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/module.h"
  24 #include "linux/blkdev.h"
  25 #include "linux/hdreg.h"
  26 #include "linux/init.h"
  27 #include "linux/cdrom.h"
  28 #include "linux/proc_fs.h"
  29 #include "linux/ctype.h"
  30 #include "linux/capability.h"
  31 #include "linux/mm.h"
  32 #include "linux/vmalloc.h"
  33 #include "linux/blkpg.h"
  34 #include "linux/genhd.h"
  35 #include "linux/spinlock.h"
  36 #include "linux/platform_device.h"
  37 #include "asm/segment.h"
  38 #include "asm/uaccess.h"
  39 #include "asm/irq.h"
  40 #include "asm/types.h"
  41 #include "asm/tlbflush.h"
  42 #include "user_util.h"
  43 #include "mem_user.h"
  44 #include "kern_util.h"
  45 #include "kern.h"
  46 #include "mconsole_kern.h"
  47 #include "init.h"
  48 #include "irq_user.h"
  49 #include "irq_kern.h"
  50 #include "ubd_user.h"
  51 #include "os.h"
  52 #include "mem.h"
  53 #include "mem_kern.h"
  54 #include "cow.h"
  55
  56 enum ubd_req { UBD_READ, UBD_WRITE };
  57
  58 struct io_thread_req {
  59         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 /* XXX - this made sense in 2.4 days, now it's only used as a boolean, and
 113  * probably it doesn't make sense even for that. */
 114 static int do_ubd;
 115
 116 static int ubd_open(struct inode * inode, struct file * filp);
 117 static int ubd_release(struct inode * inode, struct file * file);
 118 static int ubd_ioctl(struct inode * inode, struct file * file,
 119                      unsigned int cmd, unsigned long arg);
 120 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 121
 122 #define MAX_DEV (16)
 123
 124 static struct block_device_operations ubd_blops = {
 125         .owner          = THIS_MODULE,
 126         .open           = ubd_open,
 127         .release        = ubd_release,
 128         .ioctl          = ubd_ioctl,
 129         .getgeo         = ubd_getgeo,
 130 };
 131
 132 /* Protected by ubd_lock */
 133 static int fake_major = MAJOR_NR;
 134 static struct gendisk *ubd_gendisk[MAX_DEV];
 135 static struct gendisk *fake_gendisk[MAX_DEV];
 136
 137 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 138 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 139                                          .cl = 1 })
 140 #else
 141 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 142                                          .cl = 1 })
 143 #endif
 144 static struct openflags global_openflags = OPEN_FLAGS;
 145
 146 struct cow {
 147         /* backing file name */
 148         char *file;
 149         /* backing file fd */
 150         int fd;
 151         unsigned long *bitmap;
 152         unsigned long bitmap_len;
 153         int bitmap_offset;
 154         int data_offset;
 155 };
 156
 157 struct ubd {
 158         /* name (and fd, below) of the file opened for writing, either the
 159          * backing or the cow file. */
 160         char *file;
 161         int count;
 162         int fd;
 163         __u64 size;
 164         struct openflags boot_openflags;
 165         struct openflags openflags;
 166         unsigned shared:1;
 167         unsigned no_cow:1;
 168         struct cow cow;
 169         struct platform_device pdev;
 170         struct request_queue *queue;
 171         spinlock_t lock;
 172 };
 173
 174 #define DEFAULT_COW { \
 175         .file =                 NULL, \
 176         .fd =                   -1, \
 177         .bitmap =               NULL, \
 178         .bitmap_offset =        0, \
 179         .data_offset =          0, \
 180 }
 181
 182 #define DEFAULT_UBD { \
 183         .file =                 NULL, \
 184         .count =                0, \
 185         .fd =                   -1, \
 186         .size =                 -1, \
 187         .boot_openflags =       OPEN_FLAGS, \
 188         .openflags =            OPEN_FLAGS, \
 189         .no_cow =               0, \
 190         .shared =               0, \
 191         .cow =                  DEFAULT_COW, \
 192         .lock =                 SPIN_LOCK_UNLOCKED,     \
 193 }
 194
 195 /* Protected by ubd_lock */
 196 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
 197
 198 /* Only changed by fake_ide_setup which is a setup */
 199 static int fake_ide = 0;
 200 static struct proc_dir_entry *proc_ide_root = NULL;
 201 static struct proc_dir_entry *proc_ide = NULL;
 202
 203 static void make_proc_ide(void)
 204 {
 205         proc_ide_root = proc_mkdir("ide", NULL);
 206         proc_ide = proc_mkdir("ide0", proc_ide_root);
 207 }
 208
 209 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
 210                                int *eof, void *data)
 211 {
 212         int len;
 213
 214         strcpy(page, "disk\n");
 215         len = strlen("disk\n");
 216         len -= off;
 217         if (len < count){
 218                 *eof = 1;
 219                 if (len <= 0) return 0;
 220         }
 221         else len = count;
 222         *start = page + off;
 223         return len;
 224 }
 225
 226 static void make_ide_entries(char *dev_name)
 227 {
 228         struct proc_dir_entry *dir, *ent;
 229         char name[64];
 230
 231         if(proc_ide_root == NULL) make_proc_ide();
 232
 233         dir = proc_mkdir(dev_name, proc_ide);
 234         if(!dir) return;
 235
 236         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
 237         if(!ent) return;
 238         ent->data = NULL;
 239         ent->read_proc = proc_ide_read_media;
 240         ent->write_proc = NULL;
 241         sprintf(name,"ide0/%s", dev_name);
 242         proc_symlink(dev_name, proc_ide_root, name);
 243 }
 244
 245 static int fake_ide_setup(char *str)
 246 {
 247         fake_ide = 1;
 248         return(1);
 249 }
 250
 251 __setup("fake_ide", fake_ide_setup);
 252
 253 __uml_help(fake_ide_setup,
 254 "fake_ide\n"
 255 "    Create ide0 entries that map onto ubd devices.\n\n"
 256 );
 257
 258 static int parse_unit(char **ptr)
 259 {
 260         char *str = *ptr, *end;
 261         int n = -1;
 262
 263         if(isdigit(*str)) {
 264                 n = simple_strtoul(str, &end, 0);
 265                 if(end == str)
 266                         return(-1);
 267                 *ptr = end;
 268         }
 269         else if (('a' <= *str) && (*str <= 'z')) {
 270                 n = *str - 'a';
 271                 str++;
 272                 *ptr = str;
 273         }
 274         return(n);
 275 }
 276
 277 /* If *index_out == -1 at exit, the passed option was a general one;
 278  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 279  * should not be freed on exit.
