arch/um/drivers/ubd_kern.c

   1 /*
   2  * Copyright (C) 2015-2016 Anton Ivanov (aivanov@brocade.com)
   3  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
   4  * Licensed under the GPL
   5  */
   6
   7 /* 2001-09-28...2002-04-17
   8  * Partition stuff by James_McMechan@hotmail.com
   9  * old style ubd by setting UBD_SHIFT to 0
  10  * 2002-09-27...2002-10-18 massive tinkering for 2.5
  11  * partitions have changed in 2.5
  12  * 2003-01-29 more tinkering for 2.5.59-1
  13  * This should now address the sysfs problems and has
  14  * the symlink for devfs to allow for booting with
  15  * the common /dev/ubd/discX/... names rather than
  16  * only /dev/ubdN/discN this version also has lots of
  17  * clean ups preparing for ubd-many.
  18  * James McMechan
  19  */
  20
  21 #define UBD_SHIFT 4
  22
  23 #include <linux/module.h>
  24 #include <linux/init.h>
  25 #include <linux/blkdev.h>
  26 #include <linux/ata.h>
  27 #include <linux/hdreg.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/slab.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/platform_device.h>
  35 #include <linux/scatterlist.h>
  36 #include <asm/tlbflush.h>
  37 #include <kern_util.h>
  38 #include "mconsole_kern.h"
  39 #include <init.h>
  40 #include <irq_kern.h>
  41 #include "ubd.h"
  42 #include <os.h>
  43 #include "cow.h"
  44
  45 enum ubd_req { UBD_READ, UBD_WRITE, UBD_FLUSH };
  46
  47 struct io_thread_req {
  48         struct request *req;
  49         enum ubd_req op;
  50         int fds[2];
  51         unsigned long offsets[2];
  52         unsigned long long offset;
  53         unsigned long length;
  54         char *buffer;
  55         int sectorsize;
  56         unsigned long sector_mask;
  57         unsigned long long cow_offset;
  58         unsigned long bitmap_words[2];
  59         int error;
  60 };
  61
  62
  63 static struct io_thread_req * (*irq_req_buffer)[];
  64 static struct io_thread_req *irq_remainder;
  65 static int irq_remainder_size;
  66
  67 static struct io_thread_req * (*io_req_buffer)[];
  68 static struct io_thread_req *io_remainder;
  69 static int io_remainder_size;
  70
  71
  72
  73 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  74 {
  75         __u64 n;
  76         int bits, off;
  77
  78         bits = sizeof(data[0]) * 8;
  79         n = bit / bits;
  80         off = bit % bits;
  81         return (data[n] & (1 << off)) != 0;
  82 }
  83
  84 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  85 {
  86         __u64 n;
  87         int bits, off;
  88
  89         bits = sizeof(data[0]) * 8;
  90         n = bit / bits;
  91         off = bit % bits;
  92         data[n] |= (1 << off);
  93 }
  94 /*End stuff from ubd_user.h*/
  95
  96 #define DRIVER_NAME "uml-blkdev"
  97
  98 static DEFINE_MUTEX(ubd_lock);
  99 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
 100
 101 static int ubd_open(struct block_device *bdev, fmode_t mode);
 102 static void ubd_release(struct gendisk *disk, fmode_t mode);
 103 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
 104                      unsigned int cmd, unsigned long arg);
 105 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 106
 107 #define MAX_DEV (16)
 108
 109 static const struct block_device_operations ubd_blops = {
 110         .owner          = THIS_MODULE,
 111         .open           = ubd_open,
 112         .release        = ubd_release,
 113         .ioctl          = ubd_ioctl,
 114         .getgeo         = ubd_getgeo,
 115 };
 116
 117 /* Protected by ubd_lock */
 118 static int fake_major = UBD_MAJOR;
 119 static struct gendisk *ubd_gendisk[MAX_DEV];
 120 static struct gendisk *fake_gendisk[MAX_DEV];
 121
 122 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 123 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 124                                          .cl = 1 })
 125 #else
 126 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 127                                          .cl = 1 })
 128 #endif
 129 static struct openflags global_openflags = OPEN_FLAGS;
 130
 131 struct cow {
 132         /* backing file name */
 133         char *file;
 134         /* backing file fd */
 135         int fd;
 136         unsigned long *bitmap;
 137         unsigned long bitmap_len;
 138         int bitmap_offset;
 139         int data_offset;
 140 };
 141
 142 #define MAX_SG 64
 143
 144 struct ubd {
 145         struct list_head restart;
 146         /* name (and fd, below) of the file opened for writing, either the
 147          * backing or the cow file. */
 148         char *file;
 149         int count;
 150         int fd;
 151         __u64 size;
 152         struct openflags boot_openflags;
 153         struct openflags openflags;
 154         unsigned shared:1;
 155         unsigned no_cow:1;
 156         struct cow cow;
 157         struct platform_device pdev;
 158         struct request_queue *queue;
 159         spinlock_t lock;
 160         struct scatterlist sg[MAX_SG];
 161         struct request *request;
 162         int start_sg, end_sg;
 163         sector_t rq_pos;
 164 };
 165
 166 #define DEFAULT_COW { \
 167         .file =                 NULL, \
 168         .fd =                   -1,     \
 169         .bitmap =               NULL, \
 170         .bitmap_offset =        0, \
 171         .data_offset =          0, \
 172 }
 173
 174 #define DEFAULT_UBD { \
 175         .file =                 NULL, \
 176         .count =                0, \
 177         .fd =                   -1, \
 178         .size =                 -1, \
 179         .boot_openflags =       OPEN_FLAGS, \
 180         .openflags =            OPEN_FLAGS, \
 181         .no_cow =               0, \
 182         .shared =               0, \
 183         .cow =                  DEFAULT_COW, \
 184         .lock =                 __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
 185         .request =              NULL, \
 186         .start_sg =             0, \
 187         .end_sg =               0, \
 188         .rq_pos =               0, \
 189 }
 190
 191 /* Protected by ubd_lock */
 192 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
 193
 194 /* Only changed by fake_ide_setup which is a setup */
 195 static int fake_ide = 0;
 196 static struct proc_dir_entry *proc_ide_root = NULL;
 197 static struct proc_dir_entry *proc_ide = NULL;
 198
 199 static void make_proc_ide(void)
 200 {
 201         proc_ide_root = proc_mkdir("ide", NULL);
 202         proc_ide = proc_mkdir("ide0", proc_ide_root);
 203 }
 204
 205 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
 206 {
 207         seq_puts(m, "disk\n");
 208         return 0;
 209 }
 210
 211 static int fake_ide_media_proc_open(struct inode *inode, struct file *file)
 212 {
 213         return single_open(file, fake_ide_media_proc_show, NULL);
 214 }
 215
 216 static const struct file_operations fake_ide_media_proc_fops = {
 217         .owner          = THIS_MODULE,
 218         .open           = fake_ide_media_proc_open,
 219         .read           = seq_read,
 220         .llseek         = seq_lseek,
 221         .release        = single_release,
 222 };
 223
 224 static void make_ide_entries(const char *dev_name)
 225 {
 226         struct proc_dir_entry *dir, *ent;
 227         char name[64];
 228
 229         if(proc_ide_root == NULL) make_proc_ide();
 230
 231         dir = proc_mkdir(dev_name, proc_ide);
 232         if(!dir) return;
 233
 234         ent = proc_create("media", S_IRUGO, dir, &fake_ide_media_proc_fops);
