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/ata.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/seq_file.h"
  31 #include "linux/ctype.h"
  32 #include "linux/capability.h"
  33 #include "linux/mm.h"
  34 #include "linux/slab.h"
  35 #include "linux/vmalloc.h"
  36 #include "linux/mutex.h"
  37 #include "linux/blkpg.h"
  38 #include "linux/genhd.h"
  39 #include "linux/spinlock.h"
  40 #include "linux/platform_device.h"
  41 #include "linux/scatterlist.h"
  42 #include "asm/segment.h"
  43 #include "asm/uaccess.h"
  44 #include "asm/irq.h"
  45 #include "asm/types.h"
  46 #include "asm/tlbflush.h"
  47 #include "mem_user.h"
  48 #include "kern_util.h"
  49 #include "mconsole_kern.h"
  50 #include "init.h"
  51 #include "irq_user.h"
  52 #include "irq_kern.h"
  53 #include "ubd_user.h"
  54 #include "os.h"
  55 #include "mem.h"
  56 #include "cow.h"
  57
  58 enum ubd_req { UBD_READ, UBD_WRITE };
  59
  60 struct io_thread_req {
  61         struct request *req;
  62         enum ubd_req op;
  63         int fds[2];
  64         unsigned long offsets[2];
  65         unsigned long long offset;
  66         unsigned long length;
  67         char *buffer;
  68         int sectorsize;
  69         unsigned long sector_mask;
  70         unsigned long long cow_offset;
  71         unsigned long bitmap_words[2];
  72         int error;
  73 };
  74
  75 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  76 {
  77         __u64 n;
  78         int bits, off;
  79
  80         bits = sizeof(data[0]) * 8;
  81         n = bit / bits;
  82         off = bit % bits;
  83         return (data[n] & (1 << off)) != 0;
  84 }
  85
  86 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  87 {
  88         __u64 n;
  89         int bits, off;
  90
  91         bits = sizeof(data[0]) * 8;
  92         n = bit / bits;
  93         off = bit % bits;
  94         data[n] |= (1 << off);
  95 }
  96 /*End stuff from ubd_user.h*/
  97
  98 #define DRIVER_NAME "uml-blkdev"
  99
 100 static DEFINE_MUTEX(ubd_lock);
 101 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
 102
 103 static int ubd_open(struct block_device *bdev, fmode_t mode);
 104 static int ubd_release(struct gendisk *disk, fmode_t mode);
 105 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
 106                      unsigned int cmd, unsigned long arg);
 107 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 108
 109 #define MAX_DEV (16)
 110
 111 static const struct block_device_operations ubd_blops = {
 112         .owner          = THIS_MODULE,
 113         .open           = ubd_open,
 114         .release        = ubd_release,
 115         .ioctl          = ubd_ioctl,
 116         .getgeo         = ubd_getgeo,
 117 };
 118
 119 /* Protected by ubd_lock */
 120 static int fake_major = UBD_MAJOR;
 121 static struct gendisk *ubd_gendisk[MAX_DEV];
 122 static struct gendisk *fake_gendisk[MAX_DEV];
 123
 124 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 125 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 126                                          .cl = 1 })
 127 #else
 128 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 129                                          .cl = 1 })
 130 #endif
 131 static struct openflags global_openflags = OPEN_FLAGS;
 132
 133 struct cow {
 134         /* backing file name */
 135         char *file;
 136         /* backing file fd */
 137         int fd;
 138         unsigned long *bitmap;
 139         unsigned long bitmap_len;
 140         int bitmap_offset;
 141         int data_offset;
 142 };
 143
 144 #define MAX_SG 64
 145
 146 struct ubd {
 147         struct list_head restart;
 148         /* name (and fd, below) of the file opened for writing, either the
 149          * backing or the cow file. */
 150         char *file;
 151         int count;
 152         int fd;
 153         __u64 size;
 154         struct openflags boot_openflags;
 155         struct openflags openflags;
 156         unsigned shared:1;
 157         unsigned no_cow:1;
 158         struct cow cow;
 159         struct platform_device pdev;
 160         struct request_queue *queue;
 161         spinlock_t lock;
 162         struct scatterlist sg[MAX_SG];
 163         struct request *request;
 164         int start_sg, end_sg;
 165         sector_t rq_pos;
 166 };
 167
 168 #define DEFAULT_COW { \
 169         .file =                 NULL, \
 170         .fd =                   -1,     \
 171         .bitmap =               NULL, \
 172         .bitmap_offset =        0, \
 173         .data_offset =          0, \
 174 }
 175
 176 #define DEFAULT_UBD { \
 177         .file =                 NULL, \
 178         .count =                0, \
 179         .fd =                   -1, \
 180         .size =                 -1, \
 181         .boot_openflags =       OPEN_FLAGS, \
 182         .openflags =            OPEN_FLAGS, \
 183         .no_cow =               0, \
 184         .shared =               0, \
 185         .cow =                  DEFAULT_COW, \
 186         .lock =                 __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
 187         .request =              NULL, \
 188         .start_sg =             0, \
 189         .end_sg =               0, \
 190         .rq_pos =               0, \
 191 }
 192
 193 /* Protected by ubd_lock */
 194 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
 195
 196 /* Only changed by fake_ide_setup which is a setup */
 197 static int fake_ide = 0;
 198 static struct proc_dir_entry *proc_ide_root = NULL;
 199 static struct proc_dir_entry *proc_ide = NULL;
 200
 201 static void make_proc_ide(void)
 202 {
 203         proc_ide_root = proc_mkdir("ide", NULL);
 204         proc_ide = proc_mkdir("ide0", proc_ide_root);
 205 }
 206
 207 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
 208 {
 209         seq_puts(m, "disk\n");
 210         return 0;
 211 }
 212
 213 static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
 214 {
 215         return single_open(file, fake_ide_media_proc_show, NULL);
 216 }
 217
 218 static const struct file_operations fake_ide_media_proc_fops = {
 219         .owner          = THIS_MODULE,
 220         .open           = fake_ide_media_proc_open,
