Merge branch 'linus' into sched/core
[linux-3.10.git] / arch / um / drivers / ubd_kern.c
1 /*
2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 /* 2001-09-28...2002-04-17
7  * Partition stuff by James_McMechan@hotmail.com
8  * old style ubd by setting UBD_SHIFT to 0
9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
10  * partitions have changed in 2.5
11  * 2003-01-29 more tinkering for 2.5.59-1
12  * This should now address the sysfs problems and has
13  * the symlink for devfs to allow for booting with
14  * the common /dev/ubd/discX/... names rather than
15  * only /dev/ubdN/discN this version also has lots of
16  * clean ups preparing for ubd-many.
17  * James McMechan
18  */
19
20 #define UBD_SHIFT 4
21
22 #include "linux/kernel.h"
23 #include "linux/module.h"
24 #include "linux/blkdev.h"
25 #include "linux/ata.h"
26 #include "linux/hdreg.h"
27 #include "linux/init.h"
28 #include "linux/cdrom.h"
29 #include "linux/proc_fs.h"
30 #include "linux/seq_file.h"
31 #include "linux/ctype.h"
32 #include "linux/capability.h"
33 #include "linux/mm.h"
34 #include "linux/slab.h"
35 #include "linux/vmalloc.h"
36 #include "linux/blkpg.h"
37 #include "linux/genhd.h"
38 #include "linux/spinlock.h"
39 #include "linux/platform_device.h"
40 #include "linux/scatterlist.h"
41 #include "asm/segment.h"
42 #include "asm/uaccess.h"
43 #include "asm/irq.h"
44 #include "asm/types.h"
45 #include "asm/tlbflush.h"
46 #include "mem_user.h"
47 #include "kern_util.h"
48 #include "kern.h"
49 #include "mconsole_kern.h"
50 #include "init.h"
51 #include "irq_user.h"
52 #include "irq_kern.h"
53 #include "ubd_user.h"
54 #include "os.h"
55 #include "mem.h"
56 #include "mem_kern.h"
57 #include "cow.h"
58
59 enum ubd_req { UBD_READ, UBD_WRITE };
60
61 struct io_thread_req {
62         struct request *req;
63         enum ubd_req op;
64         int fds[2];
65         unsigned long offsets[2];
66         unsigned long long offset;
67         unsigned long length;
68         char *buffer;
69         int sectorsize;
70         unsigned long sector_mask;
71         unsigned long long cow_offset;
72         unsigned long bitmap_words[2];
73         int error;
74 };
75
76 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
77 {
78         __u64 n;
79         int bits, off;
80
81         bits = sizeof(data[0]) * 8;
82         n = bit / bits;
83         off = bit % bits;
84         return (data[n] & (1 << off)) != 0;
85 }
86
87 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
88 {
89         __u64 n;
90         int bits, off;
91
92         bits = sizeof(data[0]) * 8;
93         n = bit / bits;
94         off = bit % bits;
95         data[n] |= (1 << off);
96 }
97 /*End stuff from ubd_user.h*/
98
99 #define DRIVER_NAME "uml-blkdev"
100
101 static DEFINE_MUTEX(ubd_lock);
102
103 static int ubd_open(struct block_device *bdev, fmode_t mode);
104 static int ubd_release(struct gendisk *disk, fmode_t mode);
105 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
106                      unsigned int cmd, unsigned long arg);
107 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
108
109 #define MAX_DEV (16)
110
111 static const struct block_device_operations ubd_blops = {
112         .owner          = THIS_MODULE,
113         .open           = ubd_open,
114         .release        = ubd_release,
115         .ioctl          = ubd_ioctl,
116         .getgeo         = ubd_getgeo,
117 };
118
119 /* Protected by ubd_lock */
120 static int fake_major = UBD_MAJOR;
121 static struct gendisk *ubd_gendisk[MAX_DEV];
122 static struct gendisk *fake_gendisk[MAX_DEV];
123
124 #ifdef CONFIG_BLK_DEV_UBD_SYNC
125 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
126                                          .cl = 1 })
127 #else
128 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
129                                          .cl = 1 })
130 #endif
131 static struct openflags global_openflags = OPEN_FLAGS;
132
133 struct cow {
134         /* backing file name */
135         char *file;
136         /* backing file fd */
137         int fd;
138         unsigned long *bitmap;
139         unsigned long bitmap_len;
140         int bitmap_offset;
141         int data_offset;
142 };
143
144 #define MAX_SG 64
145
146 struct ubd {
147         struct list_head restart;
148         /* name (and fd, below) of the file opened for writing, either the
149          * backing or the cow file. */
150         char *file;
151         int count;
152         int fd;
153         __u64 size;
154         struct openflags boot_openflags;
155         struct openflags openflags;
156         unsigned shared:1;
157         unsigned no_cow:1;
158         struct cow cow;
159         struct platform_device pdev;
160         struct request_queue *queue;
161         spinlock_t lock;
162         struct scatterlist sg[MAX_SG];
163         struct request *request;
164         int start_sg, end_sg;
165 };
166
167 #define DEFAULT_COW { \
168         .file =                 NULL, \
169         .fd =                   -1,     \
170         .bitmap =               NULL, \
171         .bitmap_offset =        0, \
172         .data_offset =          0, \
173 }
174
175 #define DEFAULT_UBD { \
176         .file =                 NULL, \
177         .count =                0, \
178         .fd =                   -1, \
179         .size =                 -1, \
180         .boot_openflags =       OPEN_FLAGS, \
181         .openflags =            OPEN_FLAGS, \
182         .no_cow =               0, \
183         .shared =               0, \
184         .cow =                  DEFAULT_COW, \
185         .lock =                 SPIN_LOCK_UNLOCKED,     \
186         .request =              NULL, \
187         .start_sg =             0, \
188         .end_sg =               0, \
189 }
190
191 /* Protected by ubd_lock */
192 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
193
194 /* Only changed by fake_ide_setup which is a setup */
195 static int fake_ide = 0;
196 static struct proc_dir_entry *proc_ide_root = NULL;