 280  */
 281 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 282 {
 283         struct ubd *ubd_dev;
 284         struct openflags flags = global_openflags;
 285         char *backing_file;
 286         int n, err = 0, i;
 287
 288         if(index_out) *index_out = -1;
 289         n = *str;
 290         if(n == '='){
 291                 char *end;
 292                 int major;
 293
 294                 str++;
 295                 if(!strcmp(str, "sync")){
 296                         global_openflags = of_sync(global_openflags);
 297                         goto out1;
 298                 }
 299
 300                 err = -EINVAL;
 301                 major = simple_strtoul(str, &end, 0);
 302                 if((*end != '\0') || (end == str)){
 303                         *error_out = "Didn't parse major number";
 304                         goto out1;
 305                 }
 306
 307                 mutex_lock(&ubd_lock);
 308                 if(fake_major != MAJOR_NR){
 309                         *error_out = "Can't assign a fake major twice";
 310                         goto out1;
 311                 }
 312
 313                 fake_major = major;
 314
 315                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 316                        major);
 317                 err = 0;
 318         out1:
 319                 mutex_unlock(&ubd_lock);
 320                 return err;
 321         }
 322
 323         n = parse_unit(&str);
 324         if(n < 0){
 325                 *error_out = "Couldn't parse device number";
 326                 return -EINVAL;
 327         }
 328         if(n >= MAX_DEV){
 329                 *error_out = "Device number out of range";
 330                 return 1;
 331         }
 332
 333         err = -EBUSY;
 334         mutex_lock(&ubd_lock);
 335
 336         ubd_dev = &ubd_devs[n];
 337         if(ubd_dev->file != NULL){
 338                 *error_out = "Device is already configured";
 339                 goto out;
 340         }
 341
 342         if (index_out)
 343                 *index_out = n;
 344
 345         err = -EINVAL;
 346         for (i = 0; i < sizeof("rscd="); i++) {
 347                 switch (*str) {
 348                 case 'r':
 349                         flags.w = 0;
 350                         break;
 351                 case 's':
 352                         flags.s = 1;
 353                         break;
 354                 case 'd':
 355                         ubd_dev->no_cow = 1;
 356                         break;
 357                 case 'c':
 358                         ubd_dev->shared = 1;
 359                         break;
 360                 case '=':
 361                         str++;
 362                         goto break_loop;
 363                 default:
 364                         *error_out = "Expected '=' or flag letter "
 365                                 "(r, s, c, or d)";
 366                         goto out;
 367                 }
 368                 str++;
 369         }
 370
 371         if (*str == '=')
 372                 *error_out = "Too many flags specified";
 373         else
 374                 *error_out = "Missing '='";
 375         goto out;
 376
 377 break_loop:
 378         backing_file = strchr(str, ',');
 379
 380         if (backing_file == NULL)
 381                 backing_file = strchr(str, ':');
 382
 383         if(backing_file != NULL){
 384                 if(ubd_dev->no_cow){
 385                         *error_out = "Can't specify both 'd' and a cow file";
 386                         goto out;
 387                 }
 388                 else {
 389                         *backing_file = '\0';
 390                         backing_file++;
 391                 }
 392         }
 393         err = 0;
 394         ubd_dev->file = str;
 395         ubd_dev->cow.file = backing_file;
 396         ubd_dev->boot_openflags = flags;
 397 out:
 398         mutex_unlock(&ubd_lock);
 399         return err;
 400 }
 401
 402 static int ubd_setup(char *str)
 403 {
 404         char *error;
 405         int err;
 406
 407         err = ubd_setup_common(str, NULL, &error);
 408         if(err)
 409                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 410                        "%s\n", str, error);
 411         return 1;
 412 }
 413
 414 __setup("ubd", ubd_setup);
 415 __uml_help(ubd_setup,
 416 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 417 "    This is used to associate a device with a file in the underlying\n"
 418 "    filesystem. When specifying two filenames, the first one is the\n"
 419 "    COW name and the second is the backing file name. As separator you can\n"
 420 "    use either a ':' or a ',': the first one allows writing things like;\n"
 421 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 422 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 423 "    When using only one filename, UML will detect whether to treat it like\n"
 424 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 425 "    flag:\n"