 235         if(!ent) return;
 236         snprintf(name, sizeof(name), "ide0/%s", dev_name);
 237         proc_symlink(dev_name, proc_ide_root, name);
 238 }
 239
 240 static int fake_ide_setup(char *str)
 241 {
 242         fake_ide = 1;
 243         return 1;
 244 }
 245
 246 __setup("fake_ide", fake_ide_setup);
 247
 248 __uml_help(fake_ide_setup,
 249 "fake_ide\n"
 250 "    Create ide0 entries that map onto ubd devices.\n\n"
 251 );
 252
 253 static int parse_unit(char **ptr)
 254 {
 255         char *str = *ptr, *end;
 256         int n = -1;
 257
 258         if(isdigit(*str)) {
 259                 n = simple_strtoul(str, &end, 0);
 260                 if(end == str)
 261                         return -1;
 262                 *ptr = end;
 263         }
 264         else if (('a' <= *str) && (*str <= 'z')) {
 265                 n = *str - 'a';
 266                 str++;
 267                 *ptr = str;
 268         }
 269         return n;
 270 }
 271
 272 /* If *index_out == -1 at exit, the passed option was a general one;
 273  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 274  * should not be freed on exit.
 275  */
 276 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 277 {
 278         struct ubd *ubd_dev;
 279         struct openflags flags = global_openflags;
 280         char *backing_file;
 281         int n, err = 0, i;
 282
 283         if(index_out) *index_out = -1;
 284         n = *str;
 285         if(n == '='){
 286                 char *end;
 287                 int major;
 288
 289                 str++;
 290                 if(!strcmp(str, "sync")){
 291                         global_openflags = of_sync(global_openflags);
 292                         goto out1;
 293                 }
 294
 295                 err = -EINVAL;
 296                 major = simple_strtoul(str, &end, 0);
 297                 if((*end != '\0') || (end == str)){
 298                         *error_out = "Didn't parse major number";
 299                         goto out1;
 300                 }
 301
 302                 mutex_lock(&ubd_lock);
 303                 if (fake_major != UBD_MAJOR) {
 304                         *error_out = "Can't assign a fake major twice";
 305                         goto out1;
 306                 }
 307
 308                 fake_major = major;
 309
 310                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 311                        major);
 312                 err = 0;
 313         out1:
 314                 mutex_unlock(&ubd_lock);
 315                 return err;
 316         }
 317
 318         n = parse_unit(&str);
 319         if(n < 0){
 320                 *error_out = "Couldn't parse device number";
 321                 return -EINVAL;
 322         }
 323         if(n >= MAX_DEV){
 324                 *error_out = "Device number out of range";
 325                 return 1;
 326         }
 327
 328         err = -EBUSY;
 329         mutex_lock(&ubd_lock);
 330
 331         ubd_dev = &ubd_devs[n];
 332         if(ubd_dev->file != NULL){
 333                 *error_out = "Device is already configured";
 334                 goto out;
 335         }
 336
 337         if (index_out)
 338                 *index_out = n;
 339
 340         err = -EINVAL;
 341         for (i = 0; i < sizeof("rscd="); i++) {
 342                 switch (*str) {
 343                 case 'r':
 344                         flags.w = 0;
 345                         break;
 346                 case 's':
 347                         flags.s = 1;
 348                         break;
 349                 case 'd':
 350                         ubd_dev->no_cow = 1;
 351                         break;
 352                 case 'c':
 353                         ubd_dev->shared = 1;
 354                         break;
 355                 case '=':
 356                         str++;
 357                         goto break_loop;
 358                 default:
 359                         *error_out = "Expected '=' or flag letter "
 360                                 "(r, s, c, or d)";
 361                         goto out;
 362                 }
 363                 str++;
 364         }
 365
 366         if (*str == '=')
 367                 *error_out = "Too many flags specified";
 368         else
 369                 *error_out = "Missing '='";
 370         goto out;
 371
 372 break_loop:
 373         backing_file = strchr(str, ',');
 374
 375         if (backing_file == NULL)
 376                 backing_file = strchr(str, ':');
 377
 378         if(backing_file != NULL){
 379                 if(ubd_dev->no_cow){
 380                         *error_out = "Can't specify both 'd' and a cow file";
 381                         goto out;
 382                 }
 383                 else {
 384                         *backing_file = '\0';
 385                         backing_file++;
 386                 }
 387         }
 388         err = 0;
 389         ubd_dev->file = str;
 390         ubd_dev->cow.file = backing_file;
 391         ubd_dev->boot_openflags = flags;
 392 out:
 393         mutex_unlock(&ubd_lock);
 394         return err;
 395 }
 396
 397 static int ubd_setup(char *str)
 398 {
 399         char *error;
 400         int err;
 401
 402         err = ubd_setup_common(str, NULL, &error);
 403         if(err)
 404                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 405                        "%s\n", str, error);
 406         return 1;
 407 }
 408
 409 __setup("ubd", ubd_setup);
 410 __uml_help(ubd_setup,
 411 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 412 "    This is used to associate a device with a file in the underlying\n"
 413 "    filesystem. When specifying two filenames, the first one is the\n"
 414 "    COW name and the second is the backing file name. As separator you can\n"
 415 "    use either a ':' or a ',': the first one allows writing things like;\n"
 416 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 417 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 418 "    When using only one filename, UML will detect whether to treat it like\n"
 419 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 420 "    flag:\n"
 421 "       ubd0d=BackingFile\n"
 422 "    Usually, there is a filesystem in the file, but \n"
 423 "    that's not required. Swap devices containing swap files can be\n"
 424 "    specified like this. Also, a file which doesn't contain a\n"
 425 "    filesystem can have its contents read in the virtual \n"
 426 "    machine by running 'dd' on the device. <n> must be in the range\n"
 427 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 428 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 429 "    an 's' will cause data to be written to disk on the host immediately.\n"
 430 "    'c' will cause the device to be treated as being shared between multiple\n"
 431 "    UMLs and file locking will be turned off - this is appropriate for a\n"
 432 "    cluster filesystem and inappropriate at almost all other times.\n\n"
 433 );
 434
 435 static int udb_setup(char *str)
 436 {
 437         printk("udb%s specified on command line is almost certainly a ubd -> "
 438                "udb TYPO\n", str);
 439         return 1;
 440 }
 441
 442 __setup("udb", udb_setup);
 443 __uml_help(udb_setup,
 444 "udb\n"
 445 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 446 "    to impossible to catch visually unless you specifically look for\n"
 447 "    them.  The only result of any option starting with 'udb' is an error\n"