197 static struct proc_dir_entry *proc_ide = NULL;
198
199 static void make_proc_ide(void)
200 {
201         proc_ide_root = proc_mkdir("ide", NULL);
202         proc_ide = proc_mkdir("ide0", proc_ide_root);
203 }
204
205 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
206 {
207         seq_puts(m, "disk\n");
208         return 0;
209 }
210
211 static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
212 {
213         return single_open(file, fake_ide_media_proc_show, NULL);
214 }
215
216 static const struct file_operations fake_ide_media_proc_fops = {
217         .owner          = THIS_MODULE,
218         .open           = fake_ide_media_proc_open,
219         .read           = seq_read,
220         .llseek         = seq_lseek,
221         .release        = single_release,
222 };
223
224 static void make_ide_entries(const char *dev_name)
225 {
226         struct proc_dir_entry *dir, *ent;
227         char name[64];
228
229         if(proc_ide_root == NULL) make_proc_ide();
230
231         dir = proc_mkdir(dev_name, proc_ide);
232         if(!dir) return;
233
234         ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
235         if(!ent) return;
236         snprintf(name, sizeof(name), "ide0/%s", dev_name);
237         proc_symlink(dev_name, proc_ide_root, name);
238 }
239
240 static int fake_ide_setup(char *str)
241 {
242         fake_ide = 1;
243         return 1;
244 }
245
246 __setup("fake_ide", fake_ide_setup);
247
248 __uml_help(fake_ide_setup,
249 "fake_ide\n"
250 "    Create ide0 entries that map onto ubd devices.\n\n"
251 );
252
253 static int parse_unit(char **ptr)
254 {
255         char *str = *ptr, *end;
256         int n = -1;
257
258         if(isdigit(*str)) {
259                 n = simple_strtoul(str, &end, 0);
260                 if(end == str)
261                         return -1;
262                 *ptr = end;
263         }
264         else if (('a' <= *str) && (*str <= 'z')) {
265                 n = *str - 'a';
266                 str++;
267                 *ptr = str;
268         }
269         return n;
270 }
271
272 /* If *index_out == -1 at exit, the passed option was a general one;
273  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
274  * should not be freed on exit.
275  */
276 static int ubd_setup_common(char *str, int *index_out, char **error_out)
277 {
278         struct ubd *ubd_dev;
279         struct openflags flags = global_openflags;
280         char *backing_file;
281         int n, err = 0, i;
282
283         if(index_out) *index_out = -1;
284         n = *str;
285         if(n == '='){
286                 char *end;
287                 int major;
288
289                 str++;
290                 if(!strcmp(str, "sync")){
291                         global_openflags = of_sync(global_openflags);
292                         goto out1;
293                 }
294
295                 err = -EINVAL;
296                 major = simple_strtoul(str, &end, 0);
297                 if((*end != '\0') || (end == str)){
298                         *error_out = "Didn't parse major number";
299                         goto out1;
300                 }
301
302                 mutex_lock(&ubd_lock);
303                 if (fake_major != UBD_MAJOR) {
304                         *error_out = "Can't assign a fake major twice";
305                         goto out1;
306                 }
307
308                 fake_major = major;
309
310                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
311                        major);
312                 err = 0;
313         out1:
314                 mutex_unlock(&ubd_lock);
315                 return err;
316         }
317
318         n = parse_unit(&str);
319         if(n < 0){
320                 *error_out = "Couldn't parse device number";
321                 return -EINVAL;
322         }
323         if(n >= MAX_DEV){
324                 *error_out = "Device number out of range";
325                 return 1;
326         }
327
328         err = -EBUSY;
329         mutex_lock(&ubd_lock);
330
331         ubd_dev = &ubd_devs[n];
332         if(ubd_dev->file != NULL){
333                 *error_out = "Device is already configured";
334                 goto out;
335         }
336
337         if (index_out)
338                 *index_out = n;
339
340         err = -EINVAL;
341         for (i = 0; i < sizeof("rscd="); i++) {
342                 switch (*str) {
343                 case 'r':
344                         flags.w = 0;
345                         break;
346                 case 's':
347                         flags.s = 1;
348                         break;
349                 case 'd':
350                         ubd_dev->no_cow = 1;
351                         break;
352                 case 'c':
353                         ubd_dev->shared = 1;
354                         break;
355                 case '=':
356                         str++;
357                         goto break_loop;
358                 default:
359                         *error_out = "Expected '=' or flag letter "
360                                 "(r, s, c, or d)";
361                         goto out;
362                 }
363                 str++;
364         }
365
366         if (*str == '=')
367                 *error_out = "Too many flags specified";
368         else
369                 *error_out = "Missing '='";
370         goto out;
371
372 break_loop:
373         backing_file = strchr(str, ',');
374
375         if (backing_file == NULL)
376                 backing_file = strchr(str, ':');
377
378         if(backing_file != NULL){
379                 if(ubd_dev->no_cow){
380                         *error_out = "Can't specify both 'd' and a cow file";
381                         goto out;
382                 }
383                 else {
384                         *backing_file = '\0';
385                         backing_file++;
386                 }
387         }
388         err = 0;
389         ubd_dev->file = str;
390         ubd_dev->cow.file = backing_file;
