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

   1 /*
   2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
   3  * Licensed under the GPL
   4  */
   5
   6 /* 2001-09-28...2002-04-17
   7  * Partition stuff by James_McMechan@hotmail.com
   8  * old style ubd by setting UBD_SHIFT to 0
   9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
  10  * partitions have changed in 2.5
  11  * 2003-01-29 more tinkering for 2.5.59-1
  12  * This should now address the sysfs problems and has
  13  * the symlink for devfs to allow for booting with
  14  * the common /dev/ubd/discX/... names rather than
  15  * only /dev/ubdN/discN this version also has lots of
  16  * clean ups preparing for ubd-many.
  17  * James McMechan
  18  */
  19
  20 #define MAJOR_NR UBD_MAJOR
  21 #define UBD_SHIFT 4
  22
  23 #include "linux/kernel.h"
  24 #include "linux/module.h"
  25 #include "linux/blkdev.h"
  26 #include "linux/hdreg.h"
  27 #include "linux/init.h"
  28 #include "linux/cdrom.h"
  29 #include "linux/proc_fs.h"
  30 #include "linux/ctype.h"
  31 #include "linux/capability.h"
  32 #include "linux/mm.h"
  33 #include "linux/vmalloc.h"
  34 #include "linux/blkpg.h"
  35 #include "linux/genhd.h"
  36 #include "linux/spinlock.h"
  37 #include "linux/platform_device.h"
  38 #include "asm/segment.h"
  39 #include "asm/uaccess.h"
  40 #include "asm/irq.h"
  41 #include "asm/types.h"
  42 #include "asm/tlbflush.h"
  43 #include "mem_user.h"
  44 #include "kern_util.h"
  45 #include "kern.h"
  46 #include "mconsole_kern.h"
  47 #include "init.h"
  48 #include "irq_user.h"
  49 #include "irq_kern.h"
  50 #include "ubd_user.h"
  51 #include "os.h"
  52 #include "mem.h"
  53 #include "mem_kern.h"
  54 #include "cow.h"
  55
  56 enum ubd_req { UBD_READ, UBD_WRITE };
  57
  58 struct io_thread_req {
  59         struct request *req;
  60         enum ubd_req op;
  61         int fds[2];
  62         unsigned long offsets[2];
  63         unsigned long long offset;
  64         unsigned long length;
  65         char *buffer;
  66         int sectorsize;
  67         unsigned long sector_mask;
  68         unsigned long long cow_offset;
  69         unsigned long bitmap_words[2];
  70         int error;
  71 };
  72
  73 extern int open_ubd_file(char *file, struct openflags *openflags, int shared,
  74                          char **backing_file_out, int *bitmap_offset_out,
  75                          unsigned long *bitmap_len_out, int *data_offset_out,
  76                          int *create_cow_out);
  77 extern int create_cow_file(char *cow_file, char *backing_file,
  78                            struct openflags flags, int sectorsize,
  79                            int alignment, int *bitmap_offset_out,
  80                            unsigned long *bitmap_len_out,
  81                            int *data_offset_out);
  82 extern int read_cow_bitmap(int fd, void *buf, int offset, int len);
  83 extern void do_io(struct io_thread_req *req);
  84
  85 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  86 {
  87         __u64 n;
  88         int bits, off;
  89
  90         bits = sizeof(data[0]) * 8;
  91         n = bit / bits;
  92         off = bit % bits;
  93         return (data[n] & (1 << off)) != 0;
  94 }
  95
  96 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  97 {
  98         __u64 n;
  99         int bits, off;
 100
 101         bits = sizeof(data[0]) * 8;
 102         n = bit / bits;
 103         off = bit % bits;
 104         data[n] |= (1 << off);
 105 }
 106 /*End stuff from ubd_user.h*/
 107
 108 #define DRIVER_NAME "uml-blkdev"
 109
 110 static DEFINE_MUTEX(ubd_lock);
 111
 112 static int ubd_open(struct inode * inode, struct file * filp);
 113 static int ubd_release(struct inode * inode, struct file * file);
 114 static int ubd_ioctl(struct inode * inode, struct file * file,
 115                      unsigned int cmd, unsigned long arg);
 116 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 117
 118 #define MAX_DEV (16)
 119
 120 static struct block_device_operations ubd_blops = {
 121         .owner          = THIS_MODULE,
 122         .open           = ubd_open,
 123         .release        = ubd_release,
 124         .ioctl          = ubd_ioctl,
 125         .getgeo         = ubd_getgeo,
 126 };
 127
 128 /* Protected by ubd_lock */
 129 static int fake_major = MAJOR_NR;
 130 static struct gendisk *ubd_gendisk[MAX_DEV];
 131 static struct gendisk *fake_gendisk[MAX_DEV];
 132
 133 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 134 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 135                                          .cl = 1 })
 136 #else
 137 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 138                                          .cl = 1 })
 139 #endif
 140 static struct openflags global_openflags = OPEN_FLAGS;
 141
 142 struct cow {
 143         /* backing file name */
 144         char *file;
 145         /* backing file fd */
 146         int fd;
 147         unsigned long *bitmap;
 148         unsigned long bitmap_len;
 149         int bitmap_offset;
 150         int data_offset;
 151 };
 152
 153 #define MAX_SG 64
 154
 155 struct ubd {
 156         struct list_head restart;
 157         /* name (and fd, below) of the file opened for writing, either the
 158          * backing or the cow file. */
 159         char *file;
 160         int count;
 161         int fd;
 162         __u64 size;
 163         struct openflags boot_openflags;
 164         struct openflags openflags;
 165         unsigned shared:1;
 166         unsigned no_cow:1;
 167         struct cow cow;
 168         struct platform_device pdev;
 169         struct request_queue *queue;
 170         spinlock_t lock;
 171         struct scatterlist sg[MAX_SG];
 172         struct request *request;
 173         int start_sg, end_sg;
 174 };
 175
 176 #define DEFAULT_COW { \
 177         .file =                 NULL, \
 178         .fd =                   -1,     \
 179         .bitmap =               NULL, \
 180         .bitmap_offset =        0, \
 181         .data_offset =          0, \
 182 }
 183
 184 #define DEFAULT_UBD { \
 185         .file =                 NULL, \
 186         .count =                0, \
 187         .fd =                   -1, \
 188         .size =                 -1, \
 189         .boot_openflags =       OPEN_FLAGS, \
 190         .openflags =            OPEN_FLAGS, \
 191         .no_cow =               0, \
