git.oblomov.eu Git - linux-2.6/blob - arch/um/drivers/ubd_kern.c

   1 /*
   2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
   3  * Licensed under the GPL
   4  */
   5
   6 /* 2001-09-28...2002-04-17
   7  * Partition stuff by James_McMechan@hotmail.com
   8  * old style ubd by setting UBD_SHIFT to 0
   9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
  10  * partitions have changed in 2.5
  11  * 2003-01-29 more tinkering for 2.5.59-1
  12  * This should now address the sysfs problems and has
  13  * the symlink for devfs to allow for booting with
  14  * the common /dev/ubd/discX/... names rather than
  15  * only /dev/ubdN/discN this version also has lots of
  16  * clean ups preparing for ubd-many.
  17  * James McMechan
  18  */
  19
  20 #define MAJOR_NR UBD_MAJOR
  21 #define UBD_SHIFT 4
  22
  23 #include "linux/kernel.h"
  24 #include "linux/module.h"
  25 #include "linux/blkdev.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/ctype.h"
  31 #include "linux/capability.h"
  32 #include "linux/mm.h"
  33 #include "linux/vmalloc.h"
  34 #include "linux/blkpg.h"
  35 #include "linux/genhd.h"
  36 #include "linux/spinlock.h"
  37 #include "linux/platform_device.h"
  38 #include "asm/segment.h"
  39 #include "asm/uaccess.h"
  40 #include "asm/irq.h"
  41 #include "asm/types.h"
  42 #include "asm/tlbflush.h"
  43 #include "mem_user.h"
  44 #include "kern_util.h"
  45 #include "kern.h"
  46 #include "mconsole_kern.h"
  47 #include "init.h"
  48 #include "irq_user.h"
  49 #include "irq_kern.h"
  50 #include "ubd_user.h"
  51 #include "os.h"
  52 #include "mem.h"
  53 #include "mem_kern.h"
  54 #include "cow.h"
  55
  56 enum ubd_req { UBD_READ, UBD_WRITE };
  57
  58 struct io_thread_req {
  59         struct request *req;
  60         enum ubd_req op;
  61         int fds[2];
  62         unsigned long offsets[2];
  63         unsigned long long offset;
  64         unsigned long length;
  65         char *buffer;
  66         int sectorsize;
  67         unsigned long sector_mask;
  68         unsigned long long cow_offset;
  69         unsigned long bitmap_words[2];
  70         int error;
  71 };
  72
  73 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
  74                          char **backing_file_out, int *bitmap_offset_out,
  75                          unsigned long *bitmap_len_out, int *data_offset_out,
  76                          int *create_cow_out);
  77 extern int create_cow_file(char *cow_file, char *backing_file,
  78                            struct openflags flags, int sectorsize,
  79                            int alignment, int *bitmap_offset_out,
  80                            unsigned long *bitmap_len_out,
  81                            int *data_offset_out);
  82 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
  83 extern void do_io(struct io_thread_req *req);
  84
  85 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  86 {
  87         __u64 n;
  88         int bits, off;
  89
  90         bits = sizeof(data[0]) * 8;
  91         n = bit / bits;
  92         off = bit % bits;
  93         return (data[n] & (1 << off)) != 0;
  94 }
  95
  96 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  97 {
  98         __u64 n;
  99         int bits, off;
 100
 101         bits = sizeof(data[0]) * 8;
 102         n = bit / bits;
 103         off = bit % bits;
 104         data[n] |= (1 << off);
 105 }
 106 /*End stuff from ubd_user.h*/
 107
 108 #define DRIVER_NAME "uml-blkdev"
 109
 110 static DEFINE_MUTEX(ubd_lock);
 111
 112 static int ubd_open(struct inode * inode, struct file * filp);
 113 static int ubd_release(struct inode * inode, struct file * file);
 114 static int ubd_ioctl(struct inode * inode, struct file * file,
 115                      unsigned int cmd, unsigned long arg);
 116 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 117
 118 #define MAX_DEV (16)
 119
 120 static struct block_device_operations ubd_blops = {
 121         .owner          = THIS_MODULE,
 122         .open           = ubd_open,
 123         .release        = ubd_release,
 124         .ioctl          = ubd_ioctl,
 125         .getgeo         = ubd_getgeo,
 126 };
 127
 128 /* Protected by ubd_lock */
 129 static int fake_major = MAJOR_NR;
 130 static struct gendisk *ubd_gendisk[MAX_DEV];
 131 static struct gendisk *fake_gendisk[MAX_DEV];
 132
 133 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 134 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 135                                          .cl = 1 })
 136 #else
 137 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 138                                          .cl = 1 })
 139 #endif
 140 static struct openflags global_openflags = OPEN_FLAGS;
 141
 142 struct cow {
 143         /* backing file name */
 144         char *file;
 145         /* backing file fd */
 146         int fd;
 147         unsigned long *bitmap;
 148         unsigned long bitmap_len;
 149         int bitmap_offset;
 150         int data_offset;
 151 };
 152
 153 #define MAX_SG 64
 154
 155 struct ubd {
 156         struct list_head restart;
 157         /* name (and fd, below) of the file opened for writing, either the
 158          * backing or the cow file. */
 159         char *file;
 160         int count;
 161         int fd;
 162         __u64 size;
 163         struct openflags boot_openflags;
 164         struct openflags openflags;
 165         unsigned shared:1;
 166         unsigned no_cow:1;
 167         struct cow cow;
 168         struct platform_device pdev;
 169         struct request_queue *queue;
 170         spinlock_t lock;
 171         struct scatterlist sg[MAX_SG];
 172         struct request *request;
 173         int start_sg, end_sg;
 174 };
 175
 176 #define DEFAULT_COW { \
 177         .file =                 NULL, \
 178         .fd =                   -1,     \
 179         .bitmap =               NULL, \
 180         .bitmap_offset =        0, \
 181         .data_offset =          0, \
 182 }
 183
 184 #define DEFAULT_UBD { \
 185         .file =                 NULL, \
 186         .count =                0, \
 187         .fd =                   -1, \
 188         .size =                 -1, \
 189         .boot_openflags =       OPEN_FLAGS, \
 190         .openflags =            OPEN_FLAGS, \
 191         .no_cow =               0, \
 192         .shared =               0, \
