arch/um/drivers/ubd_kern.c

   1 /*
   2  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
   3  * Licensed under the GPL
   4  */
   5
   6 /* 2001-09-28...2002-04-17
   7  * Partition stuff by James_McMechan@hotmail.com
   8  * old style ubd by setting UBD_SHIFT to 0
   9  * 2002-09-27...2002-10-18 massive tinkering for 2.5
  10  * partitions have changed in 2.5
  11  * 2003-01-29 more tinkering for 2.5.59-1
  12  * This should now address the sysfs problems and has
  13  * the symlink for devfs to allow for booting with
  14  * the common /dev/ubd/discX/... names rather than
  15  * only /dev/ubdN/discN this version also has lots of
  16  * clean ups preparing for ubd-many.
  17  * James McMechan
  18  */
  19
  20 #define UBD_SHIFT 4
  21
  22 #include "linux/kernel.h"
  23 #include "linux/module.h"
  24 #include "linux/blkdev.h"
  25 #include "linux/ata.h"
  26 #include "linux/hdreg.h"
  27 #include "linux/init.h"
  28 #include "linux/cdrom.h"
  29 #include "linux/proc_fs.h"
  30 #include "linux/seq_file.h"
  31 #include "linux/ctype.h"
  32 #include "linux/capability.h"
  33 #include "linux/mm.h"
  34 #include "linux/slab.h"
  35 #include "linux/vmalloc.h"
  36 #include "linux/smp_lock.h"
  37 #include "linux/blkpg.h"
  38 #include "linux/genhd.h"
  39 #include "linux/spinlock.h"
  40 #include "linux/platform_device.h"
  41 #include "linux/scatterlist.h"
  42 #include "asm/segment.h"
  43 #include "asm/uaccess.h"
  44 #include "asm/irq.h"
  45 #include "asm/types.h"
  46 #include "asm/tlbflush.h"
  47 #include "mem_user.h"
  48 #include "kern_util.h"
  49 #include "kern.h"
  50 #include "mconsole_kern.h"
  51 #include "init.h"
  52 #include "irq_user.h"
  53 #include "irq_kern.h"
  54 #include "ubd_user.h"
  55 #include "os.h"
  56 #include "mem.h"
  57 #include "mem_kern.h"
  58 #include "cow.h"
  59
  60 enum ubd_req { UBD_READ, UBD_WRITE };
  61
  62 struct io_thread_req {
  63         struct request *req;
  64         enum ubd_req op;
  65         int fds[2];
  66         unsigned long offsets[2];
  67         unsigned long long offset;
  68         unsigned long length;
  69         char *buffer;
  70         int sectorsize;
  71         unsigned long sector_mask;
  72         unsigned long long cow_offset;
  73         unsigned long bitmap_words[2];
  74         int error;
  75 };
  76
  77 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  78 {
  79         __u64 n;
  80         int bits, off;
  81
  82         bits = sizeof(data[0]) * 8;
  83         n = bit / bits;
  84         off = bit % bits;
  85         return (data[n] & (1 << off)) != 0;
  86 }
  87
  88 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  89 {
  90         __u64 n;
  91         int bits, off;
  92
  93         bits = sizeof(data[0]) * 8;
  94         n = bit / bits;
  95         off = bit % bits;
  96         data[n] |= (1 << off);
  97 }
  98 /*End stuff from ubd_user.h*/
  99
 100 #define DRIVER_NAME "uml-blkdev"
 101
 102 static DEFINE_MUTEX(ubd_lock);
 103
 104 static int ubd_open(struct block_device *bdev, fmode_t mode);
 105 static int ubd_release(struct gendisk *disk, fmode_t mode);
 106 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
 107                      unsigned int cmd, unsigned long arg);
 108 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 109
 110 #define MAX_DEV (16)
 111
 112 static const struct block_device_operations ubd_blops = {
 113         .owner          = THIS_MODULE,
 114         .open           = ubd_open,
 115         .release        = ubd_release,
 116         .ioctl          = ubd_ioctl,
 117         .getgeo         = ubd_getgeo,
 118 };
 119
 120 /* Protected by ubd_lock */
 121 static int fake_major = UBD_MAJOR;
 122 static struct gendisk *ubd_gendisk[MAX_DEV];
 123 static struct gendisk *fake_gendisk[MAX_DEV];
 124
 125 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 126 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 127                                          .cl = 1 })
 128 #else
 129 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 130                                          .cl = 1 })
 131 #endif
 132 static struct openflags global_openflags = OPEN_FLAGS;
 133
 134 struct cow {
 135         /* backing file name */
 136         char *file;
 137         /* backing file fd */
 138         int fd;
 139         unsigned long *bitmap;
 140         unsigned long bitmap_len;
 141         int bitmap_offset;
 142         int data_offset;
 143 };
 144
 145 #define MAX_SG 64
 146
 147 struct ubd {
 148         struct list_head restart;
 149         /* name (and fd, below) of the file opened for writing, either the
 150          * backing or the cow file. */
 151         char *file;
 152         int count;
 153         int fd;
 154         __u64 size;
 155         struct openflags boot_openflags;
 156         struct openflags openflags;
 157         unsigned shared:1;
 158         unsigned no_cow:1;
 159         struct cow cow;
 160         struct platform_device pdev;
 161         struct request_queue *queue;
 162         spinlock_t lock;
 163         struct scatterlist sg[MAX_SG];
 164         struct request *request;
 165         int start_sg, end_sg;
 166         sector_t rq_pos;
 167 };
 168
 169 #define DEFAULT_COW { \
 170         .file =                 NULL, \
 171         .fd =                   -1,     \
 172         .bitmap =               NULL, \
 173         .bitmap_offset =        0, \
 174         .data_offset =          0, \
 175 }
 176
 177 #define DEFAULT_UBD { \
 178         .file =                 NULL, \
 179         .count =                0, \
 180         .fd =                   -1, \
 181         .size =                 -1, \
 182         .boot_openflags =       OPEN_FLAGS, \
 183         .openflags =            OPEN_FLAGS, \
 184         .no_cow =               0, \
 185         .shared =               0, \
 186         .cow =                  DEFAULT_COW, \
 187         .lock =                 SPIN_LOCK_UNLOCKED,     \
 188         .request =              NULL, \
 189         .start_sg =             0, \
 190         .end_sg =               0, \
 191         .rq_pos =               0, \
 192 }
 193
 194 /* Protected by ubd_lock */
