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/kernel.h"
  24 #include "linux/module.h"
  25 #include "linux/blkdev.h"
  26 #include "linux/hdreg.h"
  27 #include "linux/init.h"
  28 #include "linux/cdrom.h"
  29 #include "linux/proc_fs.h"
  30 #include "linux/ctype.h"
  31 #include "linux/capability.h"
  32 #include "linux/mm.h"
  33 #include "linux/vmalloc.h"
  34 #include "linux/blkpg.h"
  35 #include "linux/genhd.h"
  36 #include "linux/spinlock.h"
  37 #include "linux/platform_device.h"
  38 #include "linux/scatterlist.h"
  39 #include "asm/segment.h"
  40 #include "asm/uaccess.h"
  41 #include "asm/irq.h"
  42 #include "asm/types.h"
  43 #include "asm/tlbflush.h"
  44 #include "mem_user.h"
  45 #include "kern_util.h"
  46 #include "kern.h"
  47 #include "mconsole_kern.h"
  48 #include "init.h"
  49 #include "irq_user.h"
  50 #include "irq_kern.h"
  51 #include "ubd_user.h"
  52 #include "os.h"
  53 #include "mem.h"
  54 #include "mem_kern.h"
  55 #include "cow.h"
  56
  57 enum ubd_req { UBD_READ, UBD_WRITE };
  58
  59 struct io_thread_req {
  60         struct request *req;
  61         enum ubd_req op;
  62         int fds[2];
  63         unsigned long offsets[2];
  64         unsigned long long offset;
  65         unsigned long length;
  66         char *buffer;
  67         int sectorsize;
  68         unsigned long sector_mask;
  69         unsigned long long cow_offset;
  70         unsigned long bitmap_words[2];
  71         int error;
  72 };
  73
  74 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
  75                          char **backing_file_out, int *bitmap_offset_out,
  76                          unsigned long *bitmap_len_out, int *data_offset_out,
  77                          int *create_cow_out);
  78 extern int create_cow_file(char *cow_file, char *backing_file,
  79                            struct openflags flags, int sectorsize,
  80                            int alignment, int *bitmap_offset_out,
  81                            unsigned long *bitmap_len_out,
  82                            int *data_offset_out);
  83 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
  84 extern void do_io(struct io_thread_req *req);
  85
  86 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  87 {
  88         __u64 n;
  89         int bits, off;
  90
  91         bits = sizeof(data[0]) * 8;
  92         n = bit / bits;
  93         off = bit % bits;
  94         return (data[n] & (1 << off)) != 0;
  95 }
  96
  97 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  98 {
  99         __u64 n;
 100         int bits, off;
 101
 102         bits = sizeof(data[0]) * 8;
 103         n = bit / bits;
 104         off = bit % bits;
 105         data[n] |= (1 << off);
 106 }
 107 /*End stuff from ubd_user.h*/
 108
 109 #define DRIVER_NAME "uml-blkdev"
 110
 111 static DEFINE_MUTEX(ubd_lock);
 112
 113 static int ubd_open(struct inode * inode, struct file * filp);
 114 static int ubd_release(struct inode * inode, struct file * file);
 115 static int ubd_ioctl(struct inode * inode, struct file * file,
 116                      unsigned int cmd, unsigned long arg);
 117 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 118
 119 #define MAX_DEV (16)
 120
 121 static struct block_device_operations ubd_blops = {
 122         .owner          = THIS_MODULE,
 123         .open           = ubd_open,
 124         .release        = ubd_release,
 125         .ioctl          = ubd_ioctl,
 126         .getgeo         = ubd_getgeo,
 127 };
 128
 129 /* Protected by ubd_lock */
 130 static int fake_major = MAJOR_NR;
 131 static struct gendisk *ubd_gendisk[MAX_DEV];
 132 static struct gendisk *fake_gendisk[MAX_DEV];
 133
 134 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 135 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 136                                          .cl = 1 })
 137 #else
 138 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 139                                          .cl = 1 })
 140 #endif
 141 static struct openflags global_openflags = OPEN_FLAGS;
 142
 143 struct cow {
 144         /* backing file name */
 145         char *file;
 146         /* backing file fd */
 147         int fd;
 148         unsigned long *bitmap;
 149         unsigned long bitmap_len;
 150         int bitmap_offset;
 151         int data_offset;
 152 };
 153
 154 #define MAX_SG 64
 155
 156 struct ubd {
 157         struct list_head restart;
 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         struct scatterlist sg[MAX_SG];
 173         struct request *request;
 174         int start_sg, end_sg;
 175 };
 176
 177 #define DEFAULT_COW { \
 178         .file =                 NULL, \
 179         .fd =                   -1,     \
 180         .bitmap =               NULL, \
 181         .bitmap_offset =        0, \
 182         .data_offset =          0, \
 183 }
 184
 185 #define DEFAULT_UBD { \
 186         .file =                 NULL, \
 187         .count =                0, \
 188         .fd =                   -1, \
 189         .size =                 -1, \
 190         .boot_openflags =       OPEN_FLAGS, \
 191         .openflags =            OPEN_FLAGS, \
 192         .no_cow =               0, \
 193         .shared =               0, \
 194         .cow =                  DEFAULT_COW, \
 195         .lock =                 SPIN_LOCK_UNLOCKED,     \
 196         .request =              NULL, \
 197         .start_sg =             0, \
 198         .end_sg =               0, \
 199 }
 200
 201 /* Protected by ubd_lock */
 202 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
 203
 204 /* Only changed by fake_ide_setup which is a setup */
 205 static int fake_ide = 0;
 206 static struct proc_dir_entry *proc_ide_root = NULL;
 207 static struct proc_dir_entry *proc_ide = NULL;
 208
 209 static void make_proc_ide(void)
 210 {
 211         proc_ide_root = proc_mkdir("ide", NULL);
 212         proc_ide = proc_mkdir("ide0", proc_ide_root);
 213 }
 214
 215 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
 216                                int *eof, void *data)
 217 {
 218         int len;