 192         .shared =               0, \
 193         .cow =                  DEFAULT_COW, \
 194         .lock =                 SPIN_LOCK_UNLOCKED,     \
 195         .request =              NULL, \
 196         .start_sg =             0, \
 197         .end_sg =               0, \
 198 }
 199
 200 /* Protected by ubd_lock */
 201 struct ubd ubd_devs[MAX_DEV] = { [ 0 ... MAX_DEV - 1 ] = DEFAULT_UBD };
 202
 203 /* Only changed by fake_ide_setup which is a setup */
 204 static int fake_ide = 0;
 205 static struct proc_dir_entry *proc_ide_root = NULL;
 206 static struct proc_dir_entry *proc_ide = NULL;
 207
 208 static void make_proc_ide(void)
 209 {
 210         proc_ide_root = proc_mkdir("ide", NULL);
 211         proc_ide = proc_mkdir("ide0", proc_ide_root);
 212 }
 213
 214 static int proc_ide_read_media(char *page, char **start, off_t off, int count,
 215                                int *eof, void *data)
 216 {
 217         int len;
 218
 219         strcpy(page, "disk\n");
 220         len = strlen("disk\n");
 221         len -= off;
 222         if (len < count){
 223                 *eof = 1;
 224                 if (len <= 0) return 0;
 225         }
 226         else len = count;
 227         *start = page + off;
 228         return len;
 229 }
 230
 231 static void make_ide_entries(char *dev_name)
 232 {
 233         struct proc_dir_entry *dir, *ent;
 234         char name[64];
 235
 236         if(proc_ide_root == NULL) make_proc_ide();
 237
 238         dir = proc_mkdir(dev_name, proc_ide);
 239         if(!dir) return;
 240
 241         ent = create_proc_entry("media", S_IFREG|S_IRUGO, dir);
 242         if(!ent) return;
 243         ent->data = NULL;
 244         ent->read_proc = proc_ide_read_media;
 245         ent->write_proc = NULL;
 246         sprintf(name,"ide0/%s", dev_name);
 247         proc_symlink(dev_name, proc_ide_root, name);
 248 }
 249
 250 static int fake_ide_setup(char *str)
 251 {
 252         fake_ide = 1;
 253         return 1;
 254 }
 255
 256 __setup("fake_ide", fake_ide_setup);
 257
 258 __uml_help(fake_ide_setup,
 259 "fake_ide\n"
 260 "    Create ide0 entries that map onto ubd devices.\n\n"
 261 );
 262
 263 static int parse_unit(char **ptr)
 264 {
 265         char *str = *ptr, *end;
 266         int n = -1;
 267
 268         if(isdigit(*str)) {
 269                 n = simple_strtoul(str, &end, 0);
 270                 if(end == str)
 271                         return -1;
 272                 *ptr = end;
 273         }
 274         else if (('a' <= *str) && (*str <= 'z')) {
 275                 n = *str - 'a';
 276                 str++;
 277                 *ptr = str;
 278         }
 279         return n;
 280 }
 281
 282 /* If *index_out == -1 at exit, the passed option was a general one;
 283  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 284  * should not be freed on exit.
 285  */
 286 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 287 {
 288         struct ubd *ubd_dev;
 289         struct openflags flags = global_openflags;
 290         char *backing_file;
 291         int n, err = 0, i;
 292
 293         if(index_out) *index_out = -1;
 294         n = *str;
 295         if(n == '='){
 296                 char *end;
 297                 int major;
 298
 299                 str++;
 300                 if(!strcmp(str, "sync")){
 301                         global_openflags = of_sync(global_openflags);
 302                         goto out1;
 303                 }
 304
 305                 err = -EINVAL;
 306                 major = simple_strtoul(str, &end, 0);
 307                 if((*end != '\0') || (end == str)){
 308                         *error_out = "Didn't parse major number";
 309                         goto out1;
 310                 }
 311
 312                 mutex_lock(&ubd_lock);
 313                 if(fake_major != MAJOR_NR){
 314                         *error_out = "Can't assign a fake major twice";
 315                         goto out1;
 316                 }
 317
 318                 fake_major = major;
 319
 320                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 321                        major);
 322                 err = 0;
 323         out1:
 324                 mutex_unlock(&ubd_lock);
 325                 return err;
 326         }
 327
 328         n = parse_unit(&str);
 329         if(n < 0){
 330                 *error_out = "Couldn't parse device number";
 331                 return -EINVAL;
 332         }
 333         if(n >= MAX_DEV){
 334                 *error_out = "Device number out of range";
 335                 return 1;
 336         }
 337
 338         err = -EBUSY;
 339         mutex_lock(&ubd_lock);
 340
 341         ubd_dev = &ubd_devs[n];
 342         if(ubd_dev->file != NULL){
 343                 *error_out = "Device is already configured";
 344                 goto out;
 345         }
 346
 347         if (index_out)
 348                 *index_out = n;
 349
 350         err = -EINVAL;
 351         for (i = 0; i < sizeof("rscd="); i++) {
 352                 switch (*str) {
 353                 case 'r':
 354                         flags.w = 0;
 355                         break;
 356                 case 's':
 357                         flags.s = 1;
 358                         break;
 359                 case 'd':
 360                         ubd_dev->no_cow = 1;
 361                         break;
 362                 case 'c':
 363                         ubd_dev->shared = 1;
 364                         break;
 365                 case '=':
 366                         str++;
 367                         goto break_loop;
 368                 default:
 369                         *error_out = "Expected '=' or flag letter "
 370                                 "(r, s, c, or d)";
 371                         goto out;
 372                 }
 373                 str++;
 374         }
 375
 376         if (*str == '=')
 377                 *error_out = "Too many flags specified";
 378         else
 379                 *error_out = "Missing '='";
 380         goto out;
 381
 382 break_loop:
 383         backing_file = strchr(str, ',');
 384
 385         if (backing_file == NULL)
 386                 backing_file = strchr(str, ':');
 387
 388         if(backing_file != NULL){
 389                 if(ubd_dev->no_cow){
 390                         *error_out = "Can't specify both 'd' and a cow file";
 391                         goto out;