 448 "    in the boot output.\n\n"
 449 );
 450
 451 static void do_ubd_request(struct request_queue * q);
 452
 453 /* Only changed by ubd_init, which is an initcall. */
 454 static int thread_fd = -1;
 455 static LIST_HEAD(restart);
 456
 457 /* Function to read several request pointers at a time
 458 * handling fractional reads if (and as) needed
 459 */
 460
 461 static int bulk_req_safe_read(
 462         int fd,
 463         struct io_thread_req * (*request_buffer)[],
 464         struct io_thread_req **remainder,
 465         int *remainder_size,
 466         int max_recs
 467         )
 468 {
 469         int n = 0;
 470         int res = 0;
 471
 472         if (*remainder_size > 0) {
 473                 memmove(
 474                         (char *) request_buffer,
 475                         (char *) remainder, *remainder_size
 476                 );
 477                 n = *remainder_size;
 478         }
 479
 480         res = os_read_file(
 481                         fd,
 482                         ((char *) request_buffer) + *remainder_size,
 483                         sizeof(struct io_thread_req *)*max_recs
 484                                 - *remainder_size
 485                 );
 486         if (res > 0) {
 487                 n += res;
 488                 if ((n % sizeof(struct io_thread_req *)) > 0) {
 489                         /*
 490                         * Read somehow returned not a multiple of dword
 491                         * theoretically possible, but never observed in the
 492                         * wild, so read routine must be able to handle it
 493                         */
 494                         *remainder_size = n % sizeof(struct io_thread_req *);
 495                         WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
 496                         memmove(
 497                                 remainder,
 498                                 ((char *) request_buffer) +
 499                                         (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
 500                                 *remainder_size
 501                         );
 502                         n = n - *remainder_size;
 503                 }
 504         } else {
 505                 n = res;
 506         }
 507         return n;
 508 }
 509
 510 /* Called without dev->lock held, and only in interrupt context. */
 511 static void ubd_handler(void)
 512 {
 513         struct ubd *ubd;
 514         struct list_head *list, *next_ele;
 515         unsigned long flags;
 516         int n;
 517         int count;
 518
 519         while(1){
 520                 n = bulk_req_safe_read(
 521                         thread_fd,
 522                         irq_req_buffer,
 523                         &irq_remainder,
 524                         &irq_remainder_size,
 525                         UBD_REQ_BUFFER_SIZE
 526                 );
 527                 if (n < 0) {
 528                         if(n == -EAGAIN)
 529                                 break;
 530                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 531                                "err = %d\n", -n);
 532                         return;
 533                 }
 534                 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
 535                         blk_end_request(
 536                                 (*irq_req_buffer)[count]->req,
 537                                 BLK_STS_OK,
 538                                 (*irq_req_buffer)[count]->length
 539                         );
 540                         kfree((*irq_req_buffer)[count]);
 541                 }
 542         }
 543         reactivate_fd(thread_fd, UBD_IRQ);
 544
 545         list_for_each_safe(list, next_ele, &restart){
 546                 ubd = container_of(list, struct ubd, restart);
 547                 list_del_init(&ubd->restart);
 548                 spin_lock_irqsave(&ubd->lock, flags);
 549                 do_ubd_request(ubd->queue);
 550                 spin_unlock_irqrestore(&ubd->lock, flags);
 551         }
 552 }
 553
 554 static irqreturn_t ubd_intr(int irq, void *dev)
 555 {
 556         ubd_handler();
 557         return IRQ_HANDLED;
 558 }
 559
 560 /* Only changed by ubd_init, which is an initcall. */
 561 static int io_pid = -1;
 562
 563 static void kill_io_thread(void)
 564 {
 565         if(io_pid != -1)
 566                 os_kill_process(io_pid, 1);
 567 }
 568
 569 __uml_exitcall(kill_io_thread);
 570
 571 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 572 {
 573         char *file;
 574         int fd;
 575         int err;
 576
 577         __u32 version;
 578         __u32 align;
 579         char *backing_file;
 580         time_t mtime;
 581         unsigned long long size;
 582         int sector_size;
 583         int bitmap_offset;
 584
 585         if (ubd_dev->file && ubd_dev->cow.file) {
 586                 file = ubd_dev->cow.file;
 587
 588                 goto out;
 589         }
 590
 591         fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
 592         if (fd < 0)
 593                 return fd;
 594
 595         err = read_cow_header(file_reader, &fd, &version, &backing_file, \
 596                 &mtime, &size, &sector_size, &align, &bitmap_offset);
 597         os_close_file(fd);
 598
 599         if(err == -EINVAL)
 600                 file = ubd_dev->file;
 601         else
 602                 file = backing_file;
 603
 604 out:
 605         return os_file_size(file, size_out);
 606 }
 607
 608 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
 609 {
 610         int err;
 611
 612         err = os_pread_file(fd, buf, len, offset);
 613         if (err < 0)
 614                 return err;
 615
 616         return 0;
 617 }
 618
 619 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
 620 {
 621         unsigned long modtime;
 622         unsigned long long actual;
 623         int err;
 624
 625         err = os_file_modtime(file, &modtime);
 626         if (err < 0) {
 627                 printk(KERN_ERR "Failed to get modification time of backing "
 628                        "file \"%s\", err = %d\n", file, -err);
 629                 return err;
 630         }
 631
 632         err = os_file_size(file, &actual);
 633         if (err < 0) {
 634                 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
 635                        "err = %d\n", file, -err);
 636                 return err;
 637         }
 638
 639         if (actual != size) {
 640                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
 641                  * the typecast.*/
 642                 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
 643                        "vs backing file\n", (unsigned long long) size, actual);
 644                 return -EINVAL;
 645         }
 646         if (modtime != mtime) {
 647                 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
 648                        "backing file\n", mtime, modtime);
 649                 return -EINVAL;
 650         }
 651         return 0;
 652 }
 653
 654 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
 655 {
 656         struct uml_stat buf1, buf2;
 657         int err;
 658
 659         if (from_cmdline == NULL)
 660                 return 0;
 661         if (!strcmp(from_cmdline, from_cow))
 662                 return 0;
 663
 664         err = os_stat_file(from_cmdline, &buf1);
 665         if (err < 0) {
