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(const 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 int fakehd_set = 0;
 464 static int fakehd(char *str)
 465 {
 466         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
 467         fakehd_set = 1;
 468         return 1;
 469 }
 470
 471 __setup("fakehd", fakehd);
 472 __uml_help(fakehd,
 473 "fakehd\n"
 474 "    Change the ubd device name to \"hd\".\n\n"
 475 );
 476
 477 static void do_ubd_request(struct request_queue * q);
 478
 479 /* Only changed by ubd_init, which is an initcall. */
 480 int thread_fd = -1;
 481
 482 static void ubd_end_request(struct request *req, int bytes, int error)
 483 {
 484         blk_end_request(req, error, bytes);
 485 }
 486
 487 /* Callable only from interrupt context - otherwise you need to do
 488  * spin_lock_irq()/spin_lock_irqsave() */
 489 static inline void ubd_finish(struct request *req, int bytes)
 490 {
 491         if(bytes < 0){
 492                 ubd_end_request(req, 0, -EIO);
 493                 return;
 494         }
 495         ubd_end_request(req, bytes, 0);
 496 }
 497
 498 static LIST_HEAD(restart);
 499
 500 /* XXX - move this inside ubd_intr. */
 501 /* Called without dev->lock held, and only in interrupt context. */
 502 static void ubd_handler(void)
 503 {
 504         struct io_thread_req *req;
 505         struct request *rq;
 506         struct ubd *ubd;
 507         struct list_head *list, *next_ele;
 508         unsigned long flags;
 509         int n;
 510
 511         while(1){
 512                 n = os_read_file(thread_fd, &req,
 513                                  sizeof(struct io_thread_req *));
 514                 if(n != sizeof(req)){
 515                         if(n == -EAGAIN)
 516                                 break;
 517                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 518                                "err = %d\n", -n);
 519                         return;
 520                 }
 521
 522                 rq = req->req;
 523                 rq->nr_sectors -= req->length >> 9;
 524                 if(rq->nr_sectors == 0)
 525                         ubd_finish(rq, rq->hard_nr_sectors << 9);
 526                 kfree(req);
 527         }
 528         reactivate_fd(thread_fd, UBD_IRQ);
 529
 530         list_for_each_safe(list, next_ele, &restart){
 531                 ubd = container_of(list, struct ubd, restart);
 532                 list_del_init(&ubd->restart);
 533                 spin_lock_irqsave(&ubd->lock, flags);
 534                 do_ubd_request(ubd->queue);
 535                 spin_unlock_irqrestore(&ubd->lock, flags);
 536         }
 537 }
 538
 539 static irqreturn_t ubd_intr(int irq, void *dev)
 540 {
 541         ubd_handler();
 542         return IRQ_HANDLED;
 543 }
 544
 545 /* Only changed by ubd_init, which is an initcall. */
 546 static int io_pid = -1;
 547
 548 void kill_io_thread(void)
 549 {
 550         if(io_pid != -1)
 551                 os_kill_process(io_pid, 1);
 552 }
 553
 554 __uml_exitcall(kill_io_thread);
 555
 556 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 557 {
 558         char *file;
 559
 560         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
 561         return os_file_size(file, size_out);
 562 }
 563
 564 static void ubd_close_dev(struct ubd *ubd_dev)
 565 {
 566         os_close_file(ubd_dev->fd);
 567         if(ubd_dev->cow.file == NULL)
 568                 return;
 569
 570         os_close_file(ubd_dev->cow.fd);
 571         vfree(ubd_dev->cow.bitmap);
 572         ubd_dev->cow.bitmap = NULL;
 573 }
 574
 575 static int ubd_open_dev(struct ubd *ubd_dev)
 576 {
 577         struct openflags flags;
 578         char **back_ptr;
 579         int err, create_cow, *create_ptr;
 580         int fd;
 581
 582         ubd_dev->openflags = ubd_dev->boot_openflags;
 583         create_cow = 0;
 584         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 585         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 586
 587         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 588                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 589                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 590                                 create_ptr);
 591
 592         if((fd == -ENOENT) && create_cow){
 593                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 594                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 595                                           &ubd_dev->cow.bitmap_offset,
 596                                           &ubd_dev->cow.bitmap_len,
 597                                           &ubd_dev->cow.data_offset);
 598                 if(fd >= 0){
 599                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 600                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 601                 }
 602         }
 603
 604         if(fd < 0){
 605                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 606                        -fd);