 193         .cow =                  DEFAULT_COW, \
 194         .lock =                 SPIN_LOCK_UNLOCKED,     \
 195         .request =              NULL, \
 196         .start_sg =             0, \
 197         .end_sg =               0, \
 198 }
 199
 200 /* Protected by ubd_lock */
 201 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
 202
 203 /* Only changed by fake_ide_setup which is a setup */
 204 static int fake_ide = 0;
 205 static struct proc_dir_entry *proc_ide_root = NULL;
 206 static struct proc_dir_entry *proc_ide = NULL;
 207
 208 static void make_proc_ide(void)
 209 {
 210         proc_ide_root = proc_mkdir("ide", NULL);
 211         proc_ide = proc_mkdir("ide0", proc_ide_root);
 212 }
 213
 214 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
 215                                int *eof, void *data)
 216 {
 217         int len;
 218
 219         strcpy(page, "disk\n");
 220         len = strlen("disk\n");
 221         len -= off;
 222         if (len < count){
 223                 *eof = 1;
 224                 if (len <= 0) return 0;
 225         }
 226         else len = count;
 227         *start = page + off;
 228         return len;
 229 }
 230
 231 static void make_ide_entries(char *dev_name)
 232 {
 233         struct proc_dir_entry *dir, *ent;
 234         char name[64];
 235
 236         if(proc_ide_root == NULL) make_proc_ide();
 237
 238         dir = proc_mkdir(dev_name, proc_ide);
 239         if(!dir) return;
 240
 241         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
 242         if(!ent) return;
 243         ent->data = NULL;
 244         ent->read_proc = proc_ide_read_media;
 245         ent->write_proc = NULL;
 246         sprintf(name,"ide0/%s", dev_name);
 247         proc_symlink(dev_name, proc_ide_root, name);
 248 }
 249
 250 static int fake_ide_setup(char *str)
 251 {
 252         fake_ide = 1;
 253         return 1;
 254 }
 255
 256 __setup("fake_ide", fake_ide_setup);
 257
 258 __uml_help(fake_ide_setup,
 259 "fake_ide\n"
 260 "    Create ide0 entries that map onto ubd devices.\n\n"
 261 );
 262
 263 static int parse_unit(char **ptr)
 264 {
 265         char *str = *ptr, *end;
 266         int n = -1;
 267
 268         if(isdigit(*str)) {
 269                 n = simple_strtoul(str, &end, 0);
 270                 if(end == str)
 271                         return -1;
 272                 *ptr = end;
 273         }
 274         else if (('a' <= *str) && (*str <= 'z')) {
 275                 n = *str - 'a';
 276                 str++;
 277                 *ptr = str;
 278         }
 279         return n;
 280 }
 281
 282 /* If *index_out == -1 at exit, the passed option was a general one;
 283  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 284  * should not be freed on exit.
 285  */
 286 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 287 {
 288         struct ubd *ubd_dev;
 289         struct openflags flags = global_openflags;
 290         char *backing_file;
 291         int n, err = 0, i;
 292
 293         if(index_out) *index_out = -1;
 294         n = *str;
 295         if(n == '='){
 296                 char *end;
 297                 int major;
 298
 299                 str++;
 300                 if(!strcmp(str, "sync")){
 301                         global_openflags = of_sync(global_openflags);
 302                         goto out1;
 303                 }
 304
 305                 err = -EINVAL;
 306                 major = simple_strtoul(str, &end, 0);
 307                 if((*end != '\0') || (end == str)){
 308                         *error_out = "Didn't parse major number";
 309                         goto out1;
 310                 }
 311
 312                 mutex_lock(&ubd_lock);
 313                 if(fake_major != MAJOR_NR){
 314                         *error_out = "Can't assign a fake major twice";
 315                         goto out1;
 316                 }
 317
 318                 fake_major = major;
 319
 320                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 321                        major);
 322                 err = 0;
 323         out1:
 324                 mutex_unlock(&ubd_lock);
 325                 return err;
 326         }
 327
 328         n = parse_unit(&str);
 329         if(n < 0){
 330                 *error_out = "Couldn't parse device number";
 331                 return -EINVAL;
 332         }
 333         if(n >= MAX_DEV){
 334                 *error_out = "Device number out of range";
 335                 return 1;
 336         }
 337
 338         err = -EBUSY;
 339         mutex_lock(&ubd_lock);
 340
 341         ubd_dev = &ubd_devs[n];
 342         if(ubd_dev->file != NULL){
 343                 *error_out = "Device is already configured";
 344                 goto out;
 345         }
 346
 347         if (index_out)
 348                 *index_out = n;
 349
 350         err = -EINVAL;
 351         for (i = 0; i < sizeof("rscd="); i++) {
 352                 switch (*str) {
 353                 case 'r':
 354                         flags.w = 0;
 355                         break;
 356                 case 's':
 357                         flags.s = 1;
 358                         break;
 359                 case 'd':
 360                         ubd_dev->no_cow = 1;
 361                         break;
 362                 case 'c':
 363                         ubd_dev->shared = 1;
 364                         break;
 365                 case '=':
 366                         str++;
 367                         goto break_loop;
 368                 default:
 369                         *error_out = "Expected '=' or flag letter "
 370                                 "(r, s, c, or d)";
 371                         goto out;
 372                 }
 373                 str++;
 374         }
 375
 376         if (*str == '=')
 377                 *error_out = "Too many flags specified";
 378         else
 379                 *error_out = "Missing '='";
 380         goto out;
 381
 382 break_loop:
 383         backing_file = strchr(str, ',');
 384
 385         if (backing_file == NULL)
 386                 backing_file = strchr(str, ':');
 387
 388         if(backing_file != NULL){
 389                 if(ubd_dev->no_cow){
 390                         *error_out = "Can't specify both 'd' and a cow file";
 391                         goto out;
 392                 }