 666                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
 667                        -err);
 668                 return 0;
 669         }
 670         err = os_stat_file(from_cow, &buf2);
 671         if (err < 0) {
 672                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
 673                        -err);
 674                 return 1;
 675         }
 676         if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
 677                 return 0;
 678
 679         printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
 680                "\"%s\" specified in COW header of \"%s\"\n",
 681                from_cmdline, from_cow, cow);
 682         return 1;
 683 }
 684
 685 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
 686                   char **backing_file_out, int *bitmap_offset_out,
 687                   unsigned long *bitmap_len_out, int *data_offset_out,
 688                   int *create_cow_out)
 689 {
 690         time_t mtime;
 691         unsigned long long size;
 692         __u32 version, align;
 693         char *backing_file;
 694         int fd, err, sectorsize, asked_switch, mode = 0644;
 695
 696         fd = os_open_file(file, *openflags, mode);
 697         if (fd < 0) {
 698                 if ((fd == -ENOENT) && (create_cow_out != NULL))
 699                         *create_cow_out = 1;
 700                 if (!openflags->w ||
 701                     ((fd != -EROFS) && (fd != -EACCES)))
 702                         return fd;
 703                 openflags->w = 0;
 704                 fd = os_open_file(file, *openflags, mode);
 705                 if (fd < 0)
 706                         return fd;
 707         }
 708
 709         if (shared)
 710                 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
 711         else {
 712                 err = os_lock_file(fd, openflags->w);
 713                 if (err < 0) {
 714                         printk(KERN_ERR "Failed to lock '%s', err = %d\n",
 715                                file, -err);
 716                         goto out_close;
 717                 }
 718         }
 719
 720         /* Successful return case! */
 721         if (backing_file_out == NULL)
 722                 return fd;
 723
 724         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
 725                               &size, &sectorsize, &align, bitmap_offset_out);
 726         if (err && (*backing_file_out != NULL)) {
 727                 printk(KERN_ERR "Failed to read COW header from COW file "
 728                        "\"%s\", errno = %d\n", file, -err);
 729                 goto out_close;
 730         }
 731         if (err)
 732                 return fd;
 733
 734         asked_switch = path_requires_switch(*backing_file_out, backing_file,
 735                                             file);
 736
 737         /* Allow switching only if no mismatch. */
 738         if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
 739                                                    mtime)) {
 740                 printk(KERN_ERR "Switching backing file to '%s'\n",
 741                        *backing_file_out);
 742                 err = write_cow_header(file, fd, *backing_file_out,
 743                                        sectorsize, align, &size);
 744                 if (err) {
 745                         printk(KERN_ERR "Switch failed, errno = %d\n", -err);
 746                         goto out_close;
 747                 }
 748         } else {
 749                 *backing_file_out = backing_file;
 750                 err = backing_file_mismatch(*backing_file_out, size, mtime);
 751                 if (err)
 752                         goto out_close;
 753         }
 754
 755         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
 756                   bitmap_len_out, data_offset_out);
 757
 758         return fd;
 759  out_close:
 760         os_close_file(fd);
 761         return err;
 762 }
 763
 764 static int create_cow_file(char *cow_file, char *backing_file,
 765                     struct openflags flags,
 766                     int sectorsize, int alignment, int *bitmap_offset_out,
 767                     unsigned long *bitmap_len_out, int *data_offset_out)
 768 {
 769         int err, fd;
 770
 771         flags.c = 1;
 772         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
 773         if (fd < 0) {
 774                 err = fd;
 775                 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
 776                        cow_file, -err);
 777                 goto out;
 778         }
 779
 780         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
 781                             bitmap_offset_out, bitmap_len_out,
 782                             data_offset_out);
 783         if (!err)
 784                 return fd;
 785         os_close_file(fd);
 786  out:
 787         return err;
 788 }
 789
 790 static void ubd_close_dev(struct ubd *ubd_dev)
 791 {
 792         os_close_file(ubd_dev->fd);
 793         if(ubd_dev->cow.file == NULL)
 794                 return;
 795
 796         os_close_file(ubd_dev->cow.fd);
 797         vfree(ubd_dev->cow.bitmap);
 798         ubd_dev->cow.bitmap = NULL;
 799 }
 800
 801 static int ubd_open_dev(struct ubd *ubd_dev)
 802 {
 803         struct openflags flags;
 804         char **back_ptr;
 805         int err, create_cow, *create_ptr;
 806         int fd;
 807
 808         ubd_dev->openflags = ubd_dev->boot_openflags;
 809         create_cow = 0;
 810         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 811         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 812
 813         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 814                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 815                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 816                                 create_ptr);
 817
 818         if((fd == -ENOENT) && create_cow){
 819                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 820                                           ubd_dev->openflags, 1 << 9, PAGE_SIZE,
 821                                           &ubd_dev->cow.bitmap_offset,
 822                                           &ubd_dev->cow.bitmap_len,
 823                                           &ubd_dev->cow.data_offset);
 824                 if(fd >= 0){
 825                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 826                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 827                 }
 828         }
 829
 830         if(fd < 0){
 831                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 832                        -fd);
 833                 return fd;
 834         }
 835         ubd_dev->fd = fd;
 836
 837         if(ubd_dev->cow.file != NULL){
 838                 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
 839
 840                 err = -ENOMEM;
 841                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 842                 if(ubd_dev->cow.bitmap == NULL){
 843                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 844                         goto error;
 845                 }
 846                 flush_tlb_kernel_vm();
 847
 848                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 849                                       ubd_dev->cow.bitmap_offset,
 850                                       ubd_dev->cow.bitmap_len);
 851                 if(err < 0)