 195 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
 196
 197 /* Only changed by fake_ide_setup which is a setup */
 198 static int fake_ide = 0;
 199 static struct proc_dir_entry *proc_ide_root = NULL;
 200 static struct proc_dir_entry *proc_ide = NULL;
 201
 202 static void make_proc_ide(void)
 203 {
 204         proc_ide_root = proc_mkdir("ide", NULL);
 205         proc_ide = proc_mkdir("ide0", proc_ide_root);
 206 }
 207
 208 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
 209 {
 210         seq_puts(m, "disk\n");
 211         return 0;
 212 }
 213
 214 static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
 215 {
 216         return single_open(file, fake_ide_media_proc_show, NULL);
 217 }
 218
 219 static const struct file_operations fake_ide_media_proc_fops = {
 220         .owner          = THIS_MODULE,
 221         .open           = fake_ide_media_proc_open,
 222         .read           = seq_read,
 223         .llseek         = seq_lseek,
 224         .release        = single_release,
 225 };
 226
 227 static void make_ide_entries(const char *dev_name)
 228 {
 229         struct proc_dir_entry *dir, *ent;
 230         char name[64];
 231
 232         if(proc_ide_root == NULL) make_proc_ide();
 233
 234         dir = proc_mkdir(dev_name, proc_ide);
 235         if(!dir) return;
 236
 237         ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
 238         if(!ent) return;
 239         snprintf(name, sizeof(name), "ide0/%s", dev_name);
 240         proc_symlink(dev_name, proc_ide_root, name);
 241 }
 242
 243 static int fake_ide_setup(char *str)
 244 {
 245         fake_ide = 1;
 246         return 1;
 247 }
 248
 249 __setup("fake_ide", fake_ide_setup);
 250
 251 __uml_help(fake_ide_setup,
 252 "fake_ide\n"
 253 "    Create ide0 entries that map onto ubd devices.\n\n"
 254 );
 255
 256 static int parse_unit(char **ptr)
 257 {
 258         char *str = *ptr, *end;
 259         int n = -1;
 260
 261         if(isdigit(*str)) {
 262                 n = simple_strtoul(str, &end, 0);
 263                 if(end == str)
 264                         return -1;
 265                 *ptr = end;
 266         }
 267         else if (('a' <= *str) && (*str <= 'z')) {
 268                 n = *str - 'a';
 269                 str++;
 270                 *ptr = str;
 271         }
 272         return n;
 273 }
 274
 275 /* If *index_out == -1 at exit, the passed option was a general one;
 276  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 277  * should not be freed on exit.
 278  */
 279 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 280 {
 281         struct ubd *ubd_dev;
 282         struct openflags flags = global_openflags;
 283         char *backing_file;
 284         int n, err = 0, i;
 285
 286         if(index_out) *index_out = -1;
 287         n = *str;
 288         if(n == '='){
 289                 char *end;
 290                 int major;
 291
 292                 str++;
 293                 if(!strcmp(str, "sync")){
 294                         global_openflags = of_sync(global_openflags);
 295                         goto out1;
 296                 }
 297
 298                 err = -EINVAL;
 299                 major = simple_strtoul(str, &end, 0);
 300                 if((*end != '\0') || (end == str)){
 301                         *error_out = "Didn't parse major number";
 302                         goto out1;
 303                 }
 304
 305                 mutex_lock(&ubd_lock);
 306                 if (fake_major != UBD_MAJOR) {
 307                         *error_out = "Can't assign a fake major twice";
 308                         goto out1;
 309                 }
 310
 311                 fake_major = major;
 312
 313                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 314                        major);
 315                 err = 0;
 316         out1:
 317                 mutex_unlock(&ubd_lock);
 318                 return err;
 319         }
 320
 321         n = parse_unit(&str);
 322         if(n < 0){
 323                 *error_out = "Couldn't parse device number";
 324                 return -EINVAL;
 325         }
 326         if(n >= MAX_DEV){
 327                 *error_out = "Device number out of range";
 328                 return 1;
 329         }
 330
 331         err = -EBUSY;
 332         mutex_lock(&ubd_lock);
 333
 334         ubd_dev = &ubd_devs[n];
 335         if(ubd_dev->file != NULL){
 336                 *error_out = "Device is already configured";
 337                 goto out;
 338         }
 339
 340         if (index_out)
 341                 *index_out = n;
 342
 343         err = -EINVAL;
 344         for (i = 0; i < sizeof("rscd="); i++) {
 345                 switch (*str) {
 346                 case 'r':
 347                         flags.w = 0;
 348                         break;
 349                 case 's':
 350                         flags.s = 1;
 351                         break;
 352                 case 'd':
 353                         ubd_dev->no_cow = 1;
 354                         break;
 355                 case 'c':
 356                         ubd_dev->shared = 1;
 357                         break;
 358                 case '=':
 359                         str++;
 360                         goto break_loop;
 361                 default:
 362                         *error_out = "Expected '=' or flag letter "
 363                                 "(r, s, c, or d)";
 364                         goto out;
 365                 }
 366                 str++;
 367         }
 368
 369         if (*str == '=')
 370                 *error_out = "Too many flags specified";
 371         else
 372                 *error_out = "Missing '='";
 373         goto out;
 374
 375 break_loop:
 376         backing_file = strchr(str, ',');
 377
 378         if (backing_file == NULL)
 379                 backing_file = strchr(str, ':');
 380
 381         if(backing_file != NULL){
 382                 if(ubd_dev->no_cow){
 383                         *error_out = "Can't specify both 'd' and a cow file";
 384                         goto out;
 385                 }
 386                 else {
 387                         *backing_file = '\0';