 219
 220         strcpy(page, "disk\n");
 221         len = strlen("disk\n");
 222         len -= off;
 223         if (len < count){
 224                 *eof = 1;
 225                 if (len <= 0) return 0;
 226         }
 227         else len = count;
 228         *start = page + off;
 229         return len;
 230 }
 231
 232 static void make_ide_entries(const char *dev_name)
 233 {
 234         struct proc_dir_entry *dir, *ent;
 235         char name[64];
 236
 237         if(proc_ide_root == NULL) make_proc_ide();
 238
 239         dir = proc_mkdir(dev_name, proc_ide);
 240         if(!dir) return;
 241
 242         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
 243         if(!ent) return;
 244         ent->data = NULL;
 245         ent->read_proc = proc_ide_read_media;
 246         ent->write_proc = NULL;
 247         snprintf(name, sizeof(name), "ide0/%s", dev_name);
 248         proc_symlink(dev_name, proc_ide_root, name);
 249 }
 250
 251 static int fake_ide_setup(char *str)
 252 {
 253         fake_ide = 1;
 254         return 1;
 255 }
 256
 257 __setup("fake_ide", fake_ide_setup);
 258
 259 __uml_help(fake_ide_setup,
 260 "fake_ide\n"
 261 "    Create ide0 entries that map onto ubd devices.\n\n"
 262 );
 263
 264 static int parse_unit(char **ptr)
 265 {
 266         char *str = *ptr, *end;
 267         int n = -1;
 268
 269         if(isdigit(*str)) {
 270                 n = simple_strtoul(str, &end, 0);
 271                 if(end == str)
 272                         return -1;
 273                 *ptr = end;
 274         }
 275         else if (('a' <= *str) && (*str <= 'z')) {
 276                 n = *str - 'a';
 277                 str++;
 278                 *ptr = str;
 279         }
 280         return n;
 281 }
 282
 283 /* If *index_out == -1 at exit, the passed option was a general one;
 284  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 285  * should not be freed on exit.
 286  */
 287 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 288 {
 289         struct ubd *ubd_dev;
 290         struct openflags flags = global_openflags;
 291         char *backing_file;
 292         int n, err = 0, i;
 293
 294         if(index_out) *index_out = -1;
 295         n = *str;
 296         if(n == '='){
 297                 char *end;
 298                 int major;
 299
 300                 str++;
 301                 if(!strcmp(str, "sync")){
 302                         global_openflags = of_sync(global_openflags);
 303                         goto out1;
 304                 }
 305
 306                 err = -EINVAL;
 307                 major = simple_strtoul(str, &end, 0);
 308                 if((*end != '\0') || (end == str)){
 309                         *error_out = "Didn't parse major number";
 310                         goto out1;
 311                 }
 312
 313                 mutex_lock(&ubd_lock);
 314                 if(fake_major != MAJOR_NR){
 315                         *error_out = "Can't assign a fake major twice";
 316                         goto out1;
 317                 }
 318
 319                 fake_major = major;
 320
 321                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 322                        major);
 323                 err = 0;
 324         out1:
 325                 mutex_unlock(&ubd_lock);
 326                 return err;
 327         }
 328
 329         n = parse_unit(&str);
 330         if(n < 0){
 331                 *error_out = "Couldn't parse device number";
 332                 return -EINVAL;
 333         }
 334         if(n >= MAX_DEV){
 335                 *error_out = "Device number out of range";
 336                 return 1;
 337         }
 338
 339         err = -EBUSY;
 340         mutex_lock(&ubd_lock);
 341
 342         ubd_dev = &ubd_devs[n];
 343         if(ubd_dev->file != NULL){
 344                 *error_out = "Device is already configured";
 345                 goto out;
 346         }
 347
 348         if (index_out)
 349                 *index_out = n;
 350
 351         err = -EINVAL;
 352         for (i = 0; i < sizeof("rscd="); i++) {
 353                 switch (*str) {
 354                 case 'r':
 355                         flags.w = 0;
 356                         break;
 357                 case 's':
 358                         flags.s = 1;
 359                         break;
 360                 case 'd':
 361                         ubd_dev->no_cow = 1;
 362                         break;
 363                 case 'c':
 364                         ubd_dev->shared = 1;
 365                         break;
 366                 case '=':
 367                         str++;
 368                         goto break_loop;
 369                 default:
 370                         *error_out = "Expected '=' or flag letter "
 371                                 "(r, s, c, or d)";
 372                         goto out;
 373                 }
 374                 str++;
 375         }
 376
 377         if (*str == '=')
 378                 *error_out = "Too many flags specified";
 379         else
 380                 *error_out = "Missing '='";
 381         goto out;
 382
 383 break_loop:
 384         backing_file = strchr(str, ',');
 385
 386         if (backing_file == NULL)
 387                 backing_file = strchr(str, ':');
 388
 389         if(backing_file != NULL){
 390                 if(ubd_dev->no_cow){
 391                         *error_out = "Can't specify both 'd' and a cow file";
 392                         goto out;
 393                 }
 394                 else {
 395                         *backing_file = '\0';
 396                         backing_file++;
 397                 }
 398         }
 399         err = 0;
 400         ubd_dev->file = str;
 401         ubd_dev->cow.file = backing_file;
 402         ubd_dev->boot_openflags = flags;
 403 out:
 404         mutex_unlock(&ubd_lock);
 405         return err;
 406 }
 407
 408 static int ubd_setup(char *str)
 409 {
 410         char *error;
 411         int err;
 412
 413         err = ubd_setup_common(str, NULL, &error);
 414         if(err)
