEL);
712         if (str == NULL) {
713                 printk(KERN_ERR "ubd_config failed to strdup string\n");
714                 ret = 1;
715                 goto out;
716         }
717         ret = ubd_setup_common(str, &n);
718         if (ret) {
719                 ret = -1;
720                 goto err_free;
721         }
722         if (n == -1) {
723                 ret = 0;
724                 goto out;
725         }
726
727         mutex_lock(&ubd_lock);
728         ret = ubd_add(n);
729         if (ret)
730                 ubd_devs[n].file = NULL;
731         mutex_unlock(&ubd_lock);
732
733 out:
734         return ret;
735
736 err_free:
737         kfree(str);
738         goto out;
739 }
740
741 static int ubd_get_config(char *name, char *str, int size, char **error_out)
742 {
743         struct ubd *ubd_dev;
744         int n, len = 0;
745
746         n = parse_unit(&name);
747         if((n >= MAX_DEV) || (n < 0)){
748                 *error_out = "ubd_get_config : device number out of range";
749                 return(-1);
750         }
751
752         ubd_dev = &ubd_devs[n];
753         mutex_lock(&ubd_lock);
754
755         if(ubd_dev->file == NULL){
756                 CONFIG_CHUNK(str, size, len, "", 1);
757                 goto out;
758         }
759
760         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
761
762         if(ubd_dev->cow.file != NULL){
763                 CONFIG_CHUNK(str, size, len, ",", 0);
764                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
765         }
766         else CONFIG_CHUNK(str, size, len, "", 1);
767
768  out:
769         mutex_unlock(&ubd_lock);
770         return(len);
771 }
772
773 static int ubd_id(char **str, int *start_out, int *end_out)
774 {
775         int n;
776
777         n = parse_unit(str);
778         *start_out = 0;
779         *end_out = MAX_DEV - 1;
780         return n;
781 }
782
783 static int ubd_remove(int n)
784 {
785         struct ubd *ubd_dev;
786         int err = -ENODEV;
787
788         mutex_lock(&ubd_lock);
789
790         if(ubd_gendisk[n] == NULL)
791                 goto out;
792
793         ubd_dev = &ubd_devs[n];
794
795         if(ubd_dev->file == NULL)
796                 goto out;
797
798         /* you cannot remove a open disk */
799         err = -EBUSY;
800         if(ubd_dev->count > 0)
801                 goto out;
802
803         del_gendisk(ubd_gendisk[n]);
804         put_disk(ubd_gendisk[n]);
805         ubd_gendisk[n] = NULL;
806
807         if(fake_gendisk[n] != NULL){
808                 del_gendisk(fake_gendisk[n]);
809                 put_disk(fake_gendisk[n]);
810                 fake_gendisk[n] = NULL;
811         }
812
813         platform_device_unregister(&ubd_dev->pdev);
814         *ubd_dev = ((struct ubd) DEFAULT_UBD);
815         err = 0;
816 out:
817         mutex_unlock(&ubd_lock);
818         return err;
819 }
820
821 static struct mc_device ubd_mc = {
822         .name           = "ubd",
823         .config         = ubd_config,
824         .get_config     = ubd_get_config,
825         .id             = ubd_id,
826         .remove         = ubd_remove,
827 };
828
829 static int ubd_mc_init(void)
830 {
831         mconsole_register_dev(&ubd_mc);
832         return 0;
833 }
834
835 __initcall(ubd_mc_init);
836
837 static struct platform_driver ubd_driver = {
838         .driver = {
839                 .name  = DRIVER_NAME,
840         },
841 };
842
843 int ubd_init(void)
844 {
845         int i;
846
847         if (register_blkdev(MAJOR_NR, "ubd"))
848                 return -1;
849
850         ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
851         if (!ubd_queue) {
852                 unregister_blkdev(MAJOR_NR, "ubd");
853                 return -1;
854         }
855                 
856         if (fake_major != MAJOR_NR) {
857                 char name[sizeof("ubd_nnn\0")];
858
859                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
