block: Consolidate phys_segment and hw_segment limits
[linux-2.6.git] / arch / um / drivers / ubd_kern.c
1 /*
2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
3  * Licensed under the GPL
4  */
5
6 /* 2001-09-28...2002-04-17
7  * Partition stuff by James_McMechan@hotmail.com
8  * old style ubd by setting UBD_SHIFT to 0
9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
10  * partitions have changed in 2.5
11  * 2003-01-29 more tinkering for 2.5.59-1
12  * This should now address the sysfs problems and has
13  * the symlink for devfs to allow for booting with
14  * the common /dev/ubd/discX/... names rather than
15  * only /dev/ubdN/discN this version also has lots of
16  * clean ups preparing for ubd-many.
17  * James McMechan
18  */
19
20 #define 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/vmalloc.h"
35 #include "linux/blkpg.h"
36 #include "linux/genhd.h"
37 #include "linux/spinlock.h"
38 #include "linux/platform_device.h"
39 #include "linux/scatterlist.h"
40 #include "asm/segment.h"
41 #include "asm/uaccess.h"
42 #include "asm/irq.h"
43 #include "asm/types.h"
44 #include "asm/tlbflush.h"
45 #include "mem_user.h"
46 #include "kern_util.h"
47 #include "kern.h"
48 #include "mconsole_kern.h"
49 #include "init.h"
50 #include "irq_user.h"
51 #include "irq_kern.h"
52 #include "ubd_user.h"
53 #include "os.h"
54 #include "mem.h"
55 #include "mem_kern.h"
56 #include "cow.h"
57
58 enum ubd_req { UBD_READ, UBD_WRITE };
59
60 struct io_thread_req {
61         struct request *req;
62         enum ubd_req op;
63         int fds[2];
64         unsigned long offsets[2];
65         unsigned long long offset;
66         unsigned long length;
67         char *buffer;
68         int sectorsize;
69         unsigned long sector_mask;
70         unsigned long long cow_offset;
71         unsigned long bitmap_words[2];
72         int error;
73 };
74
75 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
76 {
77         __u64 n;
78         int bits, off;
79
80         bits = sizeof(data[0]) * 8;
81         n = bit / bits;
82         off = bit % bits;
83         return (data[n] & (1 << off)) != 0;
84 }
85
86 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
87 {
88         __u64 n;
89         int bits, off;
90
91         bits = sizeof(data[0]) * 8;
92         n = bit / bits;
93         off = bit % bits;
94         data[n] |= (1 << off);
95 }
96 /*End stuff from ubd_user.h*/
97
98 #define DRIVER_NAME "uml-blkdev"
99
100 static DEFINE_MUTEX(ubd_lock);
101
102 static int ubd_open(struct block_device *bdev, fmode_t mode);
103 static int ubd_release(struct gendisk *disk, fmode_t mode);
104 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
105                      unsigned int cmd, unsigned long arg);
106 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
107
108 #define MAX_DEV (16)
109
110 static const struct block_device_operations ubd_blops = {
111         .owner          = THIS_MODULE,
112         .open           = ubd_open,
113         .release        = ubd_release,
114         .ioctl          = ubd_ioctl,
115         .getgeo         = ubd_getgeo,
116 };
117
118 /* Protected by ubd_lock */
119 static int fake_major = UBD_MAJOR;
120 static struct gendisk *ubd_gendisk[MAX_DEV];
121 static struct gendisk *fake_gendisk[MAX_DEV];
122
123 #ifdef CONFIG_BLK_DEV_UBD_SYNC
124 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
125                                          .cl = 1 })
126 #else
127 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
128                                          .cl = 1 })
129 #endif
130 static struct openflags global_openflags = OPEN_FLAGS;
131
132 struct cow {
133         /* backing file name */
134         char *file;
135         /* backing file fd */
136         int fd;
137         unsigned long *bitmap;
138         unsigned long bitmap_len;
139         int bitmap_offset;
140         int data_offset;
141 };
142
143 #define MAX_SG 64
144
145 struct ubd {
146         struct list_head restart;
147         /* name (and fd, below) of the file opened for writing, either the
148          * backing or the cow file. */
149         char *file;
150         int count;
151         int fd;
152         __u64 size;
153         struct openflags boot_openflags;
154         struct openflags openflags;
155         unsigned shared:1;
156         unsigned no_cow:1;
157         struct cow cow;
158         struct platform_device pdev;
159         struct request_queue *queue;
160         spinlock_t lock;
161         struct scatterlist sg[MAX_SG];
162         struct request *request;
163         int start_sg, end_sg;
164 };
165
166 #define DEFAULT_COW { \
167         .file =                 NULL, \
168         .fd =                   -1,     \
169         .bitmap =               NULL, \
170         .bitmap_offset =        0, \
171         .data_offset =          0, \
172 }
173
174 #define DEFAULT_UBD { \
175         .file =                 NULL, \
176         .count =                0, \
177         .fd =                   -1, \
178         .size =                 -1, \
179         .boot_openflags =       OPEN_FLAGS, \
180         .openflags =            OPEN_FLAGS, \
181         .no_cow =               0, \
182         .shared =               0, \
183         .cow =                  DEFAULT_COW, \
184         .lock =                 SPIN_LOCK_UNLOCKED,     \
185         .request =              NULL, \
186         .start_sg =             0, \
187         .end_sg =               0, \
188 }
189
190 /* Protected by ubd_lock */
191 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
192
193 /* Only changed by fake_ide_setup which is a setup */
194 static int fake_ide = 0;
195 static struct proc_dir_entry *proc_ide_root = NULL;
196 static struct proc_dir_entry *proc_ide = NULL;
197
198 static void make_proc_ide(void)
199 {
200         proc_ide_root = proc_mkdir("ide", NULL);
201         proc_ide = proc_mkdir("ide0", proc_ide_root);
202 }
203
204 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