 388                         backing_file++;
 389                 }
 390         }
 391         err = 0;
 392         ubd_dev->file = str;
 393         ubd_dev->cow.file = backing_file;
 394         ubd_dev->boot_openflags = flags;
 395 out:
 396         mutex_unlock(&ubd_lock);
 397         return err;
 398 }
 399
 400 static int ubd_setup(char *str)
 401 {
 402         char *error;
 403         int err;
 404
 405         err = ubd_setup_common(str, NULL, &error);
 406         if(err)
 407                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 408                        "%s\n", str, error);
 409         return 1;
 410 }
 411
 412 __setup("ubd", ubd_setup);
 413 __uml_help(ubd_setup,
 414 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 415 "    This is used to associate a device with a file in the underlying\n"
 416 "    filesystem. When specifying two filenames, the first one is the\n"
 417 "    COW name and the second is the backing file name. As separator you can\n"
 418 "    use either a ':' or a ',': the first one allows writing things like;\n"
 419 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 420 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 421 "    When using only one filename, UML will detect whether to treat it like\n"
 422 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 423 "    flag:\n"
 424 "       ubd0d=BackingFile\n"
 425 "    Usually, there is a filesystem in the file, but \n"
 426 "    that's not required. Swap devices containing swap files can be\n"
 427 "    specified like this. Also, a file which doesn't contain a\n"
 428 "    filesystem can have its contents read in the virtual \n"
 429 "    machine by running 'dd' on the device. <n> must be in the range\n"
 430 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 431 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 432 "    an 's' will cause data to be written to disk on the host immediately.\n"
 433 "    'c' will cause the device to be treated as being shared between multiple\n"
 434 "    UMLs and file locking will be turned off - this is appropriate for a\n"
 435 "    cluster filesystem and inappropriate at almost all other times.\n\n"
 436 );
 437
 438 static int udb_setup(char *str)
 439 {
 440         printk("udb%s specified on command line is almost certainly a ubd -> "
 441                "udb TYPO\n", str);
 442         return 1;
 443 }
 444
 445 __setup("udb", udb_setup);
 446 __uml_help(udb_setup,
 447 "udb\n"
 448 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 449 "    to impossible to catch visually unless you specifically look for\n"
 450 "    them.  The only result of any option starting with 'udb' is an error\n"
 451 "    in the boot output.\n\n"
 452 );
 453
 454 static void do_ubd_request(struct request_queue * q);
 455
 456 /* Only changed by ubd_init, which is an initcall. */
 457 static int thread_fd = -1;
 458 static LIST_HEAD(restart);
 459
 460 /* XXX - move this inside ubd_intr. */
 461 /* Called without dev->lock held, and only in interrupt context. */
 462 static void ubd_handler(void)
 463 {
 464         struct io_thread_req *req;
 465         struct ubd *ubd;
 466         struct list_head *list, *next_ele;
 467         unsigned long flags;
 468         int n;
 469
 470         while(1){
 471                 n = os_read_file(thread_fd, &req,
 472                                  sizeof(struct io_thread_req *));
 473                 if(n != sizeof(req)){
 474                         if(n == -EAGAIN)
 475                                 break;
 476                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 477                                "err = %d\n", -n);
 478                         return;
 479                 }
 480
 481                 blk_end_request(req->req, 0, req->length);
 482                 kfree(req);
 483         }
 484         reactivate_fd(thread_fd, UBD_IRQ);
 485
 486         list_for_each_safe(list, next_ele, &restart){
 487                 ubd = container_of(list, struct ubd, restart);
 488                 list_del_init(&ubd->restart);
 489                 spin_lock_irqsave(&ubd->lock, flags);
 490                 do_ubd_request(ubd->queue);
 491                 spin_unlock_irqrestore(&ubd->lock, flags);
 492         }
 493 }
 494
 495 static irqreturn_t ubd_intr(int irq, void *dev)
 496 {
 497         ubd_handler();
 498         return IRQ_HANDLED;
 499 }
 500
 501 /* Only changed by ubd_init, which is an initcall. */
 502 static int io_pid = -1;
 503
 504 static void kill_io_thread(void)
 505 {
 506         if(io_pid != -1)
 507                 os_kill_process(io_pid, 1);
 508 }
 509
 510 __uml_exitcall(kill_io_thread);
 511
 512 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 513 {
 514         char *file;
 515
 516         file = ubd_dev->cow.file ? ubd_dev->cow.file : ubd_dev->file;
 517         return os_file_size(file, size_out);
 518 }
 519
 520 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
 521 {
 522         int err;
 523
 524         err = os_seek_file(fd, offset);
 525         if (err < 0)
 526                 return err;
 527
 528         err = os_read_file(fd, buf, len);
 529         if (err < 0)
 530                 return err;
 531
 532         return 0;
 533 }
 534
 535 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
 536 {
 537         unsigned long modtime;
 538         unsigned long long actual;
 539         int err;
 540
 541         err = os_file_modtime(file, &modtime);
 542         if (err < 0) {
 543                 printk(KERN_ERR "Failed to get modification time of backing "
 544                        "file \"%s\", err = %d\n", file, -err);
 545                 return err;
 546         }
 547
 548         err = os_file_size(file, &actual);
 549         if (err < 0) {
 550                 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
 551                        "err = %d\n", file, -err);
 552                 return err;
 553         }
 554
 555         if (actual != size) {
 556                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
 557                  * the typecast.*/
 558                 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
 559                        "vs backing file\n", (unsigned long long) size, actual);
