arch/um/drivers/ubd_kern.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2018 Cambridge Greys Ltd
   4  * Copyright (C) 2015-2016 Anton Ivanov (aivanov@brocade.com)
   5  * Copyright (C) 2000 Jeff Dike (jdike@karaya.com)
   6  */
   7
   8 /* 2001-09-28...2002-04-17
   9  * Partition stuff by James_McMechan@hotmail.com
  10  * old style ubd by setting UBD_SHIFT to 0
  11  * 2002-09-27...2002-10-18 massive tinkering for 2.5
  12  * partitions have changed in 2.5
  13  * 2003-01-29 more tinkering for 2.5.59-1
  14  * This should now address the sysfs problems and has
  15  * the symlink for devfs to allow for booting with
  16  * the common /dev/ubd/discX/... names rather than
  17  * only /dev/ubdN/discN this version also has lots of
  18  * clean ups preparing for ubd-many.
  19  * James McMechan
  20  */
  21
  22 #define UBD_SHIFT 4
  23
  24 #include <linux/module.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/blk-mq.h>
  28 #include <linux/ata.h>
  29 #include <linux/hdreg.h>
  30 #include <linux/major.h>
  31 #include <linux/cdrom.h>
  32 #include <linux/proc_fs.h>
  33 #include <linux/seq_file.h>
  34 #include <linux/ctype.h>
  35 #include <linux/slab.h>
  36 #include <linux/vmalloc.h>
  37 #include <linux/platform_device.h>
  38 #include <linux/scatterlist.h>
  39 #include <asm/tlbflush.h>
  40 #include <kern_util.h>
  41 #include "mconsole_kern.h"
  42 #include <init.h>
  43 #include <irq_kern.h>
  44 #include "ubd.h"
  45 #include <os.h>
  46 #include "cow.h"
  47
  48 /* Max request size is determined by sector mask - 32K */
  49 #define UBD_MAX_REQUEST (8 * sizeof(long))
  50
  51 struct io_desc {
  52         char *buffer;
  53         unsigned long length;
  54         unsigned long sector_mask;
  55         unsigned long long cow_offset;
  56         unsigned long bitmap_words[2];
  57 };
  58
  59 struct io_thread_req {
  60         struct request *req;
  61         int fds[2];
  62         unsigned long offsets[2];
  63         unsigned long long offset;
  64         int sectorsize;
  65         int error;
  66
  67         int desc_cnt;
  68         /* io_desc has to be the last element of the struct */
  69         struct io_desc io_desc[];
  70 };
  71
  72
  73 static struct io_thread_req * (*irq_req_buffer)[];
  74 static struct io_thread_req *irq_remainder;
  75 static int irq_remainder_size;
  76
  77 static struct io_thread_req * (*io_req_buffer)[];
  78 static struct io_thread_req *io_remainder;
  79 static int io_remainder_size;
  80
  81
  82
  83 static inline int ubd_test_bit(__u64 bit, unsigned char *data)
  84 {
  85         __u64 n;
  86         int bits, off;
  87
  88         bits = sizeof(data[0]) * 8;
  89         n = bit / bits;
  90         off = bit % bits;
  91         return (data[n] & (1 << off)) != 0;
  92 }
  93
  94 static inline void ubd_set_bit(__u64 bit, unsigned char *data)
  95 {
  96         __u64 n;
  97         int bits, off;
  98
  99         bits = sizeof(data[0]) * 8;
 100         n = bit / bits;
 101         off = bit % bits;
 102         data[n] |= (1 << off);
 103 }
 104 /*End stuff from ubd_user.h*/
 105
 106 #define DRIVER_NAME "uml-blkdev"
 107
 108 static DEFINE_MUTEX(ubd_lock);
 109 static DEFINE_MUTEX(ubd_mutex); /* replaces BKL, might not be needed */
 110
 111 static int ubd_open(struct block_device *bdev, fmode_t mode);
 112 static void ubd_release(struct gendisk *disk, fmode_t mode);
 113 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
 114                      unsigned int cmd, unsigned long arg);
 115 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 116
 117 #define MAX_DEV (16)
 118
 119 static const struct block_device_operations ubd_blops = {
 120         .owner          = THIS_MODULE,
 121         .open           = ubd_open,
 122         .release        = ubd_release,
 123         .ioctl          = ubd_ioctl,
 124         .compat_ioctl   = blkdev_compat_ptr_ioctl,
 125         .getgeo         = ubd_getgeo,
 126 };
 127
 128 /* Protected by ubd_lock */
 129 static struct gendisk *ubd_gendisk[MAX_DEV];
 130
 131 #ifdef CONFIG_BLK_DEV_UBD_SYNC
 132 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 1, .c = 0, \
 133                                          .cl = 1 })
 134 #else
 135 #define OPEN_FLAGS ((struct openflags) { .r = 1, .w = 1, .s = 0, .c = 0, \
 136                                          .cl = 1 })
 137 #endif
 138 static struct openflags global_openflags = OPEN_FLAGS;
 139
 140 struct cow {
 141         /* backing file name */
 142         char *file;
 143         /* backing file fd */
 144         int fd;
 145         unsigned long *bitmap;
 146         unsigned long bitmap_len;
 147         int bitmap_offset;
 148         int data_offset;
 149 };
 150
 151 #define MAX_SG 64
 152
 153 struct ubd {
 154         /* name (and fd, below) of the file opened for writing, either the
 155          * backing or the cow file. */
 156         char *file;
 157         char *serial;
 158         int count;
 159         int fd;
 160         __u64 size;
 161         struct openflags boot_openflags;
 162         struct openflags openflags;
 163         unsigned shared:1;
 164         unsigned no_cow:1;
 165         unsigned no_trim:1;
 166         struct cow cow;
 167         struct platform_device pdev;
 168         struct request_queue *queue;
 169         struct blk_mq_tag_set tag_set;
 170         spinlock_t lock;
 171 };
 172
 173 #define DEFAULT_COW { \
 174         .file =                 NULL, \
 175         .fd =                   -1,     \
 176         .bitmap =               NULL, \
 177         .bitmap_offset =        0, \
 178         .data_offset =          0, \
 179 }
 180
 181 #define DEFAULT_UBD { \
 182         .file =                 NULL, \
 183         .serial =               NULL, \
 184         .count =                0, \
 185         .fd =                   -1, \
 186         .size =                 -1, \
 187         .boot_openflags =       OPEN_FLAGS, \
 188         .openflags =            OPEN_FLAGS, \
 189         .no_cow =               0, \
 190         .no_trim =              0, \
 191         .shared =               0, \
 192         .cow =                  DEFAULT_COW, \
 193         .lock =                 __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
 194 }
 195
 196 /* Protected by ubd_lock */
 197 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
 198
 199 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
 200                                  const struct blk_mq_queue_data *bd);
 201
 202 static int fake_ide_setup(char *str)
 203 {
 204         pr_warn("The fake_ide option has been removed\n");
 205         return 1;
 206 }
 207 __setup("fake_ide", fake_ide_setup);
 208
 209 __uml_help(fake_ide_setup,
 210 "fake_ide\n"
 211 "    Obsolete stub.\n\n"
 212 );
 213
 214 static int parse_unit(char **ptr)
 215 {
 216         char *str = *ptr, *end;
 217         int n = -1;
 218
 219         if(isdigit(*str)) {
 220                 n = simple_strtoul(str, &end, 0);
 221                 if(end == str)
 222                         return -1;
 223                 *ptr = end;
 224         }
 225         else if (('a' <= *str) && (*str <= 'z')) {
 226                 n = *str - 'a';
 227                 str++;
 228                 *ptr = str;
 229         }
 230         return n;
 231 }
 232
 233 /* If *index_out == -1 at exit, the passed option was a general one;
 234  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 235  * should not be freed on exit.