 221         .read           = seq_read,
 222         .llseek         = seq_lseek,
 223         .release        = single_release,
 224 };
 225
 226 static void make_ide_entries(const char *dev_name)
 227 {
 228         struct proc_dir_entry *dir, *ent;
 229         char name[64];
 230
 231         if(proc_ide_root == NULL) make_proc_ide();
 232
 233         dir = proc_mkdir(dev_name, proc_ide);
 234         if(!dir) return;
 235
 236         ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
 237         if(!ent) return;
 238         snprintf(name, sizeof(name), "ide0/%s", dev_name);
 239         proc_symlink(dev_name, proc_ide_root, name);
 240 }
 241
 242 static int fake_ide_setup(char *str)
 243 {
 244         fake_ide = 1;
 245         return 1;
 246 }
 247
 248 __setup("fake_ide", fake_ide_setup);
 249
 250 __uml_help(fake_ide_setup,
 251 "fake_ide\n"
 252 "    Create ide0 entries that map onto ubd devices.\n\n"
 253 );
 254
 255 static int parse_unit(char **ptr)
 256 {
 257         char *str = *ptr, *end;
 258         int n = -1;
 259
 260         if(isdigit(*str)) {
 261                 n = simple_strtoul(str, &end, 0);
 262                 if(end == str)
 263                         return -1;
 264                 *ptr = end;
 265         }
 266         else if (('a' <= *str) && (*str <= 'z')) {
 267                 n = *str - 'a';
 268                 str++;
 269                 *ptr = str;
 270         }
 271         return n;
 272 }
 273
 274 /* If *index_out == -1 at exit, the passed option was a general one;
 275  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 276  * should not be freed on exit.
 277  */
 278 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 279 {
 280         struct ubd *ubd_dev;
 281         struct openflags flags = global_openflags;
 282         char *backing_file;
 283         int n, err = 0, i;
 284
 285         if(index_out) *index_out = -1;
 286         n = *str;
 287         if(n == '='){
 288                 char *end;
 289                 int major;
 290
 291                 str++;
 292                 if(!strcmp(str, "sync")){
 293                         global_openflags = of_sync(global_openflags);
 294                         goto out1;
 295                 }
 296
 297                 err = -EINVAL;
 298                 major = simple_strtoul(str, &end, 0);
 299                 if((*end != '\0') || (end == str)){
 300                         *error_out = "Didn't parse major number";
 301                         goto out1;
 302                 }
 303
 304                 mutex_lock(&ubd_lock);
 305                 if (fake_major != UBD_MAJOR) {
 306                         *error_out = "Can't assign a fake major twice";
 307                         goto out1;
 308                 }
 309
 310                 fake_major = major;
 311
 312                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 313                        major);
 314                 err = 0;
 315         out1:
 316                 mutex_unlock(&ubd_lock);
 317                 return err;
 318         }
 319
 320         n = parse_unit(&str);
 321         if(n < 0){
 322                 *error_out = "Couldn't parse device number";
 323                 return -EINVAL;
 324         }
 325         if(n >= MAX_DEV){
 326                 *error_out = "Device number out of range";
 327                 return 1;
 328         }
 329
 330         err = -EBUSY;
 331         mutex_lock(&ubd_lock);
 332
 333         ubd_dev = &ubd_devs[n];
 334         if(ubd_dev->file != NULL){
 335                 *error_out = "Device is already configured";
 336                 goto out;
 337         }
 338
 339         if (index_out)
 340                 *index_out = n;
 341
 342         err = -EINVAL;
 343         for (i = 0; i < sizeof("rscd="); i++) {
 344                 switch (*str) {
 345                 case 'r':
 346                         flags.w = 0;
 347                         break;
 348                 case 's':
 349                         flags.s = 1;
 350                         break;
 351                 case 'd':
 352                         ubd_dev->no_cow = 1;
 353                         break;
 354                 case 'c':
 355                         ubd_dev->shared = 1;
 356                         break;
 357                 case '=':
 358                         str++;
 359                         goto break_loop;
 360                 default:
 361                         *error_out = "Expected '=' or flag letter "
 362                                 "(r, s, c, or d)";
 363                         goto out;
 364                 }
 365                 str++;
 366         }
 367
 368         if (*str == '=')
 369                 *error_out = "Too many flags specified";
 370         else
 371                 *error_out = "Missing '='";
 372         goto out;
 373
 374 break_loop:
 375         backing_file = strchr(str, ',');
 376
 377         if (backing_file == NULL)
 378                 backing_file = strchr(str, ':');
 379
 380         if(backing_file != NULL){
 381                 if(ubd_dev->no_cow){
 382                         *error_out = "Can't specify both 'd' and a cow file";
 383                         goto out;
 384                 }
 385                 else {
 386                         *backing_file = '\0';
 387                         backing_file++;
 388                 }
 389         }
 390         err = 0;
 391         ubd_dev->file = str;
 392         ubd_dev->cow.file = backing_file;
 393         ubd_dev->boot_openflags = flags;
 394 out:
 395         mutex_unlock(&ubd_lock);
 396         return err;
 397 }
 398
 399 static int ubd_setup(char *str)
 400 {
 401         char *error;
 402         int err;
 403
 404         err = ubd_setup_common(str, NULL, &error);
 405         if(err)
 406                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 407                        "%s\n", str, error);
 408         return 1;
 409 }
 410
 411 __setup("ubd", ubd_setup);
 412 __uml_help(ubd_setup,
 413 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 414 "    This is used to associate a device with a file in the underlying\n"