 560                 return -EINVAL;
 561         }
 562         if (modtime != mtime) {
 563                 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
 564                        "backing file\n", mtime, modtime);
 565                 return -EINVAL;
 566         }
 567         return 0;
 568 }
 569
 570 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
 571 {
 572         struct uml_stat buf1, buf2;
 573         int err;
 574
 575         if (from_cmdline == NULL)
 576                 return 0;
 577         if (!strcmp(from_cmdline, from_cow))
 578                 return 0;
 579
 580         err = os_stat_file(from_cmdline, &buf1);
 581         if (err < 0) {
 582                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
 583                        -err);
 584                 return 0;
 585         }
 586         err = os_stat_file(from_cow, &buf2);
 587         if (err < 0) {
 588                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
 589                        -err);
 590                 return 1;
 591         }
 592         if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
 593                 return 0;
 594
 595         printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
 596                "\"%s\" specified in COW header of \"%s\"\n",
 597                from_cmdline, from_cow, cow);
 598         return 1;
 599 }
 600
 601 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
 602                   char **backing_file_out, int *bitmap_offset_out,
 603                   unsigned long *bitmap_len_out, int *data_offset_out,
 604                   int *create_cow_out)
 605 {
 606         time_t mtime;
 607         unsigned long long size;
 608         __u32 version, align;
 609         char *backing_file;
 610         int fd, err, sectorsize, asked_switch, mode = 0644;
 611
 612         fd = os_open_file(file, *openflags, mode);
 613         if (fd < 0) {
 614                 if ((fd == -ENOENT) && (create_cow_out != NULL))
 615                         *create_cow_out = 1;
 616                 if (!openflags->w ||
 617                     ((fd != -EROFS) && (fd != -EACCES)))
 618                         return fd;
 619                 openflags->w = 0;
 620                 fd = os_open_file(file, *openflags, mode);
 621                 if (fd < 0)
 622                         return fd;
 623         }
 624
 625         if (shared)
 626                 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
 627         else {
 628                 err = os_lock_file(fd, openflags->w);
 629                 if (err < 0) {
 630                         printk(KERN_ERR "Failed to lock '%s', err = %d\n",
 631                                file, -err);
 632                         goto out_close;
 633                 }
 634         }
 635
 636         /* Successful return case! */
 637         if (backing_file_out == NULL)
 638                 return fd;
 639
 640         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
 641                               &size, &sectorsize, &align, bitmap_offset_out);
 642         if (err && (*backing_file_out != NULL)) {
 643                 printk(KERN_ERR "Failed to read COW header from COW file "
 644                        "\"%s\", errno = %d\n", file, -err);
 645                 goto out_close;
 646         }
 647         if (err)
 648                 return fd;
 649
 650         asked_switch = path_requires_switch(*backing_file_out, backing_file,
 651                                             file);
 652
 653         /* Allow switching only if no mismatch. */
 654         if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
 655                                                    mtime)) {
 656                 printk(KERN_ERR "Switching backing file to '%s'\n",
 657                        *backing_file_out);
 658                 err = write_cow_header(file, fd, *backing_file_out,
 659                                        sectorsize, align, &size);
 660                 if (err) {
 661                         printk(KERN_ERR "Switch failed, errno = %d\n", -err);
 662                         goto out_close;
 663                 }
 664         } else {
 665                 *backing_file_out = backing_file;
 666                 err = backing_file_mismatch(*backing_file_out, size, mtime);
 667                 if (err)
 668                         goto out_close;
 669         }
 670
 671         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
 672                   bitmap_len_out, data_offset_out);
 673
 674         return fd;
 675  out_close:
 676         os_close_file(fd);
 677         return err;
 678 }
 679
 680 static int create_cow_file(char *cow_file, char *backing_file,
 681                     struct openflags flags,
 682                     int sectorsize, int alignment, int *bitmap_offset_out,
 683                     unsigned long *bitmap_len_out, int *data_offset_out)
 684 {
 685         int err, fd;
 686
 687         flags.c = 1;
 688         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
 689         if (fd < 0) {
 690                 err = fd;
 691                 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
 692                        cow_file, -err);
 693                 goto out;
 694         }
 695
 696         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
 697                             bitmap_offset_out, bitmap_len_out,
 698                             data_offset_out);
 699         if (!err)
 700                 return fd;
 701         os_close_file(fd);
 702  out:
 703         return err;
 704 }
 705
 706 static void ubd_close_dev(struct ubd *ubd_dev)
 707 {
 708         os_close_file(ubd_dev->fd);
 709         if(ubd_dev->cow.file == NULL)
 710                 return;
 711
 712         os_close_file(ubd_dev->cow.fd);
 713         vfree(ubd_dev->cow.bitmap);
 714         ubd_dev->cow.bitmap = NULL;
 715 }
 716
 717 static int ubd_open_dev(struct ubd *ubd_dev)
 718 {
 719         struct openflags flags;
 720         char **back_ptr;
 721         int err, create_cow, *create_ptr;
 722         int fd;
 723
 724         ubd_dev->openflags = ubd_dev->boot_openflags;