 607                 return fd;
 608         }
 609         ubd_dev->fd = fd;
 610
 611         if(ubd_dev->cow.file != NULL){
 612                 blk_queue_max_sectors(ubd_dev->queue, 8 * sizeof(long));
 613
 614                 err = -ENOMEM;
 615                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 616                 if(ubd_dev->cow.bitmap == NULL){
 617                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 618                         goto error;
 619                 }
 620                 flush_tlb_kernel_vm();
 621
 622                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 623                                       ubd_dev->cow.bitmap_offset,
 624                                       ubd_dev->cow.bitmap_len);
 625                 if(err < 0)
 626                         goto error;
 627
 628                 flags = ubd_dev->openflags;
 629                 flags.w = 0;
 630                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 631                                     NULL, NULL, NULL, NULL);
 632                 if(err < 0) goto error;
 633                 ubd_dev->cow.fd = err;
 634         }
 635         return 0;
 636  error:
 637         os_close_file(ubd_dev->fd);
 638         return err;
 639 }
 640
 641 static void ubd_device_release(struct device *dev)
 642 {
 643         struct ubd *ubd_dev = dev->driver_data;
 644
 645         blk_cleanup_queue(ubd_dev->queue);
 646         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 647 }
 648
 649 static int ubd_disk_register(int major, u64 size, int unit,
 650                              struct gendisk **disk_out)
 651 {
 652         struct gendisk *disk;
 653
 654         disk = alloc_disk(1 << UBD_SHIFT);
 655         if(disk == NULL)
 656                 return -ENOMEM;
 657
 658         disk->major = major;
 659         disk->first_minor = unit << UBD_SHIFT;
 660         disk->fops = &ubd_blops;
 661         set_capacity(disk, size / 512);
 662         if(major == MAJOR_NR)
 663                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 664         else
 665                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 666
 667         /* sysfs register (not for ide fake devices) */
 668         if (major == MAJOR_NR) {
 669                 ubd_devs[unit].pdev.id   = unit;
 670                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 671                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 672                 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
 673                 platform_device_register(&ubd_devs[unit].pdev);
 674                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 675         }
 676
 677         disk->private_data = &ubd_devs[unit];
 678         disk->queue = ubd_devs[unit].queue;
 679         add_disk(disk);
 680
 681         *disk_out = disk;
 682         return 0;
 683 }
 684
 685 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 686
 687 static int ubd_add(int n, char **error_out)
 688 {
 689         struct ubd *ubd_dev = &ubd_devs[n];
 690         int err = 0;
 691
 692         if(ubd_dev->file == NULL)
 693                 goto out;
 694
 695         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 696         if(err < 0){
 697                 *error_out = "Couldn't determine size of device's file";
 698                 goto out;
 699         }
 700
 701         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 702
 703         INIT_LIST_HEAD(&ubd_dev->restart);
 704         sg_init_table(ubd_dev->sg, MAX_SG);
 705
 706         err = -ENOMEM;
 707         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 708         if (ubd_dev->queue == NULL) {
 709                 *error_out = "Failed to initialize device queue";
 710                 goto out;
 711         }
 712         ubd_dev->queue->queuedata = ubd_dev;
 713
 714         blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG);
 715         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
 716         if(err){
 717                 *error_out = "Failed to register device";
 718                 goto out_cleanup;
 719         }
 720
 721         if(fake_major != MAJOR_NR)
 722                 ubd_disk_register(fake_major, ubd_dev->size, n,
 723                                   &fake_gendisk[n]);
 724
 725         /* perhaps this should also be under the "if (fake_major)" above */
 726         /* using the fake_disk->disk_name and also the fakehd_set name */
 727         if (fake_ide)
 728                 make_ide_entries(ubd_gendisk[n]->disk_name);
 729
 730         err = 0;
 731 out:
 732         return err;
 733
 734 out_cleanup:
 735         blk_cleanup_queue(ubd_dev->queue);
 736         goto out;
 737 }
 738
 739 static int ubd_config(char *str, char **error_out)
 740 {
 741         int n, ret;
 742
 743         /* This string is possibly broken up and stored, so it's only
 744          * freed if ubd_setup_common fails, or if only general options
 745          * were set.
