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/module.h>
  23 #include <linux/init.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/ata.h>
  26 #include <linux/hdreg.h>
  27 #include <linux/cdrom.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/seq_file.h>
  30 #include <linux/ctype.h>
  31 #include <linux/slab.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/platform_device.h>
  34 #include <linux/scatterlist.h>
  35 #include <asm/tlbflush.h>
  36 #include <kern_util.h>
  37 #include "mconsole_kern.h"
  38 #include <init.h>
  39 #include <irq_kern.h>
  40 #include "ubd.h"
  41 #include <os.h>
  42 #include "cow.h"
  43
  44 enum ubd_req { UBD_READ, UBD_WRITE, UBD_FLUSH };
  45
  46 struct io_thread_req {
  47         struct request *req;
  48         enum ubd_req op;
  49         int fds[2];
  50         unsigned long offsets[2];
  51         unsigned long long offset;
  52         unsigned long length;
  53         char *buffer;
  54         int sectorsize;
  55         unsigned long sector_mask;
  56         unsigned long long cow_offset;
  57         unsigned long bitmap_words[2];
  58         int error;
  59 };
  60
  61 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  62 {
  63         __u64 n;
  64         int bits, off;
  65
  66         bits = sizeof(data[0]) * 8;
  67         n = bit / bits;
  68         off = bit % bits;
  69         return (data[n] & (1 << off)) != 0;
  70 }
  71
  72 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  73 {
  74         __u64 n;
  75         int bits, off;
  76
  77         bits = sizeof(data[0]) * 8;
  78         n = bit / bits;
  79         off = bit % bits;
  80         data[n] |= (1 << off);
  81 }
  82 /*End stuff from ubd_user.h*/
  83
  84 #define DRIVER_NAME "uml-blkdev"
  85
  86 static DEFINE_MUTEX(ubd_lock);
  87 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
  88
  89 static int ubd_open(struct block_device *bdev, fmode_t mode);
  90 static void ubd_release(struct gendisk *disk, fmode_t mode);
  91 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
  92                      unsigned int cmd, unsigned long arg);
  93 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
  94
  95 #define MAX_DEV (16)
  96
  97 static const struct block_device_operations ubd_blops = {
  98         .owner          = THIS_MODULE,
  99         .open           = ubd_open,
 100         .release        = ubd_release,
 101         .ioctl          = ubd_ioctl,
 102         .getgeo         = ubd_getgeo,
 103 };
 104
 105 /* Protected by ubd_lock */
 106 static int fake_major = UBD_MAJOR;
 107 static struct gendisk *ubd_gendisk[MAX_DEV];
 108 static struct gendisk *fake_gendisk[MAX_DEV];
 109
 110 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 111 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 112                                          .cl = 1 })
 113 #else
 114 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 115                                          .cl = 1 })
 116 #endif
 117 static struct openflags global_openflags = OPEN_FLAGS;
 118
 119 struct cow {
 120         /* backing file name */
 121         char *file;
 122         /* backing file fd */
 123         int fd;
 124         unsigned long *bitmap;
 125         unsigned long bitmap_len;
 126         int bitmap_offset;
 127         int data_offset;
 128 };
 129
 130 #define MAX_SG 64
 131
 132 struct ubd {
 133         struct list_head restart;
 134         /* name (and fd, below) of the file opened for writing, either the
 135          * backing or the cow file. */
 136         char *file;
 137         int count;
 138         int fd;
 139         __u64 size;
 140         struct openflags boot_openflags;
 141         struct openflags openflags;
 142         unsigned shared:1;
 143         unsigned no_cow:1;
 144         struct cow cow;
 145         struct platform_device pdev;
 146         struct request_queue *queue;
 147         spinlock_t lock;
 148         struct scatterlist sg[MAX_SG];
 149         struct request *request;
 150         int start_sg, end_sg;
 151         sector_t rq_pos;
 152 };
 153
 154 #define DEFAULT_COW { \
 155         .file =                 NULL, \
 156         .fd =                   -1,     \
 157         .bitmap =               NULL, \
 158         .bitmap_offset =        0, \
 159         .data_offset =          0, \
 160 }
 161
 162 #define DEFAULT_UBD { \
 163         .file =                 NULL, \
 164         .count =                0, \
 165         .fd =                   -1, \
 166         .size =                 -1, \
 167         .boot_openflags =       OPEN_FLAGS, \
 168         .openflags =            OPEN_FLAGS, \
 169         .no_cow =               0, \
 170         .shared =               0, \
 171         .cow =                  DEFAULT_COW, \
 172         .lock =                 __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
 173         .request =              NULL, \
 174         .start_sg =             0, \
 175         .end_sg =               0, \
 176         .rq_pos =               0, \
 177 }
 178
 179 /* Protected by ubd_lock */
 180 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
 181
 182 /* Only changed by fake_ide_setup which is a setup */
 183 static int fake_ide = 0;
 184 static struct proc_dir_entry *proc_ide_root = NULL;
 185 static struct proc_dir_entry *proc_ide = NULL;
 186
 187 static void make_proc_ide(void)
 188 {
 189         proc_ide_root = proc_mkdir("ide", NULL);
 190         proc_ide = proc_mkdir("ide0", proc_ide_root);
 191 }
 192
 193 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
 194 {
 195         seq_puts(m, "disk\n");
 196         return 0;
 197 }
 198
 199 static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
 200 {
 201         return single_open(file, fake_ide_media_proc_show, NULL);
 202 }
 203
 204 static const struct file_operations fake_ide_media_proc_fops = {
 205         .owner          = THIS_MODULE,
 206         .open           = fake_ide_media_proc_open,
 207         .read           = seq_read,
 208         .llseek         = seq_lseek,
 209         .release        = single_release,
 210 };
 211
 212 static void make_ide_entries(const char *dev_name)
 213 {
 214         struct proc_dir_entry *dir, *ent;
 215         char name[64];
 216
 217         if(proc_ide_root == NULL) make_proc_ide();
 218
 219         dir = proc_mkdir(dev_name, proc_ide);
 220         if(!dir) return;
 221
 222         ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
 223         if(!ent) return;
 224         snprintf(name, sizeof(name), "ide0/%s", dev_name);
 225         proc_symlink(dev_name, proc_ide_root, name);
 226 }
 227
 228 static int fake_ide_setup(char *str)
 229 {
 230         fake_ide = 1;
 231         return 1;
 232 }
 233
 234 __setup("fake_ide", fake_ide_setup);
 235
 236 __uml_help(fake_ide_setup,
 237 "fake_ide\n"
 238 "    Create ide0 entries that map onto ubd devices.\n\n"
 239 );
 240
 241 static int parse_unit(char **ptr)
 242 {
 243         char *str = *ptr, *end;
 244         int n = -1;
 245
 246         if(isdigit(*str)) {
 247                 n = simple_strtoul(str, &end, 0);
 248                 if(end == str)
 249                         return -1;
 250                 *ptr = end;
 251         }
 252         else if (('a' <= *str) && (*str <= 'z')) {
 253                 n = *str - 'a';
 254                 str++;
 255                 *ptr = str;
 256         }
 257         return n;
 258 }
 259
 260 /* If *index_out == -1 at exit, the passed option was a general one;
 261  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 262  * should not be freed on exit.