 415                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 416                        "%s\n", str, error);
 417         return 1;
 418 }
 419
 420 __setup("ubd", ubd_setup);
 421 __uml_help(ubd_setup,
 422 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 423 "    This is used to associate a device with a file in the underlying\n"
 424 "    filesystem. When specifying two filenames, the first one is the\n"
 425 "    COW name and the second is the backing file name. As separator you can\n"
 426 "    use either a ':' or a ',': the first one allows writing things like;\n"
 427 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 428 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 429 "    When using only one filename, UML will detect whether to treat it like\n"
 430 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 431 "    flag:\n"
 432 "       ubd0d=BackingFile\n"
 433 "    Usually, there is a filesystem in the file, but \n"
 434 "    that's not required. Swap devices containing swap files can be\n"
 435 "    specified like this. Also, a file which doesn't contain a\n"
 436 "    filesystem can have its contents read in the virtual \n"
 437 "    machine by running 'dd' on the device. <n> must be in the range\n"
 438 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 439 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 440 "    an 's' will cause data to be written to disk on the host immediately.\n"
 441 "    'c' will cause the device to be treated as being shared between multiple\n"
 442 "    UMLs and file locking will be turned off - this is appropriate for a\n"
 443 "    cluster filesystem and inappropriate at almost all other times.\n\n"
 444 );
 445
 446 static int udb_setup(const char *str)
 447 {
 448         printk("udb%s specified on command line is almost certainly a ubd -> "
 449                "udb TYPO\n", str);
 450         return 1;
 451 }
 452
 453 __setup("udb", udb_setup);
 454 __uml_help(udb_setup,
 455 "udb\n"
 456 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 457 "    to impossible to catch visually unless you specifically look for\n"
 458 "    them.  The only result of any option starting with 'udb' is an error\n"
 459 "    in the boot output.\n\n"
 460 );
 461
 462 static int fakehd_set = 0;
 463 static int fakehd(char *str)
 464 {
 465         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
 466         fakehd_set = 1;
 467         return 1;
 468 }
 469
 470 __setup("fakehd", fakehd);
 471 __uml_help(fakehd,
 472 "fakehd\n"
 473 "    Change the ubd device name to \"hd\".\n\n"
 474 );
 475
 476 static void do_ubd_request(struct request_queue * q);
 477
 478 /* Only changed by ubd_init, which is an initcall. */
 479 int thread_fd = -1;
 480
 481 static void ubd_end_request(struct request *req, int bytes, int error)
 482 {
 483         blk_end_request(req, error, bytes);
 484 }
 485
 486 /* Callable only from interrupt context - otherwise you need to do
 487  * spin_lock_irq()/spin_lock_irqsave() */
 488 static inline void ubd_finish(struct request *req, int bytes)
 489 {
 490         if(bytes < 0){
 491                 ubd_end_request(req, 0, -EIO);
 492                 return;
 493         }
 494         ubd_end_request(req, bytes, 0);
 495 }
 496
 497 static LIST_HEAD(restart);
 498
 499 /* XXX - move this inside ubd_intr. */
 500 /* Called without dev->lock held, and only in interrupt context. */
 501 static void ubd_handler(void)
 502 {
 503         struct io_thread_req *req;
 504         struct request *rq;
 505         struct ubd *ubd;
 506         struct list_head *list, *next_ele;
 507         unsigned long flags;
 508         int n;
 509
 510         while(1){
 511                 n = os_read_file(thread_fd, &req,
 512                                  sizeof(struct io_thread_req *));
 513                 if(n != sizeof(req)){
 514                         if(n == -EAGAIN)
 515                                 break;
 516                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 517                                "err = %d\n", -n);
 518                         return;
 519                 }
 520
 521                 rq = req->req;
 522                 rq->nr_sectors -= req->length >> 9;
 523                 if(rq->nr_sectors == 0)
 524                         ubd_finish(rq, rq->hard_nr_sectors << 9);
 525                 kfree(req);
 526         }
 527         reactivate_fd(thread_fd, UBD_IRQ);
 528
 529         list_for_each_safe(list, next_ele, &restart){
 530                 ubd = container_of(list, struct ubd, restart);
 531                 list_del_init(&ubd->restart);
 532                 spin_lock_irqsave(&ubd->lock, flags);
 533                 do_ubd_request(ubd->queue);
 534                 spin_unlock_irqrestore(&ubd->lock, flags);
 535         }
 536 }
 537
 538 static irqreturn_t ubd_intr(int irq, void *dev)
 539 {
 540         ubd_handler();
 541         return IRQ_HANDLED;
 542 }
 543
 544 /* Only changed by ubd_init, which is an initcall. */
 545 static int io_pid = -1;
 546
 547 void kill_io_thread(void)
 548 {
 549         if(io_pid != -1)
 550                 os_kill_process(io_pid, 1);
 551 }
 552
 553 __uml_exitcall(kill_io_thread);
 554
 555 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 556 {
 557         char *file;
 558
 559         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
 560         return os_file_size(file, size_out);
 561 }
 562
 563 static void ubd_close_dev(struct ubd *ubd_dev)
 564 {
 565         os_close_file(ubd_dev->fd);
 566         if(ubd_dev->cow.file == NULL)
 567                 return;
 568
 569         os_close_file(ubd_dev->cow.fd);
 570         vfree(ubd_dev->cow.bitmap);
 571         ubd_dev->cow.bitmap = NULL;
 572 }
 573
 574 static int ubd_open_dev(struct ubd *ubd_dev)
 575 {
 576         struct openflags flags;
 577         char **back_ptr;
 578         int err, create_cow, *create_ptr;
 579         int fd;
 580
 581         ubd_dev->openflags = ubd_dev->boot_openflags;
 582         create_cow = 0;
 583         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 584         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 585