 746          */
 747         str = kstrdup(str, GFP_KERNEL);
 748         if (str == NULL) {
 749                 *error_out = "Failed to allocate memory";
 750                 return -ENOMEM;
 751         }
 752
 753         ret = ubd_setup_common(str, &n, error_out);
 754         if (ret)
 755                 goto err_free;
 756
 757         if (n == -1) {
 758                 ret = 0;
 759                 goto err_free;
 760         }
 761
 762         mutex_lock(&ubd_lock);
 763         ret = ubd_add(n, error_out);
 764         if (ret)
 765                 ubd_devs[n].file = NULL;
 766         mutex_unlock(&ubd_lock);
 767
 768 out:
 769         return ret;
 770
 771 err_free:
 772         kfree(str);
 773         goto out;
 774 }
 775
 776 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 777 {
 778         struct ubd *ubd_dev;
 779         int n, len = 0;
 780
 781         n = parse_unit(&name);
 782         if((n >= MAX_DEV) || (n < 0)){
 783                 *error_out = "ubd_get_config : device number out of range";
 784                 return -1;
 785         }
 786
 787         ubd_dev = &ubd_devs[n];
 788         mutex_lock(&ubd_lock);
 789
 790         if(ubd_dev->file == NULL){
 791                 CONFIG_CHUNK(str, size, len, "", 1);
 792                 goto out;
 793         }
 794
 795         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 796
 797         if(ubd_dev->cow.file != NULL){
 798                 CONFIG_CHUNK(str, size, len, ",", 0);
 799                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 800         }
 801         else CONFIG_CHUNK(str, size, len, "", 1);
 802
 803  out:
 804         mutex_unlock(&ubd_lock);
 805         return len;
 806 }
 807
 808 static int ubd_id(char **str, int *start_out, int *end_out)
 809 {
 810         int n;
 811
 812         n = parse_unit(str);
 813         *start_out = 0;
 814         *end_out = MAX_DEV - 1;
 815         return n;
 816 }
 817
 818 static int ubd_remove(int n, char **error_out)
 819 {
 820         struct gendisk *disk = ubd_gendisk[n];
 821         struct ubd *ubd_dev;
 822         int err = -ENODEV;
 823
 824         mutex_lock(&ubd_lock);
 825
 826         ubd_dev = &ubd_devs[n];
 827
 828         if(ubd_dev->file == NULL)
 829                 goto out;
 830
 831         /* you cannot remove a open disk */
 832         err = -EBUSY;
 833         if(ubd_dev->count > 0)
 834                 goto out;
 835
 836         ubd_gendisk[n] = NULL;
 837         if(disk != NULL){
 838                 del_gendisk(disk);
 839                 put_disk(disk);
 840         }
 841
 842         if(fake_gendisk[n] != NULL){
 843                 del_gendisk(fake_gendisk[n]);
 844                 put_disk(fake_gendisk[n]);
 845                 fake_gendisk[n] = NULL;
 846         }
 847
 848         err = 0;
 849         platform_device_unregister(&ubd_dev->pdev);
 850 out:
 851         mutex_unlock(&ubd_lock);
 852         return err;
 853 }
 854
 855 /* All these are called by mconsole in process context and without
 856  * ubd-specific locks.  The structure itself is const except for .list.