205 {
206         seq_puts(m, "disk\n");
207         return 0;
208 }
209
210 static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
211 {
212         return single_open(file, fake_ide_media_proc_show, NULL);
213 }
214
215 static const struct file_operations fake_ide_media_proc_fops = {
216         .owner          = THIS_MODULE,
217         .open           = fake_ide_media_proc_open,
218         .read           = seq_read,
219         .llseek         = seq_lseek,
220         .release        = single_release,
221 };
222
223 static void make_ide_entries(const char *dev_name)
224 {
225         struct proc_dir_entry *dir, *ent;
226         char name[64];
227
228         if(proc_ide_root == NULL) make_proc_ide();
229
230         dir = proc_mkdir(dev_name, proc_ide);
231         if(!dir) return;
232
233         ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
234         if(!ent) return;
235         snprintf(name, sizeof(name), "ide0/%s", dev_name);
236         proc_symlink(dev_name, proc_ide_root, name);
237 }
238
239 static int fake_ide_setup(char *str)
240 {
241         fake_ide = 1;
242         return 1;
243 }
244
245 __setup("fake_ide", fake_ide_setup);
246
247 __uml_help(fake_ide_setup,
248 "fake_ide\n"
249 "    Create ide0 entries that map onto ubd devices.\n\n"
250 );
251
252 static int parse_unit(char **ptr)
253 {
254         char *str = *ptr, *end;
255         int n = -1;
256
257         if(isdigit(*str)) {
258                 n = simple_strtoul(str, &end, 0);
259                 if(end == str)
260                         return -1;
261                 *ptr = end;
262         }
263         else if (('a' <= *str) && (*str <= 'z')) {
264                 n = *str - 'a';
265                 str++;
266                 *ptr = str;
267         }
268         return n;
269 }
270
271 /* If *index_out == -1 at exit, the passed option was a general one;
272  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
273  * should not be freed on exit.
274  */
275 static int ubd_setup_common(char *str, int *index_out, char **error_out)
276 {
277         struct ubd *ubd_dev;
278         struct openflags flags = global_openflags;
279         char *backing_file;
280         int n, err = 0, i;
281
282         if(index_out) *index_out = -1;
283         n = *str;
284         if(n == '='){
285                 char *end;
286                 int major;
287
288                 str++;
289                 if(!strcmp(str, "sync")){
290                         global_openflags = of_sync(global_openflags);
291                         goto out1;
292                 }
293
294                 err = -EINVAL;
295                 major = simple_strtoul(str, &end, 0);
296                 if((*end != '\0') || (end == str)){
297                         *error_out = "Didn't parse major number";
298                         goto out1;
299                 }
300
301                 mutex_lock(&ubd_lock);
302                 if (fake_major != UBD_MAJOR) {
303                         *error_out = "Can't assign a fake major twice";
304                         goto out1;
305                 }
306
307                 fake_major = major;
308
309                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
310                        major);
311                 err = 0;
312         out1:
313                 mutex_unlock(&ubd_lock);
314                 return err;
315         }
316
317         n = parse_unit(&str);
318         if(n < 0){
319                 *error_out = "Couldn't parse device number";
320                 return -EINVAL;
321         }
322         if(n >= MAX_DEV){
323                 *error_out = "Device number out of range";
324                 return 1;
325         }
326
327         err = -EBUSY;
328         mutex_lock(&ubd_lock);
329
330         ubd_dev = &ubd_devs[n];
331         if(ubd_dev->file != NULL){
332                 *error_out = "Device is already configured";
333                 goto out;
334         }
335
336         if (index_out)
337                 *index_out = n;
338
339         err = -EINVAL;
340         for (i = 0; i < sizeof("rscd="); i++) {
341                 switch (*str) {
342                 case 'r':
343                         flags.w = 0;
344                         break;
345                 case 's':
346                         flags.s = 1;
347                         break;
348                 case 'd':
349                         ubd_dev->no_cow = 1;
350                         break;
351                 case 'c':
352                         ubd_dev->shared = 1;
353                         break;
354                 case '=':
355                         str++;
356                         goto break_loop;
357                 default:
358                         *error_out = "Expected '=' or flag letter "
359                                 "(r, s, c, or d)";
360                         goto out;
361                 }
362                 str++;
363         }
364
365         if (*str == '=')
366                 *error_out = "Too many flags specified";
367         else
368                 *error_out = "Missing '='";
369         goto out;
370
371 break_loop:
372         backing_file = strchr(str, ',');
373
374         if (backing_file == NULL)
375                 backing_file = strchr(str, ':');
376
377         if(backing_file != NULL){
378                 if(ubd_dev->no_cow){
379                         *error_out = "Can't specify both 'd' and a cow file";
380                         goto out;
381                 }
382                 else {
383                         *backing_file = '\0';
384                         backing_file++;
385                 }
386         }
387         err = 0;
388         ubd_dev->file = str;
389         ubd_dev->cow.file = backing_file;
390         ubd_dev->boot_openflags = flags;
391 out:
392         mutex_unlock(&ubd_lock);
393         return err;
394 }
395
396 static int ubd_setup(char *str)
397 {
398         char *error;