 852                         goto error;
 853
 854                 flags = ubd_dev->openflags;
 855                 flags.w = 0;
 856                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 857                                     NULL, NULL, NULL, NULL);
 858                 if(err < 0) goto error;
 859                 ubd_dev->cow.fd = err;
 860         }
 861         return 0;
 862  error:
 863         os_close_file(ubd_dev->fd);
 864         return err;
 865 }
 866
 867 static void ubd_device_release(struct device *dev)
 868 {
 869         struct ubd *ubd_dev = dev_get_drvdata(dev);
 870
 871         blk_cleanup_queue(ubd_dev->queue);
 872         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 873 }
 874
 875 static int ubd_disk_register(int major, u64 size, int unit,
 876                              struct gendisk **disk_out)
 877 {
 878         struct device *parent = NULL;
 879         struct gendisk *disk;
 880
 881         disk = alloc_disk(1 << UBD_SHIFT);
 882         if(disk == NULL)
 883                 return -ENOMEM;
 884
 885         disk->major = major;
 886         disk->first_minor = unit << UBD_SHIFT;
 887         disk->fops = &ubd_blops;
 888         set_capacity(disk, size / 512);
 889         if (major == UBD_MAJOR)
 890                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 891         else
 892                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 893
 894         /* sysfs register (not for ide fake devices) */
 895         if (major == UBD_MAJOR) {
 896                 ubd_devs[unit].pdev.id   = unit;
 897                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 898                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 899                 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
 900                 platform_device_register(&ubd_devs[unit].pdev);
 901                 parent = &ubd_devs[unit].pdev.dev;
 902         }
 903
 904         disk->private_data = &ubd_devs[unit];
 905         disk->queue = ubd_devs[unit].queue;
 906         device_add_disk(parent, disk);
 907
 908         *disk_out = disk;
 909         return 0;
 910 }
 911
 912 #define ROUND_BLOCK(n) ((n + ((1 << 9) - 1)) & (-1 << 9))
 913
 914 static int ubd_add(int n, char **error_out)
 915 {
 916         struct ubd *ubd_dev = &ubd_devs[n];
 917         int err = 0;
 918
 919         if(ubd_dev->file == NULL)
 920                 goto out;
 921
 922         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 923         if(err < 0){
 924                 *error_out = "Couldn't determine size of device's file";
 925                 goto out;
 926         }
 927
 928         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 929
 930         INIT_LIST_HEAD(&ubd_dev->restart);
 931         sg_init_table(ubd_dev->sg, MAX_SG);
 932
 933         err = -ENOMEM;
 934         ubd_dev->queue = blk_init_queue(do_ubd_request, &ubd_dev->lock);
 935         if (ubd_dev->queue == NULL) {
 936                 *error_out = "Failed to initialize device queue";
 937                 goto out;
 938         }
 939         ubd_dev->queue->queuedata = ubd_dev;
 940         blk_queue_write_cache(ubd_dev->queue, true, false);
 941
 942         blk_queue_max_segments(ubd_dev->queue, MAX_SG);
 943         err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
 944         if(err){
 945                 *error_out = "Failed to register device";
 946                 goto out_cleanup;
 947         }
 948
 949         if (fake_major != UBD_MAJOR)
 950                 ubd_disk_register(fake_major, ubd_dev->size, n,
 951                                   &fake_gendisk[n]);
 952
 953         /*
 954          * Perhaps this should also be under the "if (fake_major)" above
 955          * using the fake_disk->disk_name
 956          */
 957         if (fake_ide)
 958                 make_ide_entries(ubd_gendisk[n]->disk_name);
 959
 960         err = 0;
 961 out:
 962         return err;
 963
 964 out_cleanup:
 965         blk_cleanup_queue(ubd_dev->queue);
 966         goto out;
 967 }
 968
 969 static int ubd_config(char *str, char **error_out)
 970 {
 971         int n, ret;
 972
 973         /* This string is possibly broken up and stored, so it's only
 974          * freed if ubd_setup_common fails, or if only general options
 975          * were set.
 976          */
 977         str = kstrdup(str, GFP_KERNEL);
 978         if (str == NULL) {
 979                 *error_out = "Failed to allocate memory";
 980                 return -ENOMEM;
 981         }
 982
 983         ret = ubd_setup_common(str, &n, error_out);
 984         if (ret)
 985                 goto err_free;
 986
 987         if (n == -1) {
 988                 ret = 0;
 989                 goto err_free;
 990         }
 991
 992         mutex_lock(&ubd_lock);
 993         ret = ubd_add(n, error_out);
 994         if (ret)
 995                 ubd_devs[n].file = NULL;
 996         mutex_unlock(&ubd_lock);
 997
 998 out:
 999         return ret;
1000
1001 err_free:
1002         kfree(str);
1003         goto out;
1004 }
1005
1006 static int ubd_get_config(char *name, char *str, int size, char **error_out)
1007 {
1008         struct ubd *ubd_dev;
1009         int n, len = 0;
1010
1011         n = parse_unit(&name);
1012         if((n >= MAX_DEV) || (n < 0)){
1013                 *error_out = "ubd_get_config : device number out of range";
1014                 return -1;
1015         }
1016
1017         ubd_dev = &ubd_devs[n];
1018         mutex_lock(&ubd_lock);
1019
1020         if(ubd_dev->file == NULL){
1021                 CONFIG_CHUNK(str, size, len, "", 1);
1022                 goto out;
1023         }
1024
1025         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
1026
1027         if(ubd_dev->cow.file != NULL){
1028                 CONFIG_CHUNK(str, size, len, ",", 0);
1029                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
1030         }
1031         else CONFIG_CHUNK(str, size, len, "", 1);
1032
1033  out:
1034         mutex_unlock(&ubd_lock);
1035         return len;
1036 }
1037
1038 static int ubd_id(char **str, int *start_out, int *end_out)
1039 {
1040         int n;
1041
1042         n = parse_unit(str);
1043         *start_out = 0;
1044         *end_out = MAX_DEV - 1;
1045         return n;
1046 }
1047
1048 static int ubd_remove(int n, char **error_out)
1049 {
1050         struct gendisk *disk = ubd_gendisk[n];
1051         struct ubd *ubd_dev;
1052         int err = -ENODEV;
1053
1054         mutex_lock(&ubd_lock);
1055
1056         ubd_dev = &ubd_devs[n];
1057
1058         if(ubd_dev->file == NULL)
1059                 goto out;
1060
1061         /* you cannot remove a open disk */
1062         err = -EBUSY;
1063         if(ubd_dev->count > 0)
1064                 goto out;
1065
1066         ubd_gendisk[n] = NULL;
1067         if(disk != NULL){
1068                 del_gendisk(disk);
1069                 put_disk(disk);
1070         }
1071
1072         if(fake_gendisk[n] != NULL){
1073                 del_gendisk(fake_gendisk[n]);
1074                 put_disk(fake_gendisk[n]);
1075                 fake_gendisk[n] = NULL;
1076         }
1077
1078         err = 0;
1079         platform_device_unregister(&ubd_dev->pdev);
1080 out:
1081         mutex_unlock(&ubd_lock);
1082         return err;
1083 }
1084
1085 /* All these are called by mconsole in process context and without
1086  * ubd-specific locks.  The structure itself is const except for .list.