 415 "    filesystem. When specifying two filenames, the first one is the\n"
 416 "    COW name and the second is the backing file name. As separator you can\n"
 417 "    use either a ':' or a ',': the first one allows writing things like;\n"
 418 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 419 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 420 "    When using only one filename, UML will detect whether to treat it like\n"
 421 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 422 "    flag:\n"
 423 "       ubd0d=BackingFile\n"
 424 "    Usually, there is a filesystem in the file, but \n"
 425 "    that's not required. Swap devices containing swap files can be\n"
 426 "    specified like this. Also, a file which doesn't contain a\n"
 427 "    filesystem can have its contents read in the virtual \n"
 428 "    machine by running 'dd' on the device. <n> must be in the range\n"
 429 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 430 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 431 "    an 's' will cause data to be written to disk on the host immediately.\n"
 432 "    'c' will cause the device to be treated as being shared between multiple\n"
 433 "    UMLs and file locking will be turned off - this is appropriate for a\n"
 434 "    cluster filesystem and inappropriate at almost all other times.\n\n"
 435 );
 436
 437 static int udb_setup(char *str)
 438 {
 439         printk("udb%s specified on command line is almost certainly a ubd -> "
 440                "udb TYPO\n", str);
 441         return 1;
 442 }
 443
 444 __setup("udb", udb_setup);
 445 __uml_help(udb_setup,
 446 "udb\n"
 447 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 448 "    to impossible to catch visually unless you specifically look for\n"
 449 "    them.  The only result of any option starting with 'udb' is an error\n"
 450 "    in the boot output.\n\n"
 451 );
 452
 453 static void do_ubd_request(struct request_queue * q);
 454
 455 /* Only changed by ubd_init, which is an initcall. */
 456 static int thread_fd = -1;
 457 static LIST_HEAD(restart);
 458
 459 /* XXX - move this inside ubd_intr. */
 460 /* Called without dev->lock held, and only in interrupt context. */
 461 static void ubd_handler(void)
 462 {
 463         struct io_thread_req *req;
 464         struct ubd *ubd;
 465         struct list_head *list, *next_ele;
 466         unsigned long flags;
 467         int n;
 468
 469         while(1){
 470                 n = os_read_file(thread_fd, &req,
 471                                  sizeof(struct io_thread_req *));
 472                 if(n != sizeof(req)){
 473                         if(n == -EAGAIN)
 474                                 break;
 475                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 476                                "err = %d\n", -n);
 477                         return;
 478                 }
 479
 480                 blk_end_request(req->req, 0, req->length);
 481                 kfree(req);
 482         }
 483         reactivate_fd(thread_fd, UBD_IRQ);
 484
 485         list_for_each_safe(list, next_ele, &restart){
 486                 ubd = container_of(list, struct ubd, restart);
 487                 list_del_init(&ubd->restart);
 488                 spin_lock_irqsave(&ubd->lock, flags);
 489                 do_ubd_request(ubd->queue);
 490                 spin_unlock_irqrestore(&ubd->lock, flags);
 491         }
 492 }
 493
 494 static irqreturn_t ubd_intr(int irq, void *dev)
 495 {
 496         ubd_handler();
 497         return IRQ_HANDLED;
 498 }
 499
 500 /* Only changed by ubd_init, which is an initcall. */
 501 static int io_pid = -1;
 502
 503 static void kill_io_thread(void)
 504 {
 505         if(io_pid != -1)
 506                 os_kill_process(io_pid, 1);
 507 }
 508
 509 __uml_exitcall(kill_io_thread);
 510
 511 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 512 {
 513         char *file;
 514         int fd;
 515         int err;
 516
 517         __u32 version;
 518         __u32 align;
 519         char *backing_file;
 520         time_t mtime;
 521         unsigned long long size;
 522         int sector_size;
 523         int bitmap_offset;
 524
 525         if (ubd_dev->file && ubd_dev->cow.file) {
 526                 file = ubd_dev->cow.file;
 527
 528                 goto out;
 529         }
 530
 531         fd = os_open_file(ubd_dev->file, global_openflags, 0);
 532         if (fd < 0)
 533                 return fd;
 534
 535         err = read_cow_header(file_reader, &fd, &version, &backing_file, \
 536                 &mtime, &size, &sector_size, &align, &bitmap_offset);
 537         os_close_file(fd);
 538
 539         if(err == -EINVAL)
 540                 file = ubd_dev->file;
 541         else
 542                 file = backing_file;
 543
 544 out:
 545         return os_file_size(file, size_out);
 546 }
 547
 548 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
 549 {
 550         int err;
 551
 552         err = os_seek_file(fd, offset);
 553         if (err < 0)
 554                 return err;
 555
 556         err = os_read_file(fd, buf, len);
 557         if (err < 0)
 558                 return err;
 559
 560         return 0;
 561 }
 562
 563 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
 564 {
 565         unsigned long modtime;
 566         unsigned long long actual;
 567         int err;
 568
 569         err = os_file_modtime(file, &modtime);
 570         if (err < 0) {
 571                 printk(KERN_ERR "Failed to get modification time of backing "
 572                        "file \"%s\", err = %d\n", file, -err);
 573                 return err;
 574         }
 575
 576         err = os_file_size(file, &actual);
 577         if (err < 0) {
 578                 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
 579                        "err = %d\n", file, -err);
 580                 return err;
 581         }
 582
 583         if (actual != size) {
 584                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
 585                  * the typecast.*/
 586                 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
 587                        "vs backing file\n", (unsigned long long) size, actual);