 263  */
 264 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 265 {
 266         struct ubd *ubd_dev;
 267         struct openflags flags = global_openflags;
 268         char *backing_file;
 269         int n, err = 0, i;
 270
 271         if(index_out) *index_out = -1;
 272         n = *str;
 273         if(n == '='){
 274                 char *end;
 275                 int major;
 276
 277                 str++;
 278                 if(!strcmp(str, "sync")){
 279                         global_openflags = of_sync(global_openflags);
 280                         goto out1;
 281                 }
 282
 283                 err = -EINVAL;
 284                 major = simple_strtoul(str, &end, 0);
 285                 if((*end != '\0') || (end == str)){
 286                         *error_out = "Didn't parse major number";
 287                         goto out1;
 288                 }
 289
 290                 mutex_lock(&ubd_lock);
 291                 if (fake_major != UBD_MAJOR) {
 292                         *error_out = "Can't assign a fake major twice";
 293                         goto out1;
 294                 }
 295
 296                 fake_major = major;
 297
 298                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 299                        major);
 300                 err = 0;
 301         out1:
 302                 mutex_unlock(&ubd_lock);
 303                 return err;
 304         }
 305
 306         n = parse_unit(&str);
 307         if(n < 0){
 308                 *error_out = "Couldn't parse device number";
 309                 return -EINVAL;
 310         }
 311         if(n >= MAX_DEV){
 312                 *error_out = "Device number out of range";
 313                 return 1;
 314         }
 315
 316         err = -EBUSY;
 317         mutex_lock(&ubd_lock);
 318
 319         ubd_dev = &ubd_devs[n];
 320         if(ubd_dev->file != NULL){
 321                 *error_out = "Device is already configured";
 322                 goto out;
 323         }
 324
 325         if (index_out)
 326                 *index_out = n;
 327
 328         err = -EINVAL;
 329         for (i = 0; i < sizeof("rscd="); i++) {
 330                 switch (*str) {
 331                 case 'r':
 332                         flags.w = 0;
 333                         break;
 334                 case 's':
 335                         flags.s = 1;
 336                         break;
 337                 case 'd':
 338                         ubd_dev->no_cow = 1;
 339                         break;
 340                 case 'c':
 341                         ubd_dev->shared = 1;
 342                         break;
 343                 case '=':
 344                         str++;
 345                         goto break_loop;
 346                 default:
 347                         *error_out = "Expected '=' or flag letter "
 348                                 "(r, s, c, or d)";
 349                         goto out;
 350                 }
 351                 str++;
 352         }
 353
 354         if (*str == '=')
 355                 *error_out = "Too many flags specified";
 356         else
 357                 *error_out = "Missing '='";
 358         goto out;
 359
 360 break_loop:
 361         backing_file = strchr(str, ',');
 362
 363         if (backing_file == NULL)
 364                 backing_file = strchr(str, ':');
 365
 366         if(backing_file != NULL){
 367                 if(ubd_dev->no_cow){
 368                         *error_out = "Can't specify both 'd' and a cow file";
 369                         goto out;
 370                 }
 371                 else {
 372                         *backing_file = '\0';
 373                         backing_file++;
 374                 }
 375         }
 376         err = 0;
 377         ubd_dev->file = str;
 378         ubd_dev->cow.file = backing_file;
 379         ubd_dev->boot_openflags = flags;
 380 out:
 381         mutex_unlock(&ubd_lock);
 382         return err;
 383 }
 384
 385 static int ubd_setup(char *str)
 386 {
 387         char *error;
 388         int err;
 389
 390         err = ubd_setup_common(str, NULL, &error);
 391         if(err)
 392                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 393                        "%s\n", str, error);
 394         return 1;
 395 }
 396
 397 __setup("ubd", ubd_setup);
 398 __uml_help(ubd_setup,
 399 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 400 "    This is used to associate a device with a file in the underlying\n"
 401 "    filesystem. When specifying two filenames, the first one is the\n"
 402 "    COW name and the second is the backing file name. As separator you can\n"
 403 "    use either a ':' or a ',': the first one allows writing things like;\n"
 404 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 405 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 406 "    When using only one filename, UML will detect whether to treat it like\n"
 407 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 408 "    flag:\n"
 409 "       ubd0d=BackingFile\n"
 410 "    Usually, there is a filesystem in the file, but \n"
 411 "    that's not required. Swap devices containing swap files can be\n"
 412 "    specified like this. Also, a file which doesn't contain a\n"
 413 "    filesystem can have its contents read in the virtual \n"
 414 "    machine by running 'dd' on the device. <n> must be in the range\n"
 415 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 416 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 417 "    an 's' will cause data to be written to disk on the host immediately.\n"
 418 "    'c' will cause the device to be treated as being shared between multiple\n"
 419 "    UMLs and file locking will be turned off - this is appropriate for a\n"
 420 "    cluster filesystem and inappropriate at almost all other times.\n\n"
 421 );
 422
 423 static int udb_setup(char *str)
 424 {
 425         printk("udb%s specified on command line is almost certainly a ubd -> "
 426                "udb TYPO\n", str);
 427         return 1;
 428 }
 429
 430 __setup("udb", udb_setup);
 431 __uml_help(udb_setup,
 432 "udb\n"
 433 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 434 "    to impossible to catch visually unless you specifically look for\n"
 435 "    them.  The only result of any option starting with 'udb' is an error\n"
 436 "    in the boot output.\n\n"
 437 );
 438
 439 static void do_ubd_request(struct request_queue * q);
 440
 441 /* Only changed by ubd_init, which is an initcall. */
 442 static int thread_fd = -1;
