Pull dock into test branch
[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 /* Only changed by fake_ide_setup which is a setup */
206 static int fake_ide = 0;
207 static struct proc_dir_entry *proc_ide_root = NULL;
208 static struct proc_dir_entry *proc_ide = NULL;
209
210 static void make_proc_ide(void)
211 {
212         proc_ide_root = proc_mkdir("ide", NULL);
213         proc_ide = proc_mkdir("ide0", proc_ide_root);
214 }
215
216 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
217                                int *eof, void *data)
218 {
219         int len;
220
221         strcpy(page, "disk\n");
222         len = strlen("disk\n");
223         len -= off;
224         if (len < count){
225                 *eof = 1;
226                 if (len <= 0) return 0;
227         }
228         else len = count;
229         *start = page + off;
230         return len;
231 }
232
233 static void make_ide_entries(char *dev_name)
234 {
235         struct proc_dir_entry *dir, *ent;
236         char name[64];
237
238         if(proc_ide_root == NULL) make_proc_ide();
239
240         dir = proc_mkdir(dev_name, proc_ide);
241         if(!dir) return;
242
243         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
244         if(!ent) return;
245         ent->nlink = 1;
246         ent->data = NULL;
247         ent->read_proc = proc_ide_read_media;
248         ent->write_proc = NULL;
249         sprintf(name,"ide0/%s", dev_name);
250         proc_symlink(dev_name, proc_ide_root, name);
251 }
252
253 static int fake_ide_setup(char *str)
254 {
255         fake_ide = 1;
256         return(1);
257 }
258
259 __setup("fake_ide", fake_ide_setup);
260
261 __uml_help(fake_ide_setup,
262 "fake_ide\n"
263 "    Create ide0 entries that map onto ubd devices.\n\n"
264 );
265
266 static int parse_unit(char **ptr)
267 {
268         char *str = *ptr, *end;
269         int n = -1;
270
271         if(isdigit(*str)) {
272                 n = simple_strtoul(str, &end, 0);
273                 if(end == str)
274                         return(-1);
275                 *ptr = end;
276         }
277         else if (('a' <= *str) && (*str <= 'z')) {
278                 n = *str - 'a';
279                 str++;
280                 *ptr = str;
281         }
282         return(n);
283 }
284
285 /* If *index_out == -1 at exit, the passed option was a general one;
286  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
287  * should not be freed on exit.
288  */
289 static int ubd_setup_common(char *str, int *index_out)
290 {
291         struct ubd *ubd_dev;
292         struct openflags flags = global_openflags;
293         char *backing_file;
294         int n, err, i;
295
296         if(index_out) *index_out = -1;
297         n = *str;
298         if(n == '='){
299                 char *end;
300                 int major;
301
302                 str++;
303                 if(!strcmp(str, "sync")){
304                         global_openflags = of_sync(global_openflags);
305                         return(0);
306                 }
307                 major = simple_strtoul(str, &end, 0);
308                 if((*end != '\0') || (end == str)){
309                         printk(KERN_ERR
310                                "ubd_setup : didn't parse major number\n");
311                         return(1);
312                 }
313
314                 err = 1;
315                 mutex_lock(&ubd_lock);
316                 if(fake_major != MAJOR_NR){
317                         printk(KERN_ERR "Can't assign a fake major twice\n");
318                         goto out1;
319                 }
320
321                 fake_major = major;
322
323                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
324                        major);
325                 err = 0;
326         out1:
327                 mutex_unlock(&ubd_lock);
328                 return(err);
329         }
330
331         n = parse_unit(&str);
332         if(n < 0){
333                 printk(KERN_ERR "ubd_setup : couldn't parse unit number "
334                        "'%s'\n", str);
335                 return(1);
336         }
337         if(n >= MAX_DEV){
338                 printk(KERN_ERR "ubd_setup : index %d out of range "
339                        "(%d devices, from 0 to %d)\n", n, MAX_DEV, MAX_DEV - 1);
340                 return(1);
341         }
342
343         err = 1;
344         mutex_lock(&ubd_lock);
345
346         ubd_dev = &ubd_devs[n];
347         if(ubd_dev->file != NULL){
348                 printk(KERN_ERR "ubd_setup : device already configured\n");