 393                 else {
 394                         *backing_file = '\0';
 395                         backing_file++;
 396                 }
 397         }
 398         err = 0;
 399         ubd_dev->file = str;
 400         ubd_dev->cow.file = backing_file;
 401         ubd_dev->boot_openflags = flags;
 402 out:
 403         mutex_unlock(&ubd_lock);
 404         return err;
 405 }
 406
 407 static int ubd_setup(char *str)
 408 {
 409         char *error;
 410         int err;
 411
 412         err = ubd_setup_common(str, NULL, &error);
 413         if(err)
 414                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 415                        "%s\n", str, error);
 416         return 1;
 417 }
 418
 419 __setup("ubd", ubd_setup);
 420 __uml_help(ubd_setup,
 421 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 422 "    This is used to associate a device with a file in the underlying\n"
 423 "    filesystem. When specifying two filenames, the first one is the\n"
 424 "    COW name and the second is the backing file name. As separator you can\n"
 425 "    use either a ':' or a ',': the first one allows writing things like;\n"
 426 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 427 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 428 "    When using only one filename, UML will detect whether to treat it like\n"
 429 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 430 "    flag:\n"
 431 "       ubd0d=BackingFile\n"
 432 "    Usually, there is a filesystem in the file, but \n"
 433 "    that's not required. Swap devices containing swap files can be\n"
 434 "    specified like this. Also, a file which doesn't contain a\n"
 435 "    filesystem can have its contents read in the virtual \n"
 436 "    machine by running 'dd' on the device. <n> must be in the range\n"
 437 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 438 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 439 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
 440 );
 441
 442 static int udb_setup(char *str)
 443 {
 444         printk("udb%s specified on command line is almost certainly a ubd -> "
 445                "udb TYPO\n", str);
 446         return 1;
 447 }
 448
 449 __setup("udb", udb_setup);
 450 __uml_help(udb_setup,
 451 "udb\n"
 452 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 453 "    to impossible to catch visually unless you specifically look for\n"
 454 "    them.  The only result of any option starting with 'udb' is an error\n"
 455 "    in the boot output.\n\n"
 456 );
 457
 458 static int fakehd_set = 0;
 459 static int fakehd(char *str)
 460 {
 461         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
 462         fakehd_set = 1;
 463         return 1;
 464 }
 465
 466 __setup("fakehd", fakehd);
 467 __uml_help(fakehd,
 468 "fakehd\n"
 469 "    Change the ubd device name to \"hd\".\n\n"
 470 );
 471
 472 static void do_ubd_request(request_queue_t * q);
 473
 474 /* Only changed by ubd_init, which is an initcall. */
 475 int thread_fd = -1;
 476
 477 static void ubd_end_request(struct request *req, int bytes, int uptodate)
 478 {
 479         if (!end_that_request_first(req, uptodate, bytes >> 9)) {
 480                 struct ubd *dev = req->rq_disk->private_data;
 481                 unsigned long flags;
 482
 483                 add_disk_randomness(req->rq_disk);
 484                 spin_lock_irqsave(&dev->lock, flags);
 485                 end_that_request_last(req, uptodate);
 486                 spin_unlock_irqrestore(&dev->lock, flags);
 487         }
 488 }
 489
 490 /* Callable only from interrupt context - otherwise you need to do
 491  * spin_lock_irq()/spin_lock_irqsave() */
 492 static inline void ubd_finish(struct request *req, int bytes)
 493 {
 494         if(bytes < 0){
 495                 ubd_end_request(req, 0, 0);
 496                 return;
 497         }
 498         ubd_end_request(req, bytes, 1);
 499 }
 500
 501 static LIST_HEAD(restart);
 502
 503 /* XXX - move this inside ubd_intr. */
 504 /* Called without dev->lock held, and only in interrupt context. */
 505 static void ubd_handler(void)
 506 {
 507         struct io_thread_req *req;
 508         struct request *rq;
 509         struct ubd *ubd;
 510         struct list_head *list, *next_ele;
 511         unsigned long flags;
 512         int n;
 513
 514         while(1){
 515                 n = os_read_file(thread_fd, &req,
 516                                  sizeof(struct io_thread_req *));
 517                 if(n != sizeof(req)){
 518                         if(n == -EAGAIN)
 519                                 break;
 520                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 521                                "err = %d\n", -n);
 522                         return;
 523                 }
 524
 525                 rq = req->req;
 526                 rq->nr_sectors -= req->length >> 9;
 527                 if(rq->nr_sectors == 0)
 528                         ubd_finish(rq, rq->hard_nr_sectors << 9);
 529                 kfree(req);
 530         }
 531         reactivate_fd(thread_fd, UBD_IRQ);
 532
 533         list_for_each_safe(list, next_ele, &restart){
 534                 ubd = container_of(list, struct ubd, restart);
 535                 list_del_init(&ubd->restart);
 536                 spin_lock_irqsave(&ubd->lock, flags);
 537                 do_ubd_request(ubd->queue);
 538                 spin_unlock_irqrestore(&ubd->lock, flags);
 539         }
 540 }
 541
 542 static irqreturn_t ubd_intr(int irq, void *dev)
 543 {
 544         ubd_handler();
 545         return IRQ_HANDLED;
 546 }
 547
 548 /* Only changed by ubd_init, which is an initcall. */
 549 static int io_pid = -1;
 550
 551 void kill_io_thread(void)
 552 {
 553         if(io_pid != -1)
 554                 os_kill_process(io_pid, 1);
 555 }
 556
 557 __uml_exitcall(kill_io_thread);
 558
 559 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 560 {
 561         char *file;
 562
 563         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
 564         return os_file_size(file, size_out);
 565 }
 566
 567 static void ubd_close_dev(struct ubd *ubd_dev)
 568 {
 569         os_close_file(ubd_dev->fd);
 570         if(ubd_dev->cow.file == NULL)
 571                 return;
 572
 573         os_close_file(ubd_dev->cow.fd);
 574         vfree(ubd_dev->cow.bitmap);
 575         ubd_dev->cow.bitmap = NULL;