 392                 }
 393                 else {
 394                         *backing_file = '\0';
 395                         backing_file++;
 396                 }
 397         }
 398         err = 0;
 399         ubd_dev->file = str;
 400         ubd_dev->cow.file = backing_file;
 401         ubd_dev->boot_openflags = flags;
 402 out:
 403         mutex_unlock(&ubd_lock);
 404         return err;
 405 }
 406
 407 static int ubd_setup(char *str)
 408 {
 409         char *error;
 410         int err;
 411
 412         err = ubd_setup_common(str, NULL, &error);
 413         if(err)
 414                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 415                        "%s\n", str, error);
 416         return 1;
 417 }
 418
 419 __setup("ubd", ubd_setup);
 420 __uml_help(ubd_setup,
 421 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 422 "    This is used to associate a device with a file in the underlying\n"
 423 "    filesystem. When specifying two filenames, the first one is the\n"
 424 "    COW name and the second is the backing file name. As separator you can\n"
 425 "    use either a ':' or a ',': the first one allows writing things like;\n"
 426 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 427 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 428 "    When using only one filename, UML will detect whether to treat it like\n"
 429 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 430 "    flag:\n"
 431 "       ubd0d=BackingFile\n"
 432 "    Usually, there is a filesystem in the file, but \n"
 433 "    that's not required. Swap devices containing swap files can be\n"
 434 "    specified like this. Also, a file which doesn't contain a\n"
 435 "    filesystem can have its contents read in the virtual \n"
 436 "    machine by running 'dd' on the device. <n> must be in the range\n"
 437 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 438 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 439 "    an 's' will cause data to be written to disk on the host immediately.\n\n"
 440 );
 441
 442 static int udb_setup(char *str)
 443 {
 444         printk("udb%s specified on command line is almost certainly a ubd -> "
 445                "udb TYPO\n", str);
 446         return 1;
 447 }
 448
 449 __setup("udb", udb_setup);
 450 __uml_help(udb_setup,
 451 "udb\n"
 452 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 453 "    to impossible to catch visually unless you specifically look for\n"
 454 "    them.  The only result of any option starting with 'udb' is an error\n"
 455 "    in the boot output.\n\n"
 456 );
 457
 458 static int fakehd_set = 0;
 459 static int fakehd(char *str)
 460 {
 461         printk(KERN_INFO "fakehd : Changing ubd name to \"hd\".\n");
 462         fakehd_set = 1;
 463         return 1;
 464 }
 465
 466 __setup("fakehd", fakehd);
 467 __uml_help(fakehd,
 468 "fakehd\n"
 469 "    Change the ubd device name to \"hd\".\n\n"
 470 );
 471
 472 static void do_ubd_request(struct request_queue * q);
 473
 474 /* Only changed by ubd_init, which is an initcall. */
 475 int thread_fd = -1;
 476
 477 static void ubd_end_request(struct request *req, int bytes, int uptodate)
 478 {
 479         if (!end_that_request_first(req, uptodate, bytes >> 9)) {
 480                 struct ubd *dev = req->rq_disk->private_data;
 481                 unsigned long flags;
 482
 483                 add_disk_randomness(req->rq_disk);
 484                 spin_lock_irqsave(&dev->lock, flags);
 485                 end_that_request_last(req, uptodate);
 486                 spin_unlock_irqrestore(&dev->lock, flags);
 487         }
 488 }
 489
 490 /* Callable only from interrupt context - otherwise you need to do
 491  * spin_lock_irq()/spin_lock_irqsave() */
 492 static inline void ubd_finish(struct request *req, int bytes)
 493 {
 494         if(bytes < 0){
 495                 ubd_end_request(req, 0, 0);
 496                 return;
 497         }
 498         ubd_end_request(req, bytes, 1);
 499 }
 500
 501 static LIST_HEAD(restart);
 502
 503 /* XXX - move this inside ubd_intr. */
 504 /* Called without dev->lock held, and only in interrupt context. */
 505 static void ubd_handler(void)
 506 {
 507         struct io_thread_req *req;
 508         struct request *rq;
 509         struct ubd *ubd;
 510         struct list_head *list, *next_ele;
 511         unsigned long flags;
 512         int n;
 513
 514         while(1){
 515                 n = os_read_file(thread_fd, &req,
 516                                  sizeof(struct io_thread_req *));
 517                 if(n != sizeof(req)){
 518                         if(n == -EAGAIN)
 519                                 break;
 520                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 521                                "err = %d\n", -n);
 522                         return;
 523                 }
 524
 525                 rq = req->req;
 526                 rq->nr_sectors -= req->length >> 9;
 527                 if(rq->nr_sectors == 0)
 528                         ubd_finish(rq, rq->hard_nr_sectors << 9);
 529                 kfree(req);
 530         }
 531         reactivate_fd(thread_fd, UBD_IRQ);
 532
 533         list_for_each_safe(list, next_ele, &restart){
 534                 ubd = container_of(list, struct ubd, restart);
 535                 list_del_init(&ubd->restart);
 536                 spin_lock_irqsave(&ubd->lock, flags);
 537                 do_ubd_request(ubd->queue);
 538                 spin_unlock_irqrestore(&ubd->lock, flags);
 539         }
 540 }
 541
 542 static irqreturn_t ubd_intr(int irq, void *dev)
 543 {
 544         ubd_handler();
 545         return IRQ_HANDLED;
 546 }
 547
 548 /* Only changed by ubd_init, which is an initcall. */
 549 static int io_pid = -1;
 550
 551 void kill_io_thread(void)
 552 {
 553         if(io_pid != -1)
 554                 os_kill_process(io_pid, 1);
 555 }
 556
 557 __uml_exitcall(kill_io_thread);
 558
 559 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 560 {
 561         char *file;
 562
 563         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
 564         return os_file_size(file, size_out);
 565 }
 566
 567 static void ubd_close_dev(struct ubd *ubd_dev)
 568 {
 569         os_close_file(ubd_dev->fd);
 570         if(ubd_dev->cow.file == NULL)
 571                 return;
 572
 573         os_close_file(ubd_dev->cow.fd);
 574         vfree(ubd_dev->cow.bitmap);
 575         ubd_dev->cow.bitmap = NULL;