349                 goto out;
350         }
351
352         if (index_out)
353                 *index_out = n;
354
355         for (i = 0; i < sizeof("rscd="); i++) {
356                 switch (*str) {
357                 case 'r':
358                         flags.w = 0;
359                         break;
360                 case 's':
361                         flags.s = 1;
362                         break;
363                 case 'd':
364                         ubd_dev->no_cow = 1;
365                         break;
366                 case 'c':
367                         ubd_dev->shared = 1;
368                         break;
369                 case '=':
370                         str++;
371                         goto break_loop;
372                 default:
373                         printk(KERN_ERR "ubd_setup : Expected '=' or flag letter (r, s, c, or d)\n");
374                         goto out;
375                 }
376                 str++;
377         }
378
379         if (*str == '=')
380                 printk(KERN_ERR "ubd_setup : Too many flags specified\n");
381         else
382                 printk(KERN_ERR "ubd_setup : Expected '='\n");
383         goto out;
384
385 break_loop:
386         err = 0;
387         backing_file = strchr(str, ',');
388
389         if (!backing_file) {
390                 backing_file = strchr(str, ':');
391         }
392
393         if(backing_file){
394                 if(ubd_dev->no_cow)
395                         printk(KERN_ERR "Can't specify both 'd' and a "
396                                "cow file\n");
397                 else {
398                         *backing_file = '\0';
399                         backing_file++;
400                 }
401         }
402         ubd_dev->file = str;
403         ubd_dev->cow.file = backing_file;
404         ubd_dev->boot_openflags = flags;
405 out:
406         mutex_unlock(&ubd_lock);
407         return(err);
408 }
409
410 static int ubd_setup(char *str)
411 {
412         ubd_setup_common(str, NULL);
413         return(1);
414 }
415
416 __setup("ubd", ubd_setup);
417 __uml_help(ubd_setup,
418 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
419 "    This is used to associate a device with a file in the underlying\n"
420 "    filesystem. When specifying two filenames, the first one is the\n"
421 "    COW name and the second is the backing file name. As separator you can\n"
422 "    use either a ':' or a ',': the first one allows writing things like;\n"
423 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
424 "    while with a ',' the shell would not expand the 2nd '~'.\n"
425 "    When using only one filename, UML will detect whether to thread it like\n"
426 "    a COW file or a backing file. To override this detection, add the 'd'\n"
427 "    flag:\n"
428 "       ubd0d=BackingFile\n"
429 "    Usually, there is a filesystem in the file, but \n"
430 "    that's not required. Swap devices containing swap files can be\n"
431 "    specified like this. Also, a file which doesn't contain a\n"
432 "    filesystem can have its contents read in the virtual \n"
433 "    machine by running 'dd' on the device. <n> must be in the range\n"
434 "    0 to 7. Appending an 'r' to the number will cause that device\n"
435 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
436 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
437 );
438
439 static int udb_setup(char *str)
440 {
441         printk("udb%s specified on command line is almost certainly a ubd -> "
442                "udb TYPO\n", str);
443         return(1);
444 }
445
446 __setup("udb", udb_setup);
447 __uml_help(udb_setup,
448 "udb\n"
449 "    This option is here solely to catch ubd -> udb typos, which can be\n"
450 "    to impossible to catch visually unless you specifically look for\n"
451 "    them.  The only result of any option starting with 'udb' is an error\n"
452 "    in the boot output.\n\n"
453 );
454
455 static int fakehd_set = 0;
456 static int fakehd(char *str)
457 {
458         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
459         fakehd_set = 1;
460         return 1;
461 }
462
463 __setup("fakehd", fakehd);
464 __uml_help(fakehd,
465 "fakehd\n"
466 "    Change the ubd device name to \"hd\".\n\n"
467 );
468
469 static void do_ubd_request(request_queue_t * q);
470
471 /* Only changed by ubd_init, which is an initcall. */
472 int thread_fd = -1;
473
474 /* Changed by ubd_handler, which is serialized because interrupts only
475  * happen on CPU 0.
476  * XXX: currently unused.