 857  */
 858 static struct mc_device ubd_mc = {
 859         .list           = LIST_HEAD_INIT(ubd_mc.list),
 860         .name           = "ubd",
 861         .config         = ubd_config,
 862         .get_config     = ubd_get_config,
 863         .id             = ubd_id,
 864         .remove         = ubd_remove,
 865 };
 866
 867 static int __init ubd_mc_init(void)
 868 {
 869         mconsole_register_dev(&ubd_mc);
 870         return 0;
 871 }
 872
 873 __initcall(ubd_mc_init);
 874
 875 static int __init ubd0_init(void)
 876 {
 877         struct ubd *ubd_dev = &ubd_devs[0];
 878
 879         mutex_lock(&ubd_lock);
 880         if(ubd_dev->file == NULL)
 881                 ubd_dev->file = "root_fs";
 882         mutex_unlock(&ubd_lock);
 883
 884         return 0;
 885 }
 886
 887 __initcall(ubd0_init);
 888
 889 /* Used in ubd_init, which is an initcall */
 890 static struct platform_driver ubd_driver = {
 891         .driver = {
 892                 .name  = DRIVER_NAME,
 893         },
 894 };
 895
 896 static int __init ubd_init(void)
 897 {
 898         char *error;
 899         int i, err;
 900
 901         if (register_blkdev(MAJOR_NR, "ubd"))
 902                 return -1;
 903
 904         if (fake_major != MAJOR_NR) {
 905                 char name[sizeof("ubd_nnn\0")];
 906
 907                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
 908                 if (register_blkdev(fake_major, "ubd"))
 909                         return -1;
 910         }
 911         platform_driver_register(&ubd_driver);
 912         mutex_lock(&ubd_lock);
 913         for (i = 0; i < MAX_DEV; i++){
 914                 err = ubd_add(i, &error);
 915                 if(err)
 916                         printk(KERN_ERR "Failed to initialize ubd device %d :"
 917                                "%s\n", i, error);
 918         }
 919         mutex_unlock(&ubd_lock);
 920         return 0;
 921 }
 922
 923 late_initcall(ubd_init);
 924
 925 static int __init ubd_driver_init(void){
 926         unsigned long stack;
 927         int err;
 928
 929         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
 930         if(global_openflags.s){
 931                 printk(KERN_INFO "ubd: Synchronous mode\n");
 932                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
 933                  * enough. So use anyway the io thread. */
 934         }
 935         stack = alloc_stack(0, 0);
 936         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
 937                                  &thread_fd);
 938         if(io_pid < 0){
 939                 printk(KERN_ERR
 940                        "ubd : Failed to start I/O thread (errno = %d) - "
 941                        "falling back to synchronous I/O\n", -io_pid);
 942                 io_pid = -1;
 943                 return 0;
 944         }
 945         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
 946                              IRQF_DISABLED, "ubd", ubd_devs);
 947         if(err != 0)
 948                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
 949         return 0;
 950 }
 951
 952 device_initcall(ubd_driver_init);
 953
 954 static int ubd_open(struct inode *inode, struct file *filp)
 955 {
 956         struct gendisk *disk = inode->i_bdev->bd_disk;
 957         struct ubd *ubd_dev = disk->private_data;
 958         int err = 0;
 959
 960         if(ubd_dev->count == 0){
 961                 err = ubd_open_dev(ubd_dev);
 962                 if(err){
 963                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
 964                                disk->disk_name, ubd_dev->file, -err);
 965                         goto out;
 966                 }
 967         }
 968         ubd_dev->count++;
 969         set_disk_ro(disk, !ubd_dev->openflags.w);
 970
 971         /* This should no more be needed. And it didn't work anyway to exclude
 972          * read-write remounting of filesystems.*/
 973         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
 974                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
 975                 err = -EROFS;
 976         }*/
 977  out:
 978         return err;
 979 }
 980
 981 static int ubd_release(struct inode * inode, struct file * file)
 982 {
 983         struct gendisk *disk = inode->i_bdev->bd_disk;
 984         struct ubd *ubd_dev = disk->private_data;
 985
 986         if(--ubd_dev->count == 0)
 987                 ubd_close_dev(ubd_dev);
 988         return 0;
 989 }
 990
 991 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
 992                           __u64 *cow_offset, unsigned long *bitmap,
 993                           __u64 bitmap_offset, unsigned long *bitmap_words,