 576 }
 577
 578 static int ubd_open_dev(struct ubd *ubd_dev)
 579 {
 580         struct openflags flags;
 581         char **back_ptr;
 582         int err, create_cow, *create_ptr;
 583         int fd;
 584
 585         ubd_dev->openflags = ubd_dev->boot_openflags;
 586         create_cow = 0;
 587         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 588         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 589
 590         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 591                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 592                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 593                                 create_ptr);
 594
 595         if((fd == -ENOENT) && create_cow){
 596                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 597                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 598                                           &ubd_dev->cow.bitmap_offset,
 599                                           &ubd_dev->cow.bitmap_len,
 600                                           &ubd_dev->cow.data_offset);
 601                 if(fd >= 0){
 602                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 603                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 604                 }
 605         }
 606
 607         if(fd < 0){
 608                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 609                        -fd);
 610                 return fd;
 611         }
 612         ubd_dev->fd = fd;
 613
 614         if(ubd_dev->cow.file != NULL){
 615                 blk_queue_max_sectors(ubd_dev->queue, 8 * sizeof(long));
 616
 617                 err = -ENOMEM;
 618                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 619                 if(ubd_dev->cow.bitmap == NULL){
 620                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 621                         goto error;
 622                 }
 623                 flush_tlb_kernel_vm();
 624
 625                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 626                                       ubd_dev->cow.bitmap_offset,
 627                                       ubd_dev->cow.bitmap_len);
 628                 if(err < 0)
 629                         goto error;
 630
 631                 flags = ubd_dev->openflags;
 632                 flags.w = 0;
 633                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 634                                     NULL, NULL, NULL, NULL);
 635                 if(err < 0) goto error;
 636                 ubd_dev->cow.fd = err;
 637         }
 638         return 0;
 639  error:
 640         os_close_file(ubd_dev->fd);
 641         return err;
 642 }
 643
 644 static void ubd_device_release(struct device *dev)
 645 {
 646         struct ubd *ubd_dev = dev->driver_data;
 647
 648         blk_cleanup_queue(ubd_dev->queue);
 649         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 650 }
 651
 652 static int ubd_disk_register(int major, u64 size, int unit,
 653                              struct gendisk **disk_out)
 654 {
 655         struct gendisk *disk;
 656
 657         disk = alloc_disk(1 << UBD_SHIFT);
 658         if(disk == NULL)
 659                 return -ENOMEM;
 660
 661         disk->major = major;
 662         disk->first_minor = unit << UBD_SHIFT;
 663         disk->fops = &ubd_blops;
 664         set_capacity(disk, size / 512);
 665         if(major == MAJOR_NR)
 666                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 667         else
 668                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 669
 670         /* sysfs register (not for ide fake devices) */
 671         if (major == MAJOR_NR) {
 672                 ubd_devs[unit].pdev.id   = unit;
 673                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 674                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 675                 ubd_devs[unit].pdev.dev.driver_data = &ubd_devs[unit];
 676                 platform_device_register(&ubd_devs[unit].pdev);
 677                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 678         }
 679
 680         disk->private_data = &ubd_devs[unit];
 681         disk->queue = ubd_devs[unit].queue;
 682         add_disk(disk);
 683
 684         *disk_out = disk;
 685         return 0;
 686 }
 687
 688 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 689
 690 static int ubd_add(int n, char **error_out)
 691 {
 692         struct ubd *ubd_dev = &ubd_devs[n];
 693         int err = 0;
 694
 695         if(ubd_dev->file == NULL)
 696                 goto out;
 697
 698         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 699         if(err < 0){
 700                 *error_out = "Couldn't determine size of device's file";
 701                 goto out;
 702         }
 703
 704         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 705
 706         INIT_LIST_HEAD(&ubd_dev->restart);
 707
 708         err = -ENOMEM;
 709         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 710         if (ubd_dev->queue == NULL) {
 711                 *error_out = "Failed to initialize device queue";
 712                 goto out;
 713         }
 714         ubd_dev->queue->queuedata = ubd_dev;
 715
 716         blk_queue_max_hw_segments(ubd_dev->queue, MAX_SG);
 717         err = ubd_disk_register(MAJOR_NR, ubd_dev->size, n, &ubd_gendisk[n]);
 718         if(err){
 719                 *error_out = "Failed to register device";
 720                 goto out_cleanup;
 721         }
 722
 723         if(fake_major != MAJOR_NR)
 724                 ubd_disk_register(fake_major, ubd_dev->size, n,
 725                                   &fake_gendisk[n]);
 726
 727         /* perhaps this should also be under the "if (fake_major)" above */
 728         /* using the fake_disk->disk_name and also the fakehd_set name */
 729         if (fake_ide)
 730                 make_ide_entries(ubd_gendisk[n]->disk_name);
 731
 732         err = 0;
 733 out:
 734         return err;
 735
 736 out_cleanup:
 737         blk_cleanup_queue(ubd_dev->queue);
 738         goto out;
 739 }
 740
 741 static int ubd_config(char *str, char **error_out)
 742 {
 743         int n, ret;
 744
 745         /* This string is possibly broken up and stored, so it's only
 746          * freed if ubd_setup_common fails, or if only general options
 747          * were set.