 236  */
 237 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 238 {
 239         struct ubd *ubd_dev;
 240         struct openflags flags = global_openflags;
 241         char *file, *backing_file, *serial;
 242         int n, err = 0, i;
 243
 244         if(index_out) *index_out = -1;
 245         n = *str;
 246         if(n == '='){
 247                 str++;
 248                 if(!strcmp(str, "sync")){
 249                         global_openflags = of_sync(global_openflags);
 250                         return err;
 251                 }
 252
 253                 pr_warn("fake major not supported any more\n");
 254                 return 0;
 255         }
 256
 257         n = parse_unit(&str);
 258         if(n < 0){
 259                 *error_out = "Couldn't parse device number";
 260                 return -EINVAL;
 261         }
 262         if(n >= MAX_DEV){
 263                 *error_out = "Device number out of range";
 264                 return 1;
 265         }
 266
 267         err = -EBUSY;
 268         mutex_lock(&ubd_lock);
 269
 270         ubd_dev = &ubd_devs[n];
 271         if(ubd_dev->file != NULL){
 272                 *error_out = "Device is already configured";
 273                 goto out;
 274         }
 275
 276         if (index_out)
 277                 *index_out = n;
 278
 279         err = -EINVAL;
 280         for (i = 0; i < sizeof("rscdt="); i++) {
 281                 switch (*str) {
 282                 case 'r':
 283                         flags.w = 0;
 284                         break;
 285                 case 's':
 286                         flags.s = 1;
 287                         break;
 288                 case 'd':
 289                         ubd_dev->no_cow = 1;
 290                         break;
 291                 case 'c':
 292                         ubd_dev->shared = 1;
 293                         break;
 294                 case 't':
 295                         ubd_dev->no_trim = 1;
 296                         break;
 297                 case '=':
 298                         str++;
 299                         goto break_loop;
 300                 default:
 301                         *error_out = "Expected '=' or flag letter "
 302                                 "(r, s, c, t or d)";
 303                         goto out;
 304                 }
 305                 str++;
 306         }
 307
 308         if (*str == '=')
 309                 *error_out = "Too many flags specified";
 310         else
 311                 *error_out = "Missing '='";
 312         goto out;
 313
 314 break_loop:
 315         file = strsep(&str, ",:");
 316         if (*file == '\0')
 317                 file = NULL;
 318
 319         backing_file = strsep(&str, ",:");
 320         if (backing_file && *backing_file == '\0')
 321                 backing_file = NULL;
 322
 323         serial = strsep(&str, ",:");
 324         if (serial && *serial == '\0')
 325                 serial = NULL;
 326
 327         if (backing_file && ubd_dev->no_cow) {
 328                 *error_out = "Can't specify both 'd' and a cow file";
 329                 goto out;
 330         }
 331
 332         err = 0;
 333         ubd_dev->file = file;
 334         ubd_dev->cow.file = backing_file;
 335         ubd_dev->serial = serial;
 336         ubd_dev->boot_openflags = flags;
 337 out:
 338         mutex_unlock(&ubd_lock);
 339         return err;
 340 }
 341
 342 static int ubd_setup(char *str)
 343 {
 344         char *error;
 345         int err;
 346
 347         err = ubd_setup_common(str, NULL, &error);
 348         if(err)
 349                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 350                        "%s\n", str, error);
 351         return 1;
 352 }
 353
 354 __setup("ubd", ubd_setup);
 355 __uml_help(ubd_setup,
 356 "ubd<n><flags>=<filename>[(:|,)<filename2>][(:|,)<serial>]\n"
 357 "    This is used to associate a device with a file in the underlying\n"
 358 "    filesystem. When specifying two filenames, the first one is the\n"
 359 "    COW name and the second is the backing file name. As separator you can\n"
 360 "    use either a ':' or a ',': the first one allows writing things like;\n"
 361 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 362 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 363 "    When using only one filename, UML will detect whether to treat it like\n"
 364 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 365 "    flag:\n"
 366 "       ubd0d=BackingFile\n"
 367 "    Usually, there is a filesystem in the file, but \n"
 368 "    that's not required. Swap devices containing swap files can be\n"
 369 "    specified like this. Also, a file which doesn't contain a\n"
 370 "    filesystem can have its contents read in the virtual \n"
 371 "    machine by running 'dd' on the device. <n> must be in the range\n"
 372 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 373 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 374 "    an 's' will cause data to be written to disk on the host immediately.\n"
 375 "    'c' will cause the device to be treated as being shared between multiple\n"
 376 "    UMLs and file locking will be turned off - this is appropriate for a\n"
 377 "    cluster filesystem and inappropriate at almost all other times.\n\n"
 378 "    't' will disable trim/discard support on the device (enabled by default).\n\n"
 379 "    An optional device serial number can be exposed using the serial parameter\n"
 380 "    on the cmdline which is exposed as a sysfs entry. This is particularly\n"
 381 "    useful when a unique number should be given to the device. Note when\n"
 382 "    specifying a label, the filename2 must be also presented. It can be\n"
 383 "    an empty string, in which case the backing file is not used:\n"
 384 "       ubd0=File,,Serial\n"
 385 );
 386
 387 static int udb_setup(char *str)
 388 {
 389         printk("udb%s specified on command line is almost certainly a ubd -> "
 390                "udb TYPO\n", str);
 391         return 1;
 392 }
 393
 394 __setup("udb", udb_setup);
 395 __uml_help(udb_setup,
 396 "udb\n"
 397 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 398 "    to impossible to catch visually unless you specifically look for\n"
 399 "    them.  The only result of any option starting with 'udb' is an error\n"
 400 "    in the boot output.\n\n"
 401 );
 402
 403 /* Only changed by ubd_init, which is an initcall. */
 404 static int thread_fd = -1;
 405
 406 /* Function to read several request pointers at a time
 407 * handling fractional reads if (and as) needed
 408 */
 409
 410 static int bulk_req_safe_read(
 411         int fd,
 412         struct io_thread_req * (*request_buffer)[],
 413         struct io_thread_req **remainder,
 414         int *remainder_size,
 415         int max_recs
 416         )
 417 {
 418         int n = 0;
 419         int res = 0;
 420
 421         if (*remainder_size > 0) {
 422                 memmove(
 423                         (char *) request_buffer,
 424                         (char *) remainder, *remainder_size
 425                 );
 426                 n = *remainder_size;
 427         }
 428
 429         res = os_read_file(
 430                         fd,
 431                         ((char *) request_buffer) + *remainder_size,