399         int err;
400
401         err = ubd_setup_common(str, NULL, &error);
402         if(err)
403                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
404                        "%s\n", str, error);
405         return 1;
406 }
407
408 __setup("ubd", ubd_setup);
409 __uml_help(ubd_setup,
410 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
411 "    This is used to associate a device with a file in the underlying\n"
412 "    filesystem. When specifying two filenames, the first one is the\n"
413 "    COW name and the second is the backing file name. As separator you can\n"
414 "    use either a ':' or a ',': the first one allows writing things like;\n"
415 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
416 "    while with a ',' the shell would not expand the 2nd '~'.\n"
417 "    When using only one filename, UML will detect whether to treat it like\n"
418 "    a COW file or a backing file. To override this detection, add the 'd'\n"
419 "    flag:\n"
420 "       ubd0d=BackingFile\n"
421 "    Usually, there is a filesystem in the file, but \n"
422 "    that's not required. Swap devices containing swap files can be\n"
423 "    specified like this. Also, a file which doesn't contain a\n"
424 "    filesystem can have its contents read in the virtual \n"
425 "    machine by running 'dd' on the device. <n> must be in the range\n"
426 "    0 to 7. Appending an 'r' to the number will cause that device\n"
427 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
428 "    an 's' will cause data to be written to disk on the host immediately.\n"
429 "    'c' will cause the device to be treated as being shared between multiple\n"
430 "    UMLs and file locking will be turned off - this is appropriate for a\n"
431 "    cluster filesystem and inappropriate at almost all other times.\n\n"
432 );
433
434 static int udb_setup(char *str)
435 {
436         printk("udb%s specified on command line is almost certainly a ubd -> "
437                "udb TYPO\n", str);
438         return 1;
439 }
440
441 __setup("udb", udb_setup);
442 __uml_help(udb_setup,
443 "udb\n"
444 "    This option is here solely to catch ubd -> udb typos, which can be\n"
445 "    to impossible to catch visually unless you specifically look for\n"
446 "    them.  The only result of any option starting with 'udb' is an error\n"
447 "    in the boot output.\n\n"
448 );
449
450 static void do_ubd_request(struct request_queue * q);
451
452 /* Only changed by ubd_init, which is an initcall. */
453 static int thread_fd = -1;
454 static LIST_HEAD(restart);
455
456 /* XXX - move this inside ubd_intr. */
457 /* Called without dev->lock held, and only in interrupt context. */
458 static void ubd_handler(void)
459 {
460         struct io_thread_req *req;
461         struct ubd *ubd;
462         struct list_head *list, *next_ele;
463         unsigned long flags;
464         int n;
465
466         while(1){
467                 n = os_read_file(thread_fd, &req,
468                                  sizeof(struct io_thread_req *));
469                 if(n != sizeof(req)){
470                         if(n == -EAGAIN)
471                                 break;
472                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
473                                "err = %d\n", -n);
474                         return;
475                 }
476
477                 blk_end_request(req->req, 0, req->length);
478                 kfree(req);
479         }
480         reactivate_fd(thread_fd, UBD_IRQ);
481
482         list_for_each_safe(list, next_ele, &restart){
483                 ubd = container_of(list, struct ubd, restart);
484                 list_del_init(&ubd->restart);
485                 spin_lock_irqsave(&ubd->lock, flags);
486                 do_ubd_request(ubd->queue);
487                 spin_unlock_irqrestore(&ubd->lock, flags);
488         }
489 }
490
491 static irqreturn_t ubd_intr(int irq, void *dev)
492 {
493         ubd_handler();
494         return IRQ_HANDLED;
495 }
496
497 /* Only changed by ubd_init, which is an initcall. */
498 static int io_pid = -1;
499
500 static void kill_io_thread(void)
501 {
502         if(io_pid != -1)
503                 os_kill_process(io_pid, 1);
504 }
505
506 __uml_exitcall(kill_io_thread);
507
508 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
509 {
510         char *file;
511
512         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
513         return os_file_size(file, size_out);
514 }
515
516 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
517 {
518         int err;
519
520         err = os_seek_file(fd, offset);
521         if (err < 0)
522                 return err;
523
524         err = os_read_file(fd, buf, len);
525         if (err < 0)
526                 return err;
527
528         return 0;
529 }
530
531 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
532 {
533         unsigned long modtime;
534         unsigned long long actual;
535         int err;
536
537         err = os_file_modtime(file, &modtime);
538         if (err < 0) {
539                 printk(KERN_ERR "Failed to get modification time of backing "
540                        "file \"%s\", err = %d\n", file, -err);
541                 return err;
542         }
543
544         err = os_file_size(file, &actual);
545         if (err < 0) {
546                 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
547                        "err = %d\n", file, -err);
548                 return err;
549         }
550
551         if (actual != size) {
552                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
553                  * the typecast.*/
554                 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
555                        "vs backing file\n", (unsigned long long) size, actual);
556                 return -EINVAL;
557         }
558         if (modtime != mtime) {
559                 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