 748          */
 749         str = kstrdup(str, GFP_KERNEL);
 750         if (str == NULL) {
 751                 *error_out = "Failed to allocate memory";
 752                 return -ENOMEM;
 753         }
 754
 755         ret = ubd_setup_common(str, &n, error_out);
 756         if (ret)
 757                 goto err_free;
 758
 759         if (n == -1) {
 760                 ret = 0;
 761                 goto err_free;
 762         }
 763
 764         mutex_lock(&ubd_lock);
 765         ret = ubd_add(n, error_out);
 766         if (ret)
 767                 ubd_devs[n].file = NULL;
 768         mutex_unlock(&ubd_lock);
 769
 770 out:
 771         return ret;
 772
 773 err_free:
 774         kfree(str);
 775         goto out;
 776 }
 777
 778 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 779 {
 780         struct ubd *ubd_dev;
 781         int n, len = 0;
 782
 783         n = parse_unit(&name);
 784         if((n >= MAX_DEV) || (n < 0)){
 785                 *error_out = "ubd_get_config : device number out of range";
 786                 return -1;
 787         }
 788
 789         ubd_dev = &ubd_devs[n];
 790         mutex_lock(&ubd_lock);
 791
 792         if(ubd_dev->file == NULL){
 793                 CONFIG_CHUNK(str, size, len, "", 1);
 794                 goto out;
 795         }
 796
 797         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 798
 799         if(ubd_dev->cow.file != NULL){
 800                 CONFIG_CHUNK(str, size, len, ",", 0);
 801                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 802         }
 803         else CONFIG_CHUNK(str, size, len, "", 1);
 804
 805  out:
 806         mutex_unlock(&ubd_lock);
 807         return len;
 808 }
 809
 810 static int ubd_id(char **str, int *start_out, int *end_out)
 811 {
 812         int n;
 813
 814         n = parse_unit(str);
 815         *start_out = 0;
 816         *end_out = MAX_DEV - 1;
 817         return n;
 818 }
 819
 820 static int ubd_remove(int n, char **error_out)
 821 {
 822         struct gendisk *disk = ubd_gendisk[n];
 823         struct ubd *ubd_dev;
 824         int err = -ENODEV;
 825
 826         mutex_lock(&ubd_lock);
 827
 828         ubd_dev = &ubd_devs[n];
 829
 830         if(ubd_dev->file == NULL)
 831                 goto out;
 832
 833         /* you cannot remove a open disk */
 834         err = -EBUSY;
 835         if(ubd_dev->count > 0)
 836                 goto out;
 837
 838         ubd_gendisk[n] = NULL;
 839         if(disk != NULL){
 840                 del_gendisk(disk);
 841                 put_disk(disk);
 842         }
 843
 844         if(fake_gendisk[n] != NULL){
 845                 del_gendisk(fake_gendisk[n]);
 846                 put_disk(fake_gendisk[n]);
 847                 fake_gendisk[n] = NULL;
 848         }
 849
 850         err = 0;
 851         platform_device_unregister(&ubd_dev->pdev);
 852 out:
 853         mutex_unlock(&ubd_lock);
 854         return err;
 855 }
 856
 857 /* All these are called by mconsole in process context and without
 858  * ubd-specific locks.  The structure itself is const except for .list.
 859  */
 860 static struct mc_device ubd_mc = {
 861         .list           = LIST_HEAD_INIT(ubd_mc.list),
 862         .name           = "ubd",
 863         .config         = ubd_config,
 864         .get_config     = ubd_get_config,
 865         .id             = ubd_id,
 866         .remove         = ubd_remove,
 867 };
 868
 869 static int __init ubd_mc_init(void)
 870 {
 871         mconsole_register_dev(&ubd_mc);
 872         return 0;
 873 }
 874
 875 __initcall(ubd_mc_init);
 876
 877 static int __init ubd0_init(void)
 878 {
 879         struct ubd *ubd_dev = &ubd_devs[0];
 880
 881         mutex_lock(&ubd_lock);
 882         if(ubd_dev->file == NULL)
 883                 ubd_dev->file = "root_fs";
 884         mutex_unlock(&ubd_lock);
 885
 886         return 0;
 887 }
 888
 889 __initcall(ubd0_init);
 890
 891 /* Used in ubd_init, which is an initcall */
 892 static struct platform_driver ubd_driver = {
 893         .driver = {
 894                 .name  = DRIVER_NAME,
 895         },
 896 };
 897
 898 static int __init ubd_init(void)
 899 {
 900         char *error;
 901         int i, err;
 902
 903         if (register_blkdev(MAJOR_NR, "ubd"))
 904                 return -1;
 905
 906         if (fake_major != MAJOR_NR) {
 907                 char name[sizeof("ubd_nnn\0")];
 908
 909                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
 910                 if (register_blkdev(fake_major, "ubd"))
 911                         return -1;
 912         }
 913         platform_driver_register(&ubd_driver);
 914         mutex_lock(&ubd_lock);
 915         for (i = 0; i < MAX_DEV; i++){
 916                 err = ubd_add(i, &error);
 917                 if(err)
 918                         printk(KERN_ERR "Failed to initialize ubd device %d :"
 919                                "%s\n", i, error);
 920         }
 921         mutex_unlock(&ubd_lock);
 922         return 0;
 923 }
 924
 925 late_initcall(ubd_init);
 926
 927 static int __init ubd_driver_init(void){
 928         unsigned long stack;
 929         int err;
 930
 931         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
 932         if(global_openflags.s){
 933                 printk(KERN_INFO "ubd: Synchronous mode\n");
 934                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
 935                  * enough. So use anyway the io thread. */
 936         }
 937         stack = alloc_stack(0, 0);
 938         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
 939                                  &thread_fd);
 940         if(io_pid < 0){
 941                 printk(KERN_ERR
 942                        "ubd : Failed to start I/O thread (errno = %d) - "
 943                        "falling back to synchronous I/O\n", -io_pid);
 944                 io_pid = -1;
 945                 return 0;
 946         }
 947         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
 948                              IRQF_DISABLED, "ubd", ubd_devs);
 949         if(err != 0)
 950                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
 951         return 0;
 952 }
 953
 954 device_initcall(ubd_driver_init);
 955
 956 static int ubd_open(struct inode *inode, struct file *filp)
 957 {
 958         struct gendisk *disk = inode->i_bdev->bd_disk;
 959         struct ubd *ubd_dev = disk->private_data;
 960         int err = 0;
 961
 962         if(ubd_dev->count == 0){
 963                 err = ubd_open_dev(ubd_dev);
 964                 if(err){
 965                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
 966                                disk->disk_name, ubd_dev->file, -err);