 576 }
 577
 578 static int ubd_open_dev(struct ubd *ubd_dev)
 579 {
 580         struct openflags flags;
 581         char **back_ptr;
 582         int err, create_cow, *create_ptr;
 583         int fd;
 584
 585         ubd_dev->openflags = ubd_dev->boot_openflags;
 586         create_cow = 0;
 587         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 588         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 589
 590         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 591                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 592                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 593                                 create_ptr);
 594
 595         if((fd == -ENOENT) && create_cow){
 596                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 597                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 598                                           &ubd_dev->cow.bitmap_offset,
 599                                           &ubd_dev->cow.bitmap_len,
 600                                           &ubd_dev->cow.data_offset);
 601                 if(fd >= 0){
 602                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 603                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 604                 }
 605         }
 606
 607         if(fd < 0){
 608                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 609                        -fd);
 610                 return fd;
 611         }
 612         ubd_dev->fd = fd;
 613
 614         if(ubd_dev->cow.file != NULL){
 615                 err = -ENOMEM;
 616                 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
 617                 if(ubd_dev->cow.bitmap == NULL){
 618                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 619                         goto error;
 620                 }
 621                 flush_tlb_kernel_vm();
 622
 623                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 624                                       ubd_dev->cow.bitmap_offset,
 625                                       ubd_dev->cow.bitmap_len);
 626                 if(err < 0)
 627                         goto error;
 628
 629                 flags = ubd_dev->openflags;
 630                 flags.w = 0;
 631                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 632                                     NULL, NULL, NULL, NULL);
 633                 if(err < 0) goto error;
 634                 ubd_dev->cow.fd = err;
 635         }
 636         return 0;
 637  error:
 638         os_close_file(ubd_dev->fd);
 639         return err;
 640 }
 641
 642 static void ubd_device_release(struct device *dev)
 643 {
 644         struct ubd *ubd_dev = dev->driver_data;
 645
 646         blk_cleanup_queue(ubd_dev->queue);
 647         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 648 }
 649
 650 static int ubd_disk_register(int major, u64 size, int unit,
 651                              struct gendisk **disk_out)
 652 {
 653         struct gendisk *disk;
 654
 655         disk = alloc_disk(1 << UBD_SHIFT);
 656         if(disk == NULL)
 657                 return -ENOMEM;
 658
 659         disk->major = major;
 660         disk->first_minor = unit << UBD_SHIFT;
 661         disk->fops = &ubd_blops;
 662         set_capacity(disk, size / 512);
 663         if(major == MAJOR_NR)
 664                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 665         else
 666                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 667
 668         /* sysfs register (not for ide fake devices) */
 669         if (major == MAJOR_NR) {
 670                 ubd_devs[unit].pdev.id   = unit;
 671                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 672                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 673                 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
 674                 platform_device_register(&ubd_devs[unit].pdev);
 675                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 676         }
 677
 678         disk->private_data = &ubd_devs[unit];
 679         disk->queue = ubd_devs[unit].queue;
 680         add_disk(disk);
 681
 682         *disk_out = disk;
 683         return 0;
 684 }
 685
 686 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 687
 688 static int ubd_add(int n, char **error_out)
 689 {
 690         struct ubd *ubd_dev = &ubd_devs[n];
 691         int err = 0;
 692
 693         if(ubd_dev->file == NULL)
 694                 goto out;
 695
 696         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 697         if(err < 0){
 698                 *error_out = "Couldn't determine size of device's file";
 699                 goto out;
 700         }
 701
 702         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 703
 704         INIT_LIST_HEAD(&ubd_dev->restart);
 705
 706         err = -ENOMEM;
 707         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 708         if (ubd_dev->queue == NULL) {
 709                 *error_out = "Failed to initialize device queue";
 710                 goto out;
 711         }
 712         ubd_dev->queue->queuedata = ubd_dev;
 713
 714         blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG);
 715         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
 716         if(err){
 717                 *error_out = "Failed to register device";
 718                 goto out_cleanup;
 719         }
 720
 721         if(fake_major != MAJOR_NR)
 722                 ubd_disk_register(fake_major, ubd_dev->size, n,
 723                                   &fake_gendisk[n]);
 724
 725         /* perhaps this should also be under the "if (fake_major)" above */
 726         /* using the fake_disk->disk_name and also the fakehd_set name */
 727         if (fake_ide)
 728                 make_ide_entries(ubd_gendisk[n]->disk_name);
 729
 730         err = 0;
 731 out:
 732         return err;
 733
 734 out_cleanup:
 735         blk_cleanup_queue(ubd_dev->queue);
 736         goto out;
 737 }
 738
 739 static int ubd_config(char *str, char **error_out)
 740 {
 741         int n, ret;
 742
 743         /* This string is possibly broken up and stored, so it's only
 744          * freed if ubd_setup_common fails, or if only general options
 745          * were set.