560                        "backing file\n", mtime, modtime);
561                 return -EINVAL;
562         }
563         return 0;
564 }
565
566 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
567 {
568         struct uml_stat buf1, buf2;
569         int err;
570
571         if (from_cmdline == NULL)
572                 return 0;
573         if (!strcmp(from_cmdline, from_cow))
574                 return 0;
575
576         err = os_stat_file(from_cmdline, &buf1);
577         if (err < 0) {
578                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
579                        -err);
580                 return 0;
581         }
582         err = os_stat_file(from_cow, &buf2);
583         if (err < 0) {
584                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
585                        -err);
586                 return 1;
587         }
588         if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
589                 return 0;
590
591         printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
592                "\"%s\" specified in COW header of \"%s\"\n",
593                from_cmdline, from_cow, cow);
594         return 1;
595 }
596
597 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
598                   char **backing_file_out, int *bitmap_offset_out,
599                   unsigned long *bitmap_len_out, int *data_offset_out,
600                   int *create_cow_out)
601 {
602         time_t mtime;
603         unsigned long long size;
604         __u32 version, align;
605         char *backing_file;
606         int fd, err, sectorsize, asked_switch, mode = 0644;
607
608         fd = os_open_file(file, *openflags, mode);
609         if (fd < 0) {
610                 if ((fd == -ENOENT) && (create_cow_out != NULL))
611                         *create_cow_out = 1;
612                 if (!openflags->w ||
613                     ((fd != -EROFS) && (fd != -EACCES)))
614                         return fd;
615                 openflags->w = 0;
616                 fd = os_open_file(file, *openflags, mode);
617                 if (fd < 0)
618                         return fd;
619         }
620
621         if (shared)
622                 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
623         else {
624                 err = os_lock_file(fd, openflags->w);
625                 if (err < 0) {
626                         printk(KERN_ERR "Failed to lock '%s', err = %d\n",
627                                file, -err);
628                         goto out_close;
629                 }
630         }
631
632         /* Successful return case! */
633         if (backing_file_out == NULL)
634                 return fd;
635
636         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
637                               &size, &sectorsize, &align, bitmap_offset_out);
638         if (err && (*backing_file_out != NULL)) {
639                 printk(KERN_ERR "Failed to read COW header from COW file "
640                        "\"%s\", errno = %d\n", file, -err);
641                 goto out_close;
642         }
643         if (err)
644                 return fd;
645
646         asked_switch = path_requires_switch(*backing_file_out, backing_file,
647                                             file);
648
649         /* Allow switching only if no mismatch. */
650         if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
651                                                    mtime)) {
652                 printk(KERN_ERR "Switching backing file to '%s'\n",
653                        *backing_file_out);
654                 err = write_cow_header(file, fd, *backing_file_out,
655                                        sectorsize, align, &size);
656                 if (err) {
657                         printk(KERN_ERR "Switch failed, errno = %d\n", -err);
658                         goto out_close;
659                 }
660         } else {
661                 *backing_file_out = backing_file;
662                 err = backing_file_mismatch(*backing_file_out, size, mtime);
663                 if (err)
664                         goto out_close;
665         }
666
667         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
668                   bitmap_len_out, data_offset_out);
669
670         return fd;
671  out_close:
672         os_close_file(fd);
673         return err;
674 }
675
676 static int create_cow_file(char *cow_file, char *backing_file,
677                     struct openflags flags,
678                     int sectorsize, int alignment, int *bitmap_offset_out,
679                     unsigned long *bitmap_len_out, int *data_offset_out)
680 {
681         int err, fd;
682
683         flags.c = 1;
684         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
685         if (fd < 0) {
686                 err = fd;
687                 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
688                        cow_file, -err);
689                 goto out;
690         }
691
692         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
693                             bitmap_offset_out, bitmap_len_out,
694                             data_offset_out);
695         if (!err)
696                 return fd;
697         os_close_file(fd);
698  out:
699         return err;
700 }
701
702 static void ubd_close_dev(struct ubd *ubd_dev)
703 {
704         os_close_file(ubd_dev->fd);
705         if(ubd_dev->cow.file == NULL)
706                 return;
707
708         os_close_file(ubd_dev->cow.fd);
709         vfree(ubd_dev->cow.bitmap);
710         ubd_dev->cow.bitmap = NULL;
711 }
712
713 static int ubd_open_dev(struct ubd *ubd_dev)
714 {
715         struct openflags flags;
716         char **back_ptr;
717         int err, create_cow, *create_ptr;
718         int fd;
719
720         ubd_dev->openflags = ubd_dev->boot_openflags;
721         create_cow = 0;
722         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