 967                         goto out;
 968                 }
 969         }
 970         ubd_dev->count++;
 971         set_disk_ro(disk, !ubd_dev->openflags.w);
 972
 973         /* This should no more be needed. And it didn't work anyway to exclude
 974          * read-write remounting of filesystems.*/
 975         /*if((filp->f_mode & FMODE_WRITE) && !ubd_dev->openflags.w){
 976                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
 977                 err = -EROFS;
 978         }*/
 979  out:
 980         return err;
 981 }
 982
 983 static int ubd_release(struct inode * inode, struct file * file)
 984 {
 985         struct gendisk *disk = inode->i_bdev->bd_disk;
 986         struct ubd *ubd_dev = disk->private_data;
 987
 988         if(--ubd_dev->count == 0)
 989                 ubd_close_dev(ubd_dev);
 990         return 0;
 991 }
 992
 993 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
 994                           __u64 *cow_offset, unsigned long *bitmap,
 995                           __u64 bitmap_offset, unsigned long *bitmap_words,
 996                           __u64 bitmap_len)
 997 {
 998         __u64 sector = io_offset >> 9;
 999         int i, update_bitmap = 0;
1000
1001         for(i = 0; i < length >> 9; i++){
1002                 if(cow_mask != NULL)
1003                         ubd_set_bit(i, (unsigned char *) cow_mask);
1004                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1005                         continue;
1006
1007                 update_bitmap = 1;
1008                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1009         }
1010
1011         if(!update_bitmap)
1012                 return;
1013
1014         *cow_offset = sector / (sizeof(unsigned long) * 8);
1015
1016         /* This takes care of the case where we're exactly at the end of the
1017          * device, and *cow_offset + 1 is off the end.  So, just back it up
1018          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1019          * for the original diagnosis.
1020          */
1021         if(*cow_offset == ((bitmap_len + sizeof(unsigned long) - 1) /
1022                            sizeof(unsigned long) - 1))
1023                 (*cow_offset)--;
1024
1025         bitmap_words[0] = bitmap[*cow_offset];
1026         bitmap_words[1] = bitmap[*cow_offset + 1];
1027
1028         *cow_offset *= sizeof(unsigned long);
1029         *cow_offset += bitmap_offset;
1030 }
1031
1032 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1033                        __u64 bitmap_offset, __u64 bitmap_len)
1034 {
1035         __u64 sector = req->offset >> 9;
1036         int i;
1037
1038         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1039                 panic("Operation too long");
1040
1041         if(req->op == UBD_READ) {
1042                 for(i = 0; i < req->length >> 9; i++){
1043                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1044                                 ubd_set_bit(i, (unsigned char *)
1045                                             &req->sector_mask);
1046                 }
1047         }
1048         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1049                            &req->cow_offset, bitmap, bitmap_offset,
1050                            req->bitmap_words, bitmap_len);
1051 }
1052
1053 /* Called with dev->lock held */
1054 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1055                             unsigned long long offset, int page_offset,
1056                             int len, struct page *page)
1057 {
1058         struct gendisk *disk = req->rq_disk;
1059         struct ubd *ubd_dev = disk->private_data;
1060
1061         io_req->req = req;
1062         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1063                 ubd_dev->fd;
1064         io_req->fds[1] = ubd_dev->fd;
1065         io_req->cow_offset = -1;
1066         io_req->offset = offset;
1067         io_req->length = len;
1068         io_req->error = 0;
1069         io_req->sector_mask = 0;
1070
1071         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1072         io_req->offsets[0] = 0;
1073         io_req->offsets[1] = ubd_dev->cow.data_offset;
1074         io_req->buffer = page_address(page) + page_offset;
1075         io_req->sectorsize = 1 << 9;
1076
1077         if(ubd_dev->cow.file != NULL)
1078                 cowify_req(io_req, ubd_dev->cow.bitmap,
1079                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1080
1081 }
1082
1083 /* Called with dev->lock held */
1084 static void do_ubd_request(struct request_queue *q)
1085 {
1086         struct io_thread_req *io_req;
1087         struct request *req;
1088         int n, last_sectors;
1089
1090         while(1){
1091                 struct ubd *dev = q->queuedata;
1092                 if(dev->end_sg == 0){
1093                         struct request *req = elv_next_request(q);
1094                         if(req == NULL)
1095                                 return;
1096
1097                         dev->request = req;
1098                         blkdev_dequeue_request(req);
1099                         dev->start_sg = 0;
1100                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1101                 }
1102
1103                 req = dev->request;
1104                 last_sectors = 0;
1105                 while(dev->start_sg < dev->end_sg){
1106                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1107
1108                         req->sector += last_sectors;
1109                         io_req = kmalloc(sizeof(struct io_thread_req),
1110                                          GFP_ATOMIC);
1111                         if(io_req == NULL){
1112                                 if(list_empty(&dev->restart))
1113                                         list_add(&dev->restart, &restart);
1114                                 return;
1115                         }
1116                         prepare_request(req, io_req,
1117                                         (unsigned long long) req->sector << 9,
1118                                         sg->offset, sg->length, sg_page(sg));
1119
1120                         last_sectors = sg->length >> 9;
1121                         n = os_write_file(thread_fd, &io_req,
1122                                           sizeof(struct io_thread_req *));
1123                         if(n != sizeof(struct io_thread_req *)){
1124                                 if(n != -EAGAIN)
1125                                         printk("write to io thread failed, "
1126                                                "errno = %d\n", -n);
1127                                 else if(list_empty(&dev->restart))