477  */
478 static 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 /* Callable only from interrupt context - otherwise you need to do
499  * spin_lock_irq()/spin_lock_irqsave() */
500 static inline void ubd_finish(struct request *req, int error)
501 {
502         spin_lock(&ubd_io_lock);
503         __ubd_finish(req, error);
504         spin_unlock(&ubd_io_lock);
505 }
506
507 /* XXX - move this inside ubd_intr. */
508 /* Called without ubd_io_lock held, and only in interrupt context. */
509 static void ubd_handler(void)
510 {
511         struct io_thread_req req;
512         struct request *rq = elv_next_request(ubd_queue);
513         int n;
514
515         do_ubd = 0;
516         intr_count++;
517         n = os_read_file(thread_fd, &req, sizeof(req));
518         if(n != sizeof(req)){
519                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
520                        "err = %d\n", os_getpid(), -n);
521                 spin_lock(&ubd_io_lock);
522                 end_request(rq, 0);
523                 spin_unlock(&ubd_io_lock);
524                 return;
525         }
526
527         ubd_finish(rq, req.error);
528         reactivate_fd(thread_fd, UBD_IRQ);      
529         spin_lock(&ubd_io_lock);
530         do_ubd_request(ubd_queue);
531         spin_unlock(&ubd_io_lock);
532 }
533
534 static irqreturn_t ubd_intr(int irq, void *dev)
535 {
536         ubd_handler();
537         return(IRQ_HANDLED);
538 }
539
540 /* Only changed by ubd_init, which is an initcall. */
541 static int io_pid = -1;
542
543 void kill_io_thread(void)
544 {
545         if(io_pid != -1)
546                 os_kill_process(io_pid, 1);
547 }
548
549 __uml_exitcall(kill_io_thread);
550
551 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
552 {
553         char *file;
554
555         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
556         return(os_file_size(file, size_out));
557 }
558
559 static void ubd_close_dev(struct ubd *ubd_dev)
560 {
561         os_close_file(ubd_dev->fd);
562         if(ubd_dev->cow.file == NULL)
563                 return;
564
565         os_close_file(ubd_dev->cow.fd);
566         vfree(ubd_dev->cow.bitmap);
567         ubd_dev->cow.bitmap = NULL;
568 }
569
570 static int ubd_open_dev(struct ubd *ubd_dev)
571 {
572         struct openflags flags;
573         char **back_ptr;
574         int err, create_cow, *create_ptr;
575         int fd;
576
577         ubd_dev->openflags = ubd_dev->boot_openflags;
578         create_cow = 0;
579         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
580         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
581
582         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
583                                 back_ptr, &ubd_dev->cow.bitmap_offset,
584                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
585                                 create_ptr);
586
587         if((fd == -ENOENT) && create_cow){
588                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
589                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
590                                           &ubd_dev->cow.bitmap_offset,
591                                           &ubd_dev->cow.bitmap_len,
592                                           &ubd_dev->cow.data_offset);
593                 if(fd >= 0){
594                         printk(KERN_INFO "Creating \"%s\" as COW file for "
595                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
596                 }
597         }
598
599         if(fd < 0){
600                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
601                        -fd);
602                 return fd;
603         }
604         ubd_dev->fd = fd;
605
606         if(ubd_dev->cow.file != NULL){
607                 err = -ENOMEM;
608                 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
609                 if(ubd_dev->cow.bitmap == NULL){
610                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
611                         goto error;
612                 }
613                 flush_tlb_kernel_vm();
614
615                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
616                                       ubd_dev->cow.bitmap_offset,
617                                       ubd_dev->cow.bitmap_len);
618                 if(err < 0)
619                         goto error;
620
621                 flags = ubd_dev->openflags;
622                 flags.w = 0;
623                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
624                                     NULL, NULL, NULL, NULL);
625                 if(err < 0) goto error;
626                 ubd_dev->cow.fd = err;
627         }
628         return(0);
629  error:
630         os_close_file(ubd_dev->fd);
631         return(err);
632 }
633
634 static int ubd_disk_register(int major, u64 size, int unit,