 443 static LIST_HEAD(restart);
 444
 445 /* XXX - move this inside ubd_intr. */
 446 /* Called without dev->lock held, and only in interrupt context. */
 447 static void ubd_handler(void)
 448 {
 449         struct io_thread_req *req;
 450         struct ubd *ubd;
 451         struct list_head *list, *next_ele;
 452         unsigned long flags;
 453         int n;
 454
 455         while(1){
 456                 n = os_read_file(thread_fd, &req,
 457                                  sizeof(struct io_thread_req *));
 458                 if(n != sizeof(req)){
 459                         if(n == -EAGAIN)
 460                                 break;
 461                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 462                                "err = %d\n", -n);
 463                         return;
 464                 }
 465
 466                 blk_end_request(req->req, 0, req->length);
 467                 kfree(req);
 468         }
 469         reactivate_fd(thread_fd, UBD_IRQ);
 470
 471         list_for_each_safe(list, next_ele, &restart){
 472                 ubd = container_of(list, struct ubd, restart);
 473                 list_del_init(&ubd->restart);
 474                 spin_lock_irqsave(&ubd->lock, flags);
 475                 do_ubd_request(ubd->queue);
 476                 spin_unlock_irqrestore(&ubd->lock, flags);
 477         }
 478 }
 479
 480 static irqreturn_t ubd_intr(int irq, void *dev)
 481 {
 482         ubd_handler();
 483         return IRQ_HANDLED;
 484 }
 485
 486 /* Only changed by ubd_init, which is an initcall. */
 487 static int io_pid = -1;
 488
 489 static void kill_io_thread(void)
 490 {
 491         if(io_pid != -1)
 492                 os_kill_process(io_pid, 1);
 493 }
 494
 495 __uml_exitcall(kill_io_thread);
 496
 497 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 498 {
 499         char *file;
 500         int fd;
 501         int err;
 502
 503         __u32 version;
 504         __u32 align;
 505         char *backing_file;
 506         time_t mtime;
 507         unsigned long long size;
 508         int sector_size;
 509         int bitmap_offset;
 510
 511         if (ubd_dev->file && ubd_dev->cow.file) {
 512                 file = ubd_dev->cow.file;
 513
 514                 goto out;
 515         }
 516
 517         fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
 518         if (fd < 0)
 519                 return fd;
 520
 521         err = read_cow_header(file_reader, &fd, &version, &backing_file, \
 522                 &mtime, &size, &sector_size, &align, &bitmap_offset);
 523         os_close_file(fd);
 524
 525         if(err == -EINVAL)
 526                 file = ubd_dev->file;
 527         else
 528                 file = backing_file;
 529
 530 out:
 531         return os_file_size(file, size_out);
 532 }
 533
 534 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
 535 {
 536         int err;
 537
 538         err = os_pread_file(fd, buf, len, offset);
 539         if (err < 0)
 540                 return err;
 541
 542         return 0;
 543 }
 544
 545 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
 546 {
 547         unsigned long modtime;
 548         unsigned long long actual;
 549         int err;
 550
 551         err = os_file_modtime(file, &modtime);
 552         if (err < 0) {
 553                 printk(KERN_ERR "Failed to get modification time of backing "
 554                        "file \"%s\", err = %d\n", file, -err);
 555                 return err;
 556         }
 557
 558         err = os_file_size(file, &actual);
 559         if (err < 0) {
 560                 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
 561                        "err = %d\n", file, -err);
 562                 return err;
 563         }
 564
 565         if (actual != size) {
 566                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
 567                  * the typecast.*/
 568                 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
 569                        "vs backing file\n", (unsigned long long) size, actual);
 570                 return -EINVAL;
 571         }
 572         if (modtime != mtime) {
 573                 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
 574                        "backing file\n", mtime, modtime);
 575                 return -EINVAL;
 576         }
 577         return 0;
 578 }
 579
 580 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
 581 {
 582         struct uml_stat buf1, buf2;
 583         int err;
 584
 585         if (from_cmdline == NULL)
 586                 return 0;
 587         if (!strcmp(from_cmdline, from_cow))
 588                 return 0;
 589
 590         err = os_stat_file(from_cmdline, &buf1);
 591         if (err < 0) {
 592                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
 593                        -err);
 594                 return 0;
 595         }
 596         err = os_stat_file(from_cow, &buf2);
 597         if (err < 0) {
 598                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
 599                        -err);
 600                 return 1;
 601         }
 602         if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
 603                 return 0;
 604
 605         printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
 606                "\"%s\" specified in COW header of \"%s\"\n",
 607                from_cmdline, from_cow, cow);
 608         return 1;
 609 }
 610
 611 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
 612                   char **backing_file_out, int *bitmap_offset_out,
 613                   unsigned long *bitmap_len_out, int *data_offset_out,
 614                   int *create_cow_out)
 615 {
 616         time_t mtime;
 617         unsigned long long size;
 618         __u32 version, align;
 619         char *backing_file;
 620         int fd, err, sectorsize, asked_switch, mode = 0644;
 621
 622         fd = os_open_file(file, *openflags, mode);
 623         if (fd < 0) {
 624                 if ((fd == -ENOENT) && (create_cow_out != NULL))
 625                         *create_cow_out = 1;
 626                 if (!openflags->w ||
 627                     ((fd != -EROFS) && (fd != -EACCES)))
 628                         return fd;
 629                 openflags->w = 0;
 630                 fd = os_open_file(file, *openflags, mode);
 631                 if (fd < 0)
 632                         return fd;
 633         }
 634
 635         if (shared)
 636                 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