860                 if (register_blkdev(fake_major, "ubd"))
861                         return -1;
862         }
863         platform_driver_register(&ubd_driver);
864         for (i = 0; i < MAX_DEV; i++)
865                 ubd_add(i);
866         return 0;
867 }
868
869 late_initcall(ubd_init);
870
871 int ubd_driver_init(void){
872         unsigned long stack;
873         int err;
874
875         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
876         if(global_openflags.s){
877                 printk(KERN_INFO "ubd: Synchronous mode\n");
878                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
879                  * enough. So use anyway the io thread. */
880         }
881         stack = alloc_stack(0, 0);
882         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
883                                  &thread_fd);
884         if(io_pid < 0){
885                 printk(KERN_ERR
886                        "ubd : Failed to start I/O thread (errno = %d) - "
887                        "falling back to synchronous I/O\n", -io_pid);
888                 io_pid = -1;
889                 return(0);
890         }
891         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
892                              IRQF_DISABLED, "ubd", ubd_devs);
893         if(err != 0)
894                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
895         return 0;
896 }
897
898 device_initcall(ubd_driver_init);
899
900 static int ubd_open(struct inode *inode, struct file *filp)
901 {
902         struct gendisk *disk = inode->i_bdev->bd_disk;
903         struct ubd *ubd_dev = disk->private_data;
904         int err = 0;
905
906         if(ubd_dev->count == 0){
907                 err = ubd_open_dev(ubd_dev);
908                 if(err){
909                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
910                                disk->disk_name, ubd_dev->file, -err);
911                         goto out;
912                 }
913         }
914         ubd_dev->count++;
915         set_disk_ro(disk, !ubd_dev->openflags.w);
916
917         /* This should no more be needed. And it didn't work anyway to exclude
918          * read-write remounting of filesystems.*/
919         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
920                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
921                 err = -EROFS;
922         }*/
923  out:
924         return(err);
925 }
926
927 static int ubd_release(struct inode * inode, struct file * file)
928 {
929         struct gendisk *disk = inode->i_bdev->bd_disk;
930         struct ubd *ubd_dev = disk->private_data;
931
932         if(--ubd_dev->count == 0)
933                 ubd_close_dev(ubd_dev);
934         return(0);
935 }
936
937 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
938                           __u64 *cow_offset, unsigned long *bitmap,
939                           __u64 bitmap_offset, unsigned long *bitmap_words,
940                           __u64 bitmap_len)
941 {
942         __u64 sector = io_offset >> 9;
943         int i, update_bitmap = 0;
944
945         for(i = 0; i < length >> 9; i++){
946                 if(cow_mask != NULL)
947                         ubd_set_bit(i, (unsigned char *) cow_mask);
948                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
949                         continue;
950
951                 update_bitmap = 1;
952                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
953         }
954
955         if(!update_bitmap)
956                 return;
957
958         *cow_offset = sector / (sizeof(unsigned long) * 8);
959
960         /* This takes care of the case where we're exactly at the end of the
961          * device, and *cow_offset + 1 is off the end.  So, just back it up
962          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
963          * for the original diagnosis.