 426 "       ubd0d=BackingFile\n"
 427 "    Usually, there is a filesystem in the file, but \n"
 428 "    that's not required. Swap devices containing swap files can be\n"
 429 "    specified like this. Also, a file which doesn't contain a\n"
 430 "    filesystem can have its contents read in the virtual \n"
 431 "    machine by running 'dd' on the device. <n> must be in the range\n"
 432 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 433 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 434 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
 435 );
 436
 437 static int udb_setup(char *str)
 438 {
 439         printk("udb%s specified on command line is almost certainly a ubd -> "
 440                "udb TYPO\n", str);
 441         return(1);
 442 }
 443
 444 __setup("udb", udb_setup);
 445 __uml_help(udb_setup,
 446 "udb\n"
 447 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 448 "    to impossible to catch visually unless you specifically look for\n"
 449 "    them.  The only result of any option starting with 'udb' is an error\n"
 450 "    in the boot output.\n\n"
 451 );
 452
 453 static int fakehd_set = 0;
 454 static int fakehd(char *str)
 455 {
 456         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
 457         fakehd_set = 1;
 458         return 1;
 459 }
 460
 461 __setup("fakehd", fakehd);
 462 __uml_help(fakehd,
 463 "fakehd\n"
 464 "    Change the ubd device name to \"hd\".\n\n"
 465 );
 466
 467 static void do_ubd_request(request_queue_t * q);
 468
 469 /* Only changed by ubd_init, which is an initcall. */
 470 int thread_fd = -1;
 471
 472 /* call ubd_finish if you need to serialize */
 473 static void __ubd_finish(struct request *req, int error)
 474 {
 475         int nsect;
 476
 477         if(error){
 478                 end_request(req, 0);
 479                 return;
 480         }
 481         nsect = req->current_nr_sectors;
 482         req->sector += nsect;
 483         req->buffer += nsect << 9;
 484         req->errors = 0;
 485         req->nr_sectors -= nsect;
 486         req->current_nr_sectors = 0;
 487         end_request(req, 1);
 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 error)
 493 {
 494         struct ubd *dev = req->rq_disk->private_data;
 495
 496         spin_lock(&dev->lock);
 497         __ubd_finish(req, error);
 498         spin_unlock(&dev->lock);
 499 }
 500
 501 /* XXX - move this inside ubd_intr. */
 502 /* Called without dev->lock held, and only in interrupt context. */
 503 static void ubd_handler(void)
 504 {
 505         struct io_thread_req req;
 506         struct request *rq;
 507         struct ubd *dev;
 508         int n;
 509
 510         do_ubd = 0;
 511         n = os_read_file(thread_fd, &req, sizeof(req));
 512         if(n != sizeof(req)){
 513                 printk(KERN_ERR "Pid %d - spurious interrupt in ubd_handler, "
 514                        "err = %d\n", os_getpid(), -n);
 515                 return;
 516         }
 517
 518         rq = req.req;
 519         dev = rq->rq_disk->private_data;
 520
 521         ubd_finish(rq, req.error);
 522         reactivate_fd(thread_fd, UBD_IRQ);
 523         spin_lock(&dev->lock);
 524         do_ubd_request(dev->queue);
 525         spin_unlock(&dev->lock);
 526 }
 527
 528 static irqreturn_t ubd_intr(int irq, void *dev)
 529 {
 530         ubd_handler();
 531         return(IRQ_HANDLED);
 532 }
 533
 534 /* Only changed by ubd_init, which is an initcall. */
 535 static int io_pid = -1;
 536
 537 void kill_io_thread(void)
 538 {
 539         if(io_pid != -1)
 540                 os_kill_process(io_pid, 1);
 541 }
 542
 543 __uml_exitcall(kill_io_thread);
 544
 545 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 546 {
 547         char *file;
 548
 549         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
 550         return(os_file_size(file, size_out));
 551 }
 552
 553 static void ubd_close_dev(struct ubd *ubd_dev)
 554 {
 555         os_close_file(ubd_dev->fd);
 556         if(ubd_dev->cow.file == NULL)
 557                 return;
 558
 559         os_close_file(ubd_dev->cow.fd);
 560         vfree(ubd_dev->cow.bitmap);
 561         ubd_dev->cow.bitmap = NULL;
 562 }
 563
 564 static int ubd_open_dev(struct ubd *ubd_dev)
 565 {
 566         struct openflags flags;
 567         char **back_ptr;
 568         int err, create_cow, *create_ptr;
 569         int fd;
 570
 571         ubd_dev->openflags = ubd_dev->boot_openflags;
 572         create_cow = 0;
 573         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 574         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 575
 576         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 577                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 578                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 579                                 create_ptr);
 580
 581         if((fd == -ENOENT) && create_cow){
