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