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