 586         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 587                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 588                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 589                                 create_ptr);
 590
 591         if((fd == -ENOENT) && create_cow){
 592                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 593                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 594                                           &ubd_dev->cow.bitmap_offset,
 595                                           &ubd_dev->cow.bitmap_len,
 596                                           &ubd_dev->cow.data_offset);
 597                 if(fd >= 0){
 598                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 599                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 600                 }
 601         }
 602
 603         if(fd < 0){
 604                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 605                        -fd);
 606                 return fd;
 607         }
 608         ubd_dev->fd = fd;
 609
 610         if(ubd_dev->cow.file != NULL){
 611                 blk_queue_max_sectors(ubd_dev->queue, 8 * sizeof(long));
 612
 613                 err = -ENOMEM;
 614                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 615                 if(ubd_dev->cow.bitmap == NULL){
 616                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 617                         goto error;
 618                 }
 619                 flush_tlb_kernel_vm();
 620
 621                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 622                                       ubd_dev->cow.bitmap_offset,
 623                                       ubd_dev->cow.bitmap_len);
 624                 if(err < 0)
 625                         goto error;
 626
 627                 flags = ubd_dev->openflags;
 628                 flags.w = 0;
 629                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 630                                     NULL, NULL, NULL, NULL);
 631                 if(err < 0) goto error;
 632                 ubd_dev->cow.fd = err;
 633         }
 634         return 0;
 635  error:
 636         os_close_file(ubd_dev->fd);
 637         return err;
 638 }
 639
 640 static void ubd_device_release(struct device *dev)
 641 {
 642         struct ubd *ubd_dev = dev->driver_data;
 643
 644         blk_cleanup_queue(ubd_dev->queue);
 645         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 646 }
 647
 648 static int ubd_disk_register(int major, u64 size, int unit,
 649                              struct gendisk **disk_out)
 650 {
 651         struct gendisk *disk;
 652
 653         disk = alloc_disk(1 << UBD_SHIFT);
 654         if(disk == NULL)
 655                 return -ENOMEM;
 656
 657         disk->major = major;
 658         disk->first_minor = unit << UBD_SHIFT;
 659         disk->fops = &ubd_blops;
 660         set_capacity(disk, size / 512);
 661         if(major == MAJOR_NR)
 662                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 663         else
 664                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 665
 666         /* sysfs register (not for ide fake devices) */
 667         if (major == MAJOR_NR) {
 668                 ubd_devs[unit].pdev.id   = unit;
 669                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 670                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 671                 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
 672                 platform_device_register(&ubd_devs[unit].pdev);
 673                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 674         }
 675
 676         disk->private_data = &ubd_devs[unit];
 677         disk->queue = ubd_devs[unit].queue;
 678         add_disk(disk);
 679
 680         *disk_out = disk;
 681         return 0;
 682 }
 683
 684 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 685
 686 static int ubd_add(int n, char **error_out)
 687 {
 688         struct ubd *ubd_dev = &ubd_devs[n];
 689         int err = 0;
 690
 691         if(ubd_dev->file == NULL)
 692                 goto out;
 693
 694         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 695         if(err < 0){
 696                 *error_out = "Couldn't determine size of device's file";
 697                 goto out;
 698         }
 699
 700         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 701
 702         INIT_LIST_HEAD(&ubd_dev->restart);
 703         sg_init_table(ubd_dev->sg, MAX_SG);
 704
 705         err = -ENOMEM;
 706         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 707         if (ubd_dev->queue == NULL) {
 708                 *error_out = "Failed to initialize device queue";
 709                 goto out;
 710         }
 711         ubd_dev->queue->queuedata = ubd_dev;
 712
 713         blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG);
 714         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
 715         if(err){
 716                 *error_out = "Failed to register device";
 717                 goto out_cleanup;
 718         }
 719
 720         if(fake_major != MAJOR_NR)
 721                 ubd_disk_register(fake_major, ubd_dev->size, n,
 722                                   &fake_gendisk[n]);
 723
 724         /* perhaps this should also be under the "if (fake_major)" above */
 725         /* using the fake_disk->disk_name and also the fakehd_set name */
 726         if (fake_ide)
 727                 make_ide_entries(ubd_gendisk[n]->disk_name);
 728
 729         err = 0;
 730 out:
 731         return err;
 732
 733 out_cleanup:
 734         blk_cleanup_queue(ubd_dev->queue);
 735         goto out;
 736 }
 737
 738 static int ubd_config(char *str, char **error_out)
 739 {
 740         int n, ret;
 741
 742         /* This string is possibly broken up and stored, so it's only
 743          * freed if ubd_setup_common fails, or if only general options
 744          * were set.