 588                 return -EINVAL;
 589         }
 590         if (modtime != mtime) {
 591                 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
 592                        "backing file\n", mtime, modtime);
 593                 return -EINVAL;
 594         }
 595         return 0;
 596 }
 597
 598 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
 599 {
 600         struct uml_stat buf1, buf2;
 601         int err;
 602
 603         if (from_cmdline == NULL)
 604                 return 0;
 605         if (!strcmp(from_cmdline, from_cow))
 606                 return 0;
 607
 608         err = os_stat_file(from_cmdline, &buf1);
 609         if (err < 0) {
 610                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
 611                        -err);
 612                 return 0;
 613         }
 614         err = os_stat_file(from_cow, &buf2);
 615         if (err < 0) {
 616                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
 617                        -err);
 618                 return 1;
 619         }
 620         if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
 621                 return 0;
 622
 623         printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
 624                "\"%s\" specified in COW header of \"%s\"\n",
 625                from_cmdline, from_cow, cow);
 626         return 1;
 627 }
 628
 629 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
 630                   char **backing_file_out, int *bitmap_offset_out,
 631                   unsigned long *bitmap_len_out, int *data_offset_out,
 632                   int *create_cow_out)
 633 {
 634         time_t mtime;
 635         unsigned long long size;
 636         __u32 version, align;
 637         char *backing_file;
 638         int fd, err, sectorsize, asked_switch, mode = 0644;
 639
 640         fd = os_open_file(file, *openflags, mode);
 641         if (fd < 0) {
 642                 if ((fd == -ENOENT) && (create_cow_out != NULL))
 643                         *create_cow_out = 1;
 644                 if (!openflags->w ||
 645                     ((fd != -EROFS) && (fd != -EACCES)))
 646                         return fd;
 647                 openflags->w = 0;
 648                 fd = os_open_file(file, *openflags, mode);
 649                 if (fd < 0)
 650                         return fd;
 651         }
 652
 653         if (shared)
 654                 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
 655         else {
 656                 err = os_lock_file(fd, openflags->w);
 657                 if (err < 0) {
 658                         printk(KERN_ERR "Failed to lock '%s', err = %d\n",
 659                                file, -err);
 660                         goto out_close;
 661                 }
 662         }
 663
 664         /* Successful return case! */
 665         if (backing_file_out == NULL)
 666                 return fd;
 667
 668         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
 669                               &size, &sectorsize, &align, bitmap_offset_out);
 670         if (err && (*backing_file_out != NULL)) {
 671                 printk(KERN_ERR "Failed to read COW header from COW file "
 672                        "\"%s\", errno = %d\n", file, -err);
 673                 goto out_close;
 674         }
 675         if (err)
 676                 return fd;
 677
 678         asked_switch = path_requires_switch(*backing_file_out, backing_file,
 679                                             file);
 680
 681         /* Allow switching only if no mismatch. */
 682         if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
 683                                                    mtime)) {
 684                 printk(KERN_ERR "Switching backing file to '%s'\n",
 685                        *backing_file_out);
 686                 err = write_cow_header(file, fd, *backing_file_out,
 687                                        sectorsize, align, &size);
 688                 if (err) {
 689                         printk(KERN_ERR "Switch failed, errno = %d\n", -err);
 690                         goto out_close;
 691                 }
 692         } else {
 693                 *backing_file_out = backing_file;
 694                 err = backing_file_mismatch(*backing_file_out, size, mtime);
 695                 if (err)
 696                         goto out_close;
 697         }
 698
 699         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
 700                   bitmap_len_out, data_offset_out);
 701
 702         return fd;
 703  out_close:
 704         os_close_file(fd);
 705         return err;
 706 }
 707
 708 static int create_cow_file(char *cow_file, char *backing_file,
 709                     struct openflags flags,
 710                     int sectorsize, int alignment, int *bitmap_offset_out,
 711                     unsigned long *bitmap_len_out, int *data_offset_out)
 712 {
 713         int err, fd;
 714
 715         flags.c = 1;
 716         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
 717         if (fd < 0) {
 718                 err = fd;
 719                 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
 720                        cow_file, -err);
 721                 goto out;
 722         }
 723
 724         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
 725                             bitmap_offset_out, bitmap_len_out,
 726                             data_offset_out);
 727         if (!err)
 728                 return fd;
 729         os_close_file(fd);
 730  out:
 731         return err;
 732 }
 733
 734 static void ubd_close_dev(struct ubd *ubd_dev)
 735 {
 736         os_close_file(ubd_dev->fd);
 737         if(ubd_dev->cow.file == NULL)
 738                 return;
 739
 740         os_close_file(ubd_dev->cow.fd);
 741         vfree(ubd_dev->cow.bitmap);
 742         ubd_dev->cow.bitmap = NULL;
 743 }
 744
 745 static int ubd_open_dev(struct ubd *ubd_dev)
 746 {
 747         struct openflags flags;
 748         char **back_ptr;
 749         int err, create_cow, *create_ptr;
 750         int fd;
 751
 752         ubd_dev->openflags = ubd_dev->boot_openflags;
 753         create_cow = 0;
