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