1128                                         list_add(&dev->restart, &restart);
1129                                 return;
1130                         }
1131
1132                         dev->start_sg++;
1133                 }
1134                 dev->end_sg = 0;
1135                 dev->request = NULL;
1136         }
1137 }
1138
1139 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1140 {
1141         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1142
1143         geo->heads = 128;
1144         geo->sectors = 32;
1145         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1146         return 0;
1147 }
1148
1149 static int ubd_ioctl(struct inode * inode, struct file * file,
1150                      unsigned int cmd, unsigned long arg)
1151 {
1152         struct ubd *ubd_dev = inode->i_bdev->bd_disk->private_data;
1153         struct hd_driveid ubd_id = {
1154                 .cyls           = 0,
1155                 .heads          = 128,
1156                 .sectors        = 32,
1157         };
1158
1159         switch (cmd) {
1160                 struct cdrom_volctrl volume;
1161         case HDIO_GET_IDENTITY:
1162                 ubd_id.cyls = ubd_dev->size / (128 * 32 * 512);
1163                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1164                                  sizeof(ubd_id)))
1165                         return -EFAULT;
1166                 return 0;
1167
1168         case CDROMVOLREAD:
1169                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1170                         return -EFAULT;
1171                 volume.channel0 = 255;
1172                 volume.channel1 = 255;
1173                 volume.channel2 = 255;
1174                 volume.channel3 = 255;
1175                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1176                         return -EFAULT;
1177                 return 0;
1178         }
1179         return -EINVAL;
1180 }
1181
1182 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
1183 {
1184         struct uml_stat buf1, buf2;
1185         int err;
1186
1187         if(from_cmdline == NULL)
1188                 return 0;
1189         if(!strcmp(from_cmdline, from_cow))
1190                 return 0;
1191
1192         err = os_stat_file(from_cmdline, &buf1);
1193         if(err < 0){
1194                 printk("Couldn't stat '%s', err = %d\n", from_cmdline, -err);
1195                 return 0;
1196         }
1197         err = os_stat_file(from_cow, &buf2);
1198         if(err < 0){
1199                 printk("Couldn't stat '%s', err = %d\n", from_cow, -err);
1200                 return 1;
1201         }
1202         if((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
1203                 return 0;
1204
1205         printk("Backing file mismatch - \"%s\" requested,\n"
1206                "\"%s\" specified in COW header of \"%s\"\n",
1207                from_cmdline, from_cow, cow);
1208         return 1;
1209 }
1210
1211 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
1212 {
1213         unsigned long modtime;
1214         unsigned long long actual;
1215         int err;
1216
1217         err = os_file_modtime(file, &modtime);
1218         if(err < 0){
1219                 printk("Failed to get modification time of backing file "
1220                        "\"%s\", err = %d\n", file, -err);
1221                 return err;
1222         }
1223
1224         err = os_file_size(file, &actual);
1225         if(err < 0){
1226                 printk("Failed to get size of backing file \"%s\", "
1227                        "err = %d\n", file, -err);
1228                 return err;
1229         }
1230
1231         if(actual != size){
1232                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
1233                  * the typecast.*/
1234                 printk("Size mismatch (%llu vs %llu) of COW header vs backing "
1235                        "file\n", (unsigned long long) size, actual);
1236                 return -EINVAL;
1237         }
1238         if(modtime != mtime){
1239                 printk("mtime mismatch (%ld vs %ld) of COW header vs backing "
1240                        "file\n", mtime, modtime);
1241                 return -EINVAL;
1242         }
1243         return 0;
1244 }
1245
1246 int read_cow_bitmap(int fd, void *buf, int offset, int len)
1247 {
1248         int err;
1249
1250         err = os_seek_file(fd, offset);
1251         if(err < 0)
1252                 return err;
1253
1254         err = os_read_file(fd, buf, len);
1255         if(err < 0)
1256                 return err;
1257
1258         return 0;
1259 }
1260
1261 int open_ubd_file(char *file, struct openflags *openflags, int shared,
1262                   char **backing_file_out, int *bitmap_offset_out,
1263                   unsigned long *bitmap_len_out, int *data_offset_out,
1264                   int *create_cow_out)
1265 {
1266         time_t mtime;
1267         unsigned long long size;
1268         __u32 version, align;
1269         char *backing_file;
1270         int fd, err, sectorsize, asked_switch, mode = 0644;
1271
1272         fd = os_open_file(file, *openflags, mode);
1273         if (fd < 0) {
1274                 if ((fd == -ENOENT) && (create_cow_out != NULL))
1275                         *create_cow_out = 1;
1276                 if (!openflags->w ||
1277                     ((fd != -EROFS) && (fd != -EACCES)))
1278                         return fd;
1279                 openflags->w = 0;
1280                 fd = os_open_file(file, *openflags, mode);
1281                 if (fd < 0)
1282                         return fd;
1283         }
1284
1285         if(shared)
1286                 printk("Not locking \"%s\" on the host\n", file);
1287         else {
1288                 err = os_lock_file(fd, openflags->w);
1289                 if(err < 0){
1290                         printk("Failed to lock '%s', err = %d\n", file, -err);
1291                         goto out_close;
1292                 }
1293         }
1294
1295         /* Successful return case! */
1296         if(backing_file_out == NULL)
1297                 return fd;
1298
1299         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
1300                               &size, &sectorsize, &align, bitmap_offset_out);
1301         if(err && (*backing_file_out != NULL)){
1302                 printk("Failed to read COW header from COW file \"%s\", "
1303                        "errno = %d\n", file, -err);
1304                 goto out_close;
1305         }
1306         if(err)
1307                 return fd;
1308
1309         asked_switch = path_requires_switch(*backing_file_out, backing_file, file);
1310