 746          */
 747         str = kstrdup(str, GFP_KERNEL);
 748         if (str == NULL) {
 749                 *error_out = "Failed to allocate memory";
 750                 return -ENOMEM;
 751         }
 752
 753         ret = ubd_setup_common(str, &n, error_out);
 754         if (ret)
 755                 goto err_free;
 756
 757         if (n == -1) {
 758                 ret = 0;
 759                 goto err_free;
 760         }
 761
 762         mutex_lock(&ubd_lock);
 763         ret = ubd_add(n, error_out);
 764         if (ret)
 765                 ubd_devs[n].file = NULL;
 766         mutex_unlock(&ubd_lock);
 767
 768 out:
 769         return ret;
 770
 771 err_free:
 772         kfree(str);
 773         goto out;
 774 }
 775
 776 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 777 {
 778         struct ubd *ubd_dev;
 779         int n, len = 0;
 780
 781         n = parse_unit(&name);
 782         if((n >= MAX_DEV) || (n < 0)){
 783                 *error_out = "ubd_get_config : device number out of range";
 784                 return -1;
 785         }
 786
 787         ubd_dev = &ubd_devs[n];
 788         mutex_lock(&ubd_lock);
 789
 790         if(ubd_dev->file == NULL){
 791                 CONFIG_CHUNK(str, size, len, "", 1);
 792                 goto out;
 793         }
 794
 795         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 796
 797         if(ubd_dev->cow.file != NULL){
 798                 CONFIG_CHUNK(str, size, len, ",", 0);
 799                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 800         }
 801         else CONFIG_CHUNK(str, size, len, "", 1);
 802
 803  out:
 804         mutex_unlock(&ubd_lock);
 805         return len;
 806 }
 807
 808 static int ubd_id(char **str, int *start_out, int *end_out)
 809 {
 810         int n;
 811
 812         n = parse_unit(str);
 813         *start_out = 0;
 814         *end_out = MAX_DEV - 1;
 815         return n;
 816 }
 817
 818 static int ubd_remove(int n, char **error_out)
 819 {
 820         struct gendisk *disk = ubd_gendisk[n];
 821         struct ubd *ubd_dev;
 822         int err = -ENODEV;
 823
 824         mutex_lock(&ubd_lock);
 825
 826         ubd_dev = &ubd_devs[n];
 827
 828         if(ubd_dev->file == NULL)
 829                 goto out;
 830
 831         /* you cannot remove a open disk */
 832         err = -EBUSY;
 833         if(ubd_dev->count > 0)
 834                 goto out;
 835
 836         ubd_gendisk[n] = NULL;
 837         if(disk != NULL){
 838                 del_gendisk(disk);
 839                 put_disk(disk);
 840         }
 841
 842         if(fake_gendisk[n] != NULL){
 843                 del_gendisk(fake_gendisk[n]);
 844                 put_disk(fake_gendisk[n]);
 845                 fake_gendisk[n] = NULL;
 846         }
 847
 848         err = 0;
 849         platform_device_unregister(&ubd_dev->pdev);
 850 out:
 851         mutex_unlock(&ubd_lock);
 852         return err;
 853 }
 854
 855 /* All these are called by mconsole in process context and without
 856  * ubd-specific locks.  The structure itself is const except for .list.
 857  */
 858 static struct mc_device ubd_mc = {
 859         .list           = LIST_HEAD_INIT(ubd_mc.list),
 860         .name           = "ubd",
 861         .config         = ubd_config,
 862         .get_config     = ubd_get_config,
 863         .id             = ubd_id,
 864         .remove         = ubd_remove,
 865 };
 866
 867 static int __init ubd_mc_init(void)
 868 {
 869         mconsole_register_dev(&ubd_mc);
 870         return 0;
 871 }
 872
 873 __initcall(ubd_mc_init);
 874
 875 static int __init ubd0_init(void)
 876 {
 877         struct ubd *ubd_dev = &ubd_devs[0];
 878
 879         mutex_lock(&ubd_lock);
 880         if(ubd_dev->file == NULL)
 881                 ubd_dev->file = "root_fs";
 882         mutex_unlock(&ubd_lock);
 883
 884         return 0;
 885 }
 886
 887 __initcall(ubd0_init);
 888
 889 /* Used in ubd_init, which is an initcall */
 890 static struct platform_driver ubd_driver = {
 891         .driver = {
 892                 .name  = DRIVER_NAME,
 893         },
 894 };
 895
 896 static int __init ubd_init(void)
 897 {
 898         char *error;
 899         int i, err;
 900
 901         if (register_blkdev(MAJOR_NR, "ubd"))
 902                 return -1;
 903
 904         if (fake_major != MAJOR_NR) {
 905                 char name[sizeof("ubd_nnn\0")];
 906
 907                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
 908                 if (register_blkdev(fake_major, "ubd"))
 909                         return -1;
 910         }
 911         platform_driver_register(&ubd_driver);
 912         mutex_lock(&ubd_lock);
 913         for (i = 0; i < MAX_DEV; i++){
 914                 err = ubd_add(i, &error);
 915                 if(err)
 916                         printk(KERN_ERR "Failed to initialize ubd device %d :"
 917                                "%s\n", i, error);
 918         }
 919         mutex_unlock(&ubd_lock);
 920         return 0;
 921 }
 922
 923 late_initcall(ubd_init);
 924
 925 static int __init ubd_driver_init(void){
 926         unsigned long stack;
 927         int err;
 928
 929         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
 930         if(global_openflags.s){
 931                 printk(KERN_INFO "ubd: Synchronous mode\n");
 932                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
 933                  * enough. So use anyway the io thread. */
 934         }
 935         stack = alloc_stack(0, 0);
 936         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
 937                                  &thread_fd);
 938         if(io_pid < 0){
 939                 printk(KERN_ERR
 940                        "ubd : Failed to start I/O thread (errno = %d) - "
 941                        "falling back to synchronous I/O\n", -io_pid);
 942                 io_pid = -1;
 943                 return 0;
 944         }
 945         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
 946                              IRQF_DISABLED, "ubd", ubd_devs);
 947         if(err != 0)
 948                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
 949         return 0;
 950 }
 951
 952 device_initcall(ubd_driver_init);
 953
 954 static int ubd_open(struct inode *inode, struct file *filp)
 955 {
 956         struct gendisk *disk = inode->i_bdev->bd_disk;
 957         struct ubd *ubd_dev = disk->private_data;
 958         int err = 0;
 959
 960         if(ubd_dev->count == 0){
 961                 err = ubd_open_dev(ubd_dev);
 962                 if(err){
 963                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