 725         create_cow = 0;
 726         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 727         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 728
 729         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 730                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 731                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 732                                 create_ptr);
 733
 734         if((fd == -ENOENT) && create_cow){
 735                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 736                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 737                                           &ubd_dev->cow.bitmap_offset,
 738                                           &ubd_dev->cow.bitmap_len,
 739                                           &ubd_dev->cow.data_offset);
 740                 if(fd >= 0){
 741                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 742                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 743                 }
 744         }
 745
 746         if(fd < 0){
 747                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 748                        -fd);
 749                 return fd;
 750         }
 751         ubd_dev->fd = fd;
 752
 753         if(ubd_dev->cow.file != NULL){
 754                 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
 755
 756                 err = -ENOMEM;
 757                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 758                 if(ubd_dev->cow.bitmap == NULL){
 759                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 760                         goto error;
 761                 }
 762                 flush_tlb_kernel_vm();
 763
 764                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 765                                       ubd_dev->cow.bitmap_offset,
 766                                       ubd_dev->cow.bitmap_len);
 767                 if(err < 0)
 768                         goto error;
 769
 770                 flags = ubd_dev->openflags;
 771                 flags.w = 0;
 772                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 773                                     NULL, NULL, NULL, NULL);
 774                 if(err < 0) goto error;
 775                 ubd_dev->cow.fd = err;
 776         }
 777         return 0;
 778  error:
 779         os_close_file(ubd_dev->fd);
 780         return err;
 781 }
 782
 783 static void ubd_device_release(struct device *dev)
 784 {
 785         struct ubd *ubd_dev = dev_get_drvdata(dev);
 786
 787         blk_cleanup_queue(ubd_dev->queue);
 788         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 789 }
 790
 791 static int ubd_disk_register(int major, u64 size, int unit,
 792                              struct gendisk **disk_out)
 793 {
 794         struct gendisk *disk;
 795
 796         disk = alloc_disk(1 << UBD_SHIFT);
 797         if(disk == NULL)
 798                 return -ENOMEM;
 799
 800         disk->major = major;
 801         disk->first_minor = unit << UBD_SHIFT;
 802         disk->fops = &ubd_blops;
 803         set_capacity(disk, size / 512);
 804         if (major == UBD_MAJOR)
 805                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 806         else
 807                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 808
 809         /* sysfs register (not for ide fake devices) */
 810         if (major == UBD_MAJOR) {
 811                 ubd_devs[unit].pdev.id   = unit;
 812                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 813                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 814                 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
 815                 platform_device_register(&ubd_devs[unit].pdev);
 816                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 817         }
 818
 819         disk->private_data = &ubd_devs[unit];
 820         disk->queue = ubd_devs[unit].queue;
 821         add_disk(disk);
 822
 823         *disk_out = disk;
 824         return 0;
 825 }
 826
 827 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 828
 829 static int ubd_add(int n, char **error_out)
 830 {
 831         struct ubd *ubd_dev = &ubd_devs[n];
 832         int err = 0;
 833
 834         if(ubd_dev->file == NULL)
 835                 goto out;
 836
 837         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 838         if(err < 0){
 839                 *error_out = "Couldn't determine size of device's file";
 840                 goto out;
 841         }
 842
 843         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 844
 845         INIT_LIST_HEAD(&ubd_dev->restart);
 846         sg_init_table(ubd_dev->sg, MAX_SG);
 847
 848         err = -ENOMEM;
 849         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 850         if (ubd_dev->queue == NULL) {
 851                 *error_out = "Failed to initialize device queue";
 852                 goto out;
 853         }
 854         ubd_dev->queue->queuedata = ubd_dev;
 855
 856         blk_queue_max_segments(ubd_dev->queue, MAX_SG);
 857         err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
 858         if(err){
 859                 *error_out = "Failed to register device";
 860                 goto out_cleanup;
 861         }
 862
 863         if (fake_major != UBD_MAJOR)
 864                 ubd_disk_register(fake_major, ubd_dev->size, n,
 865                                   &fake_gendisk[n]);
 866
 867         /*
 868          * Perhaps this should also be under the "if (fake_major)" above
 869          * using the fake_disk->disk_name
 870          */
 871         if (fake_ide)
 872                 make_ide_entries(ubd_gendisk[n]->disk_name);
 873
 874         err = 0;
 875 out:
 876         return err;
 877
 878 out_cleanup:
 879         blk_cleanup_queue(ubd_dev->queue);
 880         goto out;
 881 }
 882
 883 static int ubd_config(char *str, char **error_out)
 884 {
 885         int n, ret;
 886
 887         /* This string is possibly broken up and stored, so it's only
 888          * freed if ubd_setup_common fails, or if only general options
 889          * were set.