1311         /* Allow switching only if no mismatch. */
1312         if (asked_switch && !backing_file_mismatch(*backing_file_out, size, mtime)) {
1313                 printk("Switching backing file to '%s'\n", *backing_file_out);
1314                 err = write_cow_header(file, fd, *backing_file_out,
1315                                        sectorsize, align, &size);
1316                 if (err) {
1317                         printk("Switch failed, errno = %d\n", -err);
1318                         goto out_close;
1319                 }
1320         } else {
1321                 *backing_file_out = backing_file;
1322                 err = backing_file_mismatch(*backing_file_out, size, mtime);
1323                 if (err)
1324                         goto out_close;
1325         }
1326
1327         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
1328                   bitmap_len_out, data_offset_out);
1329
1330         return fd;
1331  out_close:
1332         os_close_file(fd);
1333         return err;
1334 }
1335
1336 int create_cow_file(char *cow_file, char *backing_file, struct openflags flags,
1337                     int sectorsize, int alignment, int *bitmap_offset_out,
1338                     unsigned long *bitmap_len_out, int *data_offset_out)
1339 {
1340         int err, fd;
1341
1342         flags.c = 1;
1343         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
1344         if(fd < 0){
1345                 err = fd;
1346                 printk("Open of COW file '%s' failed, errno = %d\n", cow_file,
1347                        -err);
1348                 goto out;
1349         }
1350
1351         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
1352                             bitmap_offset_out, bitmap_len_out,
1353                             data_offset_out);
1354         if(!err)
1355                 return fd;
1356         os_close_file(fd);
1357  out:
1358         return err;
1359 }
1360
1361 static int update_bitmap(struct io_thread_req *req)
1362 {
1363         int n;
1364
1365         if(req->cow_offset == -1)
1366                 return 0;
1367
1368         n = os_seek_file(req->fds[1], req->cow_offset);
1369         if(n < 0){
1370                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1371                 return 1;
1372         }
1373
1374         n = os_write_file(req->fds[1], &req->bitmap_words,
1375                           sizeof(req->bitmap_words));
1376         if(n != sizeof(req->bitmap_words)){
1377                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1378                        req->fds[1]);
1379                 return 1;
1380         }
1381
1382         return 0;
1383 }
1384
1385 void do_io(struct io_thread_req *req)
1386 {
1387         char *buf;
1388         unsigned long len;
1389         int n, nsectors, start, end, bit;
1390         int err;
1391         __u64 off;
1392
1393         nsectors = req->length / req->sectorsize;
1394         start = 0;
1395         do {
1396                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1397                 end = start;
1398                 while((end < nsectors) &&
1399                       (ubd_test_bit(end, (unsigned char *)
1400                                     &req->sector_mask) == bit))
1401                         end++;
1402
1403                 off = req->offset + req->offsets[bit] +
1404                         start * req->sectorsize;
1405                 len = (end - start) * req->sectorsize;
1406                 buf = &req->buffer[start * req->sectorsize];
1407
1408                 err = os_seek_file(req->fds[bit], off);
1409                 if(err < 0){
1410                         printk("do_io - lseek failed : err = %d\n", -err);
1411                         req->error = 1;
1412                         return;
1413                 }
1414                 if(req->op == UBD_READ){
1415                         n = 0;
1416                         do {
1417                                 buf = &buf[n];
1418                                 len -= n;
1419                                 n = os_read_file(req->fds[bit], buf, len);
1420                                 if (n < 0) {
1421                                         printk("do_io - read failed, err = %d "
1422                                                "fd = %d\n", -n, req->fds[bit]);
1423                                         req->error = 1;
1424                                         return;
1425                                 }
1426                         } while((n < len) && (n != 0));
1427                         if (n < len) memset(&buf[n], 0, len - n);
1428                 } else {
1429                         n = os_write_file(req->fds[bit], buf, len);
1430                         if(n != len){
1431                                 printk("do_io - write failed err = %d "
1432                                        "fd = %d\n", -n, req->fds[bit]);
1433                                 req->error = 1;
1434                                 return;
1435                         }
1436                 }
1437
1438                 start = end;
1439         } while(start < nsectors);
1440
1441         req->error = update_bitmap(req);
1442 }
1443
1444 /* Changed in start_io_thread, which is serialized by being called only
1445  * from ubd_init, which is an initcall.
1446  */
1447 int kernel_fd = -1;
1448
1449 /* Only changed by the io thread. XXX: currently unused. */
1450 static int io_count = 0;
1451
1452 int io_thread(void *arg)
1453 {
1454         struct io_thread_req *req;
1455         int n;
1456
1457         ignore_sigwinch_sig();
1458         while(1){
1459                 n = os_read_file(kernel_fd, &req,
1460                                  sizeof(struct io_thread_req *));
1461                 if(n != sizeof(struct io_thread_req *)){
1462                         if(n < 0)
1463                                 printk("io_thread - read failed, fd = %d, "
1464                                        "err = %d\n", kernel_fd, -n);
1465                         else {
1466                                 printk("io_thread - short read, fd = %d, "
1467                                        "length = %d\n", kernel_fd, n);
1468                         }
1469                         continue;
1470                 }
1471                 io_count++;
1472                 do_io(req);
1473                 n = os_write_file(kernel_fd, &req,
1474                                   sizeof(struct io_thread_req *));
1475                 if(n != sizeof(struct io_thread_req *))
1476                         printk("io_thread - write failed, fd = %d, err = %d\n",
1477                                kernel_fd, -n);
1478         }
1479
1480         return 0;
1481 }