 964                                disk->disk_name, ubd_dev->file, -err);
 965                         goto out;
 966                 }
 967         }
 968         ubd_dev->count++;
 969         set_disk_ro(disk, !ubd_dev->openflags.w);
 970
 971         /* This should no more be needed. And it didn't work anyway to exclude
 972          * read-write remounting of filesystems.*/
 973         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
 974                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
 975                 err = -EROFS;
 976         }*/
 977  out:
 978         return err;
 979 }
 980
 981 static int ubd_release(struct inode * inode, struct file * file)
 982 {
 983         struct gendisk *disk = inode->i_bdev->bd_disk;
 984         struct ubd *ubd_dev = disk->private_data;
 985
 986         if(--ubd_dev->count == 0)
 987                 ubd_close_dev(ubd_dev);
 988         return 0;
 989 }
 990
 991 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
 992                           __u64 *cow_offset, unsigned long *bitmap,
 993                           __u64 bitmap_offset, unsigned long *bitmap_words,
 994                           __u64 bitmap_len)
 995 {
 996         __u64 sector = io_offset >> 9;
 997         int i, update_bitmap = 0;
 998
 999         for(i = 0; i < length >> 9; i++){
1000                 if(cow_mask != NULL)
1001                         ubd_set_bit(i, (unsigned char *) cow_mask);
1002                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1003                         continue;
1004
1005                 update_bitmap = 1;
1006                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1007         }
1008
1009         if(!update_bitmap)
1010                 return;
1011
1012         *cow_offset = sector / (sizeof(unsigned long) * 8);
1013
1014         /* This takes care of the case where we're exactly at the end of the
1015          * device, and *cow_offset + 1 is off the end.  So, just back it up
1016          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1017          * for the original diagnosis.
1018          */
1019         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
1020                            sizeof(unsigned long) - 1))
1021                 (*cow_offset)--;
1022
1023         bitmap_words[0] = bitmap[*cow_offset];
1024         bitmap_words[1] = bitmap[*cow_offset + 1];
1025
1026         *cow_offset *= sizeof(unsigned long);
1027         *cow_offset += bitmap_offset;
1028 }
1029
1030 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1031                        __u64 bitmap_offset, __u64 bitmap_len)
1032 {
1033         __u64 sector = req->offset >> 9;
1034         int i;
1035
1036         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1037                 panic("Operation too long");
1038
1039         if(req->op == UBD_READ) {
1040                 for(i = 0; i < req->length >> 9; i++){
1041                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1042                                 ubd_set_bit(i, (unsigned char *)
1043                                             &req->sector_mask);
1044                 }
1045         }
1046         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1047                            &req->cow_offset, bitmap, bitmap_offset,
1048                            req->bitmap_words, bitmap_len);
1049 }
1050
1051 /* Called with dev->lock held */
1052 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1053                             unsigned long long offset, int page_offset,
1054                             int len, struct page *page)
1055 {
1056         struct gendisk *disk = req->rq_disk;
1057         struct ubd *ubd_dev = disk->private_data;
1058
1059         io_req->req = req;
1060         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1061                 ubd_dev->fd;
1062         io_req->fds[1] = ubd_dev->fd;
1063         io_req->cow_offset = -1;
1064         io_req->offset = offset;
1065         io_req->length = len;
1066         io_req->error = 0;
1067         io_req->sector_mask = 0;
1068
1069         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1070         io_req->offsets[0] = 0;
1071         io_req->offsets[1] = ubd_dev->cow.data_offset;
1072         io_req->buffer = page_address(page) + page_offset;
1073         io_req->sectorsize = 1 << 9;
1074
1075         if(ubd_dev->cow.file != NULL)
1076                 cowify_req(io_req, ubd_dev->cow.bitmap,
1077                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1078
1079 }
1080
1081 /* Called with dev->lock held */
1082 static void do_ubd_request(request_queue_t *q)
1083 {
1084         struct io_thread_req *io_req;
1085         struct request *req;
1086         int n;
1087
1088         while(1){
1089                 struct ubd *dev = q->queuedata;
1090                 if(dev->end_sg == 0){
1091                         struct request *req = elv_next_request(q);
1092                         if(req == NULL)
1093                                 return;
1094
1095                         dev->request = req;
1096                         blkdev_dequeue_request(req);
1097                         dev->start_sg = 0;
1098                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1099                 }
1100
1101                 req = dev->request;
1102                 while(dev->start_sg < dev->end_sg){
1103                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1104
1105                         io_req = kmalloc(sizeof(struct io_thread_req),
1106                                          GFP_ATOMIC);
1107                         if(io_req == NULL){
1108                                 if(list_empty(&dev->restart))
1109                                         list_add(&dev->restart, &restart);
1110                                 return;
1111                         }
1112                         prepare_request(req, io_req,
1113                                         (unsigned long long) req->sector << 9,
1114                                         sg->offset, sg->length, sg->page);
1115
1116                         n = os_write_file(thread_fd, &io_req,
1117                                           sizeof(struct io_thread_req *));
1118                         if(n != sizeof(struct io_thread_req *)){
1119                                 if(n != -EAGAIN)
1120                                         printk("write to io thread failed, "
1121                                                "errno = %d\n", -n);
1122                                 else if(list_empty(&dev->restart))