 582                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 583                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 584                                           &ubd_dev->cow.bitmap_offset,
 585                                           &ubd_dev->cow.bitmap_len,
 586                                           &ubd_dev->cow.data_offset);
 587                 if(fd >= 0){
 588                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 589                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 590                 }
 591         }
 592
 593         if(fd < 0){
 594                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 595                        -fd);
 596                 return fd;
 597         }
 598         ubd_dev->fd = fd;
 599
 600         if(ubd_dev->cow.file != NULL){
 601                 err = -ENOMEM;
 602                 ubd_dev->cow.bitmap = (void *) vmalloc(ubd_dev->cow.bitmap_len);
 603                 if(ubd_dev->cow.bitmap == NULL){
 604                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 605                         goto error;
 606                 }
 607                 flush_tlb_kernel_vm();
 608
 609                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 610                                       ubd_dev->cow.bitmap_offset,
 611                                       ubd_dev->cow.bitmap_len);
 612                 if(err < 0)
 613                         goto error;
 614
 615                 flags = ubd_dev->openflags;
 616                 flags.w = 0;
 617                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 618                                     NULL, NULL, NULL, NULL);
 619                 if(err < 0) goto error;
 620                 ubd_dev->cow.fd = err;
 621         }
 622         return(0);
 623  error:
 624         os_close_file(ubd_dev->fd);
 625         return(err);
 626 }
 627
 628 static int ubd_disk_register(int major, u64 size, int unit,
 629                              struct gendisk **disk_out)
 630 {
 631         struct gendisk *disk;
 632
 633         disk = alloc_disk(1 << UBD_SHIFT);
 634         if(disk == NULL)
 635                 return(-ENOMEM);
 636
 637         disk->major = major;
 638         disk->first_minor = unit << UBD_SHIFT;
 639         disk->fops = &ubd_blops;
 640         set_capacity(disk, size / 512);
 641         if(major == MAJOR_NR)
 642                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 643         else
 644                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 645
 646         /* sysfs register (not for ide fake devices) */
 647         if (major == MAJOR_NR) {
 648                 ubd_devs[unit].pdev.id   = unit;
 649                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 650                 platform_device_register(&ubd_devs[unit].pdev);
 651                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 652         }
 653
 654         disk->private_data = &ubd_devs[unit];
 655         disk->queue = ubd_devs[unit].queue;
 656         add_disk(disk);
 657
 658         *disk_out = disk;
 659         return 0;
 660 }
 661
 662 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 663
 664 static int ubd_add(int n, char **error_out)
 665 {
 666         struct ubd *ubd_dev = &ubd_devs[n];
 667         int err = 0;
 668
 669         if(ubd_dev->file == NULL)
 670                 goto out;
 671
 672         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 673         if(err < 0){
 674                 *error_out = "Couldn't determine size of device's file";
 675                 goto out;
 676         }
 677
 678         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 679
 680         err = -ENOMEM;
 681         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 682         if (ubd_dev->queue == NULL) {
 683                 *error_out = "Failed to initialize device queue";
 684                 goto out;
 685         }
 686         ubd_dev->queue->queuedata = ubd_dev;
 687
 688         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
 689         if(err){
 690                 *error_out = "Failed to register device";
 691                 goto out_cleanup;
 692         }
 693
 694         if(fake_major != MAJOR_NR)
 695                 ubd_disk_register(fake_major, ubd_dev->size, n,
 696                                   &fake_gendisk[n]);
 697
 698         /* perhaps this should also be under the "if (fake_major)" above */
 699         /* using the fake_disk->disk_name and also the fakehd_set name */
 700         if (fake_ide)
 701                 make_ide_entries(ubd_gendisk[n]->disk_name);
 702
 703         err = 0;
 704 out:
 705         return err;
 706
 707 out_cleanup:
 708         blk_cleanup_queue(ubd_dev->queue);
 709         goto out;
 710 }
 711
 712 static int ubd_config(char *str, char **error_out)
 713 {
 714         int n, ret;
 715
 716         /* This string is possibly broken up and stored, so it's only
 717          * freed if ubd_setup_common fails, or if only general options
 718          * were set.