 637         else {
 638                 err = os_lock_file(fd, openflags->w);
 639                 if (err < 0) {
 640                         printk(KERN_ERR "Failed to lock '%s', err = %d\n",
 641                                file, -err);
 642                         goto out_close;
 643                 }
 644         }
 645
 646         /* Successful return case! */
 647         if (backing_file_out == NULL)
 648                 return fd;
 649
 650         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
 651                               &size, &sectorsize, &align, bitmap_offset_out);
 652         if (err && (*backing_file_out != NULL)) {
 653                 printk(KERN_ERR "Failed to read COW header from COW file "
 654                        "\"%s\", errno = %d\n", file, -err);
 655                 goto out_close;
 656         }
 657         if (err)
 658                 return fd;
 659
 660         asked_switch = path_requires_switch(*backing_file_out, backing_file,
 661                                             file);
 662
 663         /* Allow switching only if no mismatch. */
 664         if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
 665                                                    mtime)) {
 666                 printk(KERN_ERR "Switching backing file to '%s'\n",
 667                        *backing_file_out);
 668                 err = write_cow_header(file, fd, *backing_file_out,
 669                                        sectorsize, align, &size);
 670                 if (err) {
 671                         printk(KERN_ERR "Switch failed, errno = %d\n", -err);
 672                         goto out_close;
 673                 }
 674         } else {
 675                 *backing_file_out = backing_file;
 676                 err = backing_file_mismatch(*backing_file_out, size, mtime);
 677                 if (err)
 678                         goto out_close;
 679         }
 680
 681         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
 682                   bitmap_len_out, data_offset_out);
 683
 684         return fd;
 685  out_close:
 686         os_close_file(fd);
 687         return err;
 688 }
 689
 690 static int create_cow_file(char *cow_file, char *backing_file,
 691                     struct openflags flags,
 692                     int sectorsize, int alignment, int *bitmap_offset_out,
 693                     unsigned long *bitmap_len_out, int *data_offset_out)
 694 {
 695         int err, fd;
 696
 697         flags.c = 1;
 698         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
 699         if (fd < 0) {
 700                 err = fd;
 701                 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
 702                        cow_file, -err);
 703                 goto out;
 704         }
 705
 706         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
 707                             bitmap_offset_out, bitmap_len_out,
 708                             data_offset_out);
 709         if (!err)
 710                 return fd;
 711         os_close_file(fd);
 712  out:
 713         return err;
 714 }
 715
 716 static void ubd_close_dev(struct ubd *ubd_dev)
 717 {
 718         os_close_file(ubd_dev->fd);
 719         if(ubd_dev->cow.file == NULL)
 720                 return;
 721
 722         os_close_file(ubd_dev->cow.fd);
 723         vfree(ubd_dev->cow.bitmap);
 724         ubd_dev->cow.bitmap = NULL;
 725 }
 726
 727 static int ubd_open_dev(struct ubd *ubd_dev)
 728 {
 729         struct openflags flags;
 730         char **back_ptr;
 731         int err, create_cow, *create_ptr;
 732         int fd;
 733
 734         ubd_dev->openflags = ubd_dev->boot_openflags;
 735         create_cow = 0;
 736         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 737         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 738
 739         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 740                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 741                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 742                                 create_ptr);
 743
 744         if((fd == -ENOENT) && create_cow){
 745                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 746                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 747                                           &ubd_dev->cow.bitmap_offset,
 748                                           &ubd_dev->cow.bitmap_len,
 749                                           &ubd_dev->cow.data_offset);
 750                 if(fd >= 0){
 751                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 752                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 753                 }
 754         }
 755
 756         if(fd < 0){
 757                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 758                        -fd);
 759                 return fd;
 760         }
 761         ubd_dev->fd = fd;
 762
 763         if(ubd_dev->cow.file != NULL){
 764                 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
 765
 766                 err = -ENOMEM;
 767                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 768                 if(ubd_dev->cow.bitmap == NULL){
 769                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 770                         goto error;
 771                 }
 772                 flush_tlb_kernel_vm();
 773
 774                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 775                                       ubd_dev->cow.bitmap_offset,
 776                                       ubd_dev->cow.bitmap_len);
 777                 if(err < 0)
 778                         goto error;
 779
 780                 flags = ubd_dev->openflags;
 781                 flags.w = 0;
 782                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 783                                     NULL, NULL, NULL, NULL);
 784                 if(err < 0) goto error;
 785                 ubd_dev->cow.fd = err;
 786         }
 787         return 0;
 788  error:
 789         os_close_file(ubd_dev->fd);
 790         return err;
 791 }
 792
 793 static void ubd_device_release(struct device *dev)
 794 {
 795         struct ubd *ubd_dev = dev_get_drvdata(dev);
 796
 797         blk_cleanup_queue(ubd_dev->queue);
 798         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 799 }
 800
 801 static int ubd_disk_register(int major, u64 size, int unit,
 802                              struct gendisk **disk_out)