 994                           __u64 bitmap_len)
 995 {
 996         __u64 sector = io_offset >> 9;
 997         int i, update_bitmap = 0;
 998
 999         for(i = 0; i < length >> 9; i++){
1000                 if(cow_mask != NULL)
1001                         ubd_set_bit(i, (unsigned char *) cow_mask);
1002                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1003                         continue;
1004
1005                 update_bitmap = 1;
1006                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1007         }
1008
1009         if(!update_bitmap)
1010                 return;
1011
1012         *cow_offset = sector / (sizeof(unsigned long) * 8);
1013
1014         /* This takes care of the case where we're exactly at the end of the
1015          * device, and *cow_offset + 1 is off the end.  So, just back it up
1016          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1017          * for the original diagnosis.
1018          */
1019         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
1020                            sizeof(unsigned long) - 1))
1021                 (*cow_offset)--;
1022
1023         bitmap_words[0] = bitmap[*cow_offset];
1024         bitmap_words[1] = bitmap[*cow_offset + 1];
1025
1026         *cow_offset *= sizeof(unsigned long);
1027         *cow_offset += bitmap_offset;
1028 }
1029
1030 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1031                        __u64 bitmap_offset, __u64 bitmap_len)
1032 {
1033         __u64 sector = req->offset >> 9;
1034         int i;
1035
1036         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1037                 panic("Operation too long");
1038
1039         if(req->op == UBD_READ) {
1040                 for(i = 0; i < req->length >> 9; i++){
1041                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1042                                 ubd_set_bit(i, (unsigned char *)
1043                                             &req->sector_mask);
1044                 }
1045         }
1046         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1047                            &req->cow_offset, bitmap, bitmap_offset,
1048                            req->bitmap_words, bitmap_len);
1049 }
1050
1051 /* Called with dev->lock held */
1052 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1053                             unsigned long long offset, int page_offset,
1054                             int len, struct page *page)
1055 {
1056         struct gendisk *disk = req->rq_disk;
1057         struct ubd *ubd_dev = disk->private_data;
1058
1059         io_req->req = req;
1060         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1061                 ubd_dev->fd;
1062         io_req->fds[1] = ubd_dev->fd;
1063         io_req->cow_offset = -1;
1064         io_req->offset = offset;
1065         io_req->length = len;
1066         io_req->error = 0;
1067         io_req->sector_mask = 0;
1068
1069         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1070         io_req->offsets[0] = 0;
1071         io_req->offsets[1] = ubd_dev->cow.data_offset;
1072         io_req->buffer = page_address(page) + page_offset;
1073         io_req->sectorsize = 1 << 9;
1074
1075         if(ubd_dev->cow.file != NULL)
1076                 cowify_req(io_req, ubd_dev->cow.bitmap,
1077                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1078
1079 }
1080
1081 /* Called with dev->lock held */
1082 static void do_ubd_request(struct request_queue *q)
1083 {
1084         struct io_thread_req *io_req;
1085         struct request *req;
1086         int n, last_sectors;
1087
1088         while(1){
1089                 struct ubd *dev = q->queuedata;
1090                 if(dev->end_sg == 0){
1091                         struct request *req = elv_next_request(q);
1092                         if(req == NULL)
1093                                 return;
1094
1095                         dev->request = req;
1096                         blkdev_dequeue_request(req);
1097                         dev->start_sg = 0;
1098                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1099                 }
1100
1101                 req = dev->request;
1102                 last_sectors = 0;
1103                 while(dev->start_sg < dev->end_sg){
1104                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1105
1106                         req->sector += last_sectors;
1107                         io_req = kmalloc(sizeof(struct io_thread_req),
1108                                          GFP_ATOMIC);
1109                         if(io_req == NULL){
1110                                 if(list_empty(&dev->restart))
1111                                         list_add(&dev->restart, &restart);