723         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
724
725         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
726                                 back_ptr, &ubd_dev->cow.bitmap_offset,
727                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
728                                 create_ptr);
729
730         if((fd == -ENOENT) && create_cow){
731                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
732                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
733                                           &ubd_dev->cow.bitmap_offset,
734                                           &ubd_dev->cow.bitmap_len,
735                                           &ubd_dev->cow.data_offset);
736                 if(fd >= 0){
737                         printk(KERN_INFO "Creating \"%s\" as COW file for "
738                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
739                 }
740         }
741
742         if(fd < 0){
743                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
744                        -fd);
745                 return fd;
746         }
747         ubd_dev->fd = fd;
748
749         if(ubd_dev->cow.file != NULL){
750                 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
751
752                 err = -ENOMEM;
753                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
754                 if(ubd_dev->cow.bitmap == NULL){
755                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
756                         goto error;
757                 }
758                 flush_tlb_kernel_vm();
759
760                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
761                                       ubd_dev->cow.bitmap_offset,
762                                       ubd_dev->cow.bitmap_len);
763                 if(err < 0)
764                         goto error;
765
766                 flags = ubd_dev->openflags;
767                 flags.w = 0;
768                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
769                                     NULL, NULL, NULL, NULL);
770                 if(err < 0) goto error;
771                 ubd_dev->cow.fd = err;
772         }
773         return 0;
774  error:
775         os_close_file(ubd_dev->fd);
776         return err;
777 }
778
779 static void ubd_device_release(struct device *dev)
780 {
781         struct ubd *ubd_dev = dev_get_drvdata(dev);
782
783         blk_cleanup_queue(ubd_dev->queue);
784         *ubd_dev = ((struct ubd) DEFAULT_UBD);
785 }
786
787 static int ubd_disk_register(int major, u64 size, int unit,
788                              struct gendisk **disk_out)
789 {
790         struct gendisk *disk;
791
792         disk = alloc_disk(1 << UBD_SHIFT);
793         if(disk == NULL)
794                 return -ENOMEM;
795
796         disk->major = major;
797         disk->first_minor = unit << UBD_SHIFT;
798         disk->fops = &ubd_blops;
799         set_capacity(disk, size / 512);
800         if (major == UBD_MAJOR)
801                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
802         else
803                 sprintf(disk->disk_name, "ubd_fake%d", unit);
804
805         /* sysfs register (not for ide fake devices) */
806         if (major == UBD_MAJOR) {
807                 ubd_devs[unit].pdev.id   = unit;
808                 ubd_devs[unit].pdev.name = DRIVER_NAME;
809                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
810                 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
811                 platform_device_register(&ubd_devs[unit].pdev);
812                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
813         }
814
815         disk->private_data = &ubd_devs[unit];
816         disk->queue = ubd_devs[unit].queue;
817         add_disk(disk);
818
819         *disk_out = disk;
820         return 0;
821 }
822
823 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
824
825 static int ubd_add(int n, char **error_out)
826 {
827         struct ubd *ubd_dev = &ubd_devs[n];
828         int err = 0;
829
830         if(ubd_dev->file == NULL)
831                 goto out;
832
833         err = ubd_file_size(ubd_dev, &ubd_dev->size);
834         if(err < 0){
835                 *error_out = "Couldn't determine size of device's file";
836                 goto out;
837         }
838
839         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
840
841         INIT_LIST_HEAD(&ubd_dev->restart);
842         sg_init_table(ubd_dev->sg, MAX_SG);
843
844         err = -ENOMEM;
845         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
846         if (ubd_dev->queue == NULL) {
847                 *error_out = "Failed to initialize device queue";
848                 goto out;
849         }
850         ubd_dev->queue->queuedata = ubd_dev;
851
852         blk_queue_max_segments(ubd_dev->queue, MAX_SG);
853         err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
854         if(err){
855                 *error_out = "Failed to register device";
856                 goto out_cleanup;
857         }
858
859         if (fake_major != UBD_MAJOR)
860                 ubd_disk_register(fake_major, ubd_dev->size, n,
861                                   &fake_gendisk[n]);
862
863         /*
864          * Perhaps this should also be under the "if (fake_major)" above
865          * using the fake_disk->disk_name
866          */
867         if (fake_ide)
868                 make_ide_entries(ubd_gendisk[n]->disk_name);
869
870         err = 0;
871 out:
872         return err;
873
874 out_cleanup:
875         blk_cleanup_queue(ubd_dev->queue);
876         goto out;
877 }
878
879 static int ubd_config(char *str, char **error_out)
880 {
881         int n, ret;
882
883         /* This string is possibly broken up and stored, so it's only
884          * freed if ubd_setup_common fails, or if only general options
885          * were set.