 803 {
 804         struct gendisk *disk;
 805
 806         disk = alloc_disk(1 << UBD_SHIFT);
 807         if(disk == NULL)
 808                 return -ENOMEM;
 809
 810         disk->major = major;
 811         disk->first_minor = unit << UBD_SHIFT;
 812         disk->fops = &ubd_blops;
 813         set_capacity(disk, size / 512);
 814         if (major == UBD_MAJOR)
 815                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 816         else
 817                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 818
 819         /* sysfs register (not for ide fake devices) */
 820         if (major == UBD_MAJOR) {
 821                 ubd_devs[unit].pdev.id   = unit;
 822                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 823                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 824                 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
 825                 platform_device_register(&ubd_devs[unit].pdev);
 826                 disk->driverfs_dev = &ubd_devs[unit].pdev.dev;
 827         }
 828
 829         disk->private_data = &ubd_devs[unit];
 830         disk->queue = ubd_devs[unit].queue;
 831         add_disk(disk);
 832
 833         *disk_out = disk;
 834         return 0;
 835 }
 836
 837 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 838
 839 static int ubd_add(int n, char **error_out)
 840 {
 841         struct ubd *ubd_dev = &ubd_devs[n];
 842         int err = 0;
 843
 844         if(ubd_dev->file == NULL)
 845                 goto out;
 846
 847         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 848         if(err < 0){
 849                 *error_out = "Couldn't determine size of device's file";
 850                 goto out;
 851         }
 852
 853         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 854
 855         INIT_LIST_HEAD(&ubd_dev->restart);
 856         sg_init_table(ubd_dev->sg, MAX_SG);
 857
 858         err = -ENOMEM;
 859         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 860         if (ubd_dev->queue == NULL) {
 861                 *error_out = "Failed to initialize device queue";
 862                 goto out;
 863         }
 864         ubd_dev->queue->queuedata = ubd_dev;
 865         blk_queue_write_cache(ubd_dev->queue, true, false);
 866
 867         blk_queue_max_segments(ubd_dev->queue, MAX_SG);
 868         err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
 869         if(err){
 870                 *error_out = "Failed to register device";
 871                 goto out_cleanup;
 872         }
 873
 874         if (fake_major != UBD_MAJOR)
 875                 ubd_disk_register(fake_major, ubd_dev->size, n,
 876                                   &fake_gendisk[n]);
 877
 878         /*
 879          * Perhaps this should also be under the "if (fake_major)" above
 880          * using the fake_disk->disk_name
 881          */
 882         if (fake_ide)
 883                 make_ide_entries(ubd_gendisk[n]->disk_name);
 884
 885         err = 0;
 886 out:
 887         return err;
 888
 889 out_cleanup:
 890         blk_cleanup_queue(ubd_dev->queue);
 891         goto out;
 892 }
 893
 894 static int ubd_config(char *str, char **error_out)
 895 {
 896         int n, ret;
 897
 898         /* This string is possibly broken up and stored, so it's only
 899          * freed if ubd_setup_common fails, or if only general options
 900          * were set.
 901          */
 902         str = kstrdup(str, GFP_KERNEL);
 903         if (str == NULL) {
 904                 *error_out = "Failed to allocate memory";
 905                 return -ENOMEM;
 906         }
 907
 908         ret = ubd_setup_common(str, &n, error_out);
 909         if (ret)
 910                 goto err_free;
 911
 912         if (n == -1) {
 913                 ret = 0;
 914                 goto err_free;
 915         }
 916
 917         mutex_lock(&ubd_lock);
 918         ret = ubd_add(n, error_out);
 919         if (ret)
 920                 ubd_devs[n].file = NULL;
 921         mutex_unlock(&ubd_lock);
 922
 923 out:
 924         return ret;
 925
 926 err_free:
 927         kfree(str);
 928         goto out;
 929 }
 930
 931 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 932 {
 933         struct ubd *ubd_dev;
 934         int n, len = 0;
 935
 936         n = parse_unit(&name);
 937         if((n >= MAX_DEV) || (n < 0)){
 938                 *error_out = "ubd_get_config : device number out of range";
 939                 return -1;
 940         }
 941
 942         ubd_dev = &ubd_devs[n];
 943         mutex_lock(&ubd_lock);
 944
 945         if(ubd_dev->file == NULL){
 946                 CONFIG_CHUNK(str, size, len, "", 1);
 947                 goto out;
 948         }
 949
 950         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 951
 952         if(ubd_dev->cow.file != NULL){
 953                 CONFIG_CHUNK(str, size, len, ",", 0);
 954                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 955         }
 956         else CONFIG_CHUNK(str, size, len, "", 1);
 957
 958  out:
 959         mutex_unlock(&ubd_lock);
 960         return len;
 961 }
 962
 963 static int ubd_id(char **str, int *start_out, int *end_out)
 964 {
 965         int n;
 966
 967         n = parse_unit(str);
 968         *start_out = 0;
 969         *end_out = MAX_DEV - 1;
 970         return n;
 971 }
 972
 973 static int ubd_remove(int n, char **error_out)
 974 {
 975         struct gendisk *disk = ubd_gendisk[n];
 976         struct ubd *ubd_dev;
 977         int err = -ENODEV;
 978
 979         mutex_lock(&ubd_lock);
 980
 981         ubd_dev = &ubd_devs[n];
 982
 983         if(ubd_dev->file == NULL)
 984                 goto out;
 985
 986         /* you cannot remove a open disk */
 987         err = -EBUSY;
 988         if(ubd_dev->count > 0)
 989                 goto out;
 990
 991         ubd_gendisk[n] = NULL;
 992         if(disk != NULL){
 993                 del_gendisk(disk);
 994                 put_disk(disk);
 995         }
 996
 997         if(fake_gendisk[n] != NULL){
 998                 del_gendisk(fake_gendisk[n]);
 999                 put_disk(fake_gendisk[n]);
1000                 fake_gendisk[n] = NULL;
1001         }
1002
1003         err = 0;
1004         platform_device_unregister(&ubd_dev->pdev);
1005 out:
1006         mutex_unlock(&ubd_lock);
1007         return err;
1008 }
1009
1010 /* All these are called by mconsole in process context and without
1011  * ubd-specific locks.  The structure itself is const except for .list.