 754         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 755         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 756
 757         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 758                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 759                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 760                                 create_ptr);
 761
 762         if((fd == -ENOENT) && create_cow){
 763                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 764                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 765                                           &ubd_dev->cow.bitmap_offset,
 766                                           &ubd_dev->cow.bitmap_len,
 767                                           &ubd_dev->cow.data_offset);
 768                 if(fd >= 0){
 769                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 770                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 771                 }
 772         }
 773
 774         if(fd < 0){
 775                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 776                        -fd);
 777                 return fd;
 778         }
 779         ubd_dev->fd = fd;
 780
 781         if(ubd_dev->cow.file != NULL){
 782                 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
 783
 784                 err = -ENOMEM;
 785                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 786                 if(ubd_dev->cow.bitmap == NULL){
 787                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 788                         goto error;
 789                 }
 790                 flush_tlb_kernel_vm();
 791
 792                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 793                                       ubd_dev->cow.bitmap_offset,
 794                                       ubd_dev->cow.bitmap_len);
 795                 if(err < 0)
 796                         goto error;
 797
 798                 flags = ubd_dev->openflags;
 799                 flags.w = 0;
 800                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 801                                     NULL, NULL, NULL, NULL);
 802                 if(err < 0) goto error;
 803                 ubd_dev->cow.fd = err;
 804         }
 805         return 0;
 806  error:
 807         os_close_file(ubd_dev->fd);
 808         return err;
 809 }
 810
 811 static void ubd_device_release(struct device *dev)
 812 {
 813         struct ubd *ubd_dev = dev_get_drvdata(dev);
 814
 815         blk_cleanup_queue(ubd_dev->queue);
 816         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 817 }
 818
 819 static int ubd_disk_register(int major, u64 size, int unit,
 820                              struct gendisk **disk_out)
 821 {
 822         struct gendisk *disk;
 823
 824         disk = alloc_disk(1 << UBD_SHIFT);
 825         if(disk == NULL)
 826                 return -ENOMEM;
 827
 828         disk->major = major;
 829         disk->first_minor = unit << UBD_SHIFT;
 830         disk->fops = &ubd_blops;
 831         set_capacity(disk, size / 512);
 832         if (major == UBD_MAJOR)
 833                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 834         else
 835                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 836
 837         /* sysfs register (not for ide fake devices) */
 838         if (major == UBD_MAJOR) {
 839                 ubd_devs[unit].pdev.id   = unit;
 840                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 841                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 842                 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
 843                 platform_device_register(&ubd_devs[unit].pdev);
 844                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 845         }
 846
 847         disk->private_data = &ubd_devs[unit];
 848         disk->queue = ubd_devs[unit].queue;
 849         add_disk(disk);
 850
 851         *disk_out = disk;
 852         return 0;
 853 }
 854
 855 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 856
 857 static int ubd_add(int n, char **error_out)
 858 {
 859         struct ubd *ubd_dev = &ubd_devs[n];
 860         int err = 0;
 861
 862         if(ubd_dev->file == NULL)
 863                 goto out;
 864
 865         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 866         if(err < 0){
 867                 *error_out = "Couldn't determine size of device's file";
 868                 goto out;
 869         }
 870
 871         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 872
 873         INIT_LIST_HEAD(&ubd_dev->restart);
 874         sg_init_table(ubd_dev->sg, MAX_SG);
 875
 876         err = -ENOMEM;
 877         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 878         if (ubd_dev->queue == NULL) {
 879                 *error_out = "Failed to initialize device queue";
 880                 goto out;
 881         }
 882         ubd_dev->queue->queuedata = ubd_dev;
 883
 884         blk_queue_max_segments(ubd_dev->queue, MAX_SG);
 885         err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
 886         if(err){
 887                 *error_out = "Failed to register device";
 888                 goto out_cleanup;
 889         }
 890
 891         if (fake_major != UBD_MAJOR)
 892                 ubd_disk_register(fake_major, ubd_dev->size, n,
 893                                   &fake_gendisk[n]);
 894
 895         /*
 896          * Perhaps this should also be under the "if (fake_major)" above
 897          * using the fake_disk->disk_name
 898          */
 899         if (fake_ide)
 900                 make_ide_entries(ubd_gendisk[n]->disk_name);
 901
 902         err = 0;
 903 out:
 904         return err;
 905
 906 out_cleanup:
 907         blk_cleanup_queue(ubd_dev->queue);
 908         goto out;
 909 }
 910
 911 static int ubd_config(char *str, char **error_out)
 912 {
 913         int n, ret;
 914
 915         /* This string is possibly broken up and stored, so it's only
 916          * freed if ubd_setup_common fails, or if only general options
 917          * were set.