 432                         sizeof(struct io_thread_req *)*max_recs
 433                                 - *remainder_size
 434                 );
 435         if (res > 0) {
 436                 n += res;
 437                 if ((n % sizeof(struct io_thread_req *)) > 0) {
 438                         /*
 439                         * Read somehow returned not a multiple of dword
 440                         * theoretically possible, but never observed in the
 441                         * wild, so read routine must be able to handle it
 442                         */
 443                         *remainder_size = n % sizeof(struct io_thread_req *);
 444                         WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
 445                         memmove(
 446                                 remainder,
 447                                 ((char *) request_buffer) +
 448                                         (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
 449                                 *remainder_size
 450                         );
 451                         n = n - *remainder_size;
 452                 }
 453         } else {
 454                 n = res;
 455         }
 456         return n;
 457 }
 458
 459 /* Called without dev->lock held, and only in interrupt context. */
 460 static void ubd_handler(void)
 461 {
 462         int n;
 463         int count;
 464
 465         while(1){
 466                 n = bulk_req_safe_read(
 467                         thread_fd,
 468                         irq_req_buffer,
 469                         &irq_remainder,
 470                         &irq_remainder_size,
 471                         UBD_REQ_BUFFER_SIZE
 472                 );
 473                 if (n < 0) {
 474                         if(n == -EAGAIN)
 475                                 break;
 476                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 477                                "err = %d\n", -n);
 478                         return;
 479                 }
 480                 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
 481                         struct io_thread_req *io_req = (*irq_req_buffer)[count];
 482
 483                         if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
 484                                 blk_queue_max_discard_sectors(io_req->req->q, 0);
 485                                 blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
 486                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, io_req->req->q);
 487                         }
 488                         blk_mq_end_request(io_req->req, io_req->error);
 489                         kfree(io_req);
 490                 }
 491         }
 492 }
 493
 494 static irqreturn_t ubd_intr(int irq, void *dev)
 495 {
 496         ubd_handler();
 497         return IRQ_HANDLED;
 498 }
 499
 500 /* Only changed by ubd_init, which is an initcall. */
 501 static int io_pid = -1;
 502
 503 static void kill_io_thread(void)
 504 {
 505         if(io_pid != -1)
 506                 os_kill_process(io_pid, 1);
 507 }
 508
 509 __uml_exitcall(kill_io_thread);
 510
 511 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 512 {
 513         char *file;
 514         int fd;
 515         int err;
 516
 517         __u32 version;
 518         __u32 align;
 519         char *backing_file;
 520         time64_t mtime;
 521         unsigned long long size;
 522         int sector_size;
 523         int bitmap_offset;
 524
 525         if (ubd_dev->file && ubd_dev->cow.file) {
 526                 file = ubd_dev->cow.file;
 527
 528                 goto out;
 529         }
 530
 531         fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
 532         if (fd < 0)
 533                 return fd;
 534
 535         err = read_cow_header(file_reader, &fd, &version, &backing_file, \
 536                 &mtime, &size, &sector_size, &align, &bitmap_offset);
 537         os_close_file(fd);
 538
 539         if(err == -EINVAL)
 540                 file = ubd_dev->file;
 541         else
 542                 file = backing_file;
 543
 544 out:
 545         return os_file_size(file, size_out);
 546 }
 547
 548 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
 549 {
 550         int err;
 551
 552         err = os_pread_file(fd, buf, len, offset);
 553         if (err < 0)
 554                 return err;
 555
 556         return 0;
 557 }
 558
 559 static int backing_file_mismatch(char *file, __u64 size, time64_t mtime)
 560 {
 561         time64_t modtime;
 562         unsigned long long actual;
 563         int err;
 564
 565         err = os_file_modtime(file, &modtime);
 566         if (err < 0) {
 567                 printk(KERN_ERR "Failed to get modification time of backing "
 568                        "file \"%s\", err = %d\n", file, -err);
 569                 return err;
 570         }
 571
 572         err = os_file_size(file, &actual);
 573         if (err < 0) {
 574                 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
 575                        "err = %d\n", file, -err);
 576                 return err;
 577         }
 578
 579         if (actual != size) {
 580                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
 581                  * the typecast.*/
 582                 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
 583                        "vs backing file\n", (unsigned long long) size, actual);
 584                 return -EINVAL;
 585         }
 586         if (modtime != mtime) {
 587                 printk(KERN_ERR "mtime mismatch (%lld vs %lld) of COW header vs "
 588                        "backing file\n", mtime, modtime);
 589                 return -EINVAL;
 590         }
 591         return 0;
 592 }
 593
 594 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
 595 {
 596         struct uml_stat buf1, buf2;
 597         int err;
 598
 599         if (from_cmdline == NULL)
 600                 return 0;
 601         if (!strcmp(from_cmdline, from_cow))
 602                 return 0;
 603
 604         err = os_stat_file(from_cmdline, &buf1);
 605         if (err < 0) {
 606                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
 607                        -err);
 608                 return 0;
 609         }
 610         err = os_stat_file(from_cow, &buf2);
 611         if (err < 0) {
 612                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
 613                        -err);
 614                 return 1;
 615         }
 616         if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
 617                 return 0;
 618
 619         printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
 620                "\"%s\" specified in COW header of \"%s\"\n",
 621                from_cmdline, from_cow, cow);
 622         return 1;
 623 }
 624
 625 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
 626                   char **backing_file_out, int *bitmap_offset_out,
 627                   unsigned long *bitmap_len_out, int *data_offset_out,
 628                   int *create_cow_out)
 629 {
 630         time64_t mtime;
 631         unsigned long long size;
 632         __u32 version, align;
 633         char *backing_file;
 634         int fd, err, sectorsize, asked_switch, mode = 0644;
 635
 636         fd = os_open_file(file, *openflags, mode);