 745          */
 746         str = kstrdup(str, GFP_KERNEL);
 747         if (str == NULL) {
 748                 *error_out = "Failed to allocate memory";
 749                 return -ENOMEM;
 750         }
 751
 752         ret = ubd_setup_common(str, &n, error_out);
 753         if (ret)
 754                 goto err_free;
 755
 756         if (n == -1) {
 757                 ret = 0;
 758                 goto err_free;
 759         }
 760
 761         mutex_lock(&ubd_lock);
 762         ret = ubd_add(n, error_out);
 763         if (ret)
 764                 ubd_devs[n].file = NULL;
 765         mutex_unlock(&ubd_lock);
 766
 767 out:
 768         return ret;
 769
 770 err_free:
 771         kfree(str);
 772         goto out;
 773 }
 774
 775 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 776 {
 777         struct ubd *ubd_dev;
 778         int n, len = 0;
 779
 780         n = parse_unit(&name);
 781         if((n >= MAX_DEV) || (n < 0)){
 782                 *error_out = "ubd_get_config : device number out of range";
 783                 return -1;
 784         }
 785
 786         ubd_dev = &ubd_devs[n];
 787         mutex_lock(&ubd_lock);
 788
 789         if(ubd_dev->file == NULL){
 790                 CONFIG_CHUNK(str, size, len, "", 1);
 791                 goto out;
 792         }
 793
 794         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 795
 796         if(ubd_dev->cow.file != NULL){
 797                 CONFIG_CHUNK(str, size, len, ",", 0);
 798                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 799         }
 800         else CONFIG_CHUNK(str, size, len, "", 1);
 801
 802  out:
 803         mutex_unlock(&ubd_lock);
 804         return len;
 805 }
 806
 807 static int ubd_id(char **str, int *start_out, int *end_out)
 808 {
 809         int n;
 810
 811         n = parse_unit(str);
 812         *start_out = 0;
 813         *end_out = MAX_DEV - 1;
 814         return n;
 815 }
 816
 817 static int ubd_remove(int n, char **error_out)
 818 {
 819         struct gendisk *disk = ubd_gendisk[n];
 820         struct ubd *ubd_dev;
 821         int err = -ENODEV;
 822
 823         mutex_lock(&ubd_lock);
 824
 825         ubd_dev = &ubd_devs[n];
 826
 827         if(ubd_dev->file == NULL)
 828                 goto out;
 829
 830         /* you cannot remove a open disk */
 831         err = -EBUSY;
 832         if(ubd_dev->count > 0)
 833                 goto out;
 834
 835         ubd_gendisk[n] = NULL;
 836         if(disk != NULL){
 837                 del_gendisk(disk);
 838                 put_disk(disk);
 839         }
 840
 841         if(fake_gendisk[n] != NULL){
 842                 del_gendisk(fake_gendisk[n]);
 843                 put_disk(fake_gendisk[n]);
 844                 fake_gendisk[n] = NULL;
 845         }
 846
 847         err = 0;
 848         platform_device_unregister(&ubd_dev->pdev);
 849 out:
 850         mutex_unlock(&ubd_lock);
 851         return err;
 852 }
 853
 854 /* All these are called by mconsole in process context and without
 855  * ubd-specific locks.  The structure itself is const except for .list.
 856  */
 857 static struct mc_device ubd_mc = {
 858         .list           = LIST_HEAD_INIT(ubd_mc.list),
 859         .name           = "ubd",
 860         .config         = ubd_config,
 861         .get_config     = ubd_get_config,
 862         .id             = ubd_id,
 863         .remove         = ubd_remove,
 864 };
 865
 866 static int __init ubd_mc_init(void)
 867 {
 868         mconsole_register_dev(&ubd_mc);
 869         return 0;
 870 }
 871
 872 __initcall(ubd_mc_init);
 873
 874 static int __init ubd0_init(void)
 875 {
 876         struct ubd *ubd_dev = &ubd_devs[0];
 877
 878         mutex_lock(&ubd_lock);
 879         if(ubd_dev->file == NULL)
 880                 ubd_dev->file = "root_fs";
 881         mutex_unlock(&ubd_lock);
 882
 883         return 0;
 884 }
 885
 886 __initcall(ubd0_init);
 887
 888 /* Used in ubd_init, which is an initcall */
 889 static struct platform_driver ubd_driver = {
 890         .driver = {
 891                 .name  = DRIVER_NAME,
 892         },
 893 };
 894
 895 static int __init ubd_init(void)
 896 {
 897         char *error;
 898         int i, err;
 899
 900         if (register_blkdev(MAJOR_NR, "ubd"))
 901                 return -1;
 902
 903         if (fake_major != MAJOR_NR) {
 904                 char name[sizeof("ubd_nnn\0")];
 905
 906                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
 907                 if (register_blkdev(fake_major, "ubd"))
 908                         return -1;
 909         }
 910         platform_driver_register(&ubd_driver);
 911         mutex_lock(&ubd_lock);
 912         for (i = 0; i < MAX_DEV; i++){
 913                 err = ubd_add(i, &error);
 914                 if(err)
 915                         printk(KERN_ERR "Failed to initialize ubd device %d :"
 916                                "%s\n", i, error);
 917         }
 918         mutex_unlock(&ubd_lock);
 919         return 0;
 920 }
 921
 922 late_initcall(ubd_init);
 923
 924 static int __init ubd_driver_init(void){
 925         unsigned long stack;
 926         int err;
 927
 928         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
 929         if(global_openflags.s){
 930                 printk(KERN_INFO "ubd: Synchronous mode\n");
 931                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
 932                  * enough. So use anyway the io thread. */
 933         }
 934         stack = alloc_stack(0, 0);
 935         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
 936                                  &thread_fd);
 937         if(io_pid < 0){
 938                 printk(KERN_ERR
 939                        "ubd : Failed to start I/O thread (errno = %d) - "
 940                        "falling back to synchronous I/O\n", -io_pid);
 941                 io_pid = -1;
 942                 return 0;
 943         }
 944         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
 945                              IRQF_DISABLED, "ubd", ubd_devs);
 946         if(err != 0)
 947                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
 948         return 0;
 949 }
 950
 951 device_initcall(ubd_driver_init);
 952
 953 static int ubd_open(struct inode *inode, struct file *filp)
 954 {
 955         struct gendisk *disk = inode->i_bdev->bd_disk;
 956         struct ubd *ubd_dev = disk->private_data;
 957         int err = 0;
 958
 959         if(ubd_dev->count == 0){
 960                 err = ubd_open_dev(ubd_dev);
 961                 if(err){
 962                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