964          */
965         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
966                            sizeof(unsigned long) - 1))
967                 (*cow_offset)--;
968
969         bitmap_words[0] = bitmap[*cow_offset];
970         bitmap_words[1] = bitmap[*cow_offset + 1];
971
972         *cow_offset *= sizeof(unsigned long);
973         *cow_offset += bitmap_offset;
974 }
975
976 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
977                        __u64 bitmap_offset, __u64 bitmap_len)
978 {
979         __u64 sector = req->offset >> 9;
980         int i;
981
982         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
983                 panic("Operation too long");
984
985         if(req->op == UBD_READ) {
986                 for(i = 0; i < req->length >> 9; i++){
987                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
988                                 ubd_set_bit(i, (unsigned char *)
989                                             &req->sector_mask);
990                 }
991         }
992         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
993                            &req->cow_offset, bitmap, bitmap_offset,
994                            req->bitmap_words, bitmap_len);
995 }
996
997 /* Called with ubd_io_lock held */
998 static int prepare_request(struct request *req, struct io_thread_req *io_req)
999 {
1000         struct gendisk *disk = req->rq_disk;
1001         struct ubd *ubd_dev = disk->private_data;
1002         __u64 offset;
1003         int len;
1004
1005         /* This should be impossible now */
1006         if((rq_data_dir(req) == WRITE) && !ubd_dev->openflags.w){
1007                 printk("Write attempted on readonly ubd device %s\n",
1008                        disk->disk_name);
1009                 end_request(req, 0);
1010                 return(1);
1011         }
1012
1013         offset = ((__u64) req->sector) << 9;
1014         len = req->current_nr_sectors << 9;
1015
1016         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd : ubd_dev->fd;
1017         io_req->fds[1] = ubd_dev->fd;
1018         io_req->cow_offset = -1;
1019         io_req->offset = offset;
1020         io_req->length = len;
1021         io_req->error = 0;
1022         io_req->sector_mask = 0;
1023
1024         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1025         io_req->offsets[0] = 0;
1026         io_req->offsets[1] = ubd_dev->cow.data_offset;
1027         io_req->buffer = req->buffer;
1028         io_req->sectorsize = 1 << 9;
1029
1030         if(ubd_dev->cow.file != NULL)
1031                 cowify_req(io_req, ubd_dev->cow.bitmap, ubd_dev->cow.bitmap_offset,
1032                            ubd_dev->cow.bitmap_len);
1033
1034         return(0);
1035 }
1036
1037 /* Called with ubd_io_lock held */
1038 static void do_ubd_request(request_queue_t *q)
1039 {
1040         struct io_thread_req io_req;
1041         struct request *req;
1042         int err, n;
1043
1044         if(thread_fd == -1){
1045                 while((req = elv_next_request(q)) != NULL){
1046                         err = prepare_request(req, &io_req);
1047                         if(!err){
1048                                 do_io(&io_req);
1049                                 __ubd_finish(req, io_req.error);
1050                         }
1051                 }
1052         }
1053         else {
1054                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1055                         return;
1056                 err = prepare_request(req, &io_req);
1057                 if(!err){
1058                         do_ubd = 1;
1059                         n = os_write_file(thread_fd, (char *) &io_req,
1060                                          sizeof(io_req));
1061                         if(n != sizeof(io_req))
1062                                 printk("write to io thread failed, "
1063                                        "errno = %d\n", -n);
1064                 }
1065         }
1066 }
1067
1068 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1069 {
1070         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1071
1072         geo->heads = 128;
1073         geo->sectors = 32;
1074         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1075         return 0;
1076 }
1077
1078 static int ubd_ioctl(struct inode * inode, struct file * file,
1079                      unsigned int cmd, unsigned long arg)
1080 {
1081         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1082         struct hd_driveid ubd_id = {
1083                 .cyls           = 0,
1084                 .heads          = 128,
1085                 .sectors        = 32,
1086         };
1087
1088         switch (cmd) {
1089                 struct cdrom_volctrl volume;
1090         case HDIO_GET_IDENTITY:
1091                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1092                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1093                                  sizeof(ubd_id)))
1094                         return(-EFAULT);
1095                 return(0);
1096                 
1097         case CDROMVOLREAD:
1098                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1099                         return(-EFAULT);
1100                 volume.channel0 = 255;
1101                 volume.channel1 = 255;
1102                 volume.channel2 = 255;
1103                 volume.channel3 = 255;
1104                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1105                         return(-EFAULT);
1106                 return(0);
1107         }