391         ubd_dev->boot_openflags = flags;
392 out:
393         mutex_unlock(&ubd_lock);
394         return err;
395 }
396
397 static int ubd_setup(char *str)
398 {
399         char *error;
400         int err;
401
402         err = ubd_setup_common(str, NULL, &error);
403         if(err)
404                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
405                        "%s\n", str, error);
406         return 1;
407 }
408
409 __setup("ubd", ubd_setup);
410 __uml_help(ubd_setup,
411 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
412 "    This is used to associate a device with a file in the underlying\n"
413 "    filesystem. When specifying two filenames, the first one is the\n"
414 "    COW name and the second is the backing file name. As separator you can\n"
415 "    use either a ':' or a ',': the first one allows writing things like;\n"
416 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
417 "    while with a ',' the shell would not expand the 2nd '~'.\n"
418 "    When using only one filename, UML will detect whether to treat it like\n"
419 "    a COW file or a backing file. To override this detection, add the 'd'\n"
420 "    flag:\n"
421 "       ubd0d=BackingFile\n"
422 "    Usually, there is a filesystem in the file, but \n"
423 "    that's not required. Swap devices containing swap files can be\n"
424 "    specified like this. Also, a file which doesn't contain a\n"
425 "    filesystem can have its contents read in the virtual \n"
426 "    machine by running 'dd' on the device. <n> must be in the range\n"
427 "    0 to 7. Appending an 'r' to the number will cause that device\n"
428 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
429 "    an 's' will cause data to be written to disk on the host immediately.\n"
430 "    'c' will cause the device to be treated as being shared between multiple\n"
431 "    UMLs and file locking will be turned off - this is appropriate for a\n"
432 "    cluster filesystem and inappropriate at almost all other times.\n\n"
433 );
434
435 static int udb_setup(char *str)
436 {
437         printk("udb%s specified on command line is almost certainly a ubd -> "
438                "udb TYPO\n", str);
439         return 1;
440 }
441
442 __setup("udb", udb_setup);
443 __uml_help(udb_setup,
444 "udb\n"
445 "    This option is here solely to catch ubd -> udb typos, which can be\n"
446 "    to impossible to catch visually unless you specifically look for\n"
447 "    them.  The only result of any option starting with 'udb' is an error\n"
448 "    in the boot output.\n\n"
449 );
450
451 static void do_ubd_request(struct request_queue * q);
452
453 /* Only changed by ubd_init, which is an initcall. */
454 static int thread_fd = -1;
455 static LIST_HEAD(restart);
456
457 /* XXX - move this inside ubd_intr. */
458 /* Called without dev->lock held, and only in interrupt context. */
459 static void ubd_handler(void)
460 {
461         struct io_thread_req *req;
462         struct ubd *ubd;
463         struct list_head *list, *next_ele;
464         unsigned long flags;
465         int n;
466
467         while(1){
468                 n = os_read_file(thread_fd, &req,
469                                  sizeof(struct io_thread_req *));
470                 if(n != sizeof(req)){
471                         if(n == -EAGAIN)
472                                 break;
473                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
474                                "err = %d\n", -n);
475                         return;
476                 }
477
478                 blk_end_request(req->req, 0, req->length);
479                 kfree(req);
480         }
481         reactivate_fd(thread_fd, UBD_IRQ);
482
483         list_for_each_safe(list, next_ele, &restart){
484                 ubd = container_of(list, struct ubd, restart);
485                 list_del_init(&ubd->restart);
486                 spin_lock_irqsave(&ubd->lock, flags);
487                 do_ubd_request(ubd->queue);
488                 spin_unlock_irqrestore(&ubd->lock, flags);
489         }
490 }
491
492 static irqreturn_t ubd_intr(int irq, void *dev)
493 {
494         ubd_handler();
495         return IRQ_HANDLED;
496 }
497
498 /* Only changed by ubd_init, which is an initcall. */
499 static int io_pid = -1;
500
501 static void kill_io_thread(void)
502 {
503         if(io_pid != -1)
504                 os_kill_process(io_pid, 1);
505 }
506
507 __uml_exitcall(kill_io_thread);
508
509 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
510 {
511         char *file;
512
513         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
514         return os_file_size(file, size_out);
515 }
516
517 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
518 {
519         int err;
520
521         err = os_seek_file(fd, offset);
522         if (err < 0)
523                 return err;
524
525         err = os_read_file(fd, buf, len);
526         if (err < 0)
527                 return err;
528
529         return 0;
530 }
531
532 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
533 {
534         unsigned long modtime;
535         unsigned long long actual;
536         int err;
537
538         err = os_file_modtime(file, &modtime);
539         if (err < 0) {
540                 printk(KERN_ERR "Failed to get modification time of backing "
541                        "file \"%s\", err = %d\n", file, -err);
542                 return err;
543         }
544
545         err = os_file_size(file, &actual);
546         if (err < 0) {
547                 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
548                        "err = %d\n", file, -err);
549                 return err;
550         }
551
552         if (actual != size) {
553                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
554                  * the typecast.*/
555                 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
556                        "vs backing file\n", (unsigned long long) size, actual);
557                 return -EINVAL;
