SAFE public projects git trees. - safe/jmp/linux-2.6/blob - arch/um/drivers/ubd_kern.c

   1 /*
   2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
   3  * Licensed under the GPL
   4  */
   5
   6 /* 2001-09-28...2002-04-17
   7  * Partition stuff by James_McMechan@hotmail.com
   8  * old style ubd by setting UBD_SHIFT to 0
   9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
  10  * partitions have changed in 2.5
  11  * 2003-01-29 more tinkering for 2.5.59-1
  12  * This should now address the sysfs problems and has
  13  * the symlink for devfs to allow for booting with
  14  * the common /dev/ubd/discX/... names rather than
  15  * only /dev/ubdN/discN this version also has lots of
  16  * clean ups preparing for ubd-many.
  17  * James McMechan
  18  */
  19
  20 #define MAJOR_NR UBD_MAJOR
  21 #define UBD_SHIFT 4
  22
  23 #include "linux/kernel.h"
  24 #include "linux/module.h"
  25 #include "linux/blkdev.h"
  26 #include "linux/hdreg.h"
  27 #include "linux/init.h"
  28 #include "linux/cdrom.h"
  29 #include "linux/proc_fs.h"
  30 #include "linux/ctype.h"
  31 #include "linux/capability.h"
  32 #include "linux/mm.h"
  33 #include "linux/vmalloc.h"
  34 #include "linux/blkpg.h"
  35 #include "linux/genhd.h"
  36 #include "linux/spinlock.h"
  37 #include "linux/platform_device.h"
  38 #include "linux/scatterlist.h"
  39 #include "asm/segment.h"
  40 #include "asm/uaccess.h"
  41 #include "asm/irq.h"
  42 #include "asm/types.h"
  43 #include "asm/tlbflush.h"
  44 #include "mem_user.h"
  45 #include "kern_util.h"
  46 #include "kern.h"
  47 #include "mconsole_kern.h"
  48 #include "init.h"
  49 #include "irq_user.h"
  50 #include "irq_kern.h"
  51 #include "ubd_user.h"
  52 #include "kern_util.h"
  53 #include "os.h"
  54 #include "mem.h"
  55 #include "mem_kern.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 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
  76                          char **backing_file_out, int *bitmap_offset_out,
  77                          unsigned long *bitmap_len_out, int *data_offset_out,
  78                          int *create_cow_out);
  79 extern int create_cow_file(char *cow_file, char *backing_file,
  80                            struct openflags flags, int sectorsize,
  81                            int alignment, int *bitmap_offset_out,
  82                            unsigned long *bitmap_len_out,
  83                            int *data_offset_out);
  84 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
  85 extern void do_io(struct io_thread_req *req);
  86
  87 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  88 {
  89         __u64 n;
  90         int bits, off;
  91
  92         bits = sizeof(data[0]) * 8;
  93         n = bit / bits;
  94         off = bit % bits;
  95         return (data[n] & (1 << off)) != 0;
  96 }
  97
  98 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  99 {
 100         __u64 n;
 101         int bits, off;
 102
 103         bits = sizeof(data[0]) * 8;
 104         n = bit / bits;
 105         off = bit % bits;
 106         data[n] |= (1 << off);
 107 }
 108 /*End stuff from ubd_user.h*/
 109
 110 #define DRIVER_NAME "uml-blkdev"
 111
 112 static DEFINE_MUTEX(ubd_lock);
 113
 114 static int ubd_open(struct inode * inode, struct file * filp);
 115 static int ubd_release(struct inode * inode, struct file * file);
 116 static int ubd_ioctl(struct inode * inode, struct file * file,
 117                      unsigned int cmd, unsigned long arg);
 118 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 119
 120 #define MAX_DEV (16)
 121
 122 static struct block_device_operations ubd_blops = {
 123         .owner          = THIS_MODULE,
 124         .open           = ubd_open,
 125         .release        = ubd_release,
 126         .ioctl          = ubd_ioctl,
 127         .getgeo         = ubd_getgeo,
 128 };
 129
 130 /* Protected by ubd_lock */
 131 static int fake_major = MAJOR_NR;
 132 static struct gendisk *ubd_gendisk[MAX_DEV];
 133 static struct gendisk *fake_gendisk[MAX_DEV];
 134
 135 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 136 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 137                                          .cl = 1 })
 138 #else
 139 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 140                                          .cl = 1 })
 141 #endif
 142 static struct openflags global_openflags = OPEN_FLAGS;
 143
 144 struct cow {
 145         /* backing file name */
 146         char *file;
 147         /* backing file fd */
 148         int fd;
 149         unsigned long *bitmap;
 150         unsigned long bitmap_len;
 151         int bitmap_offset;
 152         int data_offset;
 153 };
 154
 155 #define MAX_SG 64
 156
 157 struct ubd {
 158         struct list_head restart;
 159         /* name (and fd, below) of the file opened for writing, either the
 160          * backing or the cow file. */
 161         char *file;
 162         int count;
 163         int fd;
 164         __u64 size;
 165         struct openflags boot_openflags;
 166         struct openflags openflags;
 167         unsigned shared:1;
 168         unsigned no_cow:1;
 169         struct cow cow;
 170         struct platform_device pdev;
 171         struct request_queue *queue;
 172         spinlock_t lock;
 173         struct scatterlist sg[MAX_SG];
 174         struct request *request;
 175         int start_sg, end_sg;
 176 };
 177
 178 #define DEFAULT_COW { \
 179         .file =                 NULL, \
 180         .fd =                   -1,     \
 181         .bitmap =               NULL, \
 182         .bitmap_offset =        0, \
 183         .data_offset =          0, \
 184 }
 185
 186 #define DEFAULT_UBD { \
 187         .file =                 NULL, \
 188         .count =                0, \
 189         .fd =                   -1, \
 190         .size =                 -1, \
 191         .boot_openflags =       OPEN_FLAGS, \
 192         .openflags =            OPEN_FLAGS, \
 193         .no_cow =               0, \
 194         .shared =               0, \
 195         .cow =                  DEFAULT_COW, \
 196         .lock =                 SPIN_LOCK_UNLOCKED,     \
 197         .request =              NULL, \