1087  */
1088 static struct mc_device ubd_mc = {
1089         .list           = LIST_HEAD_INIT(ubd_mc.list),
1090         .name           = "ubd",
1091         .config         = ubd_config,
1092         .get_config     = ubd_get_config,
1093         .id             = ubd_id,
1094         .remove         = ubd_remove,
1095 };
1096
1097 static int __init ubd_mc_init(void)
1098 {
1099         mconsole_register_dev(&ubd_mc);
1100         return 0;
1101 }
1102
1103 __initcall(ubd_mc_init);
1104
1105 static int __init ubd0_init(void)
1106 {
1107         struct ubd *ubd_dev = &ubd_devs[0];
1108
1109         mutex_lock(&ubd_lock);
1110         if(ubd_dev->file == NULL)
1111                 ubd_dev->file = "root_fs";
1112         mutex_unlock(&ubd_lock);
1113
1114         return 0;
1115 }
1116
1117 __initcall(ubd0_init);
1118
1119 /* Used in ubd_init, which is an initcall */
1120 static struct platform_driver ubd_driver = {
1121         .driver = {
1122                 .name  = DRIVER_NAME,
1123         },
1124 };
1125
1126 static int __init ubd_init(void)
1127 {
1128         char *error;
1129         int i, err;
1130
1131         if (register_blkdev(UBD_MAJOR, "ubd"))
1132                 return -1;
1133
1134         if (fake_major != UBD_MAJOR) {
1135                 char name[sizeof("ubd_nnn\0")];
1136
1137                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1138                 if (register_blkdev(fake_major, "ubd"))
1139                         return -1;
1140         }
1141
1142         irq_req_buffer = kmalloc(
1143                         sizeof(struct io_thread_req *) * UBD_REQ_BUFFER_SIZE,
1144                         GFP_KERNEL
1145                 );
1146         irq_remainder = 0;
1147
1148         if (irq_req_buffer == NULL) {
1149                 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1150                 return -1;
1151         }
1152         io_req_buffer = kmalloc(
1153                         sizeof(struct io_thread_req *) * UBD_REQ_BUFFER_SIZE,
1154                         GFP_KERNEL
1155                 );
1156
1157         io_remainder = 0;
1158
1159         if (io_req_buffer == NULL) {
1160                 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1161                 return -1;
1162         }
1163         platform_driver_register(&ubd_driver);
1164         mutex_lock(&ubd_lock);
1165         for (i = 0; i < MAX_DEV; i++){
1166                 err = ubd_add(i, &error);
1167                 if(err)
1168                         printk(KERN_ERR "Failed to initialize ubd device %d :"
1169                                "%s\n", i, error);
1170         }
1171         mutex_unlock(&ubd_lock);
1172         return 0;
1173 }
1174
1175 late_initcall(ubd_init);
1176
1177 static int __init ubd_driver_init(void){
1178         unsigned long stack;
1179         int err;
1180
1181         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1182         if(global_openflags.s){
1183                 printk(KERN_INFO "ubd: Synchronous mode\n");
1184                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1185                  * enough. So use anyway the io thread. */
1186         }
1187         stack = alloc_stack(0, 0);
1188         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1189                                  &thread_fd);
1190         if(io_pid < 0){
1191                 printk(KERN_ERR
1192                        "ubd : Failed to start I/O thread (errno = %d) - "
1193                        "falling back to synchronous I/O\n", -io_pid);
1194                 io_pid = -1;
1195                 return 0;
1196         }
1197         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1198                              0, "ubd", ubd_devs);
1199         if(err != 0)
1200                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1201         return 0;
1202 }
1203
1204 device_initcall(ubd_driver_init);
1205
1206 static int ubd_open(struct block_device *bdev, fmode_t mode)
1207 {
1208         struct gendisk *disk = bdev->bd_disk;
1209         struct ubd *ubd_dev = disk->private_data;
1210         int err = 0;
1211
1212         mutex_lock(&ubd_mutex);
1213         if(ubd_dev->count == 0){
1214                 err = ubd_open_dev(ubd_dev);
1215                 if(err){
1216                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1217                                disk->disk_name, ubd_dev->file, -err);
1218                         goto out;
1219                 }
1220         }
1221         ubd_dev->count++;
1222         set_disk_ro(disk, !ubd_dev->openflags.w);
1223
1224         /* This should no more be needed. And it didn't work anyway to exclude
1225          * read-write remounting of filesystems.*/
1226         /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1227                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1228                 err = -EROFS;
1229         }*/
1230 out:
1231         mutex_unlock(&ubd_mutex);
1232         return err;
1233 }
1234
1235 static void ubd_release(struct gendisk *disk, fmode_t mode)
1236 {
1237         struct ubd *ubd_dev = disk->private_data;
1238
1239         mutex_lock(&ubd_mutex);
1240         if(--ubd_dev->count == 0)
1241                 ubd_close_dev(ubd_dev);
1242         mutex_unlock(&ubd_mutex);
1243 }
1244
1245 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1246                           __u64 *cow_offset, unsigned long *bitmap,
1247                           __u64 bitmap_offset, unsigned long *bitmap_words,
1248                           __u64 bitmap_len)
1249 {
1250         __u64 sector = io_offset >> 9;
1251         int i, update_bitmap = 0;
1252
1253         for(i = 0; i < length >> 9; i++){
1254                 if(cow_mask != NULL)
1255                         ubd_set_bit(i, (unsigned char *) cow_mask);
1256                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1257                         continue;
1258
1259                 update_bitmap = 1;
1260                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1261         }
1262
1263         if(!update_bitmap)
1264                 return;
1265
1266         *cow_offset = sector / (sizeof(unsigned long) * 8);
1267
1268         /* This takes care of the case where we're exactly at the end of the
1269          * device, and *cow_offset + 1 is off the end.  So, just back it up
1270          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1271          * for the original diagnosis.