558         }
559         if (modtime != mtime) {
560                 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
561                        "backing file\n", mtime, modtime);
562                 return -EINVAL;
563         }
564         return 0;
565 }
566
567 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
568 {
569         struct uml_stat buf1, buf2;
570         int err;
571
572         if (from_cmdline == NULL)
573                 return 0;
574         if (!strcmp(from_cmdline, from_cow))
575                 return 0;
576
577         err = os_stat_file(from_cmdline, &buf1);
578         if (err < 0) {
579                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
580                        -err);
581                 return 0;
582         }
583         err = os_stat_file(from_cow, &buf2);
584         if (err < 0) {
585                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
586                        -err);
587                 return 1;
588         }
589         if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
590                 return 0;
591
592         printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
593                "\"%s\" specified in COW header of \"%s\"\n",
594                from_cmdline, from_cow, cow);
595         return 1;
596 }
597
598 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
599                   char **backing_file_out, int *bitmap_offset_out,
600                   unsigned long *bitmap_len_out, int *data_offset_out,
601                   int *create_cow_out)
602 {
603         time_t mtime;
604         unsigned long long size;
605         __u32 version, align;
606         char *backing_file;
607         int fd, err, sectorsize, asked_switch, mode = 0644;
608
609         fd = os_open_file(file, *openflags, mode);
610         if (fd < 0) {
611                 if ((fd == -ENOENT) && (create_cow_out != NULL))
612                         *create_cow_out = 1;
613                 if (!openflags->w ||
614                     ((fd != -EROFS) && (fd != -EACCES)))
615                         return fd;
616                 openflags->w = 0;
617                 fd = os_open_file(file, *openflags, mode);
618                 if (fd < 0)
619                         return fd;
620         }
621
622         if (shared)
623                 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
624         else {
625                 err = os_lock_file(fd, openflags->w);
626                 if (err < 0) {
627                         printk(KERN_ERR "Failed to lock '%s', err = %d\n",
628                                file, -err);
629                         goto out_close;
630                 }
631         }
632
633         /* Successful return case! */
634         if (backing_file_out == NULL)
635                 return fd;
636
637         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
638                               &size, &sectorsize, &align, bitmap_offset_out);
639         if (err && (*backing_file_out != NULL)) {
640                 printk(KERN_ERR "Failed to read COW header from COW file "
641                        "\"%s\", errno = %d\n", file, -err);
642                 goto out_close;
643         }
644         if (err)
645                 return fd;
646
647         asked_switch = path_requires_switch(*backing_file_out, backing_file,
648                                             file);
649
650         /* Allow switching only if no mismatch. */
651         if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
652                                                    mtime)) {
653                 printk(KERN_ERR "Switching backing file to '%s'\n",
654                        *backing_file_out);
655                 err = write_cow_header(file, fd, *backing_file_out,
656                                        sectorsize, align, &size);
657                 if (err) {
658                         printk(KERN_ERR "Switch failed, errno = %d\n", -err);
659                         goto out_close;
660                 }
661         } else {
662                 *backing_file_out = backing_file;
663                 err = backing_file_mismatch(*backing_file_out, size, mtime);
664                 if (err)
665                         goto out_close;
666         }
667
668         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
669                   bitmap_len_out, data_offset_out);
670
671         return fd;
672  out_close:
673         os_close_file(fd);
674         return err;
675 }
676
677 static int create_cow_file(char *cow_file, char *backing_file,
678                     struct openflags flags,
679                     int sectorsize, int alignment, int *bitmap_offset_out,
680                     unsigned long *bitmap_len_out, int *data_offset_out)
681 {
682         int err, fd;
683
684         flags.c = 1;
685         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
686         if (fd < 0) {
687                 err = fd;
688                 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
689                        cow_file, -err);
690                 goto out;
691         }
692
693         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
694                             bitmap_offset_out, bitmap_len_out,
695                             data_offset_out);
696         if (!err)
697                 return fd;
698         os_close_file(fd);
699  out:
700         return err;
701 }
702
703 static void ubd_close_dev(struct ubd *ubd_dev)
704 {
705         os_close_file(ubd_dev->fd);
706         if(ubd_dev->cow.file == NULL)
707                 return;
708
709         os_close_file(ubd_dev->cow.fd);
710         vfree(ubd_dev->cow.bitmap);
711         ubd_dev->cow.bitmap = NULL;
712 }
713
714 static int ubd_open_dev(struct ubd *ubd_dev)
715 {
716         struct openflags flags;
717         char **back_ptr;
718         int err, create_cow, *create_ptr;
719         int fd;
720
721         ubd_dev->openflags = ubd_dev->boot_openflags;
722         create_cow = 0;