 198         .start_sg =             0, \
 199         .end_sg =               0, \
 200 }
 201
 202 /* Protected by ubd_lock */
 203 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
 204
 205 /* Only changed by fake_ide_setup which is a setup */
 206 static int fake_ide = 0;
 207 static struct proc_dir_entry *proc_ide_root = NULL;
 208 static struct proc_dir_entry *proc_ide = NULL;
 209
 210 static void make_proc_ide(void)
 211 {
 212         proc_ide_root = proc_mkdir("ide", NULL);
 213         proc_ide = proc_mkdir("ide0", proc_ide_root);
 214 }
 215
 216 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
 217                                int *eof, void *data)
 218 {
 219         int len;
 220
 221         strcpy(page, "disk\n");
 222         len = strlen("disk\n");
 223         len -= off;
 224         if (len < count){
 225                 *eof = 1;
 226                 if (len <= 0) return 0;
 227         }
 228         else len = count;
 229         *start = page + off;
 230         return len;
 231 }
 232
 233 static void make_ide_entries(const char *dev_name)
 234 {
 235         struct proc_dir_entry *dir, *ent;
 236         char name[64];
 237
 238         if(proc_ide_root == NULL) make_proc_ide();
 239
 240         dir = proc_mkdir(dev_name, proc_ide);
 241         if(!dir) return;
 242
 243         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
 244         if(!ent) return;
 245         ent->data = NULL;
 246         ent->read_proc = proc_ide_read_media;
 247         ent->write_proc = NULL;
 248         snprintf(name, sizeof(name), "ide0/%s", dev_name);
 249         proc_symlink(dev_name, proc_ide_root, name);
 250 }
 251
 252 static int fake_ide_setup(char *str)
 253 {
 254         fake_ide = 1;
 255         return 1;
 256 }
 257
 258 __setup("fake_ide", fake_ide_setup);
 259
 260 __uml_help(fake_ide_setup,
 261 "fake_ide\n"
 262 "    Create ide0 entries that map onto ubd devices.\n\n"
 263 );
 264
 265 static int parse_unit(char **ptr)
 266 {
 267         char *str = *ptr, *end;
 268         int n = -1;
 269
 270         if(isdigit(*str)) {
 271                 n = simple_strtoul(str, &end, 0);
 272                 if(end == str)
 273                         return -1;
 274                 *ptr = end;
 275         }
 276         else if (('a' <= *str) && (*str <= 'z')) {
 277                 n = *str - 'a';
 278                 str++;
 279                 *ptr = str;
 280         }
 281         return n;
 282 }
 283
 284 /* If *index_out == -1 at exit, the passed option was a general one;
 285  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 286  * should not be freed on exit.
 287  */
 288 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 289 {
 290         struct ubd *ubd_dev;
 291         struct openflags flags = global_openflags;
 292         char *backing_file;
 293         int n, err = 0, i;
 294
 295         if(index_out) *index_out = -1;
 296         n = *str;
 297         if(n == '='){
 298                 char *end;
 299                 int major;
 300
 301                 str++;
 302                 if(!strcmp(str, "sync")){
 303                         global_openflags = of_sync(global_openflags);
 304                         goto out1;
 305                 }
 306
 307                 err = -EINVAL;
 308                 major = simple_strtoul(str, &end, 0);
 309                 if((*end != '\0') || (end == str)){
 310                         *error_out = "Didn't parse major number";
 311                         goto out1;
 312                 }
 313
 314                 mutex_lock(&ubd_lock);
 315                 if(fake_major != MAJOR_NR){
 316                         *error_out = "Can't assign a fake major twice";
 317                         goto out1;
 318                 }
 319
 320                 fake_major = major;
 321
 322                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 323                        major);
 324                 err = 0;
 325         out1:
 326                 mutex_unlock(&ubd_lock);
 327                 return err;
 328         }
 329
 330         n = parse_unit(&str);
 331         if(n < 0){
 332                 *error_out = "Couldn't parse device number";
 333                 return -EINVAL;
 334         }
 335         if(n >= MAX_DEV){
 336                 *error_out = "Device number out of range";
 337                 return 1;
 338         }
 339
 340         err = -EBUSY;
 341         mutex_lock(&ubd_lock);
 342
 343         ubd_dev = &ubd_devs[n];
 344         if(ubd_dev->file != NULL){
 345                 *error_out = "Device is already configured";
 346                 goto out;
 347         }
 348
 349         if (index_out)
 350                 *index_out = n;
 351
 352         err = -EINVAL;
 353         for (i = 0; i < sizeof("rscd="); i++) {
 354                 switch (*str) {
 355                 case 'r':
 356                         flags.w = 0;
 357                         break;
 358                 case 's':
 359                         flags.s = 1;
 360                         break;
 361                 case 'd':
 362                         ubd_dev->no_cow = 1;
 363                         break;
 364                 case 'c':
 365                         ubd_dev->shared = 1;
 366                         break;
 367                 case '=':
 368                         str++;
 369                         goto break_loop;
 370                 default:
 371                         *error_out = "Expected '=' or flag letter "
 372                                 "(r, s, c, or d)";
 373                         goto out;
 374                 }
 375                 str++;
 376         }
 377
 378         if (*str == '=')
 379                 *error_out = "Too many flags specified";
 380         else
 381                 *error_out = "Missing '='";
 382         goto out;
 383
 384 break_loop:
 385         backing_file = strchr(str, ',');
 386
 387         if (backing_file == NULL)
 388                 backing_file = strchr(str, ':');
 389
 390         if(backing_file != NULL){
 391                 if(ubd_dev->no_cow){
 392                         *error_out = "Can't specify both 'd' and a cow file";
 393                         goto out;