 890          */
 891         str = kstrdup(str, GFP_KERNEL);
 892         if (str == NULL) {
 893                 *error_out = "Failed to allocate memory";
 894                 return -ENOMEM;
 895         }
 896
 897         ret = ubd_setup_common(str, &n, error_out);
 898         if (ret)
 899                 goto err_free;
 900
 901         if (n == -1) {
 902                 ret = 0;
 903                 goto err_free;
 904         }
 905
 906         mutex_lock(&ubd_lock);
 907         ret = ubd_add(n, error_out);
 908         if (ret)
 909                 ubd_devs[n].file = NULL;
 910         mutex_unlock(&ubd_lock);
 911
 912 out:
 913         return ret;
 914
 915 err_free:
 916         kfree(str);
 917         goto out;
 918 }
 919
 920 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 921 {
 922         struct ubd *ubd_dev;
 923         int n, len = 0;
 924
 925         n = parse_unit(&name);
 926         if((n >= MAX_DEV) || (n < 0)){
 927                 *error_out = "ubd_get_config : device number out of range";
 928                 return -1;
 929         }
 930
 931         ubd_dev = &ubd_devs[n];
 932         mutex_lock(&ubd_lock);
 933
 934         if(ubd_dev->file == NULL){
 935                 CONFIG_CHUNK(str, size, len, "", 1);
 936                 goto out;
 937         }
 938
 939         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 940
 941         if(ubd_dev->cow.file != NULL){
 942                 CONFIG_CHUNK(str, size, len, ",", 0);
 943                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 944         }
 945         else CONFIG_CHUNK(str, size, len, "", 1);
 946
 947  out:
 948         mutex_unlock(&ubd_lock);
 949         return len;
 950 }
 951
 952 static int ubd_id(char **str, int *start_out, int *end_out)
 953 {
 954         int n;
 955
 956         n = parse_unit(str);
 957         *start_out = 0;
 958         *end_out = MAX_DEV - 1;
 959         return n;
 960 }
 961
 962 static int ubd_remove(int n, char **error_out)
 963 {
 964         struct gendisk *disk = ubd_gendisk[n];
 965         struct ubd *ubd_dev;
 966         int err = -ENODEV;
 967
 968         mutex_lock(&ubd_lock);
 969
 970         ubd_dev = &ubd_devs[n];
 971
 972         if(ubd_dev->file == NULL)
 973                 goto out;
 974
 975         /* you cannot remove a open disk */
 976         err = -EBUSY;
 977         if(ubd_dev->count > 0)
 978                 goto out;
 979
 980         ubd_gendisk[n] = NULL;
 981         if(disk != NULL){
 982                 del_gendisk(disk);
 983                 put_disk(disk);
 984         }
 985
 986         if(fake_gendisk[n] != NULL){
 987                 del_gendisk(fake_gendisk[n]);
 988                 put_disk(fake_gendisk[n]);
 989                 fake_gendisk[n] = NULL;
 990         }
 991
 992         err = 0;
 993         platform_device_unregister(&ubd_dev->pdev);
 994 out:
 995         mutex_unlock(&ubd_lock);
 996         return err;
 997 }
 998
 999 /* All these are called by mconsole in process context and without
1000  * ubd-specific locks.  The structure itself is const except for .list.
1001  */
1002 static struct mc_device ubd_mc = {
1003         .list           = LIST_HEAD_INIT(ubd_mc.list),
1004         .name           = "ubd",
1005         .config         = ubd_config,
1006         .get_config     = ubd_get_config,
1007         .id             = ubd_id,
1008         .remove         = ubd_remove,
1009 };
1010
1011 static int __init ubd_mc_init(void)
1012 {
1013         mconsole_register_dev(&ubd_mc);
1014         return 0;
1015 }
1016
1017 __initcall(ubd_mc_init);
1018
1019 static int __init ubd0_init(void)
1020 {
1021         struct ubd *ubd_dev = &ubd_devs[0];
1022
1023         mutex_lock(&ubd_lock);
1024         if(ubd_dev->file == NULL)
1025                 ubd_dev->file = "root_fs";
1026         mutex_unlock(&ubd_lock);
1027
1028         return 0;
1029 }
1030
1031 __initcall(ubd0_init);
1032
1033 /* Used in ubd_init, which is an initcall */
1034 static struct platform_driver ubd_driver = {
1035         .driver = {
1036                 .name  = DRIVER_NAME,
1037         },
1038 };
1039
1040 static int __init ubd_init(void)
1041 {
1042         char *error;
1043         int i, err;
1044
1045         if (register_blkdev(UBD_MAJOR, "ubd"))
1046                 return -1;
1047
1048         if (fake_major != UBD_MAJOR) {
1049                 char name[sizeof("ubd_nnn\0")];
1050
1051                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1052                 if (register_blkdev(fake_major, "ubd"))
1053                         return -1;
1054         }
1055         platform_driver_register(&ubd_driver);
1056         mutex_lock(&ubd_lock);
1057         for (i = 0; i < MAX_DEV; i++){
1058                 err = ubd_add(i, &error);
1059                 if(err)
1060                         printk(KERN_ERR "Failed to initialize ubd device %d :"
1061                                "%s\n", i, error);
1062         }
1063         mutex_unlock(&ubd_lock);
1064         return 0;
1065 }
1066
1067 late_initcall(ubd_init);
1068
1069 static int __init ubd_driver_init(void){
1070         unsigned long stack;
1071         int err;
1072
1073         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1074         if(global_openflags.s){
1075                 printk(KERN_INFO "ubd: Synchronous mode\n");
1076                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1077                  * enough. So use anyway the io thread. */
1078         }
1079         stack = alloc_stack(0, 0);
1080         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1081                                  &thread_fd);
1082         if(io_pid < 0){
1083                 printk(KERN_ERR
1084                        "ubd : Failed to start I/O thread (errno = %d) - "
1085                        "falling back to synchronous I/O\n", -io_pid);
1086                 io_pid = -1;
1087                 return 0;
1088         }
1089         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1090                              IRQF_DISABLED, "ubd", ubd_devs);
1091         if(err != 0)
1092                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1093         return 0;
1094 }
1095
1096 device_initcall(ubd_driver_init);
1097
1098 static int ubd_open(struct block_device *bdev, fmode_t mode)
1099 {
1100         struct gendisk *disk = bdev->bd_disk;
1101         struct ubd *ubd_dev = disk->private_data;
1102         int err = 0;
1103
1104         lock_kernel();
1105         if(ubd_dev->count == 0){
1106                 err = ubd_open_dev(ubd_dev);
1107                 if(err){
1108                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1109                                disk->disk_name, ubd_dev->file, -err);