 719          */
 720         str = kstrdup(str, GFP_KERNEL);
 721         if (str == NULL) {
 722                 *error_out = "Failed to allocate memory";
 723                 return -ENOMEM;
 724         }
 725
 726         ret = ubd_setup_common(str, &n, error_out);
 727         if (ret)
 728                 goto err_free;
 729
 730         if (n == -1) {
 731                 ret = 0;
 732                 goto err_free;
 733         }
 734
 735         mutex_lock(&ubd_lock);
 736         ret = ubd_add(n, error_out);
 737         if (ret)
 738                 ubd_devs[n].file = NULL;
 739         mutex_unlock(&ubd_lock);
 740
 741 out:
 742         return ret;
 743
 744 err_free:
 745         kfree(str);
 746         goto out;
 747 }
 748
 749 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 750 {
 751         struct ubd *ubd_dev;
 752         int n, len = 0;
 753
 754         n = parse_unit(&name);
 755         if((n >= MAX_DEV) || (n < 0)){
 756                 *error_out = "ubd_get_config : device number out of range";
 757                 return(-1);
 758         }
 759
 760         ubd_dev = &ubd_devs[n];
 761         mutex_lock(&ubd_lock);
 762
 763         if(ubd_dev->file == NULL){
 764                 CONFIG_CHUNK(str, size, len, "", 1);
 765                 goto out;
 766         }
 767
 768         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 769
 770         if(ubd_dev->cow.file != NULL){
 771                 CONFIG_CHUNK(str, size, len, ",", 0);
 772                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 773         }
 774         else CONFIG_CHUNK(str, size, len, "", 1);
 775
 776  out:
 777         mutex_unlock(&ubd_lock);
 778         return(len);
 779 }
 780
 781 static int ubd_id(char **str, int *start_out, int *end_out)
 782 {
 783         int n;
 784
 785         n = parse_unit(str);
 786         *start_out = 0;
 787         *end_out = MAX_DEV - 1;
 788         return n;
 789 }
 790
 791 static int ubd_remove(int n, char **error_out)
 792 {
 793         struct ubd *ubd_dev;
 794         int err = -ENODEV;
 795
 796         mutex_lock(&ubd_lock);
 797
 798         if(ubd_gendisk[n] == NULL)
 799                 goto out;
 800
 801         ubd_dev = &ubd_devs[n];
 802
 803         if(ubd_dev->file == NULL)
 804                 goto out;
 805
 806         /* you cannot remove a open disk */
 807         err = -EBUSY;
 808         if(ubd_dev->count > 0)
 809                 goto out;
 810
 811         del_gendisk(ubd_gendisk[n]);
 812         put_disk(ubd_gendisk[n]);
 813         ubd_gendisk[n] = NULL;
 814
 815         if(fake_gendisk[n] != NULL){
 816                 del_gendisk(fake_gendisk[n]);
 817                 put_disk(fake_gendisk[n]);
 818                 fake_gendisk[n] = NULL;
 819         }
 820
 821         blk_cleanup_queue(ubd_dev->queue);
 822         platform_device_unregister(&ubd_dev->pdev);
 823         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 824         err = 0;
 825 out:
 826         mutex_unlock(&ubd_lock);
 827         return err;
 828 }
 829
 830 /* All these are called by mconsole in process context and without
 831  * ubd-specific locks.  The structure itself is const except for .list.
 832  */
 833 static struct mc_device ubd_mc = {
 834         .list           = LIST_HEAD_INIT(ubd_mc.list),
 835         .name           = "ubd",
 836         .config         = ubd_config,
 837         .get_config     = ubd_get_config,
 838         .id             = ubd_id,
 839         .remove         = ubd_remove,
 840 };
 841
 842 static int __init ubd_mc_init(void)
 843 {
 844         mconsole_register_dev(&ubd_mc);
 845         return 0;
 846 }
 847
 848 __initcall(ubd_mc_init);
 849
 850 static int __init ubd0_init(void)
 851 {
 852         struct ubd *ubd_dev = &ubd_devs[0];
 853
 854         mutex_lock(&ubd_lock);
 855         if(ubd_dev->file == NULL)
 856                 ubd_dev->file = "root_fs";
 857         mutex_unlock(&ubd_lock);
 858
 859         return(0);
 860 }
 861
 862 __initcall(ubd0_init);
 863
 864 /* Used in ubd_init, which is an initcall */
 865 static struct platform_driver ubd_driver = {
 866         .driver = {
 867                 .name  = DRIVER_NAME,
 868         },
 869 };
 870
 871 static int __init ubd_init(void)
 872 {
 873         char *error;
 874         int i, err;
 875
 876         if (register_blkdev(MAJOR_NR, "ubd"))
 877                 return -1;
 878
 879         if (fake_major != MAJOR_NR) {
 880                 char name[sizeof("ubd_nnn\0")];
 881
 882                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
 883                 if (register_blkdev(fake_major, "ubd"))
 884                         return -1;
 885         }
 886         platform_driver_register(&ubd_driver);
 887         mutex_lock(&ubd_lock);
 888         for (i = 0; i < MAX_DEV; i++){
 889                 err = ubd_add(i, &error);
 890                 if(err)
 891                         printk(KERN_ERR "Failed to initialize ubd device %d :"
 892                                "%s\n", i, error);
 893         }
 894         mutex_unlock(&ubd_lock);
 895         return 0;
 896 }
 897
 898 late_initcall(ubd_init);
 899
 900 static int __init ubd_driver_init(void){
 901         unsigned long stack;
 902         int err;
 903
 904         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
 905         if(global_openflags.s){
 906                 printk(KERN_INFO "ubd: Synchronous mode\n");
 907                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
 908                  * enough. So use anyway the io thread. */
 909         }
 910         stack = alloc_stack(0, 0);
 911         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
 912                                  &thread_fd);
 913         if(io_pid < 0){
 914                 printk(KERN_ERR
 915                        "ubd : Failed to start I/O thread (errno = %d) - "
 916                        "falling back to synchronous I/O\n", -io_pid);
 917                 io_pid = -1;
 918                 return(0);
 919         }
 920         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
 921                              IRQF_DISABLED, "ubd", ubd_devs);
 922         if(err != 0)
 923                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
 924         return 0;
 925 }
 926
 927 device_initcall(ubd_driver_init);
 928
 929 static int ubd_open(struct inode *inode, struct file *filp)