 394                 }
 395                 else {
 396                         *backing_file = '\0';
 397                         backing_file++;
 398                 }
 399         }
 400         err = 0;
 401         ubd_dev->file = str;
 402         ubd_dev->cow.file = backing_file;
 403         ubd_dev->boot_openflags = flags;
 404 out:
 405         mutex_unlock(&ubd_lock);
 406         return err;
 407 }
 408
 409 static int ubd_setup(char *str)
 410 {
 411         char *error;
 412         int err;
 413
 414         err = ubd_setup_common(str, NULL, &error);
 415         if(err)
 416                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 417                        "%s\n", str, error);
 418         return 1;
 419 }
 420
 421 __setup("ubd", ubd_setup);
 422 __uml_help(ubd_setup,
 423 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 424 "    This is used to associate a device with a file in the underlying\n"
 425 "    filesystem. When specifying two filenames, the first one is the\n"
 426 "    COW name and the second is the backing file name. As separator you can\n"
 427 "    use either a ':' or a ',': the first one allows writing things like;\n"
 428 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 429 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 430 "    When using only one filename, UML will detect whether to treat it like\n"
 431 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 432 "    flag:\n"
 433 "       ubd0d=BackingFile\n"
 434 "    Usually, there is a filesystem in the file, but \n"
 435 "    that's not required. Swap devices containing swap files can be\n"
 436 "    specified like this. Also, a file which doesn't contain a\n"
 437 "    filesystem can have its contents read in the virtual \n"
 438 "    machine by running 'dd' on the device. <n> must be in the range\n"
 439 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 440 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 441 "    an 's' will cause data to be written to disk on the host immediately.\n"
 442 "    'c' will cause the device to be treated as being shared between multiple\n"
 443 "    UMLs and file locking will be turned off - this is appropriate for a\n"
 444 "    cluster filesystem and inappropriate at almost all other times.\n\n"
 445 );
 446
 447 static int udb_setup(char *str)
 448 {
 449         printk("udb%s specified on command line is almost certainly a ubd -> "
 450                "udb TYPO\n", str);
 451         return 1;
 452 }
 453
 454 __setup("udb", udb_setup);
 455 __uml_help(udb_setup,
 456 "udb\n"
 457 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 458 "    to impossible to catch visually unless you specifically look for\n"
 459 "    them.  The only result of any option starting with 'udb' is an error\n"
 460 "    in the boot output.\n\n"
 461 );
 462
 463 static void do_ubd_request(struct request_queue * q);
 464
 465 /* Only changed by ubd_init, which is an initcall. */
 466 int thread_fd = -1;
 467
 468 static void ubd_end_request(struct request *req, int bytes, int error)
 469 {
 470         blk_end_request(req, error, bytes);
 471 }
 472
 473 /* Callable only from interrupt context - otherwise you need to do
 474  * spin_lock_irq()/spin_lock_irqsave() */
 475 static inline void ubd_finish(struct request *req, int bytes)
 476 {
 477         if(bytes < 0){
 478                 ubd_end_request(req, 0, -EIO);
 479                 return;
 480         }
 481         ubd_end_request(req, bytes, 0);
 482 }
 483
 484 static LIST_HEAD(restart);
 485
 486 /* XXX - move this inside ubd_intr. */
 487 /* Called without dev->lock held, and only in interrupt context. */
 488 static void ubd_handler(void)
 489 {
 490         struct io_thread_req *req;
 491         struct request *rq;
 492         struct ubd *ubd;
 493         struct list_head *list, *next_ele;
 494         unsigned long flags;
 495         int n;
 496
 497         while(1){
 498                 n = os_read_file(thread_fd, &req,
 499                                  sizeof(struct io_thread_req *));
 500                 if(n != sizeof(req)){
 501                         if(n == -EAGAIN)
 502                                 break;
 503                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 504                                "err = %d\n", -n);
 505                         return;
 506                 }
 507
 508                 rq = req->req;
 509                 rq->nr_sectors -= req->length >> 9;
 510                 if(rq->nr_sectors == 0)
 511                         ubd_finish(rq, rq->hard_nr_sectors << 9);
 512                 kfree(req);
 513         }
 514         reactivate_fd(thread_fd, UBD_IRQ);
 515
 516         list_for_each_safe(list, next_ele, &restart){
 517                 ubd = container_of(list, struct ubd, restart);
 518                 list_del_init(&ubd->restart);
 519                 spin_lock_irqsave(&ubd->lock, flags);
 520                 do_ubd_request(ubd->queue);
 521                 spin_unlock_irqrestore(&ubd->lock, flags);
 522         }
 523 }
 524
 525 static irqreturn_t ubd_intr(int irq, void *dev)
 526 {
 527         ubd_handler();
 528         return IRQ_HANDLED;
 529 }
 530
 531 /* Only changed by ubd_init, which is an initcall. */
 532 static int io_pid = -1;
 533
 534 void kill_io_thread(void)
 535 {
 536         if(io_pid != -1)
 537                 os_kill_process(io_pid, 1);
 538 }
 539
 540 __uml_exitcall(kill_io_thread);
 541
 542 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 543 {
 544         char *file;
 545
 546         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
 547         return os_file_size(file, size_out);
 548 }
 549
 550 static void ubd_close_dev(struct ubd *ubd_dev)
 551 {
 552         os_close_file(ubd_dev->fd);
 553         if(ubd_dev->cow.file == NULL)
 554                 return;
 555
 556         os_close_file(ubd_dev->cow.fd);
 557         vfree(ubd_dev->cow.bitmap);
 558         ubd_dev->cow.bitmap = NULL;
 559 }
 560
 561 static int ubd_open_dev(struct ubd *ubd_dev)
 562 {
 563         struct openflags flags;
 564         char **back_ptr;
 565         int err, create_cow, *create_ptr;
 566         int fd;
 567
 568         ubd_dev->openflags = ubd_dev->boot_openflags;