723         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
724         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
725
726         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
727                                 back_ptr, &ubd_dev->cow.bitmap_offset,
728                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
729                                 create_ptr);
730
731         if((fd == -ENOENT) && create_cow){
732                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
733                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
734                                           &ubd_dev->cow.bitmap_offset,
735                                           &ubd_dev->cow.bitmap_len,
736                                           &ubd_dev->cow.data_offset);
737                 if(fd >= 0){
738                         printk(KERN_INFO "Creating \"%s\" as COW file for "
739                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
740                 }
741         }
742
743         if(fd < 0){
744                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
745                        -fd);
746                 return fd;
747         }
748         ubd_dev->fd = fd;
749
750         if(ubd_dev->cow.file != NULL){
751                 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
752
753                 err = -ENOMEM;
754                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
755                 if(ubd_dev->cow.bitmap == NULL){
756                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
757                         goto error;
758                 }
759                 flush_tlb_kernel_vm();
760
761                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
762                                       ubd_dev->cow.bitmap_offset,
763                                       ubd_dev->cow.bitmap_len);
764                 if(err < 0)
765                         goto error;
766
767                 flags = ubd_dev->openflags;
768                 flags.w = 0;
769                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
770                                     NULL, NULL, NULL, NULL);
771                 if(err < 0) goto error;
772                 ubd_dev->cow.fd = err;
773         }
774         return 0;
775  error:
776         os_close_file(ubd_dev->fd);
777         return err;
778 }
779
780 static void ubd_device_release(struct device *dev)
781 {
782         struct ubd *ubd_dev = dev_get_drvdata(dev);
783
784         blk_cleanup_queue(ubd_dev->queue);
785         *ubd_dev = ((struct ubd) DEFAULT_UBD);
786 }
787
788 static int ubd_disk_register(int major, u64 size, int unit,
789                              struct gendisk **disk_out)
790 {
791         struct gendisk *disk;
792
793         disk = alloc_disk(1 << UBD_SHIFT);
794         if(disk == NULL)
795                 return -ENOMEM;
796
797         disk->major = major;
798         disk->first_minor = unit << UBD_SHIFT;
799         disk->fops = &ubd_blops;
800         set_capacity(disk, size / 512);
801         if (major == UBD_MAJOR)
802                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
803         else
804                 sprintf(disk->disk_name, "ubd_fake%d", unit);
805
806         /* sysfs register (not for ide fake devices) */
807         if (major == UBD_MAJOR) {
808                 ubd_devs[unit].pdev.id   = unit;
809                 ubd_devs[unit].pdev.name = DRIVER_NAME;
810                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
811                 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
812                 platform_device_register(&ubd_devs[unit].pdev);
813                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
814         }
815
816         disk->private_data = &ubd_devs[unit];
817         disk->queue = ubd_devs[unit].queue;
818         add_disk(disk);
819
820         *disk_out = disk;
821         return 0;
822 }
823
824 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
825
826 static int ubd_add(int n, char **error_out)
827 {
828         struct ubd *ubd_dev = &ubd_devs[n];
829         int err = 0;
830
831         if(ubd_dev->file == NULL)
832                 goto out;
833
834         err = ubd_file_size(ubd_dev, &ubd_dev->size);
835         if(err < 0){
836                 *error_out = "Couldn't determine size of device's file";
837                 goto out;
838         }
839
840         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
841
842         INIT_LIST_HEAD(&ubd_dev->restart);
843         sg_init_table(ubd_dev->sg, MAX_SG);
844
845         err = -ENOMEM;
846         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
847         if (ubd_dev->queue == NULL) {
848                 *error_out = "Failed to initialize device queue";
849                 goto out;
850         }
851         ubd_dev->queue->queuedata = ubd_dev;
852
853         blk_queue_max_segments(ubd_dev->queue, MAX_SG);
854         err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
855         if(err){
856                 *error_out = "Failed to register device";
857                 goto out_cleanup;
858         }
859
860         if (fake_major != UBD_MAJOR)
861                 ubd_disk_register(fake_major, ubd_dev->size, n,
862                                   &fake_gendisk[n]);
863
864         /*
865          * Perhaps this should also be under the "if (fake_major)" above
866          * using the fake_disk->disk_name
867          */
868         if (fake_ide)
869                 make_ide_entries(ubd_gendisk[n]->disk_name);
870
871         err = 0;
872 out:
873         return err;
874
875 out_cleanup:
876         blk_cleanup_queue(ubd_dev->queue);
877         goto out;
878 }
879
880 static int ubd_config(char *str, char **error_out)
881 {
882         int n, ret;
883
884         /* This string is possibly broken up and stored, so it's only
885          * freed if ubd_setup_common fails, or if only general options
886          * were set.