1123                                         list_add(&dev->restart, &restart);
1124                                 return;
1125                         }
1126
1127                         req->sector += sg->length >> 9;
1128                         dev->start_sg++;
1129                 }
1130                 dev->end_sg = 0;
1131                 dev->request = NULL;
1132         }
1133 }
1134
1135 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1136 {
1137         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1138
1139         geo->heads = 128;
1140         geo->sectors = 32;
1141         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1142         return 0;
1143 }
1144
1145 static int ubd_ioctl(struct inode * inode, struct file * file,
1146                      unsigned int cmd, unsigned long arg)
1147 {
1148         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1149         struct hd_driveid ubd_id = {
1150                 .cyls           = 0,
1151                 .heads          = 128,
1152                 .sectors        = 32,
1153         };
1154
1155         switch (cmd) {
1156                 struct cdrom_volctrl volume;
1157         case HDIO_GET_IDENTITY:
1158                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1159                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1160                                  sizeof(ubd_id)))
1161                         return -EFAULT;
1162                 return 0;
1163
1164         case CDROMVOLREAD:
1165                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1166                         return -EFAULT;
1167                 volume.channel0 = 255;
1168                 volume.channel1 = 255;
1169                 volume.channel2 = 255;
1170                 volume.channel3 = 255;
1171                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1172                         return -EFAULT;
1173                 return 0;
1174         }
1175         return -EINVAL;
1176 }
1177
1178 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1179 {
1180         struct uml_stat buf1, buf2;
1181         int err;
1182
1183         if(from_cmdline == NULL)
1184                 return 0;
1185         if(!strcmp(from_cmdline, from_cow))
1186                 return 0;
1187
1188         err = os_stat_file(from_cmdline, &buf1);
1189         if(err < 0){
1190                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1191                 return 0;
1192         }
1193         err = os_stat_file(from_cow, &buf2);
1194         if(err < 0){
1195                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1196                 return 1;
1197         }
1198         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1199                 return 0;
1200
1201         printk("Backing file mismatch - \"%s\" requested,\n"
1202                "\"%s\" specified in COW header of \"%s\"\n",
1203                from_cmdline, from_cow, cow);
1204         return 1;
1205 }
1206
1207 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1208 {
1209         unsigned long modtime;
1210         unsigned long long actual;
1211         int err;
1212
1213         err = os_file_modtime(file, &modtime);
1214         if(err < 0){
1215                 printk("Failed to get modification time of backing file "
1216                        "\"%s\", err = %d\n", file, -err);
1217                 return err;
1218         }
1219
1220         err = os_file_size(file, &actual);
1221         if(err < 0){
1222                 printk("Failed to get size of backing file \"%s\", "
1223                        "err = %d\n", file, -err);
1224                 return err;
1225         }
1226
1227         if(actual != size){
1228                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1229                  * the typecast.*/
1230                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1231                        "file\n", (unsigned long long) size, actual);
1232                 return -EINVAL;
1233         }
1234         if(modtime != mtime){
1235                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1236                        "file\n", mtime, modtime);
1237                 return -EINVAL;
1238         }
1239         return 0;
1240 }
1241
1242 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1243 {
1244         int err;
1245
1246         err = os_seek_file(fd, offset);
1247         if(err < 0)
1248                 return err;
1249
1250         err = os_read_file(fd, buf, len);
1251         if(err < 0)
1252                 return err;
1253
1254         return 0;
1255 }
1256
1257 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1258                   char **backing_file_out, int *bitmap_offset_out,
1259                   unsigned long *bitmap_len_out, int *data_offset_out,
1260                   int *create_cow_out)
1261 {
1262         time_t mtime;
1263         unsigned long long size;
1264         __u32 version, align;
1265         char *backing_file;
1266         int fd, err, sectorsize, asked_switch, mode = 0644;
1267
1268         fd = os_open_file(file, *openflags, mode);
1269         if (fd < 0) {
1270                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1271                         *create_cow_out = 1;
1272                 if (!openflags->w ||
1273                     ((fd != -EROFS) && (fd != -EACCES)))
1274                         return fd;
1275                 openflags->w = 0;
1276                 fd = os_open_file(file, *openflags, mode);
1277                 if (fd < 0)
1278                         return fd;
1279         }
1280
1281         if(shared)
1282                 printk("Not locking \"%s\" on the host\n", file);
1283         else {
1284                 err = os_lock_file(fd, openflags->w);
1285                 if(err < 0){
1286                         printk("Failed to lock '%s', err = %d\n", file, -err);
1287                         goto out_close;
1288                 }
1289         }
1290
1291         /* Successful return case! */
1292         if(backing_file_out == NULL)
1293                 return fd;
1294
1295         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1296                               &size, &sectorsize, &align, bitmap_offset_out);
1297         if(err && (*backing_file_out != NULL)){
1298                 printk("Failed to read COW header from COW file \"%s\", "
1299                        "errno = %d\n", file, -err);
1300                 goto out_close;
1301         }
1302         if(err)
1303                 return fd;
1304
1305         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1306