 569         create_cow = 0;
 570         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 571         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 572
 573         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 574                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 575                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 576                                 create_ptr);
 577
 578         if((fd == -ENOENT) && create_cow){
 579                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 580                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 581                                           &ubd_dev->cow.bitmap_offset,
 582                                           &ubd_dev->cow.bitmap_len,
 583                                           &ubd_dev->cow.data_offset);
 584                 if(fd >= 0){
 585                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 586                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 587                 }
 588         }
 589
 590         if(fd < 0){
 591                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 592                        -fd);
 593                 return fd;
 594         }
 595         ubd_dev->fd = fd;
 596
 597         if(ubd_dev->cow.file != NULL){
 598                 blk_queue_max_sectors(ubd_dev->queue, 8 * sizeof(long));
 599
 600                 err = -ENOMEM;
 601                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 602                 if(ubd_dev->cow.bitmap == NULL){
 603                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 604                         goto error;
 605                 }
 606                 flush_tlb_kernel_vm();
 607
 608                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 609                                       ubd_dev->cow.bitmap_offset,
 610                                       ubd_dev->cow.bitmap_len);
 611                 if(err < 0)
 612                         goto error;
 613
 614                 flags = ubd_dev->openflags;
 615                 flags.w = 0;
 616                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 617                                     NULL, NULL, NULL, NULL);
 618                 if(err < 0) goto error;
 619                 ubd_dev->cow.fd = err;
 620         }
 621         return 0;
 622  error:
 623         os_close_file(ubd_dev->fd);
 624         return err;
 625 }
 626
 627 static void ubd_device_release(struct device *dev)
 628 {
 629         struct ubd *ubd_dev = dev->driver_data;
 630
 631         blk_cleanup_queue(ubd_dev->queue);
 632         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 633 }
 634
 635 static int ubd_disk_register(int major, u64 size, int unit,
 636                              struct gendisk **disk_out)
 637 {
 638         struct gendisk *disk;
 639
 640         disk = alloc_disk(1 << UBD_SHIFT);
 641         if(disk == NULL)
 642                 return -ENOMEM;
 643
 644         disk->major = major;
 645         disk->first_minor = unit << UBD_SHIFT;
 646         disk->fops = &ubd_blops;
 647         set_capacity(disk, size / 512);
 648         if(major == MAJOR_NR)
 649                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 650         else
 651                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 652
 653         /* sysfs register (not for ide fake devices) */
 654         if (major == MAJOR_NR) {
 655                 ubd_devs[unit].pdev.id   = unit;
 656                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 657                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 658                 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
 659                 platform_device_register(&ubd_devs[unit].pdev);
 660                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 661         }
 662
 663         disk->private_data = &ubd_devs[unit];
 664         disk->queue = ubd_devs[unit].queue;
 665         add_disk(disk);
 666
 667         *disk_out = disk;
 668         return 0;
 669 }
 670
 671 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 672
 673 static int ubd_add(int n, char **error_out)
 674 {
 675         struct ubd *ubd_dev = &ubd_devs[n];
 676         int err = 0;
 677
 678         if(ubd_dev->file == NULL)
 679                 goto out;
 680
 681         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 682         if(err < 0){
 683                 *error_out = "Couldn't determine size of device's file";
 684                 goto out;
 685         }
 686
 687         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 688
 689         INIT_LIST_HEAD(&ubd_dev->restart);
 690         sg_init_table(ubd_dev->sg, MAX_SG);
 691
 692         err = -ENOMEM;
 693         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 694         if (ubd_dev->queue == NULL) {
 695                 *error_out = "Failed to initialize device queue";
 696                 goto out;
 697         }
 698         ubd_dev->queue->queuedata = ubd_dev;
 699
 700         blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG);
 701         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
 702         if(err){
 703                 *error_out = "Failed to register device";
 704                 goto out_cleanup;
 705         }
 706
 707         if(fake_major != MAJOR_NR)
 708                 ubd_disk_register(fake_major, ubd_dev->size, n,
 709                                   &fake_gendisk[n]);
 710
 711         /*
 712          * Perhaps this should also be under the "if (fake_major)" above
 713          * using the fake_disk->disk_name
 714          */
 715         if (fake_ide)
 716                 make_ide_entries(ubd_gendisk[n]->disk_name);
 717
 718         err = 0;
 719 out:
 720         return err;
 721
 722 out_cleanup:
 723         blk_cleanup_queue(ubd_dev->queue);
 724         goto out;
 725 }
 726
 727 static int ubd_config(char *str, char **error_out)
 728 {
 729         int n, ret;
 730
 731         /* This string is possibly broken up and stored, so it's only
 732          * freed if ubd_setup_common fails, or if only general options
 733          * were set.
