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         .getgeo         = ubd_getgeo,
 124 };
 125
 126 /* Protected by ubd_lock */
 127 static int fake_major = UBD_MAJOR;
 128 static struct gendisk *ubd_gendisk[MAX_DEV];
 129 static struct gendisk *fake_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         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         .count =                0, \
 183         .fd =                   -1, \
 184         .size =                 -1, \
 185         .boot_openflags =       OPEN_FLAGS, \
 186         .openflags =            OPEN_FLAGS, \
 187         .no_cow =               0, \
 188         .no_trim =              0, \
 189         .shared =               0, \
 190         .cow =                  DEFAULT_COW, \
 191         .lock =                 __SPIN_LOCK_UNLOCKED(ubd_devs.lock), \
 192 }
 193
 194 /* Protected by ubd_lock */
 195 static struct ubd ubd_devs[MAX_DEV] = { [0 ... MAX_DEV - 1] = DEFAULT_UBD };
 196
 197 /* Only changed by fake_ide_setup which is a setup */
 198 static int fake_ide = 0;
 199 static struct proc_dir_entry *proc_ide_root = NULL;
 200 static struct proc_dir_entry *proc_ide = NULL;
 201
 202 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
 203                                  const struct blk_mq_queue_data *bd);
 204
 205 static void make_proc_ide(void)
 206 {
 207         proc_ide_root = proc_mkdir("ide", NULL);
 208         proc_ide = proc_mkdir("ide0", proc_ide_root);
 209 }
 210
 211 static int fake_ide_media_proc_show(struct seq_file *m, void *v)
 212 {
 213         seq_puts(m, "disk\n");
 214         return 0;
 215 }
 216
 217 static void make_ide_entries(const char *dev_name)
 218 {
 219         struct proc_dir_entry *dir, *ent;
 220         char name[64];
 221
 222         if(proc_ide_root == NULL) make_proc_ide();
 223
 224         dir = proc_mkdir(dev_name, proc_ide);
 225         if(!dir) return;
 226
 227         ent = proc_create_single("media", S_IRUGO, dir,
 228                         fake_ide_media_proc_show);
 229         if(!ent) return;
 230         snprintf(name, sizeof(name), "ide0/%s", dev_name);
 231         proc_symlink(dev_name, proc_ide_root, name);
 232 }
 233
 234 static int fake_ide_setup(char *str)
 235 {
 236         fake_ide = 1;
 237         return 1;
 238 }
 239
 240 __setup("fake_ide", fake_ide_setup);
 241
 242 __uml_help(fake_ide_setup,
 243 "fake_ide\n"
 244 "    Create ide0 entries that map onto ubd devices.\n\n"
 245 );
 246
 247 static int parse_unit(char **ptr)
 248 {
 249         char *str = *ptr, *end;
 250         int n = -1;
 251
 252         if(isdigit(*str)) {
 253                 n = simple_strtoul(str, &end, 0);
 254                 if(end == str)
 255                         return -1;
 256                 *ptr = end;
 257         }
 258         else if (('a' <= *str) && (*str <= 'z')) {
 259                 n = *str - 'a';
 260                 str++;
 261                 *ptr = str;
 262         }
 263         return n;
 264 }
 265
 266 /* If *index_out == -1 at exit, the passed option was a general one;
 267  * otherwise, the str pointer is used (and owned) inside ubd_devs array, so it
 268  * should not be freed on exit.