 637         if (fd < 0) {
 638                 if ((fd == -ENOENT) && (create_cow_out != NULL))
 639                         *create_cow_out = 1;
 640                 if (!openflags->w ||
 641                     ((fd != -EROFS) && (fd != -EACCES)))
 642                         return fd;
 643                 openflags->w = 0;
 644                 fd = os_open_file(file, *openflags, mode);
 645                 if (fd < 0)
 646                         return fd;
 647         }
 648
 649         if (shared)
 650                 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
 651         else {
 652                 err = os_lock_file(fd, openflags->w);
 653                 if (err < 0) {
 654                         printk(KERN_ERR "Failed to lock '%s', err = %d\n",
 655                                file, -err);
 656                         goto out_close;
 657                 }
 658         }
 659
 660         /* Successful return case! */
 661         if (backing_file_out == NULL)
 662                 return fd;
 663
 664         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
 665                               &size, &sectorsize, &align, bitmap_offset_out);
 666         if (err && (*backing_file_out != NULL)) {
 667                 printk(KERN_ERR "Failed to read COW header from COW file "
 668                        "\"%s\", errno = %d\n", file, -err);
 669                 goto out_close;
 670         }
 671         if (err)
 672                 return fd;
 673
 674         asked_switch = path_requires_switch(*backing_file_out, backing_file,
 675                                             file);
 676
 677         /* Allow switching only if no mismatch. */
 678         if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
 679                                                    mtime)) {
 680                 printk(KERN_ERR "Switching backing file to '%s'\n",
 681                        *backing_file_out);
 682                 err = write_cow_header(file, fd, *backing_file_out,
 683                                        sectorsize, align, &size);
 684                 if (err) {
 685                         printk(KERN_ERR "Switch failed, errno = %d\n", -err);
 686                         goto out_close;
 687                 }
 688         } else {
 689                 *backing_file_out = backing_file;
 690                 err = backing_file_mismatch(*backing_file_out, size, mtime);
 691                 if (err)
 692                         goto out_close;
 693         }
 694
 695         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
 696                   bitmap_len_out, data_offset_out);
 697
 698         return fd;
 699  out_close:
 700         os_close_file(fd);
 701         return err;
 702 }
 703
 704 static int create_cow_file(char *cow_file, char *backing_file,
 705                     struct openflags flags,
 706                     int sectorsize, int alignment, int *bitmap_offset_out,
 707                     unsigned long *bitmap_len_out, int *data_offset_out)
 708 {
 709         int err, fd;
 710
 711         flags.c = 1;
 712         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
 713         if (fd < 0) {
 714                 err = fd;
 715                 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
 716                        cow_file, -err);
 717                 goto out;
 718         }
 719
 720         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
 721                             bitmap_offset_out, bitmap_len_out,
 722                             data_offset_out);
 723         if (!err)
 724                 return fd;
 725         os_close_file(fd);
 726  out:
 727         return err;
 728 }
 729
 730 static void ubd_close_dev(struct ubd *ubd_dev)
 731 {
 732         os_close_file(ubd_dev->fd);
 733         if(ubd_dev->cow.file == NULL)
 734                 return;
 735
 736         os_close_file(ubd_dev->cow.fd);
 737         vfree(ubd_dev->cow.bitmap);
 738         ubd_dev->cow.bitmap = NULL;
 739 }
 740
 741 static int ubd_open_dev(struct ubd *ubd_dev)
 742 {
 743         struct openflags flags;
 744         char **back_ptr;
 745         int err, create_cow, *create_ptr;
 746         int fd;
 747
 748         ubd_dev->openflags = ubd_dev->boot_openflags;
 749         create_cow = 0;
 750         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 751         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 752
 753         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 754                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 755                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 756                                 create_ptr);
 757
 758         if((fd == -ENOENT) && create_cow){
 759                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 760                                           ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
 761                                           &ubd_dev->cow.bitmap_offset,
 762                                           &ubd_dev->cow.bitmap_len,
 763                                           &ubd_dev->cow.data_offset);
 764                 if(fd >= 0){
 765                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 766                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 767                 }
 768         }
 769
 770         if(fd < 0){
 771                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 772                        -fd);
 773                 return fd;
 774         }
 775         ubd_dev->fd = fd;
 776
 777         if(ubd_dev->cow.file != NULL){
 778                 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
 779
 780                 err = -ENOMEM;
 781                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 782                 if(ubd_dev->cow.bitmap == NULL){
 783                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 784                         goto error;
 785                 }
 786                 flush_tlb_kernel_vm();
 787
 788                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 789                                       ubd_dev->cow.bitmap_offset,
 790                                       ubd_dev->cow.bitmap_len);
 791                 if(err < 0)
 792                         goto error;
 793
 794                 flags = ubd_dev->openflags;
 795                 flags.w = 0;
 796                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 797                                     NULL, NULL, NULL, NULL);
 798                 if(err < 0) goto error;
 799                 ubd_dev->cow.fd = err;
 800         }
 801         if (ubd_dev->no_trim == 0) {
 802                 ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE;
 803                 ubd_dev->queue->limits.discard_alignment = SECTOR_SIZE;
 804                 blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
 805                 blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
 806                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, ubd_dev->queue);
 807         }
 808         blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue);
 809         return 0;
 810  error:
 811         os_close_file(ubd_dev->fd);
 812         return err;
 813 }
 814