 963                                disk->disk_name, ubd_dev->file, -err);
 964                         goto out;
 965                 }
 966         }
 967         ubd_dev->count++;
 968         set_disk_ro(disk, !ubd_dev->openflags.w);
 969
 970         /* This should no more be needed. And it didn't work anyway to exclude
 971          * read-write remounting of filesystems.*/
 972         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
 973                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
 974                 err = -EROFS;
 975         }*/
 976  out:
 977         return err;
 978 }
 979
 980 static int ubd_release(struct inode * inode, struct file * file)
 981 {
 982         struct gendisk *disk = inode->i_bdev->bd_disk;
 983         struct ubd *ubd_dev = disk->private_data;
 984
 985         if(--ubd_dev->count == 0)
 986                 ubd_close_dev(ubd_dev);
 987         return 0;
 988 }
 989
 990 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
 991                           __u64 *cow_offset, unsigned long *bitmap,
 992                           __u64 bitmap_offset, unsigned long *bitmap_words,
 993                           __u64 bitmap_len)
 994 {
 995         __u64 sector = io_offset >> 9;
 996         int i, update_bitmap = 0;
 997
 998         for(i = 0; i < length >> 9; i++){
 999                 if(cow_mask != NULL)
1000                         ubd_set_bit(i, (unsigned char *) cow_mask);
1001                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1002                         continue;
1003
1004                 update_bitmap = 1;
1005                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1006         }
1007
1008         if(!update_bitmap)
1009                 return;
1010
1011         *cow_offset = sector / (sizeof(unsigned long) * 8);
1012
1013         /* This takes care of the case where we're exactly at the end of the
1014          * device, and *cow_offset + 1 is off the end.  So, just back it up
1015          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1016          * for the original diagnosis.
1017          */
1018         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
1019                            sizeof(unsigned long) - 1))
1020                 (*cow_offset)--;
1021
1022         bitmap_words[0] = bitmap[*cow_offset];
1023         bitmap_words[1] = bitmap[*cow_offset + 1];
1024
1025         *cow_offset *= sizeof(unsigned long);
1026         *cow_offset += bitmap_offset;
1027 }
1028
1029 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1030                        __u64 bitmap_offset, __u64 bitmap_len)
1031 {
1032         __u64 sector = req->offset >> 9;
1033         int i;
1034
1035         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1036                 panic("Operation too long");
1037
1038         if(req->op == UBD_READ) {
1039                 for(i = 0; i < req->length >> 9; i++){
1040                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1041                                 ubd_set_bit(i, (unsigned char *)
1042                                             &req->sector_mask);
1043                 }
1044         }
1045         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1046                            &req->cow_offset, bitmap, bitmap_offset,
1047                            req->bitmap_words, bitmap_len);
1048 }
1049
1050 /* Called with dev->lock held */
1051 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1052                             unsigned long long offset, int page_offset,
1053                             int len, struct page *page)
1054 {
1055         struct gendisk *disk = req->rq_disk;
1056         struct ubd *ubd_dev = disk->private_data;
1057
1058         io_req->req = req;
1059         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1060                 ubd_dev->fd;
1061         io_req->fds[1] = ubd_dev->fd;
1062         io_req->cow_offset = -1;
1063         io_req->offset = offset;
1064         io_req->length = len;
1065         io_req->error = 0;
1066         io_req->sector_mask = 0;
1067
1068         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1069         io_req->offsets[0] = 0;
1070         io_req->offsets[1] = ubd_dev->cow.data_offset;
1071         io_req->buffer = page_address(page) + page_offset;
1072         io_req->sectorsize = 1 << 9;
1073
1074         if(ubd_dev->cow.file != NULL)
1075                 cowify_req(io_req, ubd_dev->cow.bitmap,
1076                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1077
1078 }
1079
1080 /* Called with dev->lock held */
1081 static void do_ubd_request(struct request_queue *q)
1082 {
1083         struct io_thread_req *io_req;
1084         struct request *req;
1085         int n, last_sectors;
1086
1087         while(1){
1088                 struct ubd *dev = q->queuedata;
1089                 if(dev->end_sg == 0){
1090                         struct request *req = elv_next_request(q);
1091                         if(req == NULL)
1092                                 return;
1093
1094                         dev->request = req;
1095                         blkdev_dequeue_request(req);
1096                         dev->start_sg = 0;
1097                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1098                 }
1099
1100                 req = dev->request;
1101                 last_sectors = 0;
1102                 while(dev->start_sg < dev->end_sg){
1103                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1104
1105                         req->sector += last_sectors;
1106                         io_req = kmalloc(sizeof(struct io_thread_req),
1107                                          GFP_ATOMIC);
1108                         if(io_req == NULL){
1109                                 if(list_empty(&dev->restart))
1110                                         list_add(&dev->restart, &restart);
1111                                 return;
1112                         }
1113                         prepare_request(req, io_req,
1114                                         (unsigned long long) req->sector << 9,
1115                                         sg->offset, sg->length, sg_page(sg));
1116
1117                         last_sectors = sg->length >> 9;
1118                         n = os_write_file(thread_fd, &io_req,
1119                                           sizeof(struct io_thread_req *));
1120                         if(n != sizeof(struct io_thread_req *)){
1121                                 if(n != -EAGAIN)
1122                                         printk("write to io thread failed, "
1123                                                "errno = %d\n", -n);