 269  */
 270 static int ubd_setup_common(char *str, int *index_out, char **error_out)
 271 {
 272         struct ubd *ubd_dev;
 273         struct openflags flags = global_openflags;
 274         char *backing_file;
 275         int n, err = 0, i;
 276
 277         if(index_out) *index_out = -1;
 278         n = *str;
 279         if(n == '='){
 280                 char *end;
 281                 int major;
 282
 283                 str++;
 284                 if(!strcmp(str, "sync")){
 285                         global_openflags = of_sync(global_openflags);
 286                         return err;
 287                 }
 288
 289                 err = -EINVAL;
 290                 major = simple_strtoul(str, &end, 0);
 291                 if((*end != '\0') || (end == str)){
 292                         *error_out = "Didn't parse major number";
 293                         return err;
 294                 }
 295
 296                 mutex_lock(&ubd_lock);
 297                 if (fake_major != UBD_MAJOR) {
 298                         *error_out = "Can't assign a fake major twice";
 299                         goto out1;
 300                 }
 301
 302                 fake_major = major;
 303
 304                 printk(KERN_INFO "Setting extra ubd major number to %d\n",
 305                        major);
 306                 err = 0;
 307         out1:
 308                 mutex_unlock(&ubd_lock);
 309                 return err;
 310         }
 311
 312         n = parse_unit(&str);
 313         if(n < 0){
 314                 *error_out = "Couldn't parse device number";
 315                 return -EINVAL;
 316         }
 317         if(n >= MAX_DEV){
 318                 *error_out = "Device number out of range";
 319                 return 1;
 320         }
 321
 322         err = -EBUSY;
 323         mutex_lock(&ubd_lock);
 324
 325         ubd_dev = &ubd_devs[n];
 326         if(ubd_dev->file != NULL){
 327                 *error_out = "Device is already configured";
 328                 goto out;
 329         }
 330
 331         if (index_out)
 332                 *index_out = n;
 333
 334         err = -EINVAL;
 335         for (i = 0; i < sizeof("rscdt="); i++) {
 336                 switch (*str) {
 337                 case 'r':
 338                         flags.w = 0;
 339                         break;
 340                 case 's':
 341                         flags.s = 1;
 342                         break;
 343                 case 'd':
 344                         ubd_dev->no_cow = 1;
 345                         break;
 346                 case 'c':
 347                         ubd_dev->shared = 1;
 348                         break;
 349                 case 't':
 350                         ubd_dev->no_trim = 1;
 351                         break;
 352                 case '=':
 353                         str++;
 354                         goto break_loop;
 355                 default:
 356                         *error_out = "Expected '=' or flag letter "
 357                                 "(r, s, c, t or d)";
 358                         goto out;
 359                 }
 360                 str++;
 361         }
 362
 363         if (*str == '=')
 364                 *error_out = "Too many flags specified";
 365         else
 366                 *error_out = "Missing '='";
 367         goto out;
 368
 369 break_loop:
 370         backing_file = strchr(str, ',');
 371
 372         if (backing_file == NULL)
 373                 backing_file = strchr(str, ':');
 374
 375         if(backing_file != NULL){
 376                 if(ubd_dev->no_cow){
 377                         *error_out = "Can't specify both 'd' and a cow file";
 378                         goto out;
 379                 }
 380                 else {
 381                         *backing_file = '\0';
 382                         backing_file++;
 383                 }
 384         }
 385         err = 0;
 386         ubd_dev->file = str;
 387         ubd_dev->cow.file = backing_file;
 388         ubd_dev->boot_openflags = flags;
 389 out:
 390         mutex_unlock(&ubd_lock);
 391         return err;
 392 }
 393
 394 static int ubd_setup(char *str)
 395 {
 396         char *error;
 397         int err;
 398
 399         err = ubd_setup_common(str, NULL, &error);
 400         if(err)
 401                 printk(KERN_ERR "Failed to initialize device with \"%s\" : "
 402                        "%s\n", str, error);
 403         return 1;
 404 }
 405
 406 __setup("ubd", ubd_setup);
 407 __uml_help(ubd_setup,
 408 "ubd<n><flags>=<filename>[(:|,)<filename2>]\n"
 409 "    This is used to associate a device with a file in the underlying\n"
 410 "    filesystem. When specifying two filenames, the first one is the\n"
 411 "    COW name and the second is the backing file name. As separator you can\n"