 734          */
 735         str = kstrdup(str, GFP_KERNEL);
 736         if (str == NULL) {
 737                 *error_out = "Failed to allocate memory";
 738                 return -ENOMEM;
 739         }
 740
 741         ret = ubd_setup_common(str, &n, error_out);
 742         if (ret)
 743                 goto err_free;
 744
 745         if (n == -1) {
 746                 ret = 0;
 747                 goto err_free;
 748         }
 749
 750         mutex_lock(&ubd_lock);
 751         ret = ubd_add(n, error_out);
 752         if (ret)
 753                 ubd_devs[n].file = NULL;
 754         mutex_unlock(&ubd_lock);
 755
 756 out:
 757         return ret;
 758
 759 err_free:
 760         kfree(str);
 761         goto out;
 762 }
 763
 764 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 765 {
 766         struct ubd *ubd_dev;
 767         int n, len = 0;
 768
 769         n = parse_unit(&name);
 770         if((n >= MAX_DEV) || (n < 0)){
 771                 *error_out = "ubd_get_config : device number out of range";
 772                 return -1;
 773         }
 774
 775         ubd_dev = &ubd_devs[n];
 776         mutex_lock(&ubd_lock);
 777
 778         if(ubd_dev->file == NULL){
 779                 CONFIG_CHUNK(str, size, len, "", 1);
 780                 goto out;
 781         }
 782
 783         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 784
 785         if(ubd_dev->cow.file != NULL){
 786                 CONFIG_CHUNK(str, size, len, ",", 0);
 787                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 788         }
 789         else CONFIG_CHUNK(str, size, len, "", 1);
 790
 791  out:
 792         mutex_unlock(&ubd_lock);
 793         return len;
 794 }
 795
 796 static int ubd_id(char **str, int *start_out, int *end_out)
 797 {
 798         int n;
 799
 800         n = parse_unit(str);
 801         *start_out = 0;
 802         *end_out = MAX_DEV - 1;
 803         return n;
 804 }
 805
 806 static int ubd_remove(int n, char **error_out)
 807 {
 808         struct gendisk *disk = ubd_gendisk[n];
 809         struct ubd *ubd_dev;
 810         int err = -ENODEV;
 811
 812         mutex_lock(&ubd_lock);
 813
 814         ubd_dev = &ubd_devs[n];
 815
 816         if(ubd_dev->file == NULL)
 817                 goto out;
 818
 819         /* you cannot remove a open disk */
 820         err = -EBUSY;
 821         if(ubd_dev->count > 0)
 822                 goto out;
 823
 824         ubd_gendisk[n] = NULL;
 825         if(disk != NULL){
 826                 del_gendisk(disk);
 827                 put_disk(disk);
 828         }
 829
 830         if(fake_gendisk[n] != NULL){
 831                 del_gendisk(fake_gendisk[n]);
 832                 put_disk(fake_gendisk[n]);
 833                 fake_gendisk[n] = NULL;
 834         }
 835
 836         err = 0;
 837         platform_device_unregister(&ubd_dev->pdev);
 838 out:
 839         mutex_unlock(&ubd_lock);
 840         return err;
 841 }
 842
 843 /* All these are called by mconsole in process context and without
 844  * ubd-specific locks.  The structure itself is const except for .list.
 845  */
 846 static struct mc_device ubd_mc = {
 847         .list           = LIST_HEAD_INIT(ubd_mc.list),
 848         .name           = "ubd",
 849         .config         = ubd_config,
 850         .get_config     = ubd_get_config,
 851         .id             = ubd_id,
 852         .remove         = ubd_remove,
 853 };
 854
 855 static int __init ubd_mc_init(void)
 856 {
 857         mconsole_register_dev(&ubd_mc);
 858         return 0;
 859 }
 860
 861 __initcall(ubd_mc_init);
 862
 863 static int __init ubd0_init(void)
 864 {
 865         struct ubd *ubd_dev = &ubd_devs[0];
 866
 867         mutex_lock(&ubd_lock);
 868         if(ubd_dev->file == NULL)
 869                 ubd_dev->file = "root_fs";
 870         mutex_unlock(&ubd_lock);
 871
 872         return 0;
 873 }
 874
 875 __initcall(ubd0_init);
 876
 877 /* Used in ubd_init, which is an initcall */
 878 static struct platform_driver ubd_driver = {
 879         .driver = {
 880                 .name  = DRIVER_NAME,
 881         },
 882 };
 883
 884 static int __init ubd_init(void)
 885 {
 886         char *error;
 887         int i, err;
 888
 889         if (register_blkdev(MAJOR_NR, "ubd"))
 890                 return -1;
 891
 892         if (fake_major != MAJOR_NR) {
 893                 char name[sizeof("ubd_nnn\0")];
 894
 895                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
 896                 if (register_blkdev(fake_major, "ubd"))
 897                         return -1;
 898         }
 899         platform_driver_register(&ubd_driver);
 900         mutex_lock(&ubd_lock);
 901         for (i = 0; i < MAX_DEV; i++){
 902                 err = ubd_add(i, &error);
 903                 if(err)
 904                         printk(KERN_ERR "Failed to initialize ubd device %d :"
 905                                "%s\n", i, error);
 906         }
 907         mutex_unlock(&ubd_lock);
 908         return 0;
 909 }
 910
 911 late_initcall(ubd_init);
 912
 913 static int __init ubd_driver_init(void){
 914         unsigned long stack;
 915         int err;
 916
 917         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
 918         if(global_openflags.s){
 919                 printk(KERN_INFO "ubd: Synchronous mode\n");
 920                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
 921                  * enough. So use anyway the io thread. */
 922         }
 923         stack = alloc_stack(0, 0);
 924         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
 925                                  &thread_fd);
 926         if(io_pid < 0){
 927                 printk(KERN_ERR
 928                        "ubd : Failed to start I/O thread (errno = %d) - "
 929                        "falling back to synchronous I/O\n", -io_pid);
 930                 io_pid = -1;
 931                 return 0;
 932         }
 933         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
 934                              IRQF_DISABLED, "ubd", ubd_devs);
 935         if(err != 0)
 936                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
 937         return 0;
 938 }
 939
 940 device_initcall(ubd_driver_init);
 941
 942 static int ubd_open(struct inode *inode, struct file *filp)
 943 {
 944         struct gendisk *disk = inode->i_bdev->bd_disk;
 945         struct ubd *ubd_dev = disk->private_data;
 946         int err = 0;
 947
 948         if(ubd_dev->count == 0){
 949                 err = ubd_open_dev(ubd_dev);
 950                 if(err){
 951                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