1272          */
1273         if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1274                                          sizeof(unsigned long)) - 1))
1275                 (*cow_offset)--;
1276
1277         bitmap_words[0] = bitmap[*cow_offset];
1278         bitmap_words[1] = bitmap[*cow_offset + 1];
1279
1280         *cow_offset *= sizeof(unsigned long);
1281         *cow_offset += bitmap_offset;
1282 }
1283
1284 static void cowify_req(struct io_thread_req *req, unsigned long *bitmap,
1285                        __u64 bitmap_offset, __u64 bitmap_len)
1286 {
1287         __u64 sector = req->offset >> 9;
1288         int i;
1289
1290         if(req->length > (sizeof(req->sector_mask) * 8) << 9)
1291                 panic("Operation too long");
1292
1293         if(req->op == UBD_READ) {
1294                 for(i = 0; i < req->length >> 9; i++){
1295                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1296                                 ubd_set_bit(i, (unsigned char *)
1297                                             &req->sector_mask);
1298                 }
1299         }
1300         else cowify_bitmap(req->offset, req->length, &req->sector_mask,
1301                            &req->cow_offset, bitmap, bitmap_offset,
1302                            req->bitmap_words, bitmap_len);
1303 }
1304
1305 /* Called with dev->lock held */
1306 static void prepare_request(struct request *req, struct io_thread_req *io_req,
1307                             unsigned long long offset, int page_offset,
1308                             int len, struct page *page)
1309 {
1310         struct gendisk *disk = req->rq_disk;
1311         struct ubd *ubd_dev = disk->private_data;
1312
1313         io_req->req = req;
1314         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1315                 ubd_dev->fd;
1316         io_req->fds[1] = ubd_dev->fd;
1317         io_req->cow_offset = -1;
1318         io_req->offset = offset;
1319         io_req->length = len;
1320         io_req->error = 0;
1321         io_req->sector_mask = 0;
1322
1323         io_req->op = (rq_data_dir(req) == READ) ? UBD_READ : UBD_WRITE;
1324         io_req->offsets[0] = 0;
1325         io_req->offsets[1] = ubd_dev->cow.data_offset;
1326         io_req->buffer = page_address(page) + page_offset;
1327         io_req->sectorsize = 1 << 9;
1328
1329         if(ubd_dev->cow.file != NULL)
1330                 cowify_req(io_req, ubd_dev->cow.bitmap,
1331                            ubd_dev->cow.bitmap_offset, ubd_dev->cow.bitmap_len);
1332
1333 }
1334
1335 /* Called with dev->lock held */
1336 static void prepare_flush_request(struct request *req,
1337                                   struct io_thread_req *io_req)
1338 {
1339         struct gendisk *disk = req->rq_disk;
1340         struct ubd *ubd_dev = disk->private_data;
1341
1342         io_req->req = req;
1343         io_req->fds[0] = (ubd_dev->cow.file != NULL) ? ubd_dev->cow.fd :
1344                 ubd_dev->fd;
1345         io_req->op = UBD_FLUSH;
1346 }
1347
1348 static bool submit_request(struct io_thread_req *io_req, struct ubd *dev)
1349 {
1350         int n = os_write_file(thread_fd, &io_req,
1351                              sizeof(io_req));
1352         if (n != sizeof(io_req)) {
1353                 if (n != -EAGAIN)
1354                         printk("write to io thread failed, "
1355                                "errno = %d\n", -n);
1356                 else if (list_empty(&dev->restart))
1357                         list_add(&dev->restart, &restart);
1358
1359                 kfree(io_req);
1360                 return false;
1361         }
1362         return true;
1363 }
1364
1365 /* Called with dev->lock held */
1366 static void do_ubd_request(struct request_queue *q)
1367 {
1368         struct io_thread_req *io_req;
1369         struct request *req;
1370
1371         while(1){
1372                 struct ubd *dev = q->queuedata;
1373                 if(dev->request == NULL){
1374                         struct request *req = blk_fetch_request(q);
1375                         if(req == NULL)
1376                                 return;
1377
1378                         dev->request = req;
1379                         dev->rq_pos = blk_rq_pos(req);
1380                         dev->start_sg = 0;
1381                         dev->end_sg = blk_rq_map_sg(q, req, dev->sg);
1382                 }
1383
1384                 req = dev->request;
1385
1386                 if (req_op(req) == REQ_OP_FLUSH) {
1387                         io_req = kmalloc(sizeof(struct io_thread_req),
1388                                          GFP_ATOMIC);
1389                         if (io_req == NULL) {
1390                                 if (list_empty(&dev->restart))
1391                                         list_add(&dev->restart, &restart);
1392                                 return;
1393                         }
1394                         prepare_flush_request(req, io_req);
1395                         if (submit_request(io_req, dev) == false)
1396                                 return;
1397                 }
1398
1399                 while(dev->start_sg < dev->end_sg){
1400                         struct scatterlist *sg = &dev->sg[dev->start_sg];
1401
1402                         io_req = kmalloc(sizeof(struct io_thread_req),
1403                                          GFP_ATOMIC);
1404                         if(io_req == NULL){
1405                                 if(list_empty(&dev->restart))
1406                                         list_add(&dev->restart, &restart);
1407                                 return;
1408                         }
1409                         prepare_request(req, io_req,
1410                                         (unsigned long long)dev->rq_pos << 9,
1411                                         sg->offset, sg->length, sg_page(sg));
1412
1413                         if (submit_request(io_req, dev) == false)
1414                                 return;
1415
1416                         dev->rq_pos += sg->length >> 9;
1417                         dev->start_sg++;
1418                 }
1419                 dev->end_sg = 0;
1420                 dev->request = NULL;
1421         }
1422 }
1423
1424 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1425 {
1426         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1427
1428         geo->heads = 128;
1429         geo->sectors = 32;
1430         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1431         return 0;
1432 }
1433
1434 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1435                      unsigned int cmd, unsigned long arg)
1436 {
1437         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1438         u16 ubd_id[ATA_ID_WORDS];