 412 "    use either a ':' or a ',': the first one allows writing things like;\n"
 413 "       ubd0=~/Uml/root_cow:~/Uml/root_backing_file\n"
 414 "    while with a ',' the shell would not expand the 2nd '~'.\n"
 415 "    When using only one filename, UML will detect whether to treat it like\n"
 416 "    a COW file or a backing file. To override this detection, add the 'd'\n"
 417 "    flag:\n"
 418 "       ubd0d=BackingFile\n"
 419 "    Usually, there is a filesystem in the file, but \n"
 420 "    that's not required. Swap devices containing swap files can be\n"
 421 "    specified like this. Also, a file which doesn't contain a\n"
 422 "    filesystem can have its contents read in the virtual \n"
 423 "    machine by running 'dd' on the device. <n> must be in the range\n"
 424 "    0 to 7. Appending an 'r' to the number will cause that device\n"
 425 "    to be mounted read-only. For example ubd1r=./ext_fs. Appending\n"
 426 "    an 's' will cause data to be written to disk on the host immediately.\n"
 427 "    'c' will cause the device to be treated as being shared between multiple\n"
 428 "    UMLs and file locking will be turned off - this is appropriate for a\n"
 429 "    cluster filesystem and inappropriate at almost all other times.\n\n"
 430 "    't' will disable trim/discard support on the device (enabled by default).\n\n"
 431 );
 432
 433 static int udb_setup(char *str)
 434 {
 435         printk("udb%s specified on command line is almost certainly a ubd -> "
 436                "udb TYPO\n", str);
 437         return 1;
 438 }
 439
 440 __setup("udb", udb_setup);
 441 __uml_help(udb_setup,
 442 "udb\n"
 443 "    This option is here solely to catch ubd -> udb typos, which can be\n"
 444 "    to impossible to catch visually unless you specifically look for\n"
 445 "    them.  The only result of any option starting with 'udb' is an error\n"
 446 "    in the boot output.\n\n"
 447 );
 448
 449 /* Only changed by ubd_init, which is an initcall. */
 450 static int thread_fd = -1;
 451
 452 /* Function to read several request pointers at a time
 453 * handling fractional reads if (and as) needed
 454 */
 455
 456 static int bulk_req_safe_read(
 457         int fd,
 458         struct io_thread_req * (*request_buffer)[],
 459         struct io_thread_req **remainder,
 460         int *remainder_size,
 461         int max_recs
 462         )
 463 {
 464         int n = 0;
 465         int res = 0;
 466
 467         if (*remainder_size > 0) {
 468                 memmove(
 469                         (char *) request_buffer,
 470                         (char *) remainder, *remainder_size
 471                 );
 472                 n = *remainder_size;
 473         }
 474
 475         res = os_read_file(
 476                         fd,
 477                         ((char *) request_buffer) + *remainder_size,
 478                         sizeof(struct io_thread_req *)*max_recs
 479                                 - *remainder_size
 480                 );
 481         if (res > 0) {
 482                 n += res;
 483                 if ((n % sizeof(struct io_thread_req *)) > 0) {
 484                         /*
 485                         * Read somehow returned not a multiple of dword
 486                         * theoretically possible, but never observed in the
 487                         * wild, so read routine must be able to handle it
 488                         */
 489                         *remainder_size = n % sizeof(struct io_thread_req *);
 490                         WARN(*remainder_size > 0, "UBD IPC read returned a partial result");
 491                         memmove(
 492                                 remainder,
 493                                 ((char *) request_buffer) +
 494                                         (n/sizeof(struct io_thread_req *))*sizeof(struct io_thread_req *),
 495                                 *remainder_size
 496                         );
 497                         n = n - *remainder_size;
 498                 }
 499         } else {
 500                 n = res;
 501         }
 502         return n;
 503 }
 504
 505 /* Called without dev->lock held, and only in interrupt context. */
 506 static void ubd_handler(void)
 507 {
 508         int n;
 509         int count;
 510
 511         while(1){
 512                 n = bulk_req_safe_read(
 513                         thread_fd,
 514                         irq_req_buffer,
 515                         &irq_remainder,
 516                         &irq_remainder_size,
 517                         UBD_REQ_BUFFER_SIZE
 518                 );
 519                 if (n < 0) {
 520                         if(n == -EAGAIN)
 521                                 break;
 522                         printk(KERN_ERR "spurious interrupt in ubd_handler, "
 523                                "err = %d\n", -n);