886          */
887         str = kstrdup(str, GFP_KERNEL);
888         if (str == NULL) {
889                 *error_out = "Failed to allocate memory";
890                 return -ENOMEM;
891         }
892
893         ret = ubd_setup_common(str, &n, error_out);
894         if (ret)
895                 goto err_free;
896
897         if (n == -1) {
898                 ret = 0;
899                 goto err_free;
900         }
901
902         mutex_lock(&ubd_lock);
903         ret = ubd_add(n, error_out);
904         if (ret)
905                 ubd_devs[n].file = NULL;
906         mutex_unlock(&ubd_lock);
907
908 out:
909         return ret;
910
911 err_free:
912         kfree(str);
913         goto out;
914 }
915
916 static int ubd_get_config(char *name, char *str, int size, char **error_out)
917 {
918         struct ubd *ubd_dev;
919         int n, len = 0;
920
921         n = parse_unit(&name);
922         if((n >= MAX_DEV) || (n < 0)){
923                 *error_out = "ubd_get_config : device number out of range";
924                 return -1;
925         }
926
927         ubd_dev = &ubd_devs[n];
928         mutex_lock(&ubd_lock);
929
930         if(ubd_dev->file == NULL){
931                 CONFIG_CHUNK(str, size, len, "", 1);
932                 goto out;
933         }
934
935         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
936
937         if(ubd_dev->cow.file != NULL){
938                 CONFIG_CHUNK(str, size, len, ",", 0);
939                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
940         }
941         else CONFIG_CHUNK(str, size, len, "", 1);
942
943  out:
944         mutex_unlock(&ubd_lock);
945         return len;
946 }
947
948 static int ubd_id(char **str, int *start_out, int *end_out)
949 {
950         int n;
951
952         n = parse_unit(str);
953         *start_out = 0;
954         *end_out = MAX_DEV - 1;
955         return n;
956 }
957
958 static int ubd_remove(int n, char **error_out)
959 {
960         struct gendisk *disk = ubd_gendisk[n];
961         struct ubd *ubd_dev;
962         int err = -ENODEV;
963
964         mutex_lock(&ubd_lock);
965
966         ubd_dev = &ubd_devs[n];
967
968         if(ubd_dev->file == NULL)
969                 goto out;
970
971         /* you cannot remove a open disk */
972         err = -EBUSY;
973         if(ubd_dev->count > 0)
974                 goto out;
975
976         ubd_gendisk[n] = NULL;
977         if(disk != NULL){
978                 del_gendisk(disk);
979                 put_disk(disk);
980         }
981
982         if(fake_gendisk[n] != NULL){
983                 del_gendisk(fake_gendisk[n]);
984                 put_disk(fake_gendisk[n]);
985                 fake_gendisk[n] = NULL;
986         }
987
988         err = 0;
989         platform_device_unregister(&ubd_dev->pdev);
990 out:
991         mutex_unlock(&ubd_lock);
992         return err;
993 }
994
995 /* All these are called by mconsole in process context and without
996  * ubd-specific locks.  The structure itself is const except for .list.
997  */
998 static struct mc_device ubd_mc = {
999         .list           = LIST_HEAD_INIT(ubd_mc.list),
1000         .name           = "ubd",
1001         .config         = ubd_config,
1002         .get_config     = ubd_get_config,
1003         .id             = ubd_id,
1004         .remove         = ubd_remove,
1005 };
1006
1007 static int __init ubd_mc_init(void)
1008 {
1009         mconsole_register_dev(&ubd_mc);
1010         return 0;
1011 }
1012
1013 __initcall(ubd_mc_init);
1014
1015 static int __init ubd0_init(void)
1016 {
1017         struct ubd *ubd_dev = &ubd_devs[0];
1018
1019         mutex_lock(&ubd_lock);
1020         if(ubd_dev->file == NULL)
1021                 ubd_dev->file = "root_fs";
1022         mutex_unlock(&ubd_lock);
1023
1024         return 0;
1025 }
1026
1027 __initcall(ubd0_init);
1028
1029 /* Used in ubd_init, which is an initcall */
1030 static struct platform_driver ubd_driver = {
1031         .driver = {
1032                 .name  = DRIVER_NAME,
1033         },
1034 };
1035
1036 static int __init ubd_init(void)
1037 {
1038         char *error;
1039         int i, err;
1040
1041         if (register_blkdev(UBD_MAJOR, "ubd"))
1042                 return -1;
1043
1044         if (fake_major != UBD_MAJOR) {
1045                 char name[sizeof("ubd_nnn\0")];
1046
1047                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1048                 if (register_blkdev(fake_major, "ubd"))
1049                         return -1;
1050         }
1051         platform_driver_register(&ubd_driver);
1052         mutex_lock(&ubd_lock);
1053         for (i = 0; i < MAX_DEV; i++){
1054                 err = ubd_add(i, &error);
1055                 if(err)
1056                         printk(KERN_ERR "Failed to initialize ubd device %d :"
1057                                "%s\n", i, error);
1058         }
1059         mutex_unlock(&ubd_lock);
1060         return 0;
1061 }
1062
1063 late_initcall(ubd_init);
1064
1065 static int __init ubd_driver_init(void){
1066         unsigned long stack;
1067         int err;
1068
1069         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1070         if(global_openflags.s){
1071                 printk(KERN_INFO "ubd: Synchronous mode\n");
1072                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1073                  * enough. So use anyway the io thread. */
1074         }
1075         stack = alloc_stack(0, 0);
1076         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1077                                  &thread_fd);
1078         if(io_pid < 0){
1079                 printk(KERN_ERR
1080                        "ubd : Failed to start I/O thread (errno = %d) - "
1081                        "falling back to synchronous I/O\n", -io_pid);
1082                 io_pid = -1;
1083                 return 0;
1084         }
1085         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1086                              IRQF_DISABLED, "ubd", ubd_devs);
1087         if(err != 0)
1088                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1089         return 0;
1090 }
1091
1092 device_initcall(ubd_driver_init);
1093
1094 static int ubd_open(struct block_device *bdev, fmode_t mode)
1095 {
1096         struct gendisk *disk = bdev->bd_disk;
1097         struct ubd *ubd_dev = disk->private_data;
1098         int err = 0;
1099
1100         if(ubd_dev->count == 0){
1101                 err = ubd_open_dev(ubd_dev);
1102                 if(err){
1103                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1104                                disk->disk_name, ubd_dev->file, -err);
1105                         goto out;
1106                 }
1107         }
1108         ubd_dev->count++;
1109         set_disk_ro(disk, !ubd_dev->openflags.w);
1110
1111         /* This should no more be needed. And it didn't work anyway to exclude
1112          * read-write remounting of filesystems.*/
1113         /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1114                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1115                 err = -EROFS;
1116         }*/
1117  out:
1118         return err;
1119 }
1120
1121 static int ubd_release(struct gendisk *disk, fmode_t mode)
1122 {
1123         struct ubd *ubd_dev = disk->private_data;
1124
1125         if(--ubd_dev->count == 0)
1126                 ubd_close_dev(ubd_dev);
1127         return 0;
1128 }
1129
1130 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1131                           __u64 *cow_offset, unsigned long *bitmap,
1132                           __u64 bitmap_offset, unsigned long *bitmap_words,
1133                           __u64 bitmap_len)
1134 {
1135         __u64 sector = io_offset >> 9;
1136         int i, update_bitmap = 0;
1137
1138         for(i = 0; i < length >> 9; i++){
1139                 if(cow_mask != NULL)
1140                         ubd_set_bit(i, (unsigned char *) cow_mask);
1141                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1142                         continue;
1143
1144                 update_bitmap = 1;
1145                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1146         }
1147
1148         if(!update_bitmap)
1149                 return;
1150
1151         *cow_offset = sector / (sizeof(unsigned long) * 8);
1152
1153         /* This takes care of the case where we're exactly at the end of the
1154          * device, and *cow_offset + 1 is off the end.  So, just back it up
1155          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1156          * for the original diagnosis.