1124                                 else if(list_empty(&dev->restart))
1125                                         list_add(&dev->restart, &restart);
1126                                 kfree(io_req);
1127                                 return;
1128                         }
1129
1130                         dev->start_sg++;
1131                 }
1132                 dev->end_sg = 0;
1133                 dev->request = NULL;
1134         }
1135 }
1136
1137 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1138 {
1139         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1140
1141         geo->heads = 128;
1142         geo->sectors = 32;
1143         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1144         return 0;
1145 }
1146
1147 static int ubd_ioctl(struct inode * inode, struct file * file,
1148                      unsigned int cmd, unsigned long arg)
1149 {
1150         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1151         struct hd_driveid ubd_id = {
1152                 .cyls           = 0,
1153                 .heads          = 128,
1154                 .sectors        = 32,
1155         };
1156
1157         switch (cmd) {
1158                 struct cdrom_volctrl volume;
1159         case HDIO_GET_IDENTITY:
1160                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1161                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1162                                  sizeof(ubd_id)))
1163                         return -EFAULT;
1164                 return 0;
1165
1166         case CDROMVOLREAD:
1167                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1168                         return -EFAULT;
1169                 volume.channel0 = 255;
1170                 volume.channel1 = 255;
1171                 volume.channel2 = 255;
1172                 volume.channel3 = 255;
1173                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1174                         return -EFAULT;
1175                 return 0;
1176         }
1177         return -EINVAL;
1178 }
1179
1180 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1181 {
1182         struct uml_stat buf1, buf2;
1183         int err;
1184
1185         if(from_cmdline == NULL)
1186                 return 0;
1187         if(!strcmp(from_cmdline, from_cow))
1188                 return 0;
1189
1190         err = os_stat_file(from_cmdline, &buf1);
1191         if(err < 0){
1192                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1193                 return 0;
1194         }
1195         err = os_stat_file(from_cow, &buf2);
1196         if(err < 0){
1197                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1198                 return 1;
1199         }
1200         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1201                 return 0;
1202
1203         printk("Backing file mismatch - \"%s\" requested,\n"
1204                "\"%s\" specified in COW header of \"%s\"\n",
1205                from_cmdline, from_cow, cow);
1206         return 1;
1207 }
1208
1209 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1210 {
1211         unsigned long modtime;
1212         unsigned long long actual;
1213         int err;
1214
1215         err = os_file_modtime(file, &modtime);
1216         if(err < 0){
1217                 printk("Failed to get modification time of backing file "
1218                        "\"%s\", err = %d\n", file, -err);
1219                 return err;
1220         }
1221
1222         err = os_file_size(file, &actual);
1223         if(err < 0){
1224                 printk("Failed to get size of backing file \"%s\", "
1225                        "err = %d\n", file, -err);
1226                 return err;
1227         }
1228
1229         if(actual != size){
1230                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1231                  * the typecast.*/
1232                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1233                        "file\n", (unsigned long long) size, actual);
1234                 return -EINVAL;
1235         }
1236         if(modtime != mtime){
1237                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1238                        "file\n", mtime, modtime);
1239                 return -EINVAL;
1240         }
1241         return 0;
1242 }
1243
1244 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1245 {
1246         int err;
1247
1248         err = os_seek_file(fd, offset);
1249         if(err < 0)
1250                 return err;
1251
1252         err = os_read_file(fd, buf, len);
1253         if(err < 0)
1254                 return err;
1255
1256         return 0;
1257 }
1258
1259 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1260                   char **backing_file_out, int *bitmap_offset_out,
1261                   unsigned long *bitmap_len_out, int *data_offset_out,
1262                   int *create_cow_out)
1263 {
1264         time_t mtime;
1265         unsigned long long size;
1266         __u32 version, align;
1267         char *backing_file;
1268         int fd, err, sectorsize, asked_switch, mode = 0644;
1269
1270         fd = os_open_file(file, *openflags, mode);
1271         if (fd < 0) {
1272                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1273                         *create_cow_out = 1;
1274                 if (!openflags->w ||
1275                     ((fd != -EROFS) && (fd != -EACCES)))
1276                         return fd;
1277                 openflags->w = 0;
1278                 fd = os_open_file(file, *openflags, mode);
1279                 if (fd < 0)
1280                         return fd;
1281         }
1282
1283         if(shared)
1284                 printk("Not locking \"%s\" on the host\n", file);
1285         else {
1286                 err = os_lock_file(fd, openflags->w);
1287                 if(err < 0){
1288                         printk("Failed to lock '%s', err = %d\n", file, -err);
1289                         goto out_close;
1290                 }
1291         }
1292
1293         /* Successful return case! */
1294         if(backing_file_out == NULL)
1295                 return fd;
1296
1297         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1298                               &size, &sectorsize, &align, bitmap_offset_out);
1299         if(err && (*backing_file_out != NULL)){
1300                 printk("Failed to read COW header from COW file \"%s\", "
1301                        "errno = %d\n", file, -err);
1302                 goto out_close;
1303         }
1304         if(err)
1305                 return fd;
1306
1307         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1308