1012  */
1013 static struct mc_device ubd_mc = {
1014         .list           = LIST_HEAD_INIT(ubd_mc.list),
1015         .name           = "ubd",
1016         .config         = ubd_config,
1017         .get_config     = ubd_get_config,
1018         .id             = ubd_id,
1019         .remove         = ubd_remove,
1020 };
1021
1022 static int __init ubd_mc_init(void)
1023 {
1024         mconsole_register_dev(&ubd_mc);
1025         return 0;
1026 }
1027
1028 __initcall(ubd_mc_init);
1029
1030 static int __init ubd0_init(void)
1031 {
1032         struct ubd *ubd_dev = &ubd_devs[0];
1033
1034         mutex_lock(&ubd_lock);
1035         if(ubd_dev->file == NULL)
1036                 ubd_dev->file = "root_fs";
1037         mutex_unlock(&ubd_lock);
1038
1039         return 0;
1040 }
1041
1042 __initcall(ubd0_init);
1043
1044 /* Used in ubd_init, which is an initcall */
1045 static struct platform_driver ubd_driver = {
1046         .driver = {
1047                 .name  = DRIVER_NAME,
1048         },
1049 };
1050
1051 static int __init ubd_init(void)
1052 {
1053         char *error;
1054         int i, err;
1055
1056         if (register_blkdev(UBD_MAJOR, "ubd"))
1057                 return -1;
1058
1059         if (fake_major != UBD_MAJOR) {
1060                 char name[sizeof("ubd_nnn\0")];
1061
1062                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1063                 if (register_blkdev(fake_major, "ubd"))
1064                         return -1;
1065         }
1066         platform_driver_register(&ubd_driver);
1067         mutex_lock(&ubd_lock);
1068         for (i = 0; i < MAX_DEV; i++){
1069                 err = ubd_add(i, &error);
1070                 if(err)
1071                         printk(KERN_ERR "Failed to initialize ubd device %d :"
1072                                "%s\n", i, error);
1073         }
1074         mutex_unlock(&ubd_lock);
1075         return 0;
1076 }
1077
1078 late_initcall(ubd_init);
1079
1080 static int __init ubd_driver_init(void){
1081         unsigned long stack;
1082         int err;
1083
1084         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1085         if(global_openflags.s){
1086                 printk(KERN_INFO "ubd: Synchronous mode\n");
1087                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1088                  * enough. So use anyway the io thread. */
1089         }
1090         stack = alloc_stack(0, 0);
1091         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1092                                  &thread_fd);
1093         if(io_pid < 0){
1094                 printk(KERN_ERR
1095                        "ubd : Failed to start I/O thread (errno = %d) - "
1096                        "falling back to synchronous I/O\n", -io_pid);
1097                 io_pid = -1;
1098                 return 0;
1099         }
1100         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1101                              0, "ubd", ubd_devs);
1102         if(err != 0)
1103                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1104         return 0;
1105 }
1106
1107 device_initcall(ubd_driver_init);
1108
1109 static int ubd_open(struct block_device *bdev, fmode_t mode)
1110 {
1111         struct gendisk *disk = bdev->bd_disk;
1112         struct ubd *ubd_dev = disk->private_data;
1113         int err = 0;
1114
1115         mutex_lock(&ubd_mutex);
1116         if(ubd_dev->count == 0){
1117                 err = ubd_open_dev(ubd_dev);
1118                 if(err){
1119                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1120                                disk->disk_name, ubd_dev->file, -err);
1121                         goto out;
1122                 }
1123         }
1124         ubd_dev->count++;
1125         set_disk_ro(disk, !ubd_dev->openflags.w);
1126
1127         /* This should no more be needed. And it didn't work anyway to exclude
1128          * read-write remounting of filesystems.*/
1129         /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1130                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1131                 err = -EROFS;
1132         }*/
1133 out:
1134         mutex_unlock(&ubd_mutex);
1135         return err;
1136 }
1137
1138 static void ubd_release(struct gendisk *disk, fmode_t mode)
1139 {
1140         struct ubd *ubd_dev = disk->private_data;
1141
1142         mutex_lock(&ubd_mutex);
1143         if(--ubd_dev->count == 0)
1144                 ubd_close_dev(ubd_dev);
1145         mutex_unlock(&ubd_mutex);
1146 }
1147
1148 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1149                           __u64 *cow_offset, unsigned long *bitmap,
1150                           __u64 bitmap_offset, unsigned long *bitmap_words,
1151                           __u64 bitmap_len)
1152 {
1153         __u64 sector = io_offset >> 9;
1154         int i, update_bitmap = 0;
1155
1156         for(i = 0; i < length >> 9; i++){
1157                 if(cow_mask != NULL)
1158                         ubd_set_bit(i, (unsigned char *) cow_mask);
1159                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1160                         continue;
1161
1162                 update_bitmap = 1;
1163                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1164         }
1165
1166         if(!update_bitmap)
1167                 return;
1168
1169         *cow_offset = sector / (sizeof(unsigned long) * 8);
1170
1171         /* This takes care of the case where we're exactly at the end of the
1172          * device, and *cow_offset + 1 is off the end.  So, just back it up
1173          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1174          * for the original diagnosis.