 930 {
 931         struct gendisk *disk = inode->i_bdev->bd_disk;
 932         struct ubd *ubd_dev = disk->private_data;
 933         int err = 0;
 934
 935         if(ubd_dev->count == 0){
 936                 err = ubd_open_dev(ubd_dev);
 937                 if(err){
 938                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
 939                                disk->disk_name, ubd_dev->file, -err);
 940                         goto out;
 941                 }
 942         }
 943         ubd_dev->count++;
 944         set_disk_ro(disk, !ubd_dev->openflags.w);
 945
 946         /* This should no more be needed. And it didn't work anyway to exclude
 947          * read-write remounting of filesystems.*/
 948         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
 949                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
 950                 err = -EROFS;
 951         }*/
 952  out:
 953         return(err);
 954 }
 955
 956 static int ubd_release(struct inode * inode, struct file * file)
 957 {
 958         struct gendisk *disk = inode->i_bdev->bd_disk;
 959         struct ubd *ubd_dev = disk->private_data;
 960
 961         if(--ubd_dev->count == 0)
 962                 ubd_close_dev(ubd_dev);
 963         return(0);
 964 }
 965
 966 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
 967                           __u64 *cow_offset, unsigned long *bitmap,
 968                           __u64 bitmap_offset, unsigned long *bitmap_words,
 969                           __u64 bitmap_len)
 970 {
 971         __u64 sector = io_offset >> 9;
 972         int i, update_bitmap = 0;
 973
 974         for(i = 0; i < length >> 9; i++){
 975                 if(cow_mask != NULL)
 976                         ubd_set_bit(i, (unsigned char *) cow_mask);
 977                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
 978                         continue;
 979
 980                 update_bitmap = 1;
 981                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
 982         }
 983
 984         if(!update_bitmap)
 985                 return;
 986
 987         *cow_offset = sector / (sizeof(unsigned long) * 8);
 988
 989         /* This takes care of the case where we're exactly at the end of the
 990          * device, and *cow_offset + 1 is off the end.  So, just back it up
 991          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
 992          * for the original diagnosis.
 993          */
 994         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
 995                            sizeof(unsigned long) - 1))
 996                 (*cow_offset)--;
 997
 998         bitmap_words[0] = bitmap[*cow_offset];
 999         bitmap_words[1] = bitmap[*cow_offset + 1];
1000
1001         *cow_offset *= sizeof(unsigned long);
1002         *cow_offset += bitmap_offset;
1003 }
1004
1005 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1006                        __u64 bitmap_offset, __u64 bitmap_len)
1007 {
1008         __u64 sector = req->offset >> 9;
1009         int i;
1010
1011         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1012                 panic("Operation too long");
1013
1014         if(req->op == UBD_READ) {
1015                 for(i = 0; i < req->length >> 9; i++){
1016                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1017                                 ubd_set_bit(i, (unsigned char *)
1018                                             &req->sector_mask);
1019                 }
1020         }
1021         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1022                            &req->cow_offset, bitmap, bitmap_offset,
1023                            req->bitmap_words, bitmap_len);
1024 }
1025
1026 /* Called with dev->lock held */
1027 static int prepare_request(struct request *req, struct io_thread_req *io_req)
1028 {
1029         struct gendisk *disk = req->rq_disk;
1030         struct ubd *ubd_dev = disk->private_data;
1031         __u64 offset;
1032         int len;
1033
1034         /* This should be impossible now */
1035         if((rq_data_dir(req) == WRITE) && !ubd_dev->openflags.w){
1036                 printk("Write attempted on readonly ubd device %s\n",
1037                        disk->disk_name);
1038                 end_request(req, 0);
1039                 return(1);
1040         }
1041
1042         offset = ((__u64) req->sector) << 9;
1043         len = req->current_nr_sectors << 9;
1044
1045         io_req->req = req;
1046         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd : ubd_dev->fd;
1047         io_req->fds[1] = ubd_dev->fd;
1048         io_req->cow_offset = -1;
1049         io_req->offset = offset;
1050         io_req->length = len;
1051         io_req->error = 0;
1052         io_req->sector_mask = 0;
1053
1054         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1055         io_req->offsets[0] = 0;
1056         io_req->offsets[1] = ubd_dev->cow.data_offset;
1057         io_req->buffer = req->buffer;
1058         io_req->sectorsize = 1 << 9;
1059
1060         if(ubd_dev->cow.file != NULL)
1061                 cowify_req(io_req, ubd_dev->cow.bitmap, ubd_dev->cow.bitmap_offset,
1062                            ubd_dev->cow.bitmap_len);
1063
1064         return(0);
1065 }
1066
1067 /* Called with dev->lock held */
1068 static void do_ubd_request(request_queue_t *q)
1069 {
1070         struct io_thread_req io_req;
1071         struct request *req;
1072         int err, n;
1073
1074         if(thread_fd == -1){
1075                 while((req = elv_next_request(q)) != NULL){
1076                         err = prepare_request(req, &io_req);
1077                         if(!err){
1078                                 do_io(&io_req);
1079                                 __ubd_finish(req, io_req.error);
1080                         }
1081                 }
1082         }
1083         else {
1084                 if(do_ubd || (req = elv_next_request(q)) == NULL)
1085                         return;
1086                 err = prepare_request(req, &io_req);
1087                 if(!err){
1088                         do_ubd = 1;
1089                         n = os_write_file(thread_fd, (char *) &io_req,
1090                                          sizeof(io_req));
1091                         if(n != sizeof(io_req))
1092                                 printk("write to io thread failed, "