1112                                 return;
1113                         }
1114                         prepare_request(req, io_req,
1115                                         (unsigned long long) req->sector << 9,
1116                                         sg->offset, sg->length, sg_page(sg));
1117
1118                         last_sectors = sg->length >> 9;
1119                         n = os_write_file(thread_fd, &io_req,
1120                                           sizeof(struct io_thread_req *));
1121                         if(n != sizeof(struct io_thread_req *)){
1122                                 if(n != -EAGAIN)
1123                                         printk("write to io thread failed, "
1124                                                "errno = %d\n", -n);
1125                                 else if(list_empty(&dev->restart))
1126                                         list_add(&dev->restart, &restart);
1127                                 kfree(io_req);
1128                                 return;
1129                         }
1130
1131                         dev->start_sg++;
1132                 }
1133                 dev->end_sg = 0;
1134                 dev->request = NULL;
1135         }
1136 }
1137
1138 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1139 {
1140         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1141
1142         geo->heads = 128;
1143         geo->sectors = 32;
1144         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1145         return 0;
1146 }
1147
1148 static int ubd_ioctl(struct inode * inode, struct file * file,
1149                      unsigned int cmd, unsigned long arg)
1150 {
1151         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1152         struct hd_driveid ubd_id = {
1153                 .cyls           = 0,
1154                 .heads          = 128,
1155                 .sectors        = 32,
1156         };
1157
1158         switch (cmd) {
1159                 struct cdrom_volctrl volume;
1160         case HDIO_GET_IDENTITY:
1161                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1162                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1163                                  sizeof(ubd_id)))
1164                         return -EFAULT;
1165                 return 0;
1166
1167         case CDROMVOLREAD:
1168                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1169                         return -EFAULT;
1170                 volume.channel0 = 255;
1171                 volume.channel1 = 255;
1172                 volume.channel2 = 255;
1173                 volume.channel3 = 255;
1174                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1175                         return -EFAULT;
1176                 return 0;
1177         }
1178         return -EINVAL;
1179 }
1180
1181 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1182 {
1183         struct uml_stat buf1, buf2;
1184         int err;
1185
1186         if(from_cmdline == NULL)
1187                 return 0;
1188         if(!strcmp(from_cmdline, from_cow))
1189                 return 0;
1190
1191         err = os_stat_file(from_cmdline, &buf1);
1192         if(err < 0){
1193                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1194                 return 0;
1195         }
1196         err = os_stat_file(from_cow, &buf2);
1197         if(err < 0){
1198                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1199                 return 1;
1200         }
1201         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1202                 return 0;
1203
1204         printk("Backing file mismatch - \"%s\" requested,\n"
1205                "\"%s\" specified in COW header of \"%s\"\n",
1206                from_cmdline, from_cow, cow);
1207         return 1;
1208 }
1209
1210 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1211 {
1212         unsigned long modtime;
1213         unsigned long long actual;
1214         int err;
1215
1216         err = os_file_modtime(file, &modtime);
1217         if(err < 0){
1218                 printk("Failed to get modification time of backing file "
1219                        "\"%s\", err = %d\n", file, -err);
1220                 return err;
1221         }
1222
1223         err = os_file_size(file, &actual);
1224         if(err < 0){
1225                 printk("Failed to get size of backing file \"%s\", "
1226                        "err = %d\n", file, -err);
1227                 return err;
1228         }
1229
1230         if(actual != size){
1231                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1232                  * the typecast.*/