1175          */
1176         if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1177                                          sizeof(unsigned long)) - 1))
1178                 (*cow_offset)--;
1179
1180         bitmap_words[0] = bitmap[*cow_offset];
1181         bitmap_words[1] = bitmap[*cow_offset + 1];
1182
1183         *cow_offset *= sizeof(unsigned long);
1184         *cow_offset += bitmap_offset;
1185 }
1186
1187 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1188                        __u64 bitmap_offset, __u64 bitmap_len)
1189 {
1190         __u64 sector = req->offset >> 9;
1191         int i;
1192
1193         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1194                 panic("Operation too long");
1195
1196         if(req->op == UBD_READ) {
1197                 for(i = 0; i < req->length >> 9; i++){
1198                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1199                                 ubd_set_bit(i, (unsigned char *)
1200                                             &req->sector_mask);
1201                 }
1202         }
1203         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1204                            &req->cow_offset, bitmap, bitmap_offset,
1205                            req->bitmap_words, bitmap_len);
1206 }
1207
1208 /* Called with dev->lock held */
1209 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1210                             unsigned long long offset, int page_offset,
1211                             int len, struct page *page)
1212 {
1213         struct gendisk *disk = req->rq_disk;
1214         struct ubd *ubd_dev = disk->private_data;
1215
1216         io_req->req = req;
1217         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1218                 ubd_dev->fd;
1219         io_req->fds[1] = ubd_dev->fd;
1220         io_req->cow_offset = -1;
1221         io_req->offset = offset;
1222         io_req->length = len;
1223         io_req->error = 0;
1224         io_req->sector_mask = 0;
1225
1226         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1227         io_req->offsets[0] = 0;
1228         io_req->offsets[1] = ubd_dev->cow.data_offset;
1229         io_req->buffer = page_address(page) + page_offset;
1230         io_req->sectorsize = 1 << 9;
1231
1232         if(ubd_dev->cow.file != NULL)
1233                 cowify_req(io_req, ubd_dev->cow.bitmap,
1234                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1235
1236 }
1237
1238 /* Called with dev->lock held */
1239 static void prepare_flush_request(struct request *req,
1240                                   struct io_thread_req *io_req)
1241 {
1242         struct gendisk *disk = req->rq_disk;
1243         struct ubd *ubd_dev = disk->private_data;
1244
1245         io_req->req = req;
1246         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1247                 ubd_dev->fd;
1248         io_req->op = UBD_FLUSH;
1249 }
1250
1251 static bool submit_request(struct io_thread_req *io_req, struct ubd *dev)
1252 {
1253         int n = os_write_file(thread_fd, &io_req,
1254                              sizeof(io_req));
1255         if (n != sizeof(io_req)) {
1256                 if (n != -EAGAIN)
1257                         printk("write to io thread failed, "
1258                                "errno = %d\n", -n);
1259                 else if (list_empty(&dev->restart))
1260                         list_add(&dev->restart, &restart);
1261
1262                 kfree(io_req);
1263                 return false;
1264         }
1265         return true;
1266 }
1267
1268 /* Called with dev->lock held */
1269 static void do_ubd_request(struct request_queue *q)
1270 {
1271         struct io_thread_req *io_req;
1272         struct request *req;
1273
1274         while(1){
1275                 struct ubd *dev = q->queuedata;
1276                 if(dev->request == NULL){
1277                         struct request *req = blk_fetch_request(q);
1278                         if(req == NULL)
1279                                 return;
1280
1281                         dev->request = req;
1282                         dev->rq_pos = blk_rq_pos(req);
1283                         dev->start_sg = 0;
1284                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1285                 }
1286
1287                 req = dev->request;
1288
1289                 if (req_op(req) == REQ_OP_FLUSH) {
1290                         io_req = kmalloc(sizeof(struct io_thread_req),
1291                                          GFP_ATOMIC);
1292                         if (io_req == NULL) {
1293                                 if (list_empty(&dev->restart))
1294                                         list_add(&dev->restart, &restart);
1295                                 return;
1296                         }
1297                         prepare_flush_request(req, io_req);
1298                         if (submit_request(io_req, dev) == false)
1299                                 return;
1300                 }
1301
1302                 while(dev->start_sg < dev->end_sg){
1303                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1304
1305                         io_req = kmalloc(sizeof(struct io_thread_req),
1306                                          GFP_ATOMIC);
1307                         if(io_req == NULL){
1308                                 if(list_empty(&dev->restart))
1309                                         list_add(&dev->restart, &restart);
1310                                 return;
1311                         }
1312                         prepare_request(req, io_req,
1313                                         (unsigned long long)dev->rq_pos << 9,
1314                                         sg->offset, sg->length, sg_page(sg));
1315
1316                         if (submit_request(io_req, dev) == false)
1317                                 return;
1318
1319                         dev->rq_pos += sg->length >> 9;
1320                         dev->start_sg++;
1321                 }
1322                 dev->end_sg = 0;
1323                 dev->request = NULL;
1324         }
1325 }
1326
1327 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1328 {
1329         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1330