 815 static void ubd_device_release(struct device *dev)
 816 {
 817         struct ubd *ubd_dev = dev_get_drvdata(dev);
 818
 819         blk_mq_free_tag_set(&ubd_dev->tag_set);
 820         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 821 }
 822
 823 static ssize_t serial_show(struct device *dev,
 824                            struct device_attribute *attr, char *buf)
 825 {
 826         struct gendisk *disk = dev_to_disk(dev);
 827         struct ubd *ubd_dev = disk->private_data;
 828
 829         if (!ubd_dev)
 830                 return 0;
 831
 832         return sprintf(buf, "%s", ubd_dev->serial);
 833 }
 834
 835 static DEVICE_ATTR_RO(serial);
 836
 837 static struct attribute *ubd_attrs[] = {
 838         &dev_attr_serial.attr,
 839         NULL,
 840 };
 841
 842 static umode_t ubd_attrs_are_visible(struct kobject *kobj,
 843                                      struct attribute *a, int n)
 844 {
 845         return a->mode;
 846 }
 847
 848 static const struct attribute_group ubd_attr_group = {
 849         .attrs = ubd_attrs,
 850         .is_visible = ubd_attrs_are_visible,
 851 };
 852
 853 static const struct attribute_group *ubd_attr_groups[] = {
 854         &ubd_attr_group,
 855         NULL,
 856 };
 857
 858 static int ubd_disk_register(int major, u64 size, int unit,
 859                              struct gendisk *disk)
 860 {
 861         disk->major = major;
 862         disk->first_minor = unit << UBD_SHIFT;
 863         disk->minors = 1 << UBD_SHIFT;
 864         disk->fops = &ubd_blops;
 865         set_capacity(disk, size / 512);
 866         sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 867
 868         ubd_devs[unit].pdev.id   = unit;
 869         ubd_devs[unit].pdev.name = DRIVER_NAME;
 870         ubd_devs[unit].pdev.dev.release = ubd_device_release;
 871         dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
 872         platform_device_register(&ubd_devs[unit].pdev);
 873
 874         disk->private_data = &ubd_devs[unit];
 875         disk->queue = ubd_devs[unit].queue;
 876         return device_add_disk(&ubd_devs[unit].pdev.dev, disk, ubd_attr_groups);
 877 }
 878
 879 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
 880
 881 static const struct blk_mq_ops ubd_mq_ops = {
 882         .queue_rq = ubd_queue_rq,
 883 };
 884
 885 static int ubd_add(int n, char **error_out)
 886 {
 887         struct ubd *ubd_dev = &ubd_devs[n];
 888         struct gendisk *disk;
 889         int err = 0;
 890
 891         if(ubd_dev->file == NULL)
 892                 goto out;
 893
 894         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 895         if(err < 0){
 896                 *error_out = "Couldn't determine size of device's file";
 897                 goto out;
 898         }
 899
 900         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 901
 902         ubd_dev->tag_set.ops = &ubd_mq_ops;
 903         ubd_dev->tag_set.queue_depth = 64;
 904         ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
 905         ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 906         ubd_dev->tag_set.driver_data = ubd_dev;
 907         ubd_dev->tag_set.nr_hw_queues = 1;
 908
 909         err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
 910         if (err)
 911                 goto out;
 912
 913         disk = blk_mq_alloc_disk(&ubd_dev->tag_set, ubd_dev);
 914         if (IS_ERR(disk)) {
 915                 err = PTR_ERR(disk);
 916                 goto out_cleanup_tags;
 917         }
 918         ubd_dev->queue = disk->queue;
 919
 920         blk_queue_write_cache(ubd_dev->queue, true, false);
 921         blk_queue_max_segments(ubd_dev->queue, MAX_SG);
 922         blk_queue_segment_boundary(ubd_dev->queue, PAGE_SIZE - 1);
 923         err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, disk);
 924         if (err)
 925                 goto out_cleanup_disk;
 926
 927         ubd_gendisk[n] = disk;
 928         return 0;
 929
 930 out_cleanup_disk:
 931         blk_cleanup_disk(disk);
 932 out_cleanup_tags:
 933         blk_mq_free_tag_set(&ubd_dev->tag_set);
 934 out:
 935         return err;
 936 }
 937
 938 static int ubd_config(char *str, char **error_out)
 939 {
 940         int n, ret;
 941
 942         /* This string is possibly broken up and stored, so it's only
 943          * freed if ubd_setup_common fails, or if only general options
 944          * were set.
 945          */
 946         str = kstrdup(str, GFP_KERNEL);
 947         if (str == NULL) {
 948                 *error_out = "Failed to allocate memory";
 949                 return -ENOMEM;
 950         }
 951
 952         ret = ubd_setup_common(str, &n, error_out);
 953         if (ret)
 954                 goto err_free;
 955
 956         if (n == -1) {
 957                 ret = 0;
 958                 goto err_free;
 959         }
 960
 961         mutex_lock(&ubd_lock);
 962         ret = ubd_add(n, error_out);
 963         if (ret)
 964                 ubd_devs[n].file = NULL;
 965         mutex_unlock(&ubd_lock);
 966
 967 out:
 968         return ret;
 969
 970 err_free:
 971         kfree(str);
 972         goto out;
 973 }
 974
 975 static int ubd_get_config(char *name, char *str, int size, char **error_out)
 976 {
 977         struct ubd *ubd_dev;
 978         int n, len = 0;
 979
 980         n = parse_unit(&name);
 981         if((n >= MAX_DEV) || (n < 0)){
 982                 *error_out = "ubd_get_config : device number out of range";
 983                 return -1;
 984         }
 985
 986         ubd_dev = &ubd_devs[n];
 987         mutex_lock(&ubd_lock);
 988
 989         if(ubd_dev->file == NULL){
 990                 CONFIG_CHUNK(str, size, len, "", 1);
 991                 goto out;
 992         }
 993
 994         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
 995
 996         if(ubd_dev->cow.file != NULL){
 997                 CONFIG_CHUNK(str, size, len, ",", 0);
 998                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
 999         }
1000         else CONFIG_CHUNK(str, size, len, "", 1);
1001
1002  out:
1003         mutex_unlock(&ubd_lock);
1004         return len;
1005 }
1006
1007 static int ubd_id(char **str, int *start_out, int *end_out)
1008 {
1009         int n;
1010
1011         n = parse_unit(str);
1012         *start_out = 0;
1013         *end_out = MAX_DEV - 1;
1014         return n;
1015 }
1016
1017 static int ubd_remove(int n, char **error_out)
1018 {
1019         struct gendisk *disk = ubd_gendisk[n];
1020         struct ubd *ubd_dev;
1021         int err = -ENODEV;
1022
1023         mutex_lock(&ubd_lock);
1024
1025         ubd_dev = &ubd_devs[n];
1026
1027         if(ubd_dev->file == NULL)
1028                 goto out;
1029
1030         /* you cannot remove a open disk */
1031         err = -EBUSY;
1032         if(ubd_dev->count > 0)
1033                 goto out;
1034
1035         ubd_gendisk[n] = NULL;
1036         if(disk != NULL){
1037                 del_gendisk(disk);
1038                 blk_cleanup_disk(disk);
1039         }
1040
1041         err = 0;
1042         platform_device_unregister(&ubd_dev->pdev);
1043 out:
1044         mutex_unlock(&ubd_lock);
1045         return err;
1046 }
1047
1048 /* All these are called by mconsole in process context and without
1049  * ubd-specific locks.  The structure itself is const except for .list.