1157          */
1158         if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1159                                          sizeof(unsigned long)) - 1))
1160                 (*cow_offset)--;
1161
1162         bitmap_words[0] = bitmap[*cow_offset];
1163         bitmap_words[1] = bitmap[*cow_offset + 1];
1164
1165         *cow_offset *= sizeof(unsigned long);
1166         *cow_offset += bitmap_offset;
1167 }
1168
1169 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1170                        __u64 bitmap_offset, __u64 bitmap_len)
1171 {
1172         __u64 sector = req->offset >> 9;
1173         int i;
1174
1175         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1176                 panic("Operation too long");
1177
1178         if(req->op == UBD_READ) {
1179                 for(i = 0; i < req->length >> 9; i++){
1180                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1181                                 ubd_set_bit(i, (unsigned char *)
1182                                             &req->sector_mask);
1183                 }
1184         }
1185         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1186                            &req->cow_offset, bitmap, bitmap_offset,
1187                            req->bitmap_words, bitmap_len);
1188 }
1189
1190 /* Called with dev->lock held */
1191 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1192                             unsigned long long offset, int page_offset,
1193                             int len, struct page *page)
1194 {
1195         struct gendisk *disk = req->rq_disk;
1196         struct ubd *ubd_dev = disk->private_data;
1197
1198         io_req->req = req;
1199         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1200                 ubd_dev->fd;
1201         io_req->fds[1] = ubd_dev->fd;
1202         io_req->cow_offset = -1;
1203         io_req->offset = offset;
1204         io_req->length = len;
1205         io_req->error = 0;
1206         io_req->sector_mask = 0;
1207
1208         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1209         io_req->offsets[0] = 0;
1210         io_req->offsets[1] = ubd_dev->cow.data_offset;
1211         io_req->buffer = page_address(page) + page_offset;
1212         io_req->sectorsize = 1 << 9;
1213
1214         if(ubd_dev->cow.file != NULL)
1215                 cowify_req(io_req, ubd_dev->cow.bitmap,
1216                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1217
1218 }
1219
1220 /* Called with dev->lock held */
1221 static void do_ubd_request(struct request_queue *q)
1222 {
1223         struct io_thread_req *io_req;
1224         struct request *req;
1225         sector_t sector;
1226         int n;
1227
1228         while(1){
1229                 struct ubd *dev = q->queuedata;
1230                 if(dev->end_sg == 0){
1231                         struct request *req = blk_fetch_request(q);
1232                         if(req == NULL)
1233                                 return;
1234
1235                         dev->request = req;
1236                         dev->start_sg = 0;
1237                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1238                 }
1239
1240                 req = dev->request;
1241                 sector = blk_rq_pos(req);
1242                 while(dev->start_sg < dev->end_sg){
1243                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1244
1245                         io_req = kmalloc(sizeof(struct io_thread_req),
1246                                          GFP_ATOMIC);
1247                         if(io_req == NULL){
1248                                 if(list_empty(&dev->restart))
1249                                         list_add(&dev->restart, &restart);
1250                                 return;
1251                         }
1252                         prepare_request(req, io_req,
1253                                         (unsigned long long)sector << 9,
1254                                         sg->offset, sg->length, sg_page(sg));
1255
1256                         sector += sg->length >> 9;
1257                         n = os_write_file(thread_fd, &io_req,
1258                                           sizeof(struct io_thread_req *));
1259                         if(n != sizeof(struct io_thread_req *)){
1260                                 if(n != -EAGAIN)
1261                                         printk("write to io thread failed, "
1262                                                "errno = %d\n", -n);
1263                                 else if(list_empty(&dev->restart))
1264                                         list_add(&dev->restart, &restart);
1265                                 kfree(io_req);
1266                                 return;
1267                         }
1268
1269                         dev->start_sg++;
1270                 }
1271                 dev->end_sg = 0;
1272                 dev->request = NULL;
1273         }
1274 }
1275
1276 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1277 {
1278         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1279
1280         geo->heads = 128;
1281         geo->sectors = 32;
1282         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1283         return 0;
1284 }
1285
1286 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1287                      unsigned int cmd, unsigned long arg)
1288 {
1289         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1290         u16 ubd_id[ATA_ID_WORDS];
1291
1292         switch (cmd) {
1293                 struct cdrom_volctrl volume;
1294         case HDIO_GET_IDENTITY:
1295                 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1296                 ubd_id[ATA_ID_CYLS]     = ubd_dev->size / (128 * 32 * 512);