635                         struct gendisk **disk_out)
636                         
637 {
638         struct gendisk *disk;
639
640         disk = alloc_disk(1 << UBD_SHIFT);
641         if(disk == NULL)
642                 return(-ENOMEM);
643
644         disk->major = major;
645         disk->first_minor = unit << UBD_SHIFT;
646         disk->fops = &ubd_blops;
647         set_capacity(disk, size / 512);
648         if(major == MAJOR_NR)
649                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
650         else
651                 sprintf(disk->disk_name, "ubd_fake%d", unit);
652
653         /* sysfs register (not for ide fake devices) */
654         if (major == MAJOR_NR) {
655                 ubd_devs[unit].pdev.id   = unit;
656                 ubd_devs[unit].pdev.name = DRIVER_NAME;
657                 platform_device_register(&ubd_devs[unit].pdev);
658                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
659         }
660
661         disk->private_data = &ubd_devs[unit];
662         disk->queue = ubd_queue;
663         add_disk(disk);
664
665         *disk_out = disk;
666         return 0;
667 }
668
669 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
670
671 static int ubd_add(int n)
672 {
673         struct ubd *ubd_dev = &ubd_devs[n];
674         int err;
675
676         err = -ENODEV;
677         if(ubd_dev->file == NULL)
678                 goto out;
679
680         err = ubd_file_size(ubd_dev, &ubd_dev->size);
681         if(err < 0)
682                 goto out;
683
684         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
685
686         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
687         if(err)
688                 goto out;
689
690         if(fake_major != MAJOR_NR)
691                 ubd_disk_register(fake_major, ubd_dev->size, n,
692                              &fake_gendisk[n]);
693
694         /* perhaps this should also be under the "if (fake_major)" above */
695         /* using the fake_disk->disk_name and also the fakehd_set name */
696         if (fake_ide)
697                 make_ide_entries(ubd_gendisk[n]->disk_name);
698
699         err = 0;
700 out:
701         return err;
702 }
703
704 static int ubd_config(char *str)
705 {
706         int n, ret;
707
708         str = kstrdup(str, GFP_KERNEL);
709         if (str == NULL) {
710                 printk(KERN_ERR "ubd_config failed to strdup string\n");
711                 ret = 1;
712                 goto out;
713         }
714         ret = ubd_setup_common(str, &n);
715         if (ret) {
716                 ret = -1;
717                 goto err_free;
718         }
719         if (n == -1) {
720                 ret = 0;
721                 goto err_free;
722         }
723
724         mutex_lock(&ubd_lock);
725         ret = ubd_add(n);
726         if (ret)
727                 ubd_devs[n].file = NULL;
728         mutex_unlock(&ubd_lock);
729
730 out:
731         return ret;
732
733 err_free:
734         kfree(str);
735         goto out;
736 }
737
738 static int ubd_get_config(char *name, char *str, int size, char **error_out)
739 {
740         struct ubd *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         ubd_dev = &ubd_devs[n];
750         mutex_lock(&ubd_lock);
751
752         if(ubd_dev->file == NULL){
753                 CONFIG_CHUNK(str, size, len, "", 1);
754                 goto out;
755         }
756
757         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
758
759         if(ubd_dev->cow.file != NULL){
760                 CONFIG_CHUNK(str, size, len, ",", 0);
761                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
762         }
763         else CONFIG_CHUNK(str, size, len, "", 1);
764
765  out:
766         mutex_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 *ubd_dev;
783         int err = -ENODEV;
784
785         mutex_lock(&ubd_lock);
786
787         if(ubd_gendisk[n] == NULL)
788                 goto out;
789
790         ubd_dev = &ubd_devs[n];
791
792         if(ubd_dev->file == NULL)
793                 goto out;
794
795         /* you cannot remove a open disk */
796         err = -EBUSY;
797         if(ubd_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(&ubd_dev->pdev);
811         *ubd_dev = ((struct ubd) DEFAULT_UBD);
812         err = 0;
813 out:
814         mutex_unlock(&ubd_lock);
815         return err;
816 }
817
818 /* All these are called by mconsole in process context and without ubd-specific locks. */
819 static struct mc_device ubd_mc = {
820         .name           = "ubd",
821         .config         = ubd_config,
822         .get_config     = ubd_get_config,
823         .id             = ubd_id,
824         .remove         = ubd_remove,
825 };
826
827 static int __init ubd_mc_init(void)
828 {
829         mconsole_register_dev(&ubd_mc);
830         return 0;
831 }
832