 952                                disk->disk_name, ubd_dev->file, -err);
 953                         goto out;
 954                 }
 955         }
 956         ubd_dev->count++;
 957         set_disk_ro(disk, !ubd_dev->openflags.w);
 958
 959         /* This should no more be needed. And it didn't work anyway to exclude
 960          * read-write remounting of filesystems.*/
 961         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
 962                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
 963                 err = -EROFS;
 964         }*/
 965  out:
 966         return err;
 967 }
 968
 969 static int ubd_release(struct inode * inode, struct file * file)
 970 {
 971         struct gendisk *disk = inode->i_bdev->bd_disk;
 972         struct ubd *ubd_dev = disk->private_data;
 973
 974         if(--ubd_dev->count == 0)
 975                 ubd_close_dev(ubd_dev);
 976         return 0;
 977 }
 978
 979 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
 980                           __u64 *cow_offset, unsigned long *bitmap,
 981                           __u64 bitmap_offset, unsigned long *bitmap_words,
 982                           __u64 bitmap_len)
 983 {
 984         __u64 sector = io_offset >> 9;
 985         int i, update_bitmap = 0;
 986
 987         for(i = 0; i < length >> 9; i++){
 988                 if(cow_mask != NULL)
 989                         ubd_set_bit(i, (unsigned char *) cow_mask);
 990                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
 991                         continue;
 992
 993                 update_bitmap = 1;
 994                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
 995         }
 996
 997         if(!update_bitmap)
 998                 return;
 999
1000         *cow_offset = sector / (sizeof(unsigned long) * 8);
1001
1002         /* This takes care of the case where we're exactly at the end of the
1003          * device, and *cow_offset + 1 is off the end.  So, just back it up
1004          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1005          * for the original diagnosis.
1006          */
1007         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
1008                            sizeof(unsigned long) - 1))
1009                 (*cow_offset)--;
1010
1011         bitmap_words[0] = bitmap[*cow_offset];
1012         bitmap_words[1] = bitmap[*cow_offset + 1];
1013
1014         *cow_offset *= sizeof(unsigned long);
1015         *cow_offset += bitmap_offset;
1016 }
1017
1018 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1019                        __u64 bitmap_offset, __u64 bitmap_len)
1020 {
1021         __u64 sector = req->offset >> 9;
1022         int i;
1023
1024         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1025                 panic("Operation too long");
1026
1027         if(req->op == UBD_READ) {
1028                 for(i = 0; i < req->length >> 9; i++){
1029                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1030                                 ubd_set_bit(i, (unsigned char *)
1031                                             &req->sector_mask);
1032                 }
1033         }
1034         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1035                            &req->cow_offset, bitmap, bitmap_offset,
1036                            req->bitmap_words, bitmap_len);
1037 }
1038
1039 /* Called with dev->lock held */
1040 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1041                             unsigned long long offset, int page_offset,
1042                             int len, struct page *page)
1043 {
1044         struct gendisk *disk = req->rq_disk;
1045         struct ubd *ubd_dev = disk->private_data;
1046
1047         io_req->req = req;
1048         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1049                 ubd_dev->fd;
1050         io_req->fds[1] = ubd_dev->fd;
1051         io_req->cow_offset = -1;
1052         io_req->offset = offset;
1053         io_req->length = len;
1054         io_req->error = 0;
1055         io_req->sector_mask = 0;
1056
1057         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1058         io_req->offsets[0] = 0;
1059         io_req->offsets[1] = ubd_dev->cow.data_offset;
1060         io_req->buffer = page_address(page) + page_offset;
1061         io_req->sectorsize = 1 << 9;
1062
1063         if(ubd_dev->cow.file != NULL)
1064                 cowify_req(io_req, ubd_dev->cow.bitmap,
1065                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1066
1067 }
1068
1069 /* Called with dev->lock held */
1070 static void do_ubd_request(struct request_queue *q)
1071 {
1072         struct io_thread_req *io_req;
1073         struct request *req;
1074         int n, last_sectors;
1075
1076         while(1){
1077                 struct ubd *dev = q->queuedata;
1078                 if(dev->end_sg == 0){
1079                         struct request *req = elv_next_request(q);
1080                         if(req == NULL)
1081                                 return;
1082
1083                         dev->request = req;
1084                         blkdev_dequeue_request(req);
1085                         dev->start_sg = 0;
1086                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1087                 }
1088
1089                 req = dev->request;
1090                 last_sectors = 0;
1091                 while(dev->start_sg < dev->end_sg){
1092                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1093
1094                         req->sector += last_sectors;
1095                         io_req = kmalloc(sizeof(struct io_thread_req),
1096                                          GFP_ATOMIC);
1097                         if(io_req == NULL){
1098                                 if(list_empty(&dev->restart))
1099                                         list_add(&dev->restart, &restart);
1100                                 return;
1101                         }
1102                         prepare_request(req, io_req,
1103                                         (unsigned long long) req->sector << 9,
1104                                         sg->offset, sg->length, sg_page(sg));
1105
1106                         last_sectors = sg->length >> 9;
1107                         n = os_write_file(thread_fd, &io_req,
1108                                           sizeof(struct io_thread_req *));
1109                         if(n != sizeof(struct io_thread_req *)){
1110                                 if(n != -EAGAIN)
1111                                         printk("write to io thread failed, "
1112                                                "errno = %d\n", -n);