1110                         goto out;
1111                 }
1112         }
1113         ubd_dev->count++;
1114         set_disk_ro(disk, !ubd_dev->openflags.w);
1115
1116         /* This should no more be needed. And it didn't work anyway to exclude
1117          * read-write remounting of filesystems.*/
1118         /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1119                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1120                 err = -EROFS;
1121         }*/
1122 out:
1123         unlock_kernel();
1124         return err;
1125 }
1126
1127 static int ubd_release(struct gendisk *disk, fmode_t mode)
1128 {
1129         struct ubd *ubd_dev = disk->private_data;
1130
1131         lock_kernel();
1132         if(--ubd_dev->count == 0)
1133                 ubd_close_dev(ubd_dev);
1134         unlock_kernel();
1135         return 0;
1136 }
1137
1138 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1139                           __u64 *cow_offset, unsigned long *bitmap,
1140                           __u64 bitmap_offset, unsigned long *bitmap_words,
1141                           __u64 bitmap_len)
1142 {
1143         __u64 sector = io_offset >> 9;
1144         int i, update_bitmap = 0;
1145
1146         for(i = 0; i < length >> 9; i++){
1147                 if(cow_mask != NULL)
1148                         ubd_set_bit(i, (unsigned char *) cow_mask);
1149                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1150                         continue;
1151
1152                 update_bitmap = 1;
1153                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1154         }
1155
1156         if(!update_bitmap)
1157                 return;
1158
1159         *cow_offset = sector / (sizeof(unsigned long) * 8);
1160
1161         /* This takes care of the case where we're exactly at the end of the
1162          * device, and *cow_offset + 1 is off the end.  So, just back it up
1163          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1164          * for the original diagnosis.
1165          */
1166         if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1167                                          sizeof(unsigned long)) - 1))
1168                 (*cow_offset)--;
1169
1170         bitmap_words[0] = bitmap[*cow_offset];
1171         bitmap_words[1] = bitmap[*cow_offset + 1];
1172
1173         *cow_offset *= sizeof(unsigned long);
1174         *cow_offset += bitmap_offset;
1175 }
1176
1177 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1178                        __u64 bitmap_offset, __u64 bitmap_len)
1179 {
1180         __u64 sector = req->offset >> 9;
1181         int i;
1182
1183         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1184                 panic("Operation too long");
1185
1186         if(req->op == UBD_READ) {
1187                 for(i = 0; i < req->length >> 9; i++){
1188                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1189                                 ubd_set_bit(i, (unsigned char *)
1190                                             &req->sector_mask);
1191                 }
1192         }
1193         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1194                            &req->cow_offset, bitmap, bitmap_offset,
1195                            req->bitmap_words, bitmap_len);
1196 }
1197
1198 /* Called with dev->lock held */
1199 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1200                             unsigned long long offset, int page_offset,
1201                             int len, struct page *page)
1202 {
1203         struct gendisk *disk = req->rq_disk;
1204         struct ubd *ubd_dev = disk->private_data;
1205
1206         io_req->req = req;
1207         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1208                 ubd_dev->fd;
1209         io_req->fds[1] = ubd_dev->fd;
1210         io_req->cow_offset = -1;
1211         io_req->offset = offset;
1212         io_req->length = len;
1213         io_req->error = 0;
1214         io_req->sector_mask = 0;
1215
1216         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1217         io_req->offsets[0] = 0;
1218         io_req->offsets[1] = ubd_dev->cow.data_offset;
1219         io_req->buffer = page_address(page) + page_offset;
1220         io_req->sectorsize = 1 << 9;
1221
1222         if(ubd_dev->cow.file != NULL)
1223                 cowify_req(io_req, ubd_dev->cow.bitmap,
1224                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1225
1226 }
1227
1228 /* Called with dev->lock held */
1229 static void do_ubd_request(struct request_queue *q)
1230 {
1231         struct io_thread_req *io_req;
1232         struct request *req;
1233         int n;
1234
1235         while(1){
1236                 struct ubd *dev = q->queuedata;
1237                 if(dev->end_sg == 0){
1238                         struct request *req = blk_fetch_request(q);
1239                         if(req == NULL)
1240                                 return;
1241
1242                         dev->request = req;
1243                         dev->rq_pos = blk_rq_pos(req);
1244                         dev->start_sg = 0;
1245                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1246                 }
1247
1248                 req = dev->request;
1249                 while(dev->start_sg < dev->end_sg){
1250                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1251
1252                         io_req = kmalloc(sizeof(struct io_thread_req),
1253                                          GFP_ATOMIC);
1254                         if(io_req == NULL){
1255                                 if(list_empty(&dev->restart))
1256                                         list_add(&dev->restart, &restart);
1257                                 return;
1258                         }
1259                         prepare_request(req, io_req,
1260                                         (unsigned long long)dev->rq_pos << 9,
1261                                         sg->offset, sg->length, sg_page(sg));
1262
1263                         n = os_write_file(thread_fd, &io_req,
1264                                           sizeof(struct io_thread_req *));
1265                         if(n != sizeof(struct io_thread_req *)){
1266                                 if(n != -EAGAIN)
1267                                         printk("write to io thread failed, "
1268                                                "errno = %d\n", -n);
1269                                 else if(list_empty(&dev->restart))
1270                                         list_add(&dev->restart, &restart);
1271                                 kfree(io_req);