833 __initcall(ubd_mc_init);
834
835 static int __init ubd0_init(void)
836 {
837         struct ubd *ubd_dev = &ubd_devs[0];
838
839         if(ubd_dev->file == NULL)
840                 ubd_dev->file = "root_fs";
841         return(0);
842 }
843
844 __initcall(ubd0_init);
845
846 static struct platform_driver ubd_driver = {
847         .driver = {
848                 .name  = DRIVER_NAME,
849         },
850 };
851
852 static int __init ubd_init(void)
853 {
854         int i;
855
856         if (register_blkdev(MAJOR_NR, "ubd"))
857                 return -1;
858
859         ubd_queue = blk_init_queue(do_ubd_request, &ubd_io_lock);
860         if (!ubd_queue) {
861                 unregister_blkdev(MAJOR_NR, "ubd");
862                 return -1;
863         }
864                 
865         if (fake_major != MAJOR_NR) {
866                 char name[sizeof("ubd_nnn\0")];
867
868                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
869                 if (register_blkdev(fake_major, "ubd"))
870                         return -1;
871         }
872         platform_driver_register(&ubd_driver);
873         for (i = 0; i < MAX_DEV; i++)
874                 ubd_add(i);
875         return 0;
876 }
877
878 late_initcall(ubd_init);
879
880 static int __init ubd_driver_init(void){
881         unsigned long stack;
882         int err;
883
884         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
885         if(global_openflags.s){
886                 printk(KERN_INFO "ubd: Synchronous mode\n");
887                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
888                  * enough. So use anyway the io thread. */
889         }
890         stack = alloc_stack(0, 0);
891         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
892                                  &thread_fd);
893         if(io_pid < 0){
894                 printk(KERN_ERR
895                        "ubd : Failed to start I/O thread (errno = %d) - "
896                        "falling back to synchronous I/O\n", -io_pid);
897                 io_pid = -1;
898                 return(0);
899         }
900         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
901                              IRQF_DISABLED, "ubd", ubd_devs);
902         if(err != 0)
903                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
904         return 0;
905 }
906
907 device_initcall(ubd_driver_init);
908
909 static int ubd_open(struct inode *inode, struct file *filp)
910 {
911         struct gendisk *disk = inode->i_bdev->bd_disk;
912         struct ubd *ubd_dev = disk->private_data;
913         int err = 0;
914
915         if(ubd_dev->count == 0){
916                 err = ubd_open_dev(ubd_dev);
917                 if(err){
918                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
919                                disk->disk_name, ubd_dev->file, -err);
920                         goto out;
921                 }
922         }
923         ubd_dev->count++;
924         set_disk_ro(disk, !ubd_dev->openflags.w);
925
926         /* This should no more be needed. And it didn't work anyway to exclude
927          * read-write remounting of filesystems.*/
928         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
929                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
930                 err = -EROFS;
931         }*/
932  out:
933         return(err);
934 }
935
936 static int ubd_release(struct inode * inode, struct file * file)
937 {
938         struct gendisk *disk = inode->i_bdev->bd_disk;
939         struct ubd *ubd_dev = disk->private_data;
940
941         if(--ubd_dev->count == 0)
942                 ubd_close_dev(ubd_dev);
943         return(0);
944 }
945
946 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
947                           __u64 *cow_offset, unsigned long *bitmap,
948                           __u64 bitmap_offset, unsigned long *bitmap_words,
949                           __u64 bitmap_len)
950 {
951         __u64 sector = io_offset >> 9;
952         int i, update_bitmap = 0;
953
954         for(i = 0; i < length >> 9; i++){
955                 if(cow_mask != NULL)
956                         ubd_set_bit(i, (unsigned char *) cow_mask);
957                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
958                         continue;
959
960                 update_bitmap = 1;
961                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
962         }
963
964         if(!update_bitmap)
965                 return;
966
967         *cow_offset = sector / (sizeof(unsigned long) * 8);
968
969         /* This takes care of the case where we're exactly at the end of the
970          * device, and *cow_offset + 1 is off the end.  So, just back it up
971          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
972          * for the original diagnosis.