1093                                        "errno = %d\n", -n);
1094                 }
1095         }
1096 }
1097
1098 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1099 {
1100         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1101
1102         geo->heads = 128;
1103         geo->sectors = 32;
1104         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1105         return 0;
1106 }
1107
1108 static int ubd_ioctl(struct inode * inode, struct file * file,
1109                      unsigned int cmd, unsigned long arg)
1110 {
1111         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1112         struct hd_driveid ubd_id = {
1113                 .cyls           = 0,
1114                 .heads          = 128,
1115                 .sectors        = 32,
1116         };
1117
1118         switch (cmd) {
1119                 struct cdrom_volctrl volume;
1120         case HDIO_GET_IDENTITY:
1121                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1122                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1123                                  sizeof(ubd_id)))
1124                         return(-EFAULT);
1125                 return(0);
1126
1127         case CDROMVOLREAD:
1128                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1129                         return(-EFAULT);
1130                 volume.channel0 = 255;
1131                 volume.channel1 = 255;
1132                 volume.channel2 = 255;
1133                 volume.channel3 = 255;
1134                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1135                         return(-EFAULT);
1136                 return(0);
1137         }
1138         return(-EINVAL);
1139 }
1140
1141 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1142 {
1143         struct uml_stat buf1, buf2;
1144         int err;
1145
1146         if(from_cmdline == NULL)
1147                 return 0;
1148         if(!strcmp(from_cmdline, from_cow))
1149                 return 0;
1150
1151         err = os_stat_file(from_cmdline, &buf1);
1152         if(err < 0){
1153                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1154                 return 0;
1155         }
1156         err = os_stat_file(from_cow, &buf2);
1157         if(err < 0){
1158                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1159                 return 1;
1160         }
1161         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1162                 return 0;
1163
1164         printk("Backing file mismatch - \"%s\" requested,\n"
1165                "\"%s\" specified in COW header of \"%s\"\n",
1166                from_cmdline, from_cow, cow);
1167         return 1;
1168 }
1169
1170 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1171 {
1172         unsigned long modtime;
1173         unsigned long long actual;
1174         int err;
1175
1176         err = os_file_modtime(file, &modtime);
1177         if(err < 0){
1178                 printk("Failed to get modification time of backing file "
1179                        "\"%s\", err = %d\n", file, -err);
1180                 return(err);
1181         }
1182
1183         err = os_file_size(file, &actual);
1184         if(err < 0){
1185                 printk("Failed to get size of backing file \"%s\", "
1186                        "err = %d\n", file, -err);
1187                 return(err);
1188         }
1189
1190         if(actual != size){
1191                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1192                  * the typecast.*/
1193                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1194                        "file\n", (unsigned long long) size, actual);
1195                 return(-EINVAL);
1196         }
1197         if(modtime != mtime){
1198                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1199                        "file\n", mtime, modtime);
1200                 return(-EINVAL);
1201         }
1202         return(0);
1203 }
1204
1205 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1206 {
1207         int err;
1208
1209         err = os_seek_file(fd, offset);
1210         if(err < 0)
1211                 return(err);
1212
1213         err = os_read_file(fd, buf, len);
1214         if(err < 0)
1215                 return(err);
1216
1217         return(0);
1218 }
1219
1220 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1221                   char **backing_file_out, int *bitmap_offset_out,
1222                   unsigned long *bitmap_len_out, int *data_offset_out,
1223                   int *create_cow_out)
1224 {
1225         time_t mtime;
1226         unsigned long long size;
1227         __u32 version, align;
1228         char *backing_file;
1229         int fd, err, sectorsize, asked_switch, mode = 0644;
1230
1231         fd = os_open_file(file, *openflags, mode);
1232         if (fd < 0) {
1233                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1234                         *create_cow_out = 1;
1235                 if (!openflags->w ||
1236                    ((fd != -EROFS) && (fd != -EACCES)))
1237                         return fd;
1238                 openflags->w = 0;
1239                 fd = os_open_file(file, *openflags, mode);
1240                 if (fd < 0)
1241                         return fd;
1242         }
1243
1244         if(shared)
1245                 printk("Not locking \"%s\" on the host\n", file);
1246         else {
1247                 err = os_lock_file(fd, openflags->w);
1248                 if(err < 0){
1249                         printk("Failed to lock '%s', err = %d\n", file, -err);
1250                         goto out_close;
1251                 }
1252         }
1253
1254         /* Successful return case! */
1255         if(backing_file_out == NULL)
1256                 return(fd);
1257
1258         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1259                               &size, &sectorsize, &align, bitmap_offset_out);
1260         if(err && (*backing_file_out != NULL)){
1261                 printk("Failed to read COW header from COW file \"%s\", "
1262                        "errno = %d\n", file, -err);
1263                 goto out_close;
1264         }
1265         if(err)
1266                 return(fd);
1267
1268         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1269
1270         /* Allow switching only if no mismatch. */