1108         return(-EINVAL);
1109 }
1110
1111 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1112 {
1113         struct uml_stat buf1, buf2;
1114         int err;
1115
1116         if(from_cmdline == NULL)
1117                 return 0;
1118         if(!strcmp(from_cmdline, from_cow))
1119                 return 0;
1120
1121         err = os_stat_file(from_cmdline, &buf1);
1122         if(err < 0){
1123                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1124                 return 0;
1125         }
1126         err = os_stat_file(from_cow, &buf2);
1127         if(err < 0){
1128                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1129                 return 1;
1130         }
1131         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1132                 return 0;
1133
1134         printk("Backing file mismatch - \"%s\" requested,\n"
1135                "\"%s\" specified in COW header of \"%s\"\n",
1136                from_cmdline, from_cow, cow);
1137         return 1;
1138 }
1139
1140 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1141 {
1142         unsigned long modtime;
1143         unsigned long long actual;
1144         int err;
1145
1146         err = os_file_modtime(file, &modtime);
1147         if(err < 0){
1148                 printk("Failed to get modification time of backing file "
1149                        "\"%s\", err = %d\n", file, -err);
1150                 return(err);
1151         }
1152
1153         err = os_file_size(file, &actual);
1154         if(err < 0){
1155                 printk("Failed to get size of backing file \"%s\", "
1156                        "err = %d\n", file, -err);
1157                 return(err);
1158         }
1159
1160         if(actual != size){
1161                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1162                  * the typecast.*/
1163                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1164                        "file\n", (unsigned long long) size, actual);
1165                 return(-EINVAL);
1166         }
1167         if(modtime != mtime){
1168                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1169                        "file\n", mtime, modtime);
1170                 return(-EINVAL);
1171         }
1172         return(0);
1173 }
1174
1175 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1176 {
1177         int err;
1178
1179         err = os_seek_file(fd, offset);
1180         if(err < 0)
1181                 return(err);
1182
1183         err = os_read_file(fd, buf, len);
1184         if(err < 0)
1185                 return(err);
1186
1187         return(0);
1188 }
1189
1190 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1191                   char **backing_file_out, int *bitmap_offset_out,
1192                   unsigned long *bitmap_len_out, int *data_offset_out,
1193                   int *create_cow_out)
1194 {
1195         time_t mtime;
1196         unsigned long long size;
1197         __u32 version, align;
1198         char *backing_file;
1199         int fd, err, sectorsize, asked_switch, mode = 0644;
1200
1201         fd = os_open_file(file, *openflags, mode);
1202         if (fd < 0) {
1203                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1204                         *create_cow_out = 1;
1205                 if (!openflags->w ||
1206                    ((fd != -EROFS) && (fd != -EACCES)))
1207                         return fd;
1208                 openflags->w = 0;
1209                 fd = os_open_file(file, *openflags, mode);
1210                 if (fd < 0)
1211                         return fd;
1212         }
1213
1214         if(shared)
1215                 printk("Not locking \"%s\" on the host\n", file);
1216         else {
1217                 err = os_lock_file(fd, openflags->w);
1218                 if(err < 0){
1219                         printk("Failed to lock '%s', err = %d\n", file, -err);
1220                         goto out_close;
1221                 }
1222         }
1223
1224         /* Successful return case! */
1225         if(backing_file_out == NULL)
1226                 return(fd);
1227
1228         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1229                               &size, &sectorsize, &align, bitmap_offset_out);
1230         if(err && (*backing_file_out != NULL)){
1231                 printk("Failed to read COW header from COW file \"%s\", "
1232                        "errno = %d\n", file, -err);
1233                 goto out_close;
1234         }
1235         if(err)
1236                 return(fd);
1237
1238         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1239
1240         /* Allow switching only if no mismatch. */
1241         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1242                 printk("Switching backing file to '%s'\n", *backing_file_out);
1243                 err = write_cow_header(file, fd, *backing_file_out,
1244                                        sectorsize, align, &size);
1245                 if (err) {
1246                         printk("Switch failed, errno = %d\n", -err);
1247                         goto out_close;
1248                 }
1249         } else {
1250                 *backing_file_out = backing_file;
1251                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1252                 if (err)
1253                         goto out_close;
1254         }
1255
1256         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1257                   bitmap_len_out, data_offset_out);
1258