973          */
974         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
975                            sizeof(unsigned long) - 1))
976                 (*cow_offset)--;
977
978         bitmap_words[0] = bitmap[*cow_offset];
979         bitmap_words[1] = bitmap[*cow_offset + 1];
980
981         *cow_offset *= sizeof(unsigned long);
982         *cow_offset += bitmap_offset;
983 }
984
985 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
986                        __u64 bitmap_offset, __u64 bitmap_len)
987 {
988         __u64 sector = req->offset >> 9;
989         int i;
990
991         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
992                 panic("Operation too long");
993
994         if(req->op == UBD_READ) {
995                 for(i = 0; i < req->length >> 9; i++){
996                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
997                                 ubd_set_bit(i, (unsigned char *)
998                                             &req->sector_mask);
999                 }
1000         }
1001         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1002                            &req->cow_offset, bitmap, bitmap_offset,
1003                            req->bitmap_words, bitmap_len);
1004 }
1005
1006 /* Called with ubd_io_lock held */
1007 static int prepare_request(struct request *req, struct io_thread_req *io_req)
1008 {
1009         struct gendisk *disk = req->rq_disk;
1010         struct ubd *ubd_dev = disk->private_data;
1011         __u64 offset;
1012         int len;
1013
1014         /* This should be impossible now */
1015         if((rq_data_dir(req) == WRITE) && !ubd_dev->openflags.w){
1016                 printk("Write attempted on readonly ubd device %s\n",
1017                        disk->disk_name);
1018                 end_request(req, 0);
1019                 return(1);
1020         }
1021
1022         offset = ((__u64) req->sector) << 9;
1023         len = req->current_nr_sectors << 9;
1024
1025         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd : ubd_dev->fd;
1026         io_req->fds[1] = ubd_dev->fd;
1027         io_req->cow_offset = -1;
1028         io_req->offset = offset;
1029         io_req->length = len;
1030         io_req->error = 0;
1031         io_req->sector_mask = 0;
1032
1033         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1034         io_req->offsets[0] = 0;
1035         io_req->offsets[1] = ubd_dev->cow.data_offset;
1036         io_req->buffer = req->buffer;
1037         io_req->sectorsize = 1 << 9;
1038
1039         if(ubd_dev->cow.file != NULL)
1040                 cowify_req(io_req, ubd_dev->cow.bitmap, ubd_dev->cow.bitmap_offset,
1041                            ubd_dev->cow.bitmap_len);
1042
1043         return(0);
1044 }
1045
1046 /* Called with ubd_io_lock held */
1047 static void do_ubd_request(request_queue_t *q)
1048 {
1049         struct io_thread_req io_req;
1050         struct request *req;
1051         int err, n;
1052
1053         if(thread_fd == -1){
1054                 while((req = elv_next_request(q)) != NULL){
1055                         err = prepare_request(req, &io_req);
1056                         if(!err){
1057                                 do_io(&io_req);
1058                                 __ubd_finish(req, io_req.error);
1059                         }
1060                 }
1061         }
1062         else {
1063                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1064                         return;
1065                 err = prepare_request(req, &io_req);
1066                 if(!err){
1067                         do_ubd = 1;
1068                         n = os_write_file(thread_fd, (char *) &io_req,
1069                                          sizeof(io_req));
1070                         if(n != sizeof(io_req))
1071                                 printk("write to io thread failed, "
1072                                        "errno = %d\n", -n);
1073                 }
1074         }
1075 }
1076
1077 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1078 {
1079         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1080
1081         geo->heads = 128;
1082         geo->sectors = 32;
1083         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1084         return 0;
1085 }
1086
1087 static int ubd_ioctl(struct inode * inode, struct file * file,
1088                      unsigned int cmd, unsigned long arg)
1089 {
1090         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1091         struct hd_driveid ubd_id = {
1092                 .cyls           = 0,
1093                 .heads          = 128,
1094                 .sectors        = 32,
1095         };
1096
1097         switch (cmd) {
1098                 struct cdrom_volctrl volume;
1099         case HDIO_GET_IDENTITY:
1100                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1101                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1102                                  sizeof(ubd_id)))
1103                         return(-EFAULT);
1104                 return(0);
1105                 
1106         case CDROMVOLREAD:
1107                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1108                         return(-EFAULT);
1109                 volume.channel0 = 255;
1110                 volume.channel1 = 255;
1111                 volume.channel2 = 255;
1112                 volume.channel3 = 255;
1113                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1114                         return(-EFAULT);
1115                 return(0);
1116         }