1050  */
1051 static struct mc_device ubd_mc = {
1052         .list           = LIST_HEAD_INIT(ubd_mc.list),
1053         .name           = "ubd",
1054         .config         = ubd_config,
1055         .get_config     = ubd_get_config,
1056         .id             = ubd_id,
1057         .remove         = ubd_remove,
1058 };
1059
1060 static int __init ubd_mc_init(void)
1061 {
1062         mconsole_register_dev(&ubd_mc);
1063         return 0;
1064 }
1065
1066 __initcall(ubd_mc_init);
1067
1068 static int __init ubd0_init(void)
1069 {
1070         struct ubd *ubd_dev = &ubd_devs[0];
1071
1072         mutex_lock(&ubd_lock);
1073         if(ubd_dev->file == NULL)
1074                 ubd_dev->file = "root_fs";
1075         mutex_unlock(&ubd_lock);
1076
1077         return 0;
1078 }
1079
1080 __initcall(ubd0_init);
1081
1082 /* Used in ubd_init, which is an initcall */
1083 static struct platform_driver ubd_driver = {
1084         .driver = {
1085                 .name  = DRIVER_NAME,
1086         },
1087 };
1088
1089 static int __init ubd_init(void)
1090 {
1091         char *error;
1092         int i, err;
1093
1094         if (register_blkdev(UBD_MAJOR, "ubd"))
1095                 return -1;
1096
1097         irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1098                                        sizeof(struct io_thread_req *),
1099                                        GFP_KERNEL
1100                 );
1101         irq_remainder = 0;
1102
1103         if (irq_req_buffer == NULL) {
1104                 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1105                 return -1;
1106         }
1107         io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1108                                       sizeof(struct io_thread_req *),
1109                                       GFP_KERNEL
1110                 );
1111
1112         io_remainder = 0;
1113
1114         if (io_req_buffer == NULL) {
1115                 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1116                 return -1;
1117         }
1118         platform_driver_register(&ubd_driver);
1119         mutex_lock(&ubd_lock);
1120         for (i = 0; i < MAX_DEV; i++){
1121                 err = ubd_add(i, &error);
1122                 if(err)
1123                         printk(KERN_ERR "Failed to initialize ubd device %d :"
1124                                "%s\n", i, error);
1125         }
1126         mutex_unlock(&ubd_lock);
1127         return 0;
1128 }
1129
1130 late_initcall(ubd_init);
1131
1132 static int __init ubd_driver_init(void){
1133         unsigned long stack;
1134         int err;
1135
1136         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1137         if(global_openflags.s){
1138                 printk(KERN_INFO "ubd: Synchronous mode\n");
1139                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1140                  * enough. So use anyway the io thread. */
1141         }
1142         stack = alloc_stack(0, 0);
1143         io_pid = start_io_thread(stack + PAGE_SIZE, &thread_fd);
1144         if(io_pid < 0){
1145                 printk(KERN_ERR
1146                        "ubd : Failed to start I/O thread (errno = %d) - "
1147                        "falling back to synchronous I/O\n", -io_pid);
1148                 io_pid = -1;
1149                 return 0;
1150         }
1151         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1152                              0, "ubd", ubd_devs);
1153         if(err < 0)
1154                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1155         return 0;
1156 }
1157
1158 device_initcall(ubd_driver_init);
1159
1160 static int ubd_open(struct block_device *bdev, fmode_t mode)
1161 {
1162         struct gendisk *disk = bdev->bd_disk;
1163         struct ubd *ubd_dev = disk->private_data;
1164         int err = 0;
1165
1166         mutex_lock(&ubd_mutex);
1167         if(ubd_dev->count == 0){
1168                 err = ubd_open_dev(ubd_dev);
1169                 if(err){
1170                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1171                                disk->disk_name, ubd_dev->file, -err);
1172                         goto out;
1173                 }
1174         }
1175         ubd_dev->count++;
1176         set_disk_ro(disk, !ubd_dev->openflags.w);
1177
1178         /* This should no more be needed. And it didn't work anyway to exclude
1179          * read-write remounting of filesystems.*/
1180         /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1181                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1182                 err = -EROFS;
1183         }*/
1184 out:
1185         mutex_unlock(&ubd_mutex);
1186         return err;
1187 }
1188
1189 static void ubd_release(struct gendisk *disk, fmode_t mode)
1190 {
1191         struct ubd *ubd_dev = disk->private_data;
1192
1193         mutex_lock(&ubd_mutex);
1194         if(--ubd_dev->count == 0)
1195                 ubd_close_dev(ubd_dev);
1196         mutex_unlock(&ubd_mutex);
1197 }
1198
1199 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1200                           __u64 *cow_offset, unsigned long *bitmap,
1201                           __u64 bitmap_offset, unsigned long *bitmap_words,
1202                           __u64 bitmap_len)
1203 {
1204         __u64 sector = io_offset >> SECTOR_SHIFT;
1205         int i, update_bitmap = 0;
1206
1207         for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1208                 if(cow_mask != NULL)
1209                         ubd_set_bit(i, (unsigned char *) cow_mask);
1210                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1211                         continue;
1212
1213                 update_bitmap = 1;
1214                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1215         }
1216
1217         if(!update_bitmap)
1218                 return;
1219
1220         *cow_offset = sector / (sizeof(unsigned long) * 8);
1221
1222         /* This takes care of the case where we're exactly at the end of the
1223          * device, and *cow_offset + 1 is off the end.  So, just back it up
1224          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1225          * for the original diagnosis.