 524                         return;
 525                 }
 526                 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
 527                         struct io_thread_req *io_req = (*irq_req_buffer)[count];
 528
 529                         if ((io_req->error == BLK_STS_NOTSUPP) && (req_op(io_req->req) == REQ_OP_DISCARD)) {
 530                                 blk_queue_max_discard_sectors(io_req->req->q, 0);
 531                                 blk_queue_max_write_zeroes_sectors(io_req->req->q, 0);
 532                                 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, io_req->req->q);
 533                         }
 534                         blk_mq_end_request(io_req->req, io_req->error);
 535                         kfree(io_req);
 536                 }
 537         }
 538 }
 539
 540 static irqreturn_t ubd_intr(int irq, void *dev)
 541 {
 542         ubd_handler();
 543         return IRQ_HANDLED;
 544 }
 545
 546 /* Only changed by ubd_init, which is an initcall. */
 547 static int io_pid = -1;
 548
 549 static void kill_io_thread(void)
 550 {
 551         if(io_pid != -1)
 552                 os_kill_process(io_pid, 1);
 553 }
 554
 555 __uml_exitcall(kill_io_thread);
 556
 557 static inline int ubd_file_size(struct ubd *ubd_dev, __u64 *size_out)
 558 {
 559         char *file;
 560         int fd;
 561         int err;
 562
 563         __u32 version;
 564         __u32 align;
 565         char *backing_file;
 566         time_t mtime;
 567         unsigned long long size;
 568         int sector_size;
 569         int bitmap_offset;
 570
 571         if (ubd_dev->file && ubd_dev->cow.file) {
 572                 file = ubd_dev->cow.file;
 573
 574                 goto out;
 575         }
 576
 577         fd = os_open_file(ubd_dev->file, of_read(OPENFLAGS()), 0);
 578         if (fd < 0)
 579                 return fd;
 580
 581         err = read_cow_header(file_reader, &fd, &version, &backing_file, \
 582                 &mtime, &size, &sector_size, &align, &bitmap_offset);
 583         os_close_file(fd);
 584
 585         if(err == -EINVAL)
 586                 file = ubd_dev->file;
 587         else
 588                 file = backing_file;
 589
 590 out:
 591         return os_file_size(file, size_out);
 592 }
 593
 594 static int read_cow_bitmap(int fd, void *buf, int offset, int len)
 595 {
 596         int err;
 597
 598         err = os_pread_file(fd, buf, len, offset);
 599         if (err < 0)
 600                 return err;
 601
 602         return 0;
 603 }
 604
 605 static int backing_file_mismatch(char *file, __u64 size, time_t mtime)
 606 {
 607         unsigned long modtime;
 608         unsigned long long actual;
 609         int err;
 610
 611         err = os_file_modtime(file, &modtime);
 612         if (err < 0) {
 613                 printk(KERN_ERR "Failed to get modification time of backing "
 614                        "file \"%s\", err = %d\n", file, -err);
 615                 return err;
 616         }
 617
 618         err = os_file_size(file, &actual);
 619         if (err < 0) {
 620                 printk(KERN_ERR "Failed to get size of backing file \"%s\", "
 621                        "err = %d\n", file, -err);
 622                 return err;
 623         }
 624
 625         if (actual != size) {
 626                 /*__u64 can be a long on AMD64 and with %lu GCC complains; so
 627                  * the typecast.*/
 628                 printk(KERN_ERR "Size mismatch (%llu vs %llu) of COW header "
 629                        "vs backing file\n", (unsigned long long) size, actual);
 630                 return -EINVAL;
 631         }
 632         if (modtime != mtime) {
 633                 printk(KERN_ERR "mtime mismatch (%ld vs %ld) of COW header vs "
 634                        "backing file\n", mtime, modtime);
 635                 return -EINVAL;
 636         }
 637         return 0;
 638 }
 639
 640 static int path_requires_switch(char *from_cmdline, char *from_cow, char *cow)
 641 {
 642         struct uml_stat buf1, buf2;
 643         int err;
 644
 645         if (from_cmdline == NULL)
 646                 return 0;
 647         if (!strcmp(from_cmdline, from_cow))
 648                 return 0;
 649
 650         err = os_stat_file(from_cmdline, &buf1);
 651         if (err < 0) {
 652                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cmdline,
 653                        -err);
 654                 return 0;
 655         }
 656         err = os_stat_file(from_cow, &buf2);
 657         if (err < 0) {
 658                 printk(KERN_ERR "Couldn't stat '%s', err = %d\n", from_cow,
 659                        -err);
 660                 return 1;
 661         }
 662         if ((buf1.ust_dev == buf2.ust_dev) && (buf1.ust_ino == buf2.ust_ino))
 663                 return 0;
 664