1117         return(-EINVAL);
1118 }
1119
1120 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1121 {
1122         struct uml_stat buf1, buf2;
1123         int err;
1124
1125         if(from_cmdline == NULL)
1126                 return 0;
1127         if(!strcmp(from_cmdline, from_cow))
1128                 return 0;
1129
1130         err = os_stat_file(from_cmdline, &buf1);
1131         if(err < 0){
1132                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1133                 return 0;
1134         }
1135         err = os_stat_file(from_cow, &buf2);
1136         if(err < 0){
1137                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1138                 return 1;
1139         }
1140         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1141                 return 0;
1142
1143         printk("Backing file mismatch - \"%s\" requested,\n"
1144                "\"%s\" specified in COW header of \"%s\"\n",
1145                from_cmdline, from_cow, cow);
1146         return 1;
1147 }
1148
1149 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1150 {
1151         unsigned long modtime;
1152         unsigned long long actual;
1153         int err;
1154
1155         err = os_file_modtime(file, &modtime);
1156         if(err < 0){
1157                 printk("Failed to get modification time of backing file "
1158                        "\"%s\", err = %d\n", file, -err);
1159                 return(err);
1160         }
1161
1162         err = os_file_size(file, &actual);
1163         if(err < 0){
1164                 printk("Failed to get size of backing file \"%s\", "
1165                        "err = %d\n", file, -err);
1166                 return(err);
1167         }
1168
1169         if(actual != size){
1170                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1171                  * the typecast.*/
1172                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1173                        "file\n", (unsigned long long) size, actual);
1174                 return(-EINVAL);
1175         }
1176         if(modtime != mtime){
1177                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1178                        "file\n", mtime, modtime);
1179                 return(-EINVAL);
1180         }
1181         return(0);
1182 }
1183
1184 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1185 {
1186         int err;
1187
1188         err = os_seek_file(fd, offset);
1189         if(err < 0)
1190                 return(err);
1191
1192         err = os_read_file(fd, buf, len);
1193         if(err < 0)
1194                 return(err);
1195
1196         return(0);
1197 }
1198
1199 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1200                   char **backing_file_out, int *bitmap_offset_out,
1201                   unsigned long *bitmap_len_out, int *data_offset_out,
1202                   int *create_cow_out)
1203 {
1204         time_t mtime;
1205         unsigned long long size;
1206         __u32 version, align;
1207         char *backing_file;
1208         int fd, err, sectorsize, asked_switch, mode = 0644;
1209
1210         fd = os_open_file(file, *openflags, mode);
1211         if (fd < 0) {
1212                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1213                         *create_cow_out = 1;
1214                 if (!openflags->w ||
1215                    ((fd != -EROFS) && (fd != -EACCES)))
1216                         return fd;
1217                 openflags->w = 0;
1218                 fd = os_open_file(file, *openflags, mode);
1219                 if (fd < 0)
1220                         return fd;
1221         }
1222
1223         if(shared)
1224                 printk("Not locking \"%s\" on the host\n", file);
1225         else {
1226                 err = os_lock_file(fd, openflags->w);
1227                 if(err < 0){
1228                         printk("Failed to lock '%s', err = %d\n", file, -err);
1229                         goto out_close;
1230                 }
1231         }
1232
1233         /* Successful return case! */
1234         if(backing_file_out == NULL)
1235                 return(fd);
1236
1237         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1238                               &size, &sectorsize, &align, bitmap_offset_out);
1239         if(err && (*backing_file_out != NULL)){
1240                 printk("Failed to read COW header from COW file \"%s\", "
1241                        "errno = %d\n", file, -err);
1242                 goto out_close;
1243         }
1244         if(err)
1245                 return(fd);
1246
1247         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1248
1249         /* Allow switching only if no mismatch. */
1250         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1251                 printk("Switching backing file to '%s'\n", *backing_file_out);
1252                 err = write_cow_header(file, fd, *backing_file_out,
1253                                        sectorsize, align, &size);
1254                 if (err) {
1255                         printk("Switch failed, errno = %d\n", -err);
1256                         goto out_close;
1257                 }
1258         } else {
1259                 *backing_file_out = backing_file;
1260                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1261                 if (err)
1262                         goto out_close;
1263         }
1264
1265         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1266                   bitmap_len_out, data_offset_out);
1267
1268         return fd;