1307         /* Allow switching only if no mismatch. */
1308         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1309                 printk("Switching backing file to '%s'\n", *backing_file_out);
1310                 err = write_cow_header(file, fd, *backing_file_out,
1311                                        sectorsize, align, &size);
1312                 if (err) {
1313                         printk("Switch failed, errno = %d\n", -err);
1314                         goto out_close;
1315                 }
1316         } else {
1317                 *backing_file_out = backing_file;
1318                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1319                 if (err)
1320                         goto out_close;
1321         }
1322
1323         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1324                   bitmap_len_out, data_offset_out);
1325
1326         return fd;
1327  out_close:
1328         os_close_file(fd);
1329         return err;
1330 }
1331
1332 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1333                     int sectorsize, int alignment, int *bitmap_offset_out,
1334                     unsigned long *bitmap_len_out, int *data_offset_out)
1335 {
1336         int err, fd;
1337
1338         flags.c = 1;
1339         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1340         if(fd < 0){
1341                 err = fd;
1342                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1343                        -err);
1344                 goto out;
1345         }
1346
1347         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1348                             bitmap_offset_out, bitmap_len_out,
1349                             data_offset_out);
1350         if(!err)
1351                 return fd;
1352         os_close_file(fd);
1353  out:
1354         return err;
1355 }
1356
1357 static int update_bitmap(struct io_thread_req *req)
1358 {
1359         int n;
1360
1361         if(req->cow_offset == -1)
1362                 return 0;
1363
1364         n = os_seek_file(req->fds[1], req->cow_offset);
1365         if(n < 0){
1366                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1367                 return 1;
1368         }
1369
1370         n = os_write_file(req->fds[1], &req->bitmap_words,
1371                           sizeof(req->bitmap_words));
1372         if(n != sizeof(req->bitmap_words)){
1373                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1374                        req->fds[1]);
1375                 return 1;
1376         }
1377
1378         return 0;
1379 }
1380
1381 void do_io(struct io_thread_req *req)
1382 {
1383         char *buf;
1384         unsigned long len;
1385         int n, nsectors, start, end, bit;
1386         int err;
1387         __u64 off;
1388
1389         nsectors = req->length / req->sectorsize;
1390         start = 0;
1391         do {
1392                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1393                 end = start;
1394                 while((end < nsectors) &&
1395                       (ubd_test_bit(end, (unsigned char *)
1396                                     &req->sector_mask) == bit))
1397                         end++;
1398
1399                 off = req->offset + req->offsets[bit] +
1400                         start * req->sectorsize;
1401                 len = (end - start) * req->sectorsize;
1402                 buf = &req->buffer[start * req->sectorsize];
1403
1404                 err = os_seek_file(req->fds[bit], off);
1405                 if(err < 0){
1406                         printk("do_io - lseek failed : err = %d\n", -err);
1407                         req->error = 1;
1408                         return;
1409                 }
1410                 if(req->op == UBD_READ){
1411                         n = 0;
1412                         do {
1413                                 buf = &buf[n];
1414                                 len -= n;
1415                                 n = os_read_file(req->fds[bit], buf, len);
1416                                 if (n < 0) {
1417                                         printk("do_io - read failed, err = %d "
1418                                                "fd = %d\n", -n, req->fds[bit]);
1419                                         req->error = 1;
1420                                         return;
1421                                 }
1422                         } while((n < len) && (n != 0));
1423                         if (n < len) memset(&buf[n], 0, len - n);
1424                 } else {
1425                         n = os_write_file(req->fds[bit], buf, len);
1426                         if(n != len){
1427                                 printk("do_io - write failed err = %d "
1428                                        "fd = %d\n", -n, req->fds[bit]);
1429                                 req->error = 1;
1430                                 return;
1431                         }
1432                 }
1433
1434                 start = end;
1435         } while(start < nsectors);
1436
1437         req->error = update_bitmap(req);
1438 }
1439
1440 /* Changed in start_io_thread, which is serialized by being called only
1441  * from ubd_init, which is an initcall.
1442  */
1443 int kernel_fd = -1;
1444
1445 /* Only changed by the io thread. XXX: currently unused. */
1446 static int io_count = 0;
1447
1448 int io_thread(void *arg)
1449 {
1450         struct io_thread_req *req;
1451         int n;
1452
1453         ignore_sigwinch_sig();
1454         while(1){
1455                 n = os_read_file(kernel_fd, &req,
1456                                  sizeof(struct io_thread_req *));
1457                 if(n != sizeof(struct io_thread_req *)){
1458                         if(n < 0)
1459                                 printk("io_thread - read failed, fd = %d, "
1460                                        "err = %d\n", kernel_fd, -n);
1461                         else {
1462                                 printk("io_thread - short read, fd = %d, "
1463                                        "length = %d\n", kernel_fd, n);
1464                         }
1465                         continue;
1466                 }
1467                 io_count++;
1468                 do_io(req);
1469                 n = os_write_file(kernel_fd, &req,
1470                                   sizeof(struct io_thread_req *));
1471                 if(n != sizeof(struct io_thread_req *))
1472                         printk("io_thread - write failed, fd = %d, err = %d\n",
1473                                kernel_fd, -n);
1474         }
1475
1476         return 0;
1477 }