 918          */
 919         str = kstrdup(str, GFP_KERNEL);
 920         if (str == NULL) {
 921                 *error_out = "Failed to allocate memory";
 922                 return -ENOMEM;
 923         }
 924
 925         ret = ubd_setup_common(str, &n, error_out);
 926         if (ret)
 927                 goto err_free;
 928
 929         if (n == -1) {
 930                 ret = 0;
 931                 goto err_free;
 932         }
 933
 934         mutex_lock(&ubd_lock);
 935         ret = ubd_add(n, error_out);
 936         if (ret)
 937                 ubd_devs[n].file = NULL;
 938         mutex_unlock(&ubd_lock);
 939
 940 out:
 941         return ret;
 942
 943 err_free:
 944         kfree(str);
 945         goto out;
 946 }
 947
 948 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 949 {
 950         struct ubd *ubd_dev;
 951         int n, len = 0;
 952
 953         n = parse_unit(&name);
 954         if((n >= MAX_DEV) || (n < 0)){
 955                 *error_out = "ubd_get_config : device number out of range";
 956                 return -1;
 957         }
 958
 959         ubd_dev = &ubd_devs[n];
 960         mutex_lock(&ubd_lock);
 961
 962         if(ubd_dev->file == NULL){
 963                 CONFIG_CHUNK(str, size, len, "", 1);
 964                 goto out;
 965         }
 966
 967         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 968
 969         if(ubd_dev->cow.file != NULL){
 970                 CONFIG_CHUNK(str, size, len, ",", 0);
 971                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 972         }
 973         else CONFIG_CHUNK(str, size, len, "", 1);
 974
 975  out:
 976         mutex_unlock(&ubd_lock);
 977         return len;
 978 }
 979
 980 static int ubd_id(char **str, int *start_out, int *end_out)
 981 {
 982         int n;
 983
 984         n = parse_unit(str);
 985         *start_out = 0;
 986         *end_out = MAX_DEV - 1;
 987         return n;
 988 }
 989
 990 static int ubd_remove(int n, char **error_out)
 991 {
 992         struct gendisk *disk = ubd_gendisk[n];
 993         struct ubd *ubd_dev;
 994         int err = -ENODEV;
 995
 996         mutex_lock(&ubd_lock);
 997
 998         ubd_dev = &ubd_devs[n];
 999
1000         if(ubd_dev->file == NULL)
1001                 goto out;
1002
1003         /* you cannot remove a open disk */
1004         err = -EBUSY;
1005         if(ubd_dev->count > 0)
1006                 goto out;
1007
1008         ubd_gendisk[n] = NULL;
1009         if(disk != NULL){
1010                 del_gendisk(disk);
1011                 put_disk(disk);
1012         }
1013
1014         if(fake_gendisk[n] != NULL){
1015                 del_gendisk(fake_gendisk[n]);
1016                 put_disk(fake_gendisk[n]);
1017                 fake_gendisk[n] = NULL;
1018         }
1019
1020         err = 0;
1021         platform_device_unregister(&ubd_dev->pdev);
1022 out:
1023         mutex_unlock(&ubd_lock);
1024         return err;
1025 }
1026
1027 /* All these are called by mconsole in process context and without
1028  * ubd-specific locks.  The structure itself is const except for .list.
1029  */
1030 static struct mc_device ubd_mc = {
1031         .list           = LIST_HEAD_INIT(ubd_mc.list),
1032         .name           = "ubd",
1033         .config         = ubd_config,
1034         .get_config     = ubd_get_config,
1035         .id             = ubd_id,
1036         .remove         = ubd_remove,
1037 };
1038
1039 static int __init ubd_mc_init(void)
1040 {
1041         mconsole_register_dev(&ubd_mc);
1042         return 0;
1043 }
1044
1045 __initcall(ubd_mc_init);
1046
1047 static int __init ubd0_init(void)
1048 {
1049         struct ubd *ubd_dev = &ubd_devs[0];
1050
1051         mutex_lock(&ubd_lock);
1052         if(ubd_dev->file == NULL)
1053                 ubd_dev->file = "root_fs";
1054         mutex_unlock(&ubd_lock);
1055
1056         return 0;
1057 }
1058
1059 __initcall(ubd0_init);
1060
1061 /* Used in ubd_init, which is an initcall */
1062 static struct platform_driver ubd_driver = {
1063         .driver = {
1064                 .name  = DRIVER_NAME,
1065         },
1066 };
1067
1068 static int __init ubd_init(void)
1069 {
1070         char *error;
1071         int i, err;
1072
1073         if (register_blkdev(UBD_MAJOR, "ubd"))
1074                 return -1;
1075
1076         if (fake_major != UBD_MAJOR) {
1077                 char name[sizeof("ubd_nnn\0")];
1078
1079                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1080                 if (register_blkdev(fake_major, "ubd"))
1081                         return -1;
1082         }
1083         platform_driver_register(&ubd_driver);
1084         mutex_lock(&ubd_lock);
1085         for (i = 0; i < MAX_DEV; i++){
1086                 err = ubd_add(i, &error);
1087                 if(err)
1088                         printk(KERN_ERR "Failed to initialize ubd device %d :"
1089                                "%s\n", i, error);
1090         }
1091         mutex_unlock(&ubd_lock);
1092         return 0;
1093 }
1094
1095 late_initcall(ubd_init);
1096
1097 static int __init ubd_driver_init(void){
1098         unsigned long stack;
1099         int err;
1100
1101         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1102         if(global_openflags.s){
1103                 printk(KERN_INFO "ubd: Synchronous mode\n");
1104                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1105                  * enough. So use anyway the io thread. */
1106         }
1107         stack = alloc_stack(0, 0);
1108         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1109                                  &thread_fd);
1110         if(io_pid < 0){
1111                 printk(KERN_ERR
1112                        "ubd : Failed to start I/O thread (errno = %d) - "
1113                        "falling back to synchronous I/O\n", -io_pid);
1114                 io_pid = -1;
1115                 return 0;
1116         }
1117         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1118                              IRQF_DISABLED, "ubd", ubd_devs);
1119         if(err != 0)
1120                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1121         return 0;
1122 }
1123
1124 device_initcall(ubd_driver_init);
1125
1126 static int ubd_open(struct block_device *bdev, fmode_t mode)
1127 {
1128         struct gendisk *disk = bdev->bd_disk;
1129         struct ubd *ubd_dev = disk->private_data;
1130         int err = 0;
1131
1132         mutex_lock(&ubd_mutex);
1133         if(ubd_dev->count == 0){
1134                 err = ubd_open_dev(ubd_dev);
1135                 if(err){
1136                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1137                                disk->disk_name, ubd_dev->file, -err);