 665         printk(KERN_ERR "Backing file mismatch - \"%s\" requested, "
 666                "\"%s\" specified in COW header of \"%s\"\n",
 667                from_cmdline, from_cow, cow);
 668         return 1;
 669 }
 670
 671 static int open_ubd_file(char *file, struct openflags *openflags, int shared,
 672                   char **backing_file_out, int *bitmap_offset_out,
 673                   unsigned long *bitmap_len_out, int *data_offset_out,
 674                   int *create_cow_out)
 675 {
 676         time_t mtime;
 677         unsigned long long size;
 678         __u32 version, align;
 679         char *backing_file;
 680         int fd, err, sectorsize, asked_switch, mode = 0644;
 681
 682         fd = os_open_file(file, *openflags, mode);
 683         if (fd < 0) {
 684                 if ((fd == -ENOENT) && (create_cow_out != NULL))
 685                         *create_cow_out = 1;
 686                 if (!openflags->w ||
 687                     ((fd != -EROFS) && (fd != -EACCES)))
 688                         return fd;
 689                 openflags->w = 0;
 690                 fd = os_open_file(file, *openflags, mode);
 691                 if (fd < 0)
 692                         return fd;
 693         }
 694
 695         if (shared)
 696                 printk(KERN_INFO "Not locking \"%s\" on the host\n", file);
 697         else {
 698                 err = os_lock_file(fd, openflags->w);
 699                 if (err < 0) {
 700                         printk(KERN_ERR "Failed to lock '%s', err = %d\n",
 701                                file, -err);
 702                         goto out_close;
 703                 }
 704         }
 705
 706         /* Successful return case! */
 707         if (backing_file_out == NULL)
 708                 return fd;
 709
 710         err = read_cow_header(file_reader, &fd, &version, &backing_file, &mtime,
 711                               &size, &sectorsize, &align, bitmap_offset_out);
 712         if (err && (*backing_file_out != NULL)) {
 713                 printk(KERN_ERR "Failed to read COW header from COW file "
 714                        "\"%s\", errno = %d\n", file, -err);
 715                 goto out_close;
 716         }
 717         if (err)
 718                 return fd;
 719
 720         asked_switch = path_requires_switch(*backing_file_out, backing_file,
 721                                             file);
 722
 723         /* Allow switching only if no mismatch. */
 724         if (asked_switch && !backing_file_mismatch(*backing_file_out, size,
 725                                                    mtime)) {
 726                 printk(KERN_ERR "Switching backing file to '%s'\n",
 727                        *backing_file_out);
 728                 err = write_cow_header(file, fd, *backing_file_out,
 729                                        sectorsize, align, &size);
 730                 if (err) {
 731                         printk(KERN_ERR "Switch failed, errno = %d\n", -err);
 732                         goto out_close;
 733                 }
 734         } else {
 735                 *backing_file_out = backing_file;
 736                 err = backing_file_mismatch(*backing_file_out, size, mtime);
 737                 if (err)
 738                         goto out_close;
 739         }
 740
 741         cow_sizes(version, size, sectorsize, align, *bitmap_offset_out,
 742                   bitmap_len_out, data_offset_out);
 743
 744         return fd;
 745  out_close:
 746         os_close_file(fd);
 747         return err;
 748 }
 749
 750 static int create_cow_file(char *cow_file, char *backing_file,
 751                     struct openflags flags,
 752                     int sectorsize, int alignment, int *bitmap_offset_out,
 753                     unsigned long *bitmap_len_out, int *data_offset_out)
 754 {
 755         int err, fd;
 756
 757         flags.c = 1;
 758         fd = open_ubd_file(cow_file, &flags, 0, NULL, NULL, NULL, NULL, NULL);
 759         if (fd < 0) {
 760                 err = fd;
 761                 printk(KERN_ERR "Open of COW file '%s' failed, errno = %d\n",
 762                        cow_file, -err);
 763                 goto out;
 764         }
 765
 766         err = init_cow_file(fd, cow_file, backing_file, sectorsize, alignment,
 767                             bitmap_offset_out, bitmap_len_out,
 768                             data_offset_out);
 769         if (!err)
 770                 return fd;
 771         os_close_file(fd);
 772  out:
 773         return err;
 774 }
 775
 776 static void ubd_close_dev(struct ubd *ubd_dev)
 777 {
 778         os_close_file(ubd_dev->fd);
 779         if(ubd_dev->cow.file == NULL)
 780                 return;
 781
 782         os_close_file(ubd_dev->cow.fd);
 783         vfree(ubd_dev->cow.bitmap);
 784         ubd_dev->cow.bitmap = NULL;
 785 }
 786