1331         geo->heads = 128;
1332         geo->sectors = 32;
1333         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1334         return 0;
1335 }
1336
1337 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1338                      unsigned int cmd, unsigned long arg)
1339 {
1340         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1341         u16 ubd_id[ATA_ID_WORDS];
1342
1343         switch (cmd) {
1344                 struct cdrom_volctrl volume;
1345         case HDIO_GET_IDENTITY:
1346                 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1347                 ubd_id[ATA_ID_CYLS]     = ubd_dev->size / (128 * 32 * 512);
1348                 ubd_id[ATA_ID_HEADS]    = 128;
1349                 ubd_id[ATA_ID_SECTORS]  = 32;
1350                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1351                                  sizeof(ubd_id)))
1352                         return -EFAULT;
1353                 return 0;
1354
1355         case CDROMVOLREAD:
1356                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1357                         return -EFAULT;
1358                 volume.channel0 = 255;
1359                 volume.channel1 = 255;
1360                 volume.channel2 = 255;
1361                 volume.channel3 = 255;
1362                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1363                         return -EFAULT;
1364                 return 0;
1365         }
1366         return -EINVAL;
1367 }
1368
1369 static int update_bitmap(struct io_thread_req *req)
1370 {
1371         int n;
1372
1373         if(req->cow_offset == -1)
1374                 return 0;
1375
1376         n = os_pwrite_file(req->fds[1], &req->bitmap_words,
1377                           sizeof(req->bitmap_words), req->cow_offset);
1378         if(n != sizeof(req->bitmap_words)){
1379                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1380                        req->fds[1]);
1381                 return 1;
1382         }
1383
1384         return 0;
1385 }
1386
1387 static void do_io(struct io_thread_req *req)
1388 {
1389         char *buf;
1390         unsigned long len;
1391         int n, nsectors, start, end, bit;
1392         __u64 off;
1393
1394         if (req->op == UBD_FLUSH) {
1395                 /* fds[0] is always either the rw image or our cow file */
1396                 n = os_sync_file(req->fds[0]);
1397                 if (n != 0) {
1398                         printk("do_io - sync failed err = %d "
1399                                "fd = %d\n", -n, req->fds[0]);
1400                         req->error = 1;
1401                 }
1402                 return;
1403         }
1404
1405         nsectors = req->length / req->sectorsize;
1406         start = 0;
1407         do {
1408                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1409                 end = start;
1410                 while((end < nsectors) &&
1411                       (ubd_test_bit(end, (unsigned char *)
1412                                     &req->sector_mask) == bit))
1413                         end++;
1414
1415                 off = req->offset + req->offsets[bit] +
1416                         start * req->sectorsize;
1417                 len = (end - start) * req->sectorsize;
1418                 buf = &req->buffer[start * req->sectorsize];
1419
1420                 if(req->op == UBD_READ){
1421                         n = 0;
1422                         do {
1423                                 buf = &buf[n];
1424                                 len -= n;
1425                                 n = os_pread_file(req->fds[bit], buf, len, off);
1426                                 if (n < 0) {
1427                                         printk("do_io - read failed, err = %d "
1428                                                "fd = %d\n", -n, req->fds[bit]);
1429                                         req->error = 1;
1430                                         return;
1431                                 }
1432                         } while((n < len) && (n != 0));
1433                         if (n < len) memset(&buf[n], 0, len - n);
1434                 } else {
1435                         n = os_pwrite_file(req->fds[bit], buf, len, off);
1436                         if(n != len){
1437                                 printk("do_io - write failed err = %d "
1438                                        "fd = %d\n", -n, req->fds[bit]);
1439                                 req->error = 1;
1440                                 return;
1441                         }
1442                 }
1443
1444                 start = end;
1445         } while(start < nsectors);
1446
1447         req->error = update_bitmap(req);
1448 }
1449
1450 /* Changed in start_io_thread, which is serialized by being called only
1451  * from ubd_init, which is an initcall.
1452  */
1453 int kernel_fd = -1;
1454
1455 /* Only changed by the io thread. XXX: currently unused. */
1456 static int io_count = 0;
1457
1458 int io_thread(void *arg)
1459 {
1460         struct io_thread_req *req;
1461         int n;
1462
1463         os_fix_helper_signals();
1464
1465         while(1){
1466                 n = os_read_file(kernel_fd, &req,
1467                                  sizeof(struct io_thread_req *));
1468                 if(n != sizeof(struct io_thread_req *)){
1469                         if(n < 0)
1470                                 printk("io_thread - read failed, fd = %d, "
1471                                        "err = %d\n", kernel_fd, -n);
1472                         else {
1473                                 printk("io_thread - short read, fd = %d, "
1474                                        "length = %d\n", kernel_fd, n);
1475                         }
1476                         continue;
1477                 }
1478                 io_count++;
1479                 do_io(req);
1480                 n = os_write_file(kernel_fd, &req,
1481                                   sizeof(struct io_thread_req *));
1482                 if(n != sizeof(struct io_thread_req *))
1483                         printk("io_thread - write failed, fd = %d, err = %d\n",
1484                                kernel_fd, -n);
1485         }
1486
1487         return 0;
1488 }