1439
1440         switch (cmd) {
1441                 struct cdrom_volctrl volume;
1442         case HDIO_GET_IDENTITY:
1443                 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1444                 ubd_id[ATA_ID_CYLS]     = ubd_dev->size / (128 * 32 * 512);
1445                 ubd_id[ATA_ID_HEADS]    = 128;
1446                 ubd_id[ATA_ID_SECTORS]  = 32;
1447                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1448                                  sizeof(ubd_id)))
1449                         return -EFAULT;
1450                 return 0;
1451
1452         case CDROMVOLREAD:
1453                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1454                         return -EFAULT;
1455                 volume.channel0 = 255;
1456                 volume.channel1 = 255;
1457                 volume.channel2 = 255;
1458                 volume.channel3 = 255;
1459                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1460                         return -EFAULT;
1461                 return 0;
1462         }
1463         return -EINVAL;
1464 }
1465
1466 static int update_bitmap(struct io_thread_req *req)
1467 {
1468         int n;
1469
1470         if(req->cow_offset == -1)
1471                 return 0;
1472
1473         n = os_pwrite_file(req->fds[1], &req->bitmap_words,
1474                           sizeof(req->bitmap_words), req->cow_offset);
1475         if(n != sizeof(req->bitmap_words)){
1476                 printk("do_io - bitmap update failed, err = %d fd = %d\n", -n,
1477                        req->fds[1]);
1478                 return 1;
1479         }
1480
1481         return 0;
1482 }
1483
1484 static void do_io(struct io_thread_req *req)
1485 {
1486         char *buf;
1487         unsigned long len;
1488         int n, nsectors, start, end, bit;
1489         __u64 off;
1490
1491         if (req->op == UBD_FLUSH) {
1492                 /* fds[0] is always either the rw image or our cow file */
1493                 n = os_sync_file(req->fds[0]);
1494                 if (n != 0) {
1495                         printk("do_io - sync failed err = %d "
1496                                "fd = %d\n", -n, req->fds[0]);
1497                         req->error = 1;
1498                 }
1499                 return;
1500         }
1501
1502         nsectors = req->length / req->sectorsize;
1503         start = 0;
1504         do {
1505                 bit = ubd_test_bit(start, (unsigned char *) &req->sector_mask);
1506                 end = start;
1507                 while((end < nsectors) &&
1508                       (ubd_test_bit(end, (unsigned char *)
1509                                     &req->sector_mask) == bit))
1510                         end++;
1511
1512                 off = req->offset + req->offsets[bit] +
1513                         start * req->sectorsize;
1514                 len = (end - start) * req->sectorsize;
1515                 buf = &req->buffer[start * req->sectorsize];
1516
1517                 if(req->op == UBD_READ){
1518                         n = 0;
1519                         do {
1520                                 buf = &buf[n];
1521                                 len -= n;
1522                                 n = os_pread_file(req->fds[bit], buf, len, off);
1523                                 if (n < 0) {
1524                                         printk("do_io - read failed, err = %d "
1525                                                "fd = %d\n", -n, req->fds[bit]);
1526                                         req->error = 1;
1527                                         return;
1528                                 }
1529                         } while((n < len) && (n != 0));
1530                         if (n < len) memset(&buf[n], 0, len - n);
1531                 } else {
1532                         n = os_pwrite_file(req->fds[bit], buf, len, off);
1533                         if(n != len){
1534                                 printk("do_io - write failed err = %d "
1535                                        "fd = %d\n", -n, req->fds[bit]);
1536                                 req->error = 1;
1537                                 return;
1538                         }
1539                 }
1540
1541                 start = end;
1542         } while(start < nsectors);
1543
1544         req->error = update_bitmap(req);
1545 }
1546
1547 /* Changed in start_io_thread, which is serialized by being called only
1548  * from ubd_init, which is an initcall.
1549  */
1550 int kernel_fd = -1;
1551
1552 /* Only changed by the io thread. XXX: currently unused. */
1553 static int io_count = 0;
1554
1555 int io_thread(void *arg)
1556 {
1557         int n, count, written, res;
1558
1559         os_fix_helper_signals();
1560
1561         while(1){
1562                 n = bulk_req_safe_read(
1563                         kernel_fd,
1564                         io_req_buffer,
1565                         &io_remainder,
1566                         &io_remainder_size,
1567                         UBD_REQ_BUFFER_SIZE
1568                 );
1569                 if (n < 0) {
1570                         if (n == -EAGAIN) {
1571                                 ubd_read_poll(-1);
1572                                 continue;
1573                         } else {
1574                                 printk("io_thread - read failed, fd = %d, "
1575                                        "err = %d,"
1576                                        "reminder = %d\n",
1577                                        kernel_fd, -n, io_remainder_size);
1578                         }
1579                 }
1580
1581                 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1582                         io_count++;
1583                         do_io((*io_req_buffer)[count]);
1584                 }
1585
1586                 written = 0;
1587
1588                 do {
1589                         res = os_write_file(kernel_fd, ((char *) io_req_buffer) + written, n);
1590                         if (res > 0) {
1591                                 written += res;
1592                         } else {
1593                                 if (res != -EAGAIN) {
1594                                         printk("io_thread - read failed, fd = %d, "
1595                                                "err = %d\n", kernel_fd, -n);
1596                                 }
1597                         }
1598                         if (written < n) {
1599                                 ubd_write_poll(-1);
1600                         }
1601                 } while (written < n);
1602         }
1603
1604         return 0;
1605 }