887          */
888         str = kstrdup(str, GFP_KERNEL);
889         if (str == NULL) {
890                 *error_out = "Failed to allocate memory";
891                 return -ENOMEM;
892         }
893
894         ret = ubd_setup_common(str, &n, error_out);
895         if (ret)
896                 goto err_free;
897
898         if (n == -1) {
899                 ret = 0;
900                 goto err_free;
901         }
902
903         mutex_lock(&ubd_lock);
904         ret = ubd_add(n, error_out);
905         if (ret)
906                 ubd_devs[n].file = NULL;
907         mutex_unlock(&ubd_lock);
908
909 out:
910         return ret;
911
912 err_free:
913         kfree(str);
914         goto out;
915 }
916
917 static int ubd_get_config(char *name, char *str, int size, char **error_out)
918 {
919         struct ubd *ubd_dev;
920         int n, len = 0;
921
922         n = parse_unit(&name);
923         if((n >= MAX_DEV) || (n < 0)){
924                 *error_out = "ubd_get_config : device number out of range";
925                 return -1;
926         }
927
928         ubd_dev = &ubd_devs[n];
929         mutex_lock(&ubd_lock);
930
931         if(ubd_dev->file == NULL){
932                 CONFIG_CHUNK(str, size, len, "", 1);
933                 goto out;
934         }
935
936         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
937
938         if(ubd_dev->cow.file != NULL){
939                 CONFIG_CHUNK(str, size, len, ",", 0);
940                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
941         }
942         else CONFIG_CHUNK(str, size, len, "", 1);
943
944  out:
945         mutex_unlock(&ubd_lock);
946         return len;
947 }
948
949 static int ubd_id(char **str, int *start_out, int *end_out)
950 {
951         int n;
952
953         n = parse_unit(str);
954         *start_out = 0;
955         *end_out = MAX_DEV - 1;
956         return n;
957 }
958
959 static int ubd_remove(int n, char **error_out)
960 {
961         struct gendisk *disk = ubd_gendisk[n];
962         struct ubd *ubd_dev;
963         int err = -ENODEV;
964
965         mutex_lock(&ubd_lock);
966
967         ubd_dev = &ubd_devs[n];
968
969         if(ubd_dev->file == NULL)
970                 goto out;
971
972         /* you cannot remove a open disk */
973         err = -EBUSY;
974         if(ubd_dev->count > 0)
975                 goto out;
976
977         ubd_gendisk[n] = NULL;
978         if(disk != NULL){
979                 del_gendisk(disk);
980                 put_disk(disk);
981         }
982
983         if(fake_gendisk[n] != NULL){
984                 del_gendisk(fake_gendisk[n]);
985                 put_disk(fake_gendisk[n]);
986                 fake_gendisk[n] = NULL;
987         }
988
989         err = 0;
990         platform_device_unregister(&ubd_dev->pdev);
991 out:
992         mutex_unlock(&ubd_lock);
993         return err;
994 }
995
996 /* All these are called by mconsole in process context and without
997  * ubd-specific locks.  The structure itself is const except for .list.
998  */
999 static struct mc_device ubd_mc = {
1000         .list           = LIST_HEAD_INIT(ubd_mc.list),
1001         .name           = "ubd",
1002         .config         = ubd_config,
1003         .get_config     = ubd_get_config,
1004         .id             = ubd_id,
1005         .remove         = ubd_remove,
1006 };
1007
1008 static int __init ubd_mc_init(void)
1009 {
1010         mconsole_register_dev(&ubd_mc);
1011         return 0;
1012 }
1013
1014 __initcall(ubd_mc_init);
1015
1016 static int __init ubd0_init(void)
1017 {
1018         struct ubd *ubd_dev = &ubd_devs[0];
1019
1020         mutex_lock(&ubd_lock);
1021         if(ubd_dev->file == NULL)
1022                 ubd_dev->file = "root_fs";
1023         mutex_unlock(&ubd_lock);
1024
1025         return 0;
1026 }
1027
1028 __initcall(ubd0_init);
1029
1030 /* Used in ubd_init, which is an initcall */
1031 static struct platform_driver ubd_driver = {
1032         .driver = {
1033                 .name  = DRIVER_NAME,
1034         },
1035 };
1036
1037 static int __init ubd_init(void)
1038 {
1039         char *error;
1040         int i, err;
1041
1042         if (register_blkdev(UBD_MAJOR, "ubd"))
1043                 return -1;
1044
1045         if (fake_major != UBD_MAJOR) {
1046                 char name[sizeof("ubd_nnn\0")];
1047
1048                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1049                 if (register_blkdev(fake_major, "ubd"))
1050                         return -1;
1051         }
1052         platform_driver_register(&ubd_driver);
1053         mutex_lock(&ubd_lock);
1054         for (i = 0; i < MAX_DEV; i++){
1055                 err = ubd_add(i, &error);
1056                 if(err)
1057                         printk(KERN_ERR "Failed to initialize ubd device %d :"
1058                                "%s\n", i, error);
1059         }
1060         mutex_unlock(&ubd_lock);
1061         return 0;
1062 }
1063
1064 late_initcall(ubd_init);
1065
1066 static int __init ubd_driver_init(void){
1067         unsigned long stack;
1068         int err;
1069
1070         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1071         if(global_openflags.s){
1072                 printk(KERN_INFO "ubd: Synchronous mode\n");
1073                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1074                  * enough. So use anyway the io thread. */
1075         }
1076         stack = alloc_stack(0, 0);
1077         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1078                                  &thread_fd);
1079         if(io_pid < 0){
1080                 printk(KERN_ERR
1081                        "ubd : Failed to start I/O thread (errno = %d) - "
1082                        "falling back to synchronous I/O\n", -io_pid);
1083                 io_pid = -1;
1084                 return 0;
1085         }
1086         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1087                              IRQF_DISABLED, "ubd", ubd_devs);
1088         if(err != 0)
1089                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1090         return 0;
1091 }
1092
1093 device_initcall(ubd_driver_init);
1094
1095 static int ubd_open(struct block_device *bdev, fmode_t mode)
1096 {
1097         struct gendisk *disk = bdev->bd_disk;
1098         struct ubd *ubd_dev = disk->private_data;
1099         int err = 0;
1100
1101         if(ubd_dev->count == 0){
1102                 err = ubd_open_dev(ubd_dev);
1103                 if(err){
1104                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1105                                disk->disk_name, ubd_dev->file, -err);
1106                         goto out;