1233                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1234                        "file\n", (unsigned long long) size, actual);
1235                 return -EINVAL;
1236         }
1237         if(modtime != mtime){
1238                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1239                        "file\n", mtime, modtime);
1240                 return -EINVAL;
1241         }
1242         return 0;
1243 }
1244
1245 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1246 {
1247         int err;
1248
1249         err = os_seek_file(fd, offset);
1250         if(err < 0)
1251                 return err;
1252
1253         err = os_read_file(fd, buf, len);
1254         if(err < 0)
1255                 return err;
1256
1257         return 0;
1258 }
1259
1260 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1261                   char **backing_file_out, int *bitmap_offset_out,
1262                   unsigned long *bitmap_len_out, int *data_offset_out,
1263                   int *create_cow_out)
1264 {
1265         time_t mtime;
1266         unsigned long long size;
1267         __u32 version, align;
1268         char *backing_file;
1269         int fd, err, sectorsize, asked_switch, mode = 0644;
1270
1271         fd = os_open_file(file, *openflags, mode);
1272         if (fd < 0) {
1273                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1274                         *create_cow_out = 1;
1275                 if (!openflags->w ||
1276                     ((fd != -EROFS) && (fd != -EACCES)))
1277                         return fd;
1278                 openflags->w = 0;
1279                 fd = os_open_file(file, *openflags, mode);
1280                 if (fd < 0)
1281                         return fd;
1282         }
1283
1284         if(shared)
1285                 printk("Not locking \"%s\" on the host\n", file);
1286         else {
1287                 err = os_lock_file(fd, openflags->w);
1288                 if(err < 0){
1289                         printk("Failed to lock '%s', err = %d\n", file, -err);
1290                         goto out_close;
1291                 }
1292         }
1293
1294         /* Successful return case! */
1295         if(backing_file_out == NULL)
1296                 return fd;
1297
1298         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1299                               &size, &sectorsize, &align, bitmap_offset_out);
1300         if(err && (*backing_file_out != NULL)){
1301                 printk("Failed to read COW header from COW file \"%s\", "
1302                        "errno = %d\n", file, -err);
1303                 goto out_close;
1304         }
1305         if(err)
1306                 return fd;
1307
1308         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1309
1310         /* Allow switching only if no mismatch. */
1311         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1312                 printk("Switching backing file to '%s'\n", *backing_file_out);
1313                 err = write_cow_header(file, fd, *backing_file_out,
1314                                        sectorsize, align, &size);
1315                 if (err) {
1316                         printk("Switch failed, errno = %d\n", -err);
1317                         goto out_close;
1318                 }
1319         } else {
1320                 *backing_file_out = backing_file;
1321                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1322                 if (err)
1323                         goto out_close;
1324         }
1325
1326         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1327                   bitmap_len_out, data_offset_out);
1328
1329         return fd;
1330  out_close:
1331         os_close_file(fd);
1332         return err;
1333 }
1334
1335 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1336                     int sectorsize, int alignment, int *bitmap_offset_out,
1337                     unsigned long *bitmap_len_out, int *data_offset_out)
1338 {
1339         int err, fd;
1340
1341         flags.c = 1;
1342         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1343         if(fd < 0){
1344                 err = fd;
1345                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1346                        -err);
1347                 goto out;
1348         }
1349
1350         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1351                             bitmap_offset_out, bitmap_len_out,
1352                             data_offset_out);
1353         if(!err)
1354                 return fd;
1355         os_close_file(fd);
1356  out:
1357         return err;
1358 }
1359
1360 static int update_bitmap(struct io_thread_req *req)
1361 {