1297                 ubd_id[ATA_ID_HEADS]    = 128;
1298                 ubd_id[ATA_ID_SECTORS]  = 32;
1299                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1300                                  sizeof(ubd_id)))
1301                         return -EFAULT;
1302                 return 0;
1303
1304         case CDROMVOLREAD:
1305                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1306                         return -EFAULT;
1307                 volume.channel0 = 255;
1308                 volume.channel1 = 255;
1309                 volume.channel2 = 255;
1310                 volume.channel3 = 255;
1311                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1312                         return -EFAULT;
1313                 return 0;
1314         }
1315         return -EINVAL;
1316 }
1317
1318 static int update_bitmap(struct io_thread_req *req)
1319 {
1320         int n;
1321
1322         if(req->cow_offset == -1)
1323                 return 0;
1324
1325         n = os_seek_file(req->fds[1], req->cow_offset);
1326         if(n < 0){
1327                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1328                 return 1;
1329         }
1330
1331         n = os_write_file(req->fds[1], &req->bitmap_words,
1332                           sizeof(req->bitmap_words));
1333         if(n != sizeof(req->bitmap_words)){
1334                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1335                        req->fds[1]);
1336                 return 1;
1337         }
1338
1339         return 0;
1340 }
1341
1342 static void do_io(struct io_thread_req *req)
1343 {
1344         char *buf;
1345         unsigned long len;
1346         int n, nsectors, start, end, bit;
1347         int err;
1348         __u64 off;
1349
1350         nsectors = req->length / req->sectorsize;
1351         start = 0;
1352         do {
1353                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1354                 end = start;
1355                 while((end < nsectors) &&
1356                       (ubd_test_bit(end, (unsigned char *)
1357                                     &req->sector_mask) == bit))
1358                         end++;
1359
1360                 off = req->offset + req->offsets[bit] +
1361                         start * req->sectorsize;
1362                 len = (end - start) * req->sectorsize;
1363                 buf = &req->buffer[start * req->sectorsize];
1364
1365                 err = os_seek_file(req->fds[bit], off);
1366                 if(err < 0){
1367                         printk("do_io - lseek failed : err = %d\n", -err);
1368                         req->error = 1;
1369                         return;
1370                 }
1371                 if(req->op == UBD_READ){
1372                         n = 0;
1373                         do {
1374                                 buf = &buf[n];
1375                                 len -= n;
1376                                 n = os_read_file(req->fds[bit], buf, len);
1377                                 if (n < 0) {
1378                                         printk("do_io - read failed, err = %d "
1379                                                "fd = %d\n", -n, req->fds[bit]);
1380                                         req->error = 1;
1381                                         return;
1382                                 }
1383                         } while((n < len) && (n != 0));
1384                         if (n < len) memset(&buf[n], 0, len - n);
1385                 } else {
1386                         n = os_write_file(req->fds[bit], buf, len);
1387                         if(n != len){
1388                                 printk("do_io - write failed err = %d "
1389                                        "fd = %d\n", -n, req->fds[bit]);
1390                                 req->error = 1;
1391                                 return;
1392                         }
1393                 }
1394
1395                 start = end;
1396         } while(start < nsectors);
1397
1398         req->error = update_bitmap(req);
1399 }
1400
1401 /* Changed in start_io_thread, which is serialized by being called only
1402  * from ubd_init, which is an initcall.
1403  */
1404 int kernel_fd = -1;
1405
1406 /* Only changed by the io thread. XXX: currently unused. */
1407 static int io_count = 0;
1408
1409 int io_thread(void *arg)
1410 {
1411         struct io_thread_req *req;
1412         int n;
1413
1414         ignore_sigwinch_sig();
1415         while(1){
1416                 n = os_read_file(kernel_fd, &req,
1417                                  sizeof(struct io_thread_req *));
1418                 if(n != sizeof(struct io_thread_req *)){
1419                         if(n < 0)
1420                                 printk("io_thread - read failed, fd = %d, "
1421                                        "err = %d\n", kernel_fd, -n);
1422                         else {
1423                                 printk("io_thread - short read, fd = %d, "
1424                                        "length = %d\n", kernel_fd, n);
1425                         }
1426                         continue;
1427                 }
1428                 io_count++;
1429                 do_io(req);
1430                 n = os_write_file(kernel_fd, &req,
1431                                   sizeof(struct io_thread_req *));
1432                 if(n != sizeof(struct io_thread_req *))
1433                         printk("io_thread - write failed, fd = %d, err = %d\n",
1434                                kernel_fd, -n);
1435         }
1436
1437         return 0;
1438 }