1107                 }
1108         }
1109         ubd_dev->count++;
1110         set_disk_ro(disk, !ubd_dev->openflags.w);
1111
1112         /* This should no more be needed. And it didn't work anyway to exclude
1113          * read-write remounting of filesystems.*/
1114         /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1115                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1116                 err = -EROFS;
1117         }*/
1118  out:
1119         return err;
1120 }
1121
1122 static int ubd_release(struct gendisk *disk, fmode_t mode)
1123 {
1124         struct ubd *ubd_dev = disk->private_data;
1125
1126         if(--ubd_dev->count == 0)
1127                 ubd_close_dev(ubd_dev);
1128         return 0;
1129 }
1130
1131 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1132                           __u64 *cow_offset, unsigned long *bitmap,
1133                           __u64 bitmap_offset, unsigned long *bitmap_words,
1134                           __u64 bitmap_len)
1135 {
1136         __u64 sector = io_offset >> 9;
1137         int i, update_bitmap = 0;
1138
1139         for(i = 0; i < length >> 9; i++){
1140                 if(cow_mask != NULL)
1141                         ubd_set_bit(i, (unsigned char *) cow_mask);
1142                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1143                         continue;
1144
1145                 update_bitmap = 1;
1146                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1147         }
1148
1149         if(!update_bitmap)
1150                 return;
1151
1152         *cow_offset = sector / (sizeof(unsigned long) * 8);
1153
1154         /* This takes care of the case where we're exactly at the end of the
1155          * device, and *cow_offset + 1 is off the end.  So, just back it up
1156          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1157          * for the original diagnosis.
1158          */
1159         if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1160                                          sizeof(unsigned long)) - 1))
1161                 (*cow_offset)--;
1162
1163         bitmap_words[0] = bitmap[*cow_offset];
1164         bitmap_words[1] = bitmap[*cow_offset + 1];
1165
1166         *cow_offset *= sizeof(unsigned long);
1167         *cow_offset += bitmap_offset;
1168 }
1169
1170 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1171                        __u64 bitmap_offset, __u64 bitmap_len)
1172 {
1173         __u64 sector = req->offset >> 9;
1174         int i;
1175
1176         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1177                 panic("Operation too long");
1178
1179         if(req->op == UBD_READ) {
1180                 for(i = 0; i < req->length >> 9; i++){
1181                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1182                                 ubd_set_bit(i, (unsigned char *)
1183                                             &req->sector_mask);
1184                 }
1185         }
1186         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1187                            &req->cow_offset, bitmap, bitmap_offset,
1188                            req->bitmap_words, bitmap_len);
1189 }
1190
1191 /* Called with dev->lock held */
1192 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1193                             unsigned long long offset, int page_offset,
1194                             int len, struct page *page)
1195 {
1196         struct gendisk *disk = req->rq_disk;
1197         struct ubd *ubd_dev = disk->private_data;
1198
1199         io_req->req = req;
1200         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1201                 ubd_dev->fd;
1202         io_req->fds[1] = ubd_dev->fd;
1203         io_req->cow_offset = -1;
1204         io_req->offset = offset;
1205         io_req->length = len;
1206         io_req->error = 0;
1207         io_req->sector_mask = 0;
1208
1209         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1210         io_req->offsets[0] = 0;
1211         io_req->offsets[1] = ubd_dev->cow.data_offset;
1212         io_req->buffer = page_address(page) + page_offset;
1213         io_req->sectorsize = 1 << 9;
1214
1215         if(ubd_dev->cow.file != NULL)
1216                 cowify_req(io_req, ubd_dev->cow.bitmap,
1217                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1218
1219 }
1220
1221 /* Called with dev->lock held */
1222 static void do_ubd_request(struct request_queue *q)
1223 {
1224         struct io_thread_req *io_req;
1225         struct request *req;
1226         sector_t sector;
1227         int n;
1228
1229         while(1){
1230                 struct ubd *dev = q->queuedata;
1231                 if(dev->end_sg == 0){
1232                         struct request *req = blk_fetch_request(q);
1233                         if(req == NULL)
1234                                 return;
1235
1236                         dev->request = req;
1237                         dev->start_sg = 0;
1238                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1239                 }
1240
1241                 req = dev->request;
1242                 sector = blk_rq_pos(req);
1243                 while(dev->start_sg < dev->end_sg){
1244                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1245
1246                         io_req = kmalloc(sizeof(struct io_thread_req),
1247                                          GFP_ATOMIC);
1248                         if(io_req == NULL){
1249                                 if(list_empty(&dev->restart))
1250                                         list_add(&dev->restart, &restart);
1251                                 return;
1252                         }
1253                         prepare_request(req, io_req,
1254                                         (unsigned long long)sector << 9,
1255                                         sg->offset, sg->length, sg_page(sg));
1256
1257                         sector += sg->length >> 9;
1258                         n = os_write_file(thread_fd, &io_req,
1259                                           sizeof(struct io_thread_req *));
1260                         if(n != sizeof(struct io_thread_req *)){
1261                                 if(n != -EAGAIN)
1262                                         printk("write to io thread failed, "
1263                                                "errno = %d\n", -n);
1264                                 else if(list_empty(&dev->restart))
1265                                         list_add(&dev->restart, &restart);
1266                                 kfree(io_req);