1362         int n;
1363
1364         if(req->cow_offset == -1)
1365                 return 0;
1366
1367         n = os_seek_file(req->fds[1], req->cow_offset);
1368         if(n < 0){
1369                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1370                 return 1;
1371         }
1372
1373         n = os_write_file(req->fds[1], &req->bitmap_words,
1374                           sizeof(req->bitmap_words));
1375         if(n != sizeof(req->bitmap_words)){
1376                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1377                        req->fds[1]);
1378                 return 1;
1379         }
1380
1381         return 0;
1382 }
1383
1384 void do_io(struct io_thread_req *req)
1385 {
1386         char *buf;
1387         unsigned long len;
1388         int n, nsectors, start, end, bit;
1389         int err;
1390         __u64 off;
1391
1392         nsectors = req->length / req->sectorsize;
1393         start = 0;
1394         do {
1395                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1396                 end = start;
1397                 while((end < nsectors) &&
1398                       (ubd_test_bit(end, (unsigned char *)
1399                                     &req->sector_mask) == bit))
1400                         end++;
1401
1402                 off = req->offset + req->offsets[bit] +
1403                         start * req->sectorsize;
1404                 len = (end - start) * req->sectorsize;
1405                 buf = &req->buffer[start * req->sectorsize];
1406
1407                 err = os_seek_file(req->fds[bit], off);
1408                 if(err < 0){
1409                         printk("do_io - lseek failed : err = %d\n", -err);
1410                         req->error = 1;
1411                         return;
1412                 }
1413                 if(req->op == UBD_READ){
1414                         n = 0;
1415                         do {
1416                                 buf = &buf[n];
1417                                 len -= n;
1418                                 n = os_read_file(req->fds[bit], buf, len);
1419                                 if (n < 0) {
1420                                         printk("do_io - read failed, err = %d "
1421                                                "fd = %d\n", -n, req->fds[bit]);
1422                                         req->error = 1;
1423                                         return;
1424                                 }
1425                         } while((n < len) && (n != 0));
1426                         if (n < len) memset(&buf[n], 0, len - n);
1427                 } else {
1428                         n = os_write_file(req->fds[bit], buf, len);
1429                         if(n != len){
1430                                 printk("do_io - write failed err = %d "
1431                                        "fd = %d\n", -n, req->fds[bit]);
1432                                 req->error = 1;
1433                                 return;
1434                         }
1435                 }
1436
1437                 start = end;
1438         } while(start < nsectors);
1439
1440         req->error = update_bitmap(req);
1441 }
1442
1443 /* Changed in start_io_thread, which is serialized by being called only
1444  * from ubd_init, which is an initcall.
1445  */
1446 int kernel_fd = -1;
1447
1448 /* Only changed by the io thread. XXX: currently unused. */
1449 static int io_count = 0;
1450
1451 int io_thread(void *arg)
1452 {
1453         struct io_thread_req *req;
1454         int n;
1455
1456         ignore_sigwinch_sig();
1457         while(1){
1458                 n = os_read_file(kernel_fd, &req,
1459                                  sizeof(struct io_thread_req *));
1460                 if(n != sizeof(struct io_thread_req *)){
1461                         if(n < 0)
1462                                 printk("io_thread - read failed, fd = %d, "
1463                                        "err = %d\n", kernel_fd, -n);
1464                         else {
1465                                 printk("io_thread - short read, fd = %d, "
1466                                        "length = %d\n", kernel_fd, n);
1467                         }
1468                         continue;
1469                 }
1470                 io_count++;
1471                 do_io(req);
1472                 n = os_write_file(kernel_fd, &req,
1473                                   sizeof(struct io_thread_req *));
1474                 if(n != sizeof(struct io_thread_req *))
1475                         printk("io_thread - write failed, fd = %d, err = %d\n",
1476                                kernel_fd, -n);
1477         }
1478
1479         return 0;
1480 }