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

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