1272                                 return;
1273                         }
1274
1275                         dev->rq_pos += sg->length >> 9;
1276                         dev->start_sg++;
1277                 }
1278                 dev->end_sg = 0;
1279                 dev->request = NULL;
1280         }
1281 }
1282
1283 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1284 {
1285         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1286
1287         geo->heads = 128;
1288         geo->sectors = 32;
1289         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1290         return 0;
1291 }
1292
1293 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1294                      unsigned int cmd, unsigned long arg)
1295 {
1296         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1297         u16 ubd_id[ATA_ID_WORDS];
1298
1299         switch (cmd) {
1300                 struct cdrom_volctrl volume;
1301         case HDIO_GET_IDENTITY:
1302                 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1303                 ubd_id[ATA_ID_CYLS]     = ubd_dev->size / (128 * 32 * 512);
1304                 ubd_id[ATA_ID_HEADS]    = 128;
1305                 ubd_id[ATA_ID_SECTORS]  = 32;
1306                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1307                                  sizeof(ubd_id)))
1308                         return -EFAULT;
1309                 return 0;
1310
1311         case CDROMVOLREAD:
1312                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1313                         return -EFAULT;
1314                 volume.channel0 = 255;
1315                 volume.channel1 = 255;
1316                 volume.channel2 = 255;
1317                 volume.channel3 = 255;
1318                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1319                         return -EFAULT;
1320                 return 0;
1321         }
1322         return -EINVAL;
1323 }
1324
1325 static int update_bitmap(struct io_thread_req *req)
1326 {
1327         int n;
1328
1329         if(req->cow_offset == -1)
1330                 return 0;
1331
1332         n = os_seek_file(req->fds[1], req->cow_offset);
1333         if(n < 0){
1334                 printk("do_io - bitmap lseek failed : err = %d\n", -n);
1335                 return 1;
1336         }
1337
1338         n = os_write_file(req->fds[1], &req->bitmap_words,
1339                           sizeof(req->bitmap_words));
1340         if(n != sizeof(req->bitmap_words)){
1341                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1342                        req->fds[1]);
1343                 return 1;
1344         }
1345
1346         return 0;
1347 }
1348
1349 static void do_io(struct io_thread_req *req)
1350 {
1351         char *buf;
1352         unsigned long len;
1353         int n, nsectors, start, end, bit;
1354         int err;
1355         __u64 off;
1356
1357         nsectors = req->length / req->sectorsize;
1358         start = 0;
1359         do {
1360                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1361                 end = start;
1362                 while((end < nsectors) &&
1363                       (ubd_test_bit(end, (unsigned char *)
1364                                     &req->sector_mask) == bit))
1365                         end++;
1366
1367                 off = req->offset + req->offsets[bit] +
1368                         start * req->sectorsize;
1369                 len = (end - start) * req->sectorsize;
1370                 buf = &req->buffer[start * req->sectorsize];
1371
1372                 err = os_seek_file(req->fds[bit], off);
1373                 if(err < 0){
1374                         printk("do_io - lseek failed : err = %d\n", -err);
1375                         req->error = 1;
1376                         return;
1377                 }
1378                 if(req->op == UBD_READ){
1379                         n = 0;
1380                         do {
1381                                 buf = &buf[n];
1382                                 len -= n;
1383                                 n = os_read_file(req->fds[bit], buf, len);
1384                                 if (n < 0) {
1385                                         printk("do_io - read failed, err = %d "
1386                                                "fd = %d\n", -n, req->fds[bit]);
1387                                         req->error = 1;
1388                                         return;
1389                                 }
1390                         } while((n < len) && (n != 0));
1391                         if (n < len) memset(&buf[n], 0, len - n);
1392                 } else {
1393                         n = os_write_file(req->fds[bit], buf, len);
1394                         if(n != len){
1395                                 printk("do_io - write failed err = %d "
1396                                        "fd = %d\n", -n, req->fds[bit]);
1397                                 req->error = 1;
1398                                 return;
1399                         }
1400                 }
1401
1402                 start = end;
1403         } while(start < nsectors);
1404
1405         req->error = update_bitmap(req);
1406 }
1407
1408 /* Changed in start_io_thread, which is serialized by being called only
1409  * from ubd_init, which is an initcall.
1410  */
1411 int kernel_fd = -1;
1412
1413 /* Only changed by the io thread. XXX: currently unused. */
1414 static int io_count = 0;
1415
1416 int io_thread(void *arg)
1417 {
1418         struct io_thread_req *req;
1419         int n;
1420
1421         ignore_sigwinch_sig();
1422         while(1){
1423                 n = os_read_file(kernel_fd, &req,
1424                                  sizeof(struct io_thread_req *));
1425                 if(n != sizeof(struct io_thread_req *)){
1426                         if(n < 0)
1427                                 printk("io_thread - read failed, fd = %d, "
1428                                        "err = %d\n", kernel_fd, -n);
1429                         else {
1430                                 printk("io_thread - short read, fd = %d, "
1431                                        "length = %d\n", kernel_fd, n);
1432                         }
1433                         continue;
1434                 }
1435                 io_count++;
1436                 do_io(req);
1437                 n = os_write_file(kernel_fd, &req,
1438                                   sizeof(struct io_thread_req *));
1439                 if(n != sizeof(struct io_thread_req *))
1440                         printk("io_thread - write failed, fd = %d, err = %d\n",
1441                                kernel_fd, -n);
1442         }
1443
1444         return 0;
1445 }