1269  out_close:
1270         os_close_file(fd);
1271         return err;
1272 }
1273
1274 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1275                     int sectorsize, int alignment, int *bitmap_offset_out,
1276                     unsigned long *bitmap_len_out, int *data_offset_out)
1277 {
1278         int err, fd;
1279
1280         flags.c = 1;
1281         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1282         if(fd < 0){
1283                 err = fd;
1284                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1285                        -err);
1286                 goto out;
1287         }
1288
1289         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1290                             bitmap_offset_out, bitmap_len_out,
1291                             data_offset_out);
1292         if(!err)
1293                 return(fd);
1294         os_close_file(fd);
1295  out:
1296         return(err);
1297 }
1298
1299 static int update_bitmap(struct io_thread_req *req)
1300 {
1301         int n;
1302
1303         if(req->cow_offset == -1)
1304                 return(0);
1305
1306         n = os_seek_file(req->fds[1], req->cow_offset);
1307         if(n < 0){
1308                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1309                 return(1);
1310         }
1311
1312         n = os_write_file(req->fds[1], &req->bitmap_words,
1313                           sizeof(req->bitmap_words));
1314         if(n != sizeof(req->bitmap_words)){
1315                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1316                        req->fds[1]);
1317                 return(1);
1318         }
1319
1320         return(0);
1321 }
1322
1323 void do_io(struct io_thread_req *req)
1324 {
1325         char *buf;
1326         unsigned long len;
1327         int n, nsectors, start, end, bit;
1328         int err;
1329         __u64 off;
1330
1331         nsectors = req->length / req->sectorsize;
1332         start = 0;
1333         do {
1334                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1335                 end = start;
1336                 while((end < nsectors) &&
1337                       (ubd_test_bit(end, (unsigned char *)
1338                                     &req->sector_mask) == bit))
1339                         end++;
1340
1341                 off = req->offset + req->offsets[bit] +
1342                         start * req->sectorsize;
1343                 len = (end - start) * req->sectorsize;
1344                 buf = &req->buffer[start * req->sectorsize];
1345
1346                 err = os_seek_file(req->fds[bit], off);
1347                 if(err < 0){
1348                         printk("do_io - lseek failed : err = %d\n", -err);
1349                         req->error = 1;
1350                         return;
1351                 }
1352                 if(req->op == UBD_READ){
1353                         n = 0;
1354                         do {
1355                                 buf = &buf[n];
1356                                 len -= n;
1357                                 n = os_read_file(req->fds[bit], buf, len);
1358                                 if (n < 0) {
1359                                         printk("do_io - read failed, err = %d "
1360                                                "fd = %d\n", -n, req->fds[bit]);
1361                                         req->error = 1;
1362                                         return;
1363                                 }
1364                         } while((n < len) && (n != 0));
1365                         if (n < len) memset(&buf[n], 0, len - n);
1366                 } else {
1367                         n = os_write_file(req->fds[bit], buf, len);
1368                         if(n != len){
1369                                 printk("do_io - write failed err = %d "
1370                                        "fd = %d\n", -n, req->fds[bit]);
1371                                 req->error = 1;
1372                                 return;
1373                         }
1374                 }
1375
1376                 start = end;
1377         } while(start < nsectors);
1378
1379         req->error = update_bitmap(req);
1380 }
1381
1382 /* Changed in start_io_thread, which is serialized by being called only
1383  * from ubd_init, which is an initcall.
1384  */
1385 int kernel_fd = -1;
1386
1387 /* Only changed by the io thread. XXX: currently unused. */
1388 static int io_count = 0;
1389
1390 int io_thread(void *arg)
1391 {
1392         struct io_thread_req req;
1393         int n;
1394
1395         ignore_sigwinch_sig();
1396         while(1){
1397                 n = os_read_file(kernel_fd, &req, sizeof(req));
1398                 if(n != sizeof(req)){
1399                         if(n < 0)
1400                                 printk("io_thread - read failed, fd = %d, "
1401                                        "err = %d\n", kernel_fd, -n);
1402                         else {
1403                                 printk("io_thread - short read, fd = %d, "
1404                                        "length = %d\n", kernel_fd, n);
1405                         }
1406                         continue;
1407                 }
1408                 io_count++;
1409                 do_io(&req);
1410                 n = os_write_file(kernel_fd, &req, sizeof(req));
1411                 if(n != sizeof(req))
1412                         printk("io_thread - write failed, fd = %d, err = %d\n",
1413                                kernel_fd, -n);
1414         }
1415
1416         return 0;
1417 }