1271         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1272                 printk("Switching backing file to '%s'\n", *backing_file_out);
1273                 err = write_cow_header(file, fd, *backing_file_out,
1274                                        sectorsize, align, &size);
1275                 if (err) {
1276                         printk("Switch failed, errno = %d\n", -err);
1277                         goto out_close;
1278                 }
1279         } else {
1280                 *backing_file_out = backing_file;
1281                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1282                 if (err)
1283                         goto out_close;
1284         }
1285
1286         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1287                   bitmap_len_out, data_offset_out);
1288
1289         return fd;
1290  out_close:
1291         os_close_file(fd);
1292         return err;
1293 }
1294
1295 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1296                     int sectorsize, int alignment, int *bitmap_offset_out,
1297                     unsigned long *bitmap_len_out, int *data_offset_out)
1298 {
1299         int err, fd;
1300
1301         flags.c = 1;
1302         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1303         if(fd < 0){
1304                 err = fd;
1305                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1306                        -err);
1307                 goto out;
1308         }
1309
1310         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1311                             bitmap_offset_out, bitmap_len_out,
1312                             data_offset_out);
1313         if(!err)
1314                 return(fd);
1315         os_close_file(fd);
1316  out:
1317         return(err);
1318 }
1319
1320 static int update_bitmap(struct io_thread_req *req)
1321 {
1322         int n;
1323
1324         if(req->cow_offset == -1)
1325                 return(0);
1326
1327         n = os_seek_file(req->fds[1], req->cow_offset);
1328         if(n < 0){
1329                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1330                 return(1);
1331         }
1332
1333         n = os_write_file(req->fds[1], &req->bitmap_words,
1334                           sizeof(req->bitmap_words));
1335         if(n != sizeof(req->bitmap_words)){
1336                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1337                        req->fds[1]);
1338                 return(1);
1339         }
1340
1341         return(0);
1342 }
1343
1344 void do_io(struct io_thread_req *req)
1345 {
1346         char *buf;
1347         unsigned long len;
1348         int n, nsectors, start, end, bit;
1349         int err;
1350         __u64 off;
1351
1352         nsectors = req->length / req->sectorsize;
1353         start = 0;
1354         do {
1355                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1356                 end = start;
1357                 while((end < nsectors) &&
1358                       (ubd_test_bit(end, (unsigned char *)
1359                                     &req->sector_mask) == bit))
1360                         end++;
1361
1362                 off = req->offset + req->offsets[bit] +
1363                         start * req->sectorsize;
1364                 len = (end - start) * req->sectorsize;
1365                 buf = &req->buffer[start * req->sectorsize];
1366
1367                 err = os_seek_file(req->fds[bit], off);
1368                 if(err < 0){
1369                         printk("do_io - lseek failed : err = %d\n", -err);
1370                         req->error = 1;
1371                         return;
1372                 }
1373                 if(req->op == UBD_READ){
1374                         n = 0;
1375                         do {
1376                                 buf = &buf[n];
1377                                 len -= n;
1378                                 n = os_read_file(req->fds[bit], buf, len);
1379                                 if (n < 0) {
1380                                         printk("do_io - read failed, err = %d "
1381                                                "fd = %d\n", -n, req->fds[bit]);
1382                                         req->error = 1;
1383                                         return;
1384                                 }
1385                         } while((n < len) && (n != 0));
1386                         if (n < len) memset(&buf[n], 0, len - n);
1387                 } else {
1388                         n = os_write_file(req->fds[bit], buf, len);
1389                         if(n != len){
1390                                 printk("do_io - write failed err = %d "
1391                                        "fd = %d\n", -n, req->fds[bit]);
1392                                 req->error = 1;
1393                                 return;
1394                         }
1395                 }
1396
1397                 start = end;
1398         } while(start < nsectors);
1399
1400         req->error = update_bitmap(req);
1401 }
1402
1403 /* Changed in start_io_thread, which is serialized by being called only
1404  * from ubd_init, which is an initcall.
1405  */
1406 int kernel_fd = -1;
1407
1408 /* Only changed by the io thread. XXX: currently unused. */
1409 static int io_count = 0;
1410
1411 int io_thread(void *arg)
1412 {
1413         struct io_thread_req req;
1414         int n;
1415
1416         ignore_sigwinch_sig();
1417         while(1){
1418                 n = os_read_file(kernel_fd, &req, sizeof(req));
1419                 if(n != sizeof(req)){
1420                         if(n < 0)
1421                                 printk("io_thread - read failed, fd = %d, "
1422                                        "err = %d\n", kernel_fd, -n);
1423                         else {
1424                                 printk("io_thread - short read, fd = %d, "
1425                                        "length = %d\n", kernel_fd, n);
1426                         }
1427                         continue;
1428                 }
1429                 io_count++;
1430                 do_io(&req);
1431                 n = os_write_file(kernel_fd, &req, sizeof(req));
1432                 if(n != sizeof(req))
1433                         printk("io_thread - write failed, fd = %d, err = %d\n",
1434                                kernel_fd, -n);
1435         }
1436
1437         return 0;
1438 }