1267                                 return;
1268                         }
1269
1270                         dev->start_sg++;
1271                 }
1272                 dev->end_sg = 0;
1273                 dev->request = NULL;
1274         }
1275 }
1276
1277 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1278 {
1279         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1280
1281         geo->heads = 128;
1282         geo->sectors = 32;
1283         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1284         return 0;
1285 }
1286
1287 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1288                      unsigned int cmd, unsigned long arg)
1289 {
1290         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1291         u16 ubd_id[ATA_ID_WORDS];
1292
1293         switch (cmd) {
1294                 struct cdrom_volctrl volume;
1295         case HDIO_GET_IDENTITY:
1296                 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1297                 ubd_id[ATA_ID_CYLS]     = ubd_dev->size / (128 * 32 * 512);
1298                 ubd_id[ATA_ID_HEADS]    = 128;
1299                 ubd_id[ATA_ID_SECTORS]  = 32;
1300                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1301                                  sizeof(ubd_id)))
1302                         return -EFAULT;
1303                 return 0;
1304
1305         case CDROMVOLREAD:
1306                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1307                         return -EFAULT;
1308                 volume.channel0 = 255;
1309                 volume.channel1 = 255;
1310                 volume.channel2 = 255;
1311                 volume.channel3 = 255;
1312                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1313                         return -EFAULT;
1314                 return 0;
1315         }
1316         return -EINVAL;
1317 }
1318
1319 static int update_bitmap(struct io_thread_req *req)
1320 {
1321         int n;
1322
1323         if(req->cow_offset == -1)
1324                 return 0;
1325
1326         n = os_seek_file(req->fds[1], req->cow_offset);
1327         if(n < 0){
1328                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1329                 return 1;
1330         }
1331
1332         n = os_write_file(req->fds[1], &req->bitmap_words,
1333                           sizeof(req->bitmap_words));
1334         if(n != sizeof(req->bitmap_words)){
1335                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1336                        req->fds[1]);
1337                 return 1;
1338         }
1339
1340         return 0;
1341 }
1342
1343 static void do_io(struct io_thread_req *req)
1344 {
1345         char *buf;
1346         unsigned long len;
1347         int n, nsectors, start, end, bit;
1348         int err;
1349         __u64 off;
1350
1351         nsectors = req->length / req->sectorsize;
1352         start = 0;
1353         do {
1354                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1355                 end = start;
1356                 while((end < nsectors) &&
1357                       (ubd_test_bit(end, (unsigned char *)
1358                                     &req->sector_mask) == bit))
1359                         end++;
1360
1361                 off = req->offset + req->offsets[bit] +
1362                         start * req->sectorsize;
1363                 len = (end - start) * req->sectorsize;
1364                 buf = &req->buffer[start * req->sectorsize];
1365
1366                 err = os_seek_file(req->fds[bit], off);
1367                 if(err < 0){
1368                         printk("do_io - lseek failed : err = %d\n", -err);
1369                         req->error = 1;
1370                         return;
1371                 }
1372                 if(req->op == UBD_READ){
1373                         n = 0;
1374                         do {
1375                                 buf = &buf[n];
1376                                 len -= n;
1377                                 n = os_read_file(req->fds[bit], buf, len);
1378                                 if (n < 0) {
1379                                         printk("do_io - read failed, err = %d "
1380                                                "fd = %d\n", -n, req->fds[bit]);
1381                                         req->error = 1;
1382                                         return;
1383                                 }
1384                         } while((n < len) && (n != 0));
1385                         if (n < len) memset(&buf[n], 0, len - n);
1386                 } else {
1387                         n = os_write_file(req->fds[bit], buf, len);
1388                         if(n != len){
1389                                 printk("do_io - write failed err = %d "
1390                                        "fd = %d\n", -n, req->fds[bit]);
1391                                 req->error = 1;
1392                                 return;
1393                         }
1394                 }
1395
1396                 start = end;
1397         } while(start < nsectors);
1398
1399         req->error = update_bitmap(req);
1400 }
1401
1402 /* Changed in start_io_thread, which is serialized by being called only
1403  * from ubd_init, which is an initcall.
1404  */
1405 int kernel_fd = -1;
1406
1407 /* Only changed by the io thread. XXX: currently unused. */
1408 static int io_count = 0;
1409
1410 int io_thread(void *arg)
1411 {
1412         struct io_thread_req *req;
1413         int n;
1414
1415         ignore_sigwinch_sig();
1416         while(1){
1417                 n = os_read_file(kernel_fd, &req,
1418                                  sizeof(struct io_thread_req *));
1419                 if(n != sizeof(struct io_thread_req *)){
1420                         if(n < 0)
1421                                 printk("io_thread - read failed, fd = %d, "
1422                                        "err = %d\n", kernel_fd, -n);
1423                         else {
1424                                 printk("io_thread - short read, fd = %d, "
1425                                        "length = %d\n", kernel_fd, n);
1426                         }
1427                         continue;
1428                 }
1429                 io_count++;
1430                 do_io(req);
1431                 n = os_write_file(kernel_fd, &req,
1432                                   sizeof(struct io_thread_req *));
1433                 if(n != sizeof(struct io_thread_req *))
1434                         printk("io_thread - write failed, fd = %d, err = %d\n",
1435                                kernel_fd, -n);
1436         }
1437
1438         return 0;
1439 }