1113                                 else if(list_empty(&dev->restart))
1114                                         list_add(&dev->restart, &restart);
1115                                 kfree(io_req);
1116                                 return;
1117                         }
1118
1119                         dev->start_sg++;
1120                 }
1121                 dev->end_sg = 0;
1122                 dev->request = NULL;
1123         }
1124 }
1125
1126 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1127 {
1128         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1129
1130         geo->heads = 128;
1131         geo->sectors = 32;
1132         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1133         return 0;
1134 }
1135
1136 static int ubd_ioctl(struct inode * inode, struct file * file,
1137                      unsigned int cmd, unsigned long arg)
1138 {
1139         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1140         struct hd_driveid ubd_id = {
1141                 .cyls           = 0,
1142                 .heads          = 128,
1143                 .sectors        = 32,
1144         };
1145
1146         switch (cmd) {
1147                 struct cdrom_volctrl volume;
1148         case HDIO_GET_IDENTITY:
1149                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1150                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1151                                  sizeof(ubd_id)))
1152                         return -EFAULT;
1153                 return 0;
1154
1155         case CDROMVOLREAD:
1156                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1157                         return -EFAULT;
1158                 volume.channel0 = 255;
1159                 volume.channel1 = 255;
1160                 volume.channel2 = 255;
1161                 volume.channel3 = 255;
1162                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1163                         return -EFAULT;
1164                 return 0;
1165         }
1166         return -EINVAL;
1167 }
1168
1169 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1170 {
1171         struct uml_stat buf1, buf2;
1172         int err;
1173
1174         if(from_cmdline == NULL)
1175                 return 0;
1176         if(!strcmp(from_cmdline, from_cow))
1177                 return 0;
1178
1179         err = os_stat_file(from_cmdline, &buf1);
1180         if(err < 0){
1181                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1182                 return 0;
1183         }
1184         err = os_stat_file(from_cow, &buf2);
1185         if(err < 0){
1186                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1187                 return 1;
1188         }
1189         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1190                 return 0;
1191
1192         printk("Backing file mismatch - \"%s\" requested,\n"
1193                "\"%s\" specified in COW header of \"%s\"\n",
1194                from_cmdline, from_cow, cow);
1195         return 1;
1196 }
1197
1198 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1199 {
1200         unsigned long modtime;
1201         unsigned long long actual;
1202         int err;
1203
1204         err = os_file_modtime(file, &modtime);
1205         if(err < 0){
1206                 printk("Failed to get modification time of backing file "
1207                        "\"%s\", err = %d\n", file, -err);
1208                 return err;
1209         }
1210
1211         err = os_file_size(file, &actual);
1212         if(err < 0){
1213                 printk("Failed to get size of backing file \"%s\", "
1214                        "err = %d\n", file, -err);
1215                 return err;
1216         }
1217
1218         if(actual != size){
1219                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1220                  * the typecast.*/
1221                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1222                        "file\n", (unsigned long long) size, actual);
1223                 return -EINVAL;
1224         }
1225         if(modtime != mtime){
1226                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1227                        "file\n", mtime, modtime);
1228                 return -EINVAL;
1229         }
1230         return 0;
1231 }
1232
1233 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1234 {
1235         int err;
1236
1237         err = os_seek_file(fd, offset);
1238         if(err < 0)
1239                 return err;
1240
1241         err = os_read_file(fd, buf, len);
1242         if(err < 0)
1243                 return err;
1244
1245         return 0;
1246 }
1247
1248 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1249                   char **backing_file_out, int *bitmap_offset_out,
1250                   unsigned long *bitmap_len_out, int *data_offset_out,
1251                   int *create_cow_out)
1252 {
1253         time_t mtime;
1254         unsigned long long size;
1255         __u32 version, align;
1256         char *backing_file;
1257         int fd, err, sectorsize, asked_switch, mode = 0644;
1258
1259         fd = os_open_file(file, *openflags, mode);
1260         if (fd < 0) {
1261                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1262                         *create_cow_out = 1;
1263                 if (!openflags->w ||
1264                     ((fd != -EROFS) && (fd != -EACCES)))
1265                         return fd;
1266                 openflags->w = 0;
1267                 fd = os_open_file(file, *openflags, mode);
1268                 if (fd < 0)
1269                         return fd;
1270         }
1271
1272         if(shared)
1273                 printk("Not locking \"%s\" on the host\n", file);
1274         else {
1275                 err = os_lock_file(fd, openflags->w);
1276                 if(err < 0){
1277                         printk("Failed to lock '%s', err = %d\n", file, -err);
1278                         goto out_close;
1279                 }
1280         }
1281
1282         /* Successful return case! */
1283         if(backing_file_out == NULL)
1284                 return fd;
1285
1286         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1287                               &size, &sectorsize, &align, bitmap_offset_out);
1288         if(err && (*backing_file_out != NULL)){
1289                 printk("Failed to read COW header from COW file \"%s\", "
1290                        "errno = %d\n", file, -err);
1291                 goto out_close;
1292         }
1293         if(err)
1294                 return fd;
1295
1296         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1297