1226          */
1227         if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1228                                          sizeof(unsigned long)) - 1))
1229                 (*cow_offset)--;
1230
1231         bitmap_words[0] = bitmap[*cow_offset];
1232         bitmap_words[1] = bitmap[*cow_offset + 1];
1233
1234         *cow_offset *= sizeof(unsigned long);
1235         *cow_offset += bitmap_offset;
1236 }
1237
1238 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1239                        unsigned long offset, unsigned long *bitmap,
1240                        __u64 bitmap_offset, __u64 bitmap_len)
1241 {
1242         __u64 sector = offset >> SECTOR_SHIFT;
1243         int i;
1244
1245         if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1246                 panic("Operation too long");
1247
1248         if (req_op(req->req) == REQ_OP_READ) {
1249                 for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1250                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1251                                 ubd_set_bit(i, (unsigned char *)
1252                                             &segment->sector_mask);
1253                 }
1254         } else {
1255                 cowify_bitmap(offset, segment->length, &segment->sector_mask,
1256                               &segment->cow_offset, bitmap, bitmap_offset,
1257                               segment->bitmap_words, bitmap_len);
1258         }
1259 }
1260
1261 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1262                         struct request *req)
1263 {
1264         struct bio_vec bvec;
1265         struct req_iterator iter;
1266         int i = 0;
1267         unsigned long byte_offset = io_req->offset;
1268         int op = req_op(req);
1269
1270         if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1271                 io_req->io_desc[0].buffer = NULL;
1272                 io_req->io_desc[0].length = blk_rq_bytes(req);
1273         } else {
1274                 rq_for_each_segment(bvec, req, iter) {
1275                         BUG_ON(i >= io_req->desc_cnt);
1276
1277                         io_req->io_desc[i].buffer = bvec_virt(&bvec);
1278                         io_req->io_desc[i].length = bvec.bv_len;
1279                         i++;
1280                 }
1281         }
1282
1283         if (dev->cow.file) {
1284                 for (i = 0; i < io_req->desc_cnt; i++) {
1285                         cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1286                                    dev->cow.bitmap, dev->cow.bitmap_offset,
1287                                    dev->cow.bitmap_len);
1288                         byte_offset += io_req->io_desc[i].length;
1289                 }
1290
1291         }
1292 }
1293
1294 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1295                                            int desc_cnt)
1296 {
1297         struct io_thread_req *io_req;
1298         int i;
1299
1300         io_req = kmalloc(sizeof(*io_req) +
1301                          (desc_cnt * sizeof(struct io_desc)),
1302                          GFP_ATOMIC);
1303         if (!io_req)
1304                 return NULL;
1305
1306         io_req->req = req;
1307         if (dev->cow.file)
1308                 io_req->fds[0] = dev->cow.fd;
1309         else
1310                 io_req->fds[0] = dev->fd;
1311         io_req->error = 0;
1312         io_req->sectorsize = SECTOR_SIZE;
1313         io_req->fds[1] = dev->fd;
1314         io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1315         io_req->offsets[0] = 0;
1316         io_req->offsets[1] = dev->cow.data_offset;
1317
1318         for (i = 0 ; i < desc_cnt; i++) {
1319                 io_req->io_desc[i].sector_mask = 0;
1320                 io_req->io_desc[i].cow_offset = -1;
1321         }
1322
1323         return io_req;
1324 }
1325
1326 static int ubd_submit_request(struct ubd *dev, struct request *req)
1327 {
1328         int segs = 0;
1329         struct io_thread_req *io_req;
1330         int ret;
1331         int op = req_op(req);
1332
1333         if (op == REQ_OP_FLUSH)
1334                 segs = 0;
1335         else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1336                 segs = 1;
1337         else
1338                 segs = blk_rq_nr_phys_segments(req);
1339
1340         io_req = ubd_alloc_req(dev, req, segs);
1341         if (!io_req)
1342                 return -ENOMEM;
1343
1344         io_req->desc_cnt = segs;
1345         if (segs)
1346                 ubd_map_req(dev, io_req, req);
1347
1348         ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1349         if (ret != sizeof(io_req)) {
1350                 if (ret != -EAGAIN)
1351                         pr_err("write to io thread failed: %d\n", -ret);
1352                 kfree(io_req);
1353         }
1354         return ret;
1355 }
1356
1357 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1358                                  const struct blk_mq_queue_data *bd)
1359 {
1360         struct ubd *ubd_dev = hctx->queue->queuedata;
1361         struct request *req = bd->rq;
1362         int ret = 0, res = BLK_STS_OK;
1363
1364         blk_mq_start_request(req);
1365
1366         spin_lock_irq(&ubd_dev->lock);
1367
1368         switch (req_op(req)) {
1369         case REQ_OP_FLUSH:
1370         case REQ_OP_READ:
1371         case REQ_OP_WRITE:
1372         case REQ_OP_DISCARD:
1373         case REQ_OP_WRITE_ZEROES:
1374                 ret = ubd_submit_request(ubd_dev, req);
1375                 break;
1376         default:
1377                 WARN_ON_ONCE(1);
1378                 res = BLK_STS_NOTSUPP;
1379         }
1380
1381         spin_unlock_irq(&ubd_dev->lock);
1382
1383         if (ret < 0) {
1384                 if (ret == -ENOMEM)
1385                         res = BLK_STS_RESOURCE;
1386                 else
1387                         res = BLK_STS_DEV_RESOURCE;
1388         }
1389
1390         return res;
1391 }
1392
1393 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1394 {
1395         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1396
1397         geo->heads = 128;
1398         geo->sectors = 32;
1399         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1400         return 0;
1401 }
1402
1403 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1404                      unsigned int cmd, unsigned long arg)
1405 {
1406         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1407         u16 ubd_id[ATA_ID_WORDS];
1408
1409         switch (cmd) {
1410                 struct cdrom_volctrl volume;
1411         case HDIO_GET_IDENTITY:
1412                 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1413                 ubd_id[ATA_ID_CYLS]     = ubd_dev->size / (128 * 32 * 512);
1414                 ubd_id[ATA_ID_HEADS]    = 128;
1415                 ubd_id[ATA_ID_SECTORS]  = 32;
1416                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1417                                  sizeof(ubd_id)))
1418                         return -EFAULT;
1419                 return 0;
1420
1421         case CDROMVOLREAD:
1422                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1423                         return -EFAULT;
1424                 volume.channel0 = 255;
1425                 volume.channel1 = 255;
1426                 volume.channel2 = 255;
1427                 volume.channel3 = 255;
1428                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1429                         return -EFAULT;
1430                 return 0;
1431         }
1432         return -EINVAL;
1433 }
1434
1435 static int map_error(int error_code)
1436 {
1437         switch (error_code) {
1438         case 0:
1439                 return BLK_STS_OK;
1440         case ENOSYS:
1441         case EOPNOTSUPP:
1442                 return BLK_STS_NOTSUPP;
1443         case ENOSPC:
1444                 return BLK_STS_NOSPC;
1445         }
1446         return BLK_STS_IOERR;
1447 }
1448
1449 /*
1450  * Everything from here onwards *IS NOT PART OF THE KERNEL*
1451  *
1452  * The following functions are part of UML hypervisor code.