1138                         goto out;
1139                 }
1140         }
1141         ubd_dev->count++;
1142         set_disk_ro(disk, !ubd_dev->openflags.w);
1143
1144         /* This should no more be needed. And it didn't work anyway to exclude
1145          * read-write remounting of filesystems.*/
1146         /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1147                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1148                 err = -EROFS;
1149         }*/
1150 out:
1151         mutex_unlock(&ubd_mutex);
1152         return err;
1153 }
1154
1155 static int ubd_release(struct gendisk *disk, fmode_t mode)
1156 {
1157         struct ubd *ubd_dev = disk->private_data;
1158
1159         mutex_lock(&ubd_mutex);
1160         if(--ubd_dev->count == 0)
1161                 ubd_close_dev(ubd_dev);
1162         mutex_unlock(&ubd_mutex);
1163         return 0;
1164 }
1165
1166 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1167                           __u64 *cow_offset, unsigned long *bitmap,
1168                           __u64 bitmap_offset, unsigned long *bitmap_words,
1169                           __u64 bitmap_len)
1170 {
1171         __u64 sector = io_offset >> 9;
1172         int i, update_bitmap = 0;
1173
1174         for(i = 0; i < length >> 9; i++){
1175                 if(cow_mask != NULL)
1176                         ubd_set_bit(i, (unsigned char *) cow_mask);
1177                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1178                         continue;
1179
1180                 update_bitmap = 1;
1181                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1182         }
1183
1184         if(!update_bitmap)
1185                 return;
1186
1187         *cow_offset = sector / (sizeof(unsigned long) * 8);
1188
1189         /* This takes care of the case where we're exactly at the end of the
1190          * device, and *cow_offset + 1 is off the end.  So, just back it up
1191          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1192          * for the original diagnosis.
1193          */
1194         if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1195                                          sizeof(unsigned long)) - 1))
1196                 (*cow_offset)--;
1197
1198         bitmap_words[0] = bitmap[*cow_offset];
1199         bitmap_words[1] = bitmap[*cow_offset + 1];
1200
1201         *cow_offset *= sizeof(unsigned long);
1202         *cow_offset += bitmap_offset;
1203 }
1204
1205 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1206                        __u64 bitmap_offset, __u64 bitmap_len)
1207 {
1208         __u64 sector = req->offset >> 9;
1209         int i;
1210
1211         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1212                 panic("Operation too long");
1213
1214         if(req->op == UBD_READ) {
1215                 for(i = 0; i < req->length >> 9; i++){
1216                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1217                                 ubd_set_bit(i, (unsigned char *)
1218                                             &req->sector_mask);
1219                 }
1220         }
1221         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1222                            &req->cow_offset, bitmap, bitmap_offset,
1223                            req->bitmap_words, bitmap_len);
1224 }
1225
1226 /* Called with dev->lock held */
1227 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1228                             unsigned long long offset, int page_offset,
1229                             int len, struct page *page)
1230 {
1231         struct gendisk *disk = req->rq_disk;
1232         struct ubd *ubd_dev = disk->private_data;
1233
1234         io_req->req = req;
1235         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1236                 ubd_dev->fd;
1237         io_req->fds[1] = ubd_dev->fd;
1238         io_req->cow_offset = -1;
1239         io_req->offset = offset;
1240         io_req->length = len;
1241         io_req->error = 0;
1242         io_req->sector_mask = 0;
1243
1244         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1245         io_req->offsets[0] = 0;
1246         io_req->offsets[1] = ubd_dev->cow.data_offset;
1247         io_req->buffer = page_address(page) + page_offset;
1248         io_req->sectorsize = 1 << 9;
1249
1250         if(ubd_dev->cow.file != NULL)
1251                 cowify_req(io_req, ubd_dev->cow.bitmap,
1252                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1253
1254 }
1255
1256 /* Called with dev->lock held */
1257 static void do_ubd_request(struct request_queue *q)
1258 {
1259         struct io_thread_req *io_req;
1260         struct request *req;
1261         int n;
1262
1263         while(1){
1264                 struct ubd *dev = q->queuedata;
1265                 if(dev->end_sg == 0){
1266                         struct request *req = blk_fetch_request(q);
1267                         if(req == NULL)
1268                                 return;
1269
1270                         dev->request = req;
1271                         dev->rq_pos = blk_rq_pos(req);
1272                         dev->start_sg = 0;
1273                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1274                 }
1275
1276                 req = dev->request;
1277                 while(dev->start_sg < dev->end_sg){
1278                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1279
1280                         io_req = kmalloc(sizeof(struct io_thread_req),
1281                                          GFP_ATOMIC);
1282                         if(io_req == NULL){
1283                                 if(list_empty(&dev->restart))
1284                                         list_add(&dev->restart, &restart);
1285                                 return;
1286                         }
1287                         prepare_request(req, io_req,
1288                                         (unsigned long long)dev->rq_pos << 9,
1289                                         sg->offset, sg->length, sg_page(sg));
1290
1291                         n = os_write_file(thread_fd, &io_req,
1292                                           sizeof(struct io_thread_req *));
1293                         if(n != sizeof(struct io_thread_req *)){
1294                                 if(n != -EAGAIN)
1295                                         printk("write to io thread failed, "
1296                                                "errno = %d\n", -n);
1297                                 else if(list_empty(&dev->restart))
1298                                         list_add(&dev->restart, &restart);
1299                                 kfree(io_req);