1259         return fd;
1260  out_close:
1261         os_close_file(fd);
1262         return err;
1263 }
1264
1265 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1266                     int sectorsize, int alignment, int *bitmap_offset_out,
1267                     unsigned long *bitmap_len_out, int *data_offset_out)
1268 {
1269         int err, fd;
1270
1271         flags.c = 1;
1272         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1273         if(fd < 0){
1274                 err = fd;
1275                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1276                        -err);
1277                 goto out;
1278         }
1279
1280         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1281                             bitmap_offset_out, bitmap_len_out,
1282                             data_offset_out);
1283         if(!err)
1284                 return(fd);
1285         os_close_file(fd);
1286  out:
1287         return(err);
1288 }
1289
1290 static int update_bitmap(struct io_thread_req *req)
1291 {
1292         int n;
1293
1294         if(req->cow_offset == -1)
1295                 return(0);
1296
1297         n = os_seek_file(req->fds[1], req->cow_offset);
1298         if(n < 0){
1299                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1300                 return(1);
1301         }
1302
1303         n = os_write_file(req->fds[1], &req->bitmap_words,
1304                           sizeof(req->bitmap_words));
1305         if(n != sizeof(req->bitmap_words)){
1306                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1307                        req->fds[1]);
1308                 return(1);
1309         }
1310
1311         return(0);
1312 }
1313
1314 void do_io(struct io_thread_req *req)
1315 {
1316         char *buf;
1317         unsigned long len;
1318         int n, nsectors, start, end, bit;
1319         int err;
1320         __u64 off;
1321
1322         nsectors = req->length / req->sectorsize;
1323         start = 0;
1324         do {
1325                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1326                 end = start;
1327                 while((end < nsectors) &&
1328                       (ubd_test_bit(end, (unsigned char *)
1329                                     &req->sector_mask) == bit))
1330                         end++;
1331
1332                 off = req->offset + req->offsets[bit] +
1333                         start * req->sectorsize;
1334                 len = (end - start) * req->sectorsize;
1335                 buf = &req->buffer[start * req->sectorsize];
1336
1337                 err = os_seek_file(req->fds[bit], off);
1338                 if(err < 0){
1339                         printk("do_io - lseek failed : err = %d\n", -err);
1340                         req->error = 1;
1341                         return;
1342                 }
1343                 if(req->op == UBD_READ){
1344                         n = 0;
1345                         do {
1346                                 buf = &buf[n];
1347                                 len -= n;
1348                                 n = os_read_file(req->fds[bit], buf, len);
1349                                 if (n < 0) {
1350                                         printk("do_io - read failed, err = %d "
1351                                                "fd = %d\n", -n, req->fds[bit]);
1352                                         req->error = 1;
1353                                         return;
1354                                 }
1355                         } while((n < len) && (n != 0));
1356                         if (n < len) memset(&buf[n], 0, len - n);
1357                 } else {
1358                         n = os_write_file(req->fds[bit], buf, len);
1359                         if(n != len){
1360                                 printk("do_io - write failed err = %d "
1361                                        "fd = %d\n", -n, req->fds[bit]);
1362                                 req->error = 1;
1363                                 return;
1364                         }
1365                 }
1366
1367                 start = end;
1368         } while(start < nsectors);
1369
1370         req->error = update_bitmap(req);
1371 }
1372
1373 /* Changed in start_io_thread, which is serialized by being called only
1374  * from ubd_init, which is an initcall.
1375  */
1376 int kernel_fd = -1;
1377
1378 /* Only changed by the io thread */
1379 int io_count = 0;
1380
1381 int io_thread(void *arg)
1382 {
1383         struct io_thread_req req;
1384         int n;
1385
1386         ignore_sigwinch_sig();
1387         while(1){
1388                 n = os_read_file(kernel_fd, &req, sizeof(req));
1389                 if(n != sizeof(req)){
1390                         if(n < 0)
1391                                 printk("io_thread - read failed, fd = %d, "
1392                                        "err = %d\n", kernel_fd, -n);
1393                         else {
1394                                 printk("io_thread - short read, fd = %d, "
1395                                        "length = %d\n", kernel_fd, n);
1396                         }
1397                         continue;
1398                 }
1399                 io_count++;
1400                 do_io(&req);
1401                 n = os_write_file(kernel_fd, &req, sizeof(req));
1402                 if(n != sizeof(req))
1403                         printk("io_thread - write failed, fd = %d, err = %d\n",
1404                                kernel_fd, -n);
1405         }
1406
1407         return 0;
1408 }