1453  * All functions from here onwards are executed as a helper
1454  * thread and are not allowed to execute any kernel functions.
1455  *
1456  * Any communication must occur strictly via shared memory and IPC.
1457  *
1458  * Do not add printks, locks, kernel memory operations, etc - it
1459  * will result in unpredictable behaviour and/or crashes.
1460  */
1461
1462 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1463 {
1464         int n;
1465
1466         if (segment->cow_offset == -1)
1467                 return map_error(0);
1468
1469         n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1470                           sizeof(segment->bitmap_words), segment->cow_offset);
1471         if (n != sizeof(segment->bitmap_words))
1472                 return map_error(-n);
1473
1474         return map_error(0);
1475 }
1476
1477 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1478 {
1479         char *buf = NULL;
1480         unsigned long len;
1481         int n, nsectors, start, end, bit;
1482         __u64 off;
1483
1484         /* FLUSH is really a special case, we cannot "case" it with others */
1485
1486         if (req_op(req->req) == REQ_OP_FLUSH) {
1487                 /* fds[0] is always either the rw image or our cow file */
1488                 req->error = map_error(-os_sync_file(req->fds[0]));
1489                 return;
1490         }
1491
1492         nsectors = desc->length / req->sectorsize;
1493         start = 0;
1494         do {
1495                 bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1496                 end = start;
1497                 while((end < nsectors) &&
1498                       (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1499                         end++;
1500
1501                 off = req->offset + req->offsets[bit] +
1502                         start * req->sectorsize;
1503                 len = (end - start) * req->sectorsize;
1504                 if (desc->buffer != NULL)
1505                         buf = &desc->buffer[start * req->sectorsize];
1506
1507                 switch (req_op(req->req)) {
1508                 case REQ_OP_READ:
1509                         n = 0;
1510                         do {
1511                                 buf = &buf[n];
1512                                 len -= n;
1513                                 n = os_pread_file(req->fds[bit], buf, len, off);
1514                                 if (n < 0) {
1515                                         req->error = map_error(-n);
1516                                         return;
1517                                 }
1518                         } while((n < len) && (n != 0));
1519                         if (n < len) memset(&buf[n], 0, len - n);
1520                         break;
1521                 case REQ_OP_WRITE:
1522                         n = os_pwrite_file(req->fds[bit], buf, len, off);
1523                         if(n != len){
1524                                 req->error = map_error(-n);
1525                                 return;
1526                         }
1527                         break;
1528                 case REQ_OP_DISCARD:
1529                 case REQ_OP_WRITE_ZEROES:
1530                         n = os_falloc_punch(req->fds[bit], off, len);
1531                         if (n) {
1532                                 req->error = map_error(-n);
1533                                 return;
1534                         }
1535                         break;
1536                 default:
1537                         WARN_ON_ONCE(1);
1538                         req->error = BLK_STS_NOTSUPP;
1539                         return;
1540                 }
1541
1542                 start = end;
1543         } while(start < nsectors);
1544
1545         req->offset += len;
1546         req->error = update_bitmap(req, desc);
1547 }
1548
1549 /* Changed in start_io_thread, which is serialized by being called only
1550  * from ubd_init, which is an initcall.
1551  */
1552 int kernel_fd = -1;
1553
1554 /* Only changed by the io thread. XXX: currently unused. */
1555 static int io_count = 0;
1556
1557 int io_thread(void *arg)
1558 {
1559         int n, count, written, res;
1560
1561         os_fix_helper_signals();
1562
1563         while(1){
1564                 n = bulk_req_safe_read(
1565                         kernel_fd,
1566                         io_req_buffer,
1567                         &io_remainder,
1568                         &io_remainder_size,
1569                         UBD_REQ_BUFFER_SIZE
1570                 );
1571                 if (n <= 0) {
1572                         if (n == -EAGAIN)
1573                                 ubd_read_poll(-1);
1574
1575                         continue;
1576                 }
1577
1578                 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1579                         struct io_thread_req *req = (*io_req_buffer)[count];
1580                         int i;
1581
1582                         io_count++;
1583                         for (i = 0; !req->error && i < req->desc_cnt; i++)
1584                                 do_io(req, &(req->io_desc[i]));
1585
1586                 }
1587
1588                 written = 0;
1589
1590                 do {
1591                         res = os_write_file(kernel_fd,
1592                                             ((char *) io_req_buffer) + written,
1593                                             n - written);
1594                         if (res >= 0) {
1595                                 written += res;
1596                         }
1597                         if (written < n) {
1598                                 ubd_write_poll(-1);
1599                         }
1600                 } while (written < n);
1601         }
1602
1603         return 0;
1604 }