1298         /* Allow switching only if no mismatch. */
1299         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1300                 printk("Switching backing file to '%s'\n", *backing_file_out);
1301                 err = write_cow_header(file, fd, *backing_file_out,
1302                                        sectorsize, align, &size);
1303                 if (err) {
1304                         printk("Switch failed, errno = %d\n", -err);
1305                         goto out_close;
1306                 }
1307         } else {
1308                 *backing_file_out = backing_file;
1309                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1310                 if (err)
1311                         goto out_close;
1312         }
1313
1314         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1315                   bitmap_len_out, data_offset_out);
1316
1317         return fd;
1318  out_close:
1319         os_close_file(fd);
1320         return err;
1321 }
1322
1323 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1324                     int sectorsize, int alignment, int *bitmap_offset_out,
1325                     unsigned long *bitmap_len_out, int *data_offset_out)
1326 {
1327         int err, fd;
1328
1329         flags.c = 1;
1330         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1331         if(fd < 0){
1332                 err = fd;
1333                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1334                        -err);
1335                 goto out;
1336         }
1337
1338         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1339                             bitmap_offset_out, bitmap_len_out,
1340                             data_offset_out);
1341         if(!err)
1342                 return fd;
1343         os_close_file(fd);
1344  out:
1345         return err;
1346 }
1347
1348 static int update_bitmap(struct io_thread_req *req)
1349 {
1350         int n;
1351
1352         if(req->cow_offset == -1)
1353                 return 0;
1354
1355         n = os_seek_file(req->fds[1], req->cow_offset);
1356         if(n < 0){
1357                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1358                 return 1;
1359         }
1360
1361         n = os_write_file(req->fds[1], &req->bitmap_words,
1362                           sizeof(req->bitmap_words));
1363         if(n != sizeof(req->bitmap_words)){
1364                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1365                        req->fds[1]);
1366                 return 1;
1367         }
1368
1369         return 0;
1370 }
1371
1372 void do_io(struct io_thread_req *req)
1373 {
1374         char *buf;
1375         unsigned long len;
1376         int n, nsectors, start, end, bit;
1377         int err;
1378         __u64 off;
1379
1380         nsectors = req->length / req->sectorsize;
1381         start = 0;
1382         do {
1383                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1384                 end = start;
1385                 while((end < nsectors) &&
1386                       (ubd_test_bit(end, (unsigned char *)
1387                                     &req->sector_mask) == bit))
1388                         end++;
1389
1390                 off = req->offset + req->offsets[bit] +
1391                         start * req->sectorsize;
1392                 len = (end - start) * req->sectorsize;
1393                 buf = &req->buffer[start * req->sectorsize];
1394
1395                 err = os_seek_file(req->fds[bit], off);
1396                 if(err < 0){
1397                         printk("do_io - lseek failed : err = %d\n", -err);
1398                         req->error = 1;
1399                         return;
1400                 }
1401                 if(req->op == UBD_READ){
1402                         n = 0;
1403                         do {
1404                                 buf = &buf[n];
1405                                 len -= n;
1406                                 n = os_read_file(req->fds[bit], buf, len);
1407                                 if (n < 0) {
1408                                         printk("do_io - read failed, err = %d "
1409                                                "fd = %d\n", -n, req->fds[bit]);
1410                                         req->error = 1;
1411                                         return;
1412                                 }
1413                         } while((n < len) && (n != 0));
1414                         if (n < len) memset(&buf[n], 0, len - n);
1415                 } else {
1416                         n = os_write_file(req->fds[bit], buf, len);
1417                         if(n != len){
1418                                 printk("do_io - write failed err = %d "
1419                                        "fd = %d\n", -n, req->fds[bit]);
1420                                 req->error = 1;
1421                                 return;
1422                         }
1423                 }
1424
1425                 start = end;
1426         } while(start < nsectors);
1427
1428         req->error = update_bitmap(req);
1429 }
1430
1431 /* Changed in start_io_thread, which is serialized by being called only
1432  * from ubd_init, which is an initcall.
1433  */
1434 int kernel_fd = -1;
1435
1436 /* Only changed by the io thread. XXX: currently unused. */
1437 static int io_count = 0;
1438
1439 int io_thread(void *arg)
1440 {
1441         struct io_thread_req *req;
1442         int n;
1443
1444         ignore_sigwinch_sig();
1445         while(1){
1446                 n = os_read_file(kernel_fd, &req,
1447                                  sizeof(struct io_thread_req *));
1448                 if(n != sizeof(struct io_thread_req *)){
1449                         if(n < 0)
1450                                 printk("io_thread - read failed, fd = %d, "
1451                                        "err = %d\n", kernel_fd, -n);
1452                         else {
1453                                 printk("io_thread - short read, fd = %d, "
1454                                        "length = %d\n", kernel_fd, n);
1455                         }
1456                         continue;
1457                 }
1458                 io_count++;
1459                 do_io(req);
1460                 n = os_write_file(kernel_fd, &req,
1461                                   sizeof(struct io_thread_req *));
1462                 if(n != sizeof(struct io_thread_req *))
1463                         printk("io_thread - write failed, fd = %d, err = %d\n",
1464                                kernel_fd, -n);
1465         }
1466
1467         return 0;
1468 }