1300                                 return;
1301                         }
1302
1303                         dev->rq_pos += sg->length >> 9;
1304                         dev->start_sg++;
1305                 }
1306                 dev->end_sg = 0;
1307                 dev->request = NULL;
1308         }
1309 }
1310
1311 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1312 {
1313         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1314
1315         geo->heads = 128;
1316         geo->sectors = 32;
1317         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1318         return 0;
1319 }
1320
1321 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1322                      unsigned int cmd, unsigned long arg)
1323 {
1324         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1325         u16 ubd_id[ATA_ID_WORDS];
1326
1327         switch (cmd) {
1328                 struct cdrom_volctrl volume;
1329         case HDIO_GET_IDENTITY:
1330                 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1331                 ubd_id[ATA_ID_CYLS]     = ubd_dev->size / (128 * 32 * 512);
1332                 ubd_id[ATA_ID_HEADS]    = 128;
1333                 ubd_id[ATA_ID_SECTORS]  = 32;
1334                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1335                                  sizeof(ubd_id)))
1336                         return -EFAULT;
1337                 return 0;
1338
1339         case CDROMVOLREAD:
1340                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1341                         return -EFAULT;
1342                 volume.channel0 = 255;
1343                 volume.channel1 = 255;
1344                 volume.channel2 = 255;
1345                 volume.channel3 = 255;
1346                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1347                         return -EFAULT;
1348                 return 0;
1349         }
1350         return -EINVAL;
1351 }
1352
1353 static int update_bitmap(struct io_thread_req *req)
1354 {
1355         int n;
1356
1357         if(req->cow_offset == -1)
1358                 return 0;
1359
1360         n = os_seek_file(req->fds[1], req->cow_offset);
1361         if(n < 0){
1362                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1363                 return 1;
1364         }
1365
1366         n = os_write_file(req->fds[1], &req->bitmap_words,
1367                           sizeof(req->bitmap_words));
1368         if(n != sizeof(req->bitmap_words)){
1369                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1370                        req->fds[1]);
1371                 return 1;
1372         }
1373
1374         return 0;
1375 }
1376
1377 static void do_io(struct io_thread_req *req)
1378 {
1379         char *buf;
1380         unsigned long len;
1381         int n, nsectors, start, end, bit;
1382         int err;
1383         __u64 off;
1384
1385         nsectors = req->length / req->sectorsize;
1386         start = 0;
1387         do {
1388                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1389                 end = start;
1390                 while((end < nsectors) &&
1391                       (ubd_test_bit(end, (unsigned char *)
1392                                     &req->sector_mask) == bit))
1393                         end++;
1394
1395                 off = req->offset + req->offsets[bit] +
1396                         start * req->sectorsize;
1397                 len = (end - start) * req->sectorsize;
1398                 buf = &req->buffer[start * req->sectorsize];
1399
1400                 err = os_seek_file(req->fds[bit], off);
1401                 if(err < 0){
1402                         printk("do_io - lseek failed : err = %d\n", -err);
1403                         req->error = 1;
1404                         return;
1405                 }
1406                 if(req->op == UBD_READ){
1407                         n = 0;
1408                         do {
1409                                 buf = &buf[n];
1410                                 len -= n;
1411                                 n = os_read_file(req->fds[bit], buf, len);
1412                                 if (n < 0) {
1413                                         printk("do_io - read failed, err = %d "
1414                                                "fd = %d\n", -n, req->fds[bit]);
1415                                         req->error = 1;
1416                                         return;
1417                                 }
1418                         } while((n < len) && (n != 0));
1419                         if (n < len) memset(&buf[n], 0, len - n);
1420                 } else {
1421                         n = os_write_file(req->fds[bit], buf, len);
1422                         if(n != len){
1423                                 printk("do_io - write failed err = %d "
1424                                        "fd = %d\n", -n, req->fds[bit]);
1425                                 req->error = 1;
1426                                 return;
1427                         }
1428                 }
1429
1430                 start = end;
1431         } while(start < nsectors);
1432
1433         req->error = update_bitmap(req);
1434 }
1435
1436 /* Changed in start_io_thread, which is serialized by being called only
1437  * from ubd_init, which is an initcall.
1438  */
1439 int kernel_fd = -1;
1440
1441 /* Only changed by the io thread. XXX: currently unused. */
1442 static int io_count = 0;
1443
1444 int io_thread(void *arg)
1445 {
1446         struct io_thread_req *req;
1447         int n;
1448
1449         ignore_sigwinch_sig();
1450         while(1){
1451                 n = os_read_file(kernel_fd, &req,
1452                                  sizeof(struct io_thread_req *));
1453                 if(n != sizeof(struct io_thread_req *)){
1454                         if(n < 0)
1455                                 printk("io_thread - read failed, fd = %d, "
1456                                        "err = %d\n", kernel_fd, -n);
1457                         else {
1458                                 printk("io_thread - short read, fd = %d, "
1459                                        "length = %d\n", kernel_fd, n);
1460                         }
1461                         continue;
1462                 }
1463                 io_count++;
1464                 do_io(req);
1465                 n = os_write_file(kernel_fd, &req,
1466                                   sizeof(struct io_thread_req *));
1467                 if(n != sizeof(struct io_thread_req *))
1468                         printk("io_thread - write failed, fd = %d, err = %d\n",
1469                                kernel_fd, -n);
1470         }
1471
1472         return 0;
1473 }