1309         /* Allow switching only if no mismatch. */
1310         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1311                 printk("Switching backing file to '%s'\n", *backing_file_out);
1312                 err = write_cow_header(file, fd, *backing_file_out,
1313                                        sectorsize, align, &size);
1314                 if (err) {
1315                         printk("Switch failed, errno = %d\n", -err);
1316                         goto out_close;
1317                 }
1318         } else {
1319                 *backing_file_out = backing_file;
1320                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1321                 if (err)
1322                         goto out_close;
1323         }
1324
1325         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1326                   bitmap_len_out, data_offset_out);
1327
1328         return fd;
1329  out_close:
1330         os_close_file(fd);
1331         return err;
1332 }
1333
1334 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1335                     int sectorsize, int alignment, int *bitmap_offset_out,
1336                     unsigned long *bitmap_len_out, int *data_offset_out)
1337 {
1338         int err, fd;
1339
1340         flags.c = 1;
1341         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1342         if(fd < 0){
1343                 err = fd;
1344                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1345                        -err);
1346                 goto out;
1347         }
1348
1349         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1350                             bitmap_offset_out, bitmap_len_out,
1351                             data_offset_out);
1352         if(!err)
1353                 return fd;
1354         os_close_file(fd);
1355  out:
1356         return err;
1357 }
1358
1359 static int update_bitmap(struct io_thread_req *req)
1360 {
1361         int n;
1362
1363         if(req->cow_offset == -1)
1364                 return 0;
1365
1366         n = os_seek_file(req->fds[1], req->cow_offset);
1367         if(n < 0){
1368                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1369                 return 1;
1370         }
1371
1372         n = os_write_file(req->fds[1], &req->bitmap_words,
1373                           sizeof(req->bitmap_words));
1374         if(n != sizeof(req->bitmap_words)){
1375                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1376                        req->fds[1]);
1377                 return 1;
1378         }
1379
1380         return 0;
1381 }
1382
1383 void do_io(struct io_thread_req *req)
1384 {
1385         char *buf;
1386         unsigned long len;
1387         int n, nsectors, start, end, bit;
1388         int err;
1389         __u64 off;
1390
1391         nsectors = req->length / req->sectorsize;
1392         start = 0;
1393         do {
1394                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1395                 end = start;
1396                 while((end < nsectors) &&
1397                       (ubd_test_bit(end, (unsigned char *)
1398                                     &req->sector_mask) == bit))
1399                         end++;
1400
1401                 off = req->offset + req->offsets[bit] +
1402                         start * req->sectorsize;
1403                 len = (end - start) * req->sectorsize;
1404                 buf = &req->buffer[start * req->sectorsize];
1405
1406                 err = os_seek_file(req->fds[bit], off);
1407                 if(err < 0){
1408                         printk("do_io - lseek failed : err = %d\n", -err);
1409                         req->error = 1;
1410                         return;
1411                 }
1412                 if(req->op == UBD_READ){
1413                         n = 0;
1414                         do {
1415                                 buf = &buf[n];
1416                                 len -= n;
1417                                 n = os_read_file(req->fds[bit], buf, len);
1418                                 if (n < 0) {
1419                                         printk("do_io - read failed, err = %d "
1420                                                "fd = %d\n", -n, req->fds[bit]);
1421                                         req->error = 1;
1422                                         return;
1423                                 }
1424                         } while((n < len) && (n != 0));
1425                         if (n < len) memset(&buf[n], 0, len - n);
1426                 } else {
1427                         n = os_write_file(req->fds[bit], buf, len);
1428                         if(n != len){
1429                                 printk("do_io - write failed err = %d "
1430                                        "fd = %d\n", -n, req->fds[bit]);
1431                                 req->error = 1;
1432                                 return;
1433                         }
1434                 }
1435
1436                 start = end;
1437         } while(start < nsectors);
1438
1439         req->error = update_bitmap(req);
1440 }
1441
1442 /* Changed in start_io_thread, which is serialized by being called only
1443  * from ubd_init, which is an initcall.
1444  */
1445 int kernel_fd = -1;
1446
1447 /* Only changed by the io thread. XXX: currently unused. */
1448 static int io_count = 0;
1449
1450 int io_thread(void *arg)
1451 {
1452         struct io_thread_req *req;
1453         int n;
1454
1455         ignore_sigwinch_sig();
1456         while(1){
1457                 n = os_read_file(kernel_fd, &req,
1458                                  sizeof(struct io_thread_req *));
1459                 if(n != sizeof(struct io_thread_req *)){
1460                         if(n < 0)
1461                                 printk("io_thread - read failed, fd = %d, "
1462                                        "err = %d\n", kernel_fd, -n);
1463                         else {
1464                                 printk("io_thread - short read, fd = %d, "
1465                                        "length = %d\n", kernel_fd, n);
1466                         }
1467                         continue;
1468                 }
1469                 io_count++;
1470                 do_io(req);
1471                 n = os_write_file(kernel_fd, &req,
1472                                   sizeof(struct io_thread_req *));
1473                 if(n != sizeof(struct io_thread_req *))
1474                         printk("io_thread - write failed, fd = %d, err = %d\n",
1475                                kernel_fd, -n);
1476         }
1477
1478         return 0;
1479 }