 787 static int ubd_open_dev(struct ubd *ubd_dev)
 788 {
 789         struct openflags flags;
 790         char **back_ptr;
 791         int err, create_cow, *create_ptr;
 792         int fd;
 793
 794         ubd_dev->openflags = ubd_dev->boot_openflags;
 795         create_cow = 0;
 796         create_ptr = (ubd_dev->cow.file != NULL) ? &create_cow : NULL;
 797         back_ptr = ubd_dev->no_cow ? NULL : &ubd_dev->cow.file;
 798
 799         fd = open_ubd_file(ubd_dev->file, &ubd_dev->openflags, ubd_dev->shared,
 800                                 back_ptr, &ubd_dev->cow.bitmap_offset,
 801                                 &ubd_dev->cow.bitmap_len, &ubd_dev->cow.data_offset,
 802                                 create_ptr);
 803
 804         if((fd == -ENOENT) && create_cow){
 805                 fd = create_cow_file(ubd_dev->file, ubd_dev->cow.file,
 806                                           ubd_dev->openflags, SECTOR_SIZE, PAGE_SIZE,
 807                                           &ubd_dev->cow.bitmap_offset,
 808                                           &ubd_dev->cow.bitmap_len,
 809                                           &ubd_dev->cow.data_offset);
 810                 if(fd >= 0){
 811                         printk(KERN_INFO "Creating \"%s\" as COW file for "
 812                                "\"%s\"\n", ubd_dev->file, ubd_dev->cow.file);
 813                 }
 814         }
 815
 816         if(fd < 0){
 817                 printk("Failed to open '%s', errno = %d\n", ubd_dev->file,
 818                        -fd);
 819                 return fd;
 820         }
 821         ubd_dev->fd = fd;
 822
 823         if(ubd_dev->cow.file != NULL){
 824                 blk_queue_max_hw_sectors(ubd_dev->queue, 8 * sizeof(long));
 825
 826                 err = -ENOMEM;
 827                 ubd_dev->cow.bitmap = vmalloc(ubd_dev->cow.bitmap_len);
 828                 if(ubd_dev->cow.bitmap == NULL){
 829                         printk(KERN_ERR "Failed to vmalloc COW bitmap\n");
 830                         goto error;
 831                 }
 832                 flush_tlb_kernel_vm();
 833
 834                 err = read_cow_bitmap(ubd_dev->fd, ubd_dev->cow.bitmap,
 835                                       ubd_dev->cow.bitmap_offset,
 836                                       ubd_dev->cow.bitmap_len);
 837                 if(err < 0)
 838                         goto error;
 839
 840                 flags = ubd_dev->openflags;
 841                 flags.w = 0;
 842                 err = open_ubd_file(ubd_dev->cow.file, &flags, ubd_dev->shared, NULL,
 843                                     NULL, NULL, NULL, NULL);
 844                 if(err < 0) goto error;
 845                 ubd_dev->cow.fd = err;
 846         }
 847         if (ubd_dev->no_trim == 0) {
 848                 ubd_dev->queue->limits.discard_granularity = SECTOR_SIZE;
 849                 ubd_dev->queue->limits.discard_alignment = SECTOR_SIZE;
 850                 blk_queue_max_discard_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
 851                 blk_queue_max_write_zeroes_sectors(ubd_dev->queue, UBD_MAX_REQUEST);
 852                 blk_queue_flag_set(QUEUE_FLAG_DISCARD, ubd_dev->queue);
 853         }
 854         blk_queue_flag_set(QUEUE_FLAG_NONROT, ubd_dev->queue);
 855         return 0;
 856  error:
 857         os_close_file(ubd_dev->fd);
 858         return err;
 859 }
 860
 861 static void ubd_device_release(struct device *dev)
 862 {
 863         struct ubd *ubd_dev = dev_get_drvdata(dev);
 864
 865         blk_cleanup_queue(ubd_dev->queue);
 866         blk_mq_free_tag_set(&ubd_dev->tag_set);
 867         *ubd_dev = ((struct ubd) DEFAULT_UBD);
 868 }
 869
 870 static int ubd_disk_register(int major, u64 size, int unit,
 871                              struct gendisk **disk_out)
 872 {
 873         struct device *parent = NULL;
 874         struct gendisk *disk;
 875
 876         disk = alloc_disk(1 << UBD_SHIFT);
 877         if(disk == NULL)
 878                 return -ENOMEM;
 879
 880         disk->major = major;
 881         disk->first_minor = unit << UBD_SHIFT;
 882         disk->fops = &ubd_blops;
 883         set_capacity(disk, size / 512);
 884         if (major == UBD_MAJOR)
 885                 sprintf(disk->disk_name, "ubd%c", 'a' + unit);
 886         else
 887                 sprintf(disk->disk_name, "ubd_fake%d", unit);
 888
 889         /* sysfs register (not for ide fake devices) */
 890         if (major == UBD_MAJOR) {
 891                 ubd_devs[unit].pdev.id   = unit;
 892                 ubd_devs[unit].pdev.name = DRIVER_NAME;
 893                 ubd_devs[unit].pdev.dev.release = ubd_device_release;
 894                 dev_set_drvdata(&ubd_devs[unit].pdev.dev, &ubd_devs[unit]);
 895                 platform_device_register(&ubd_devs[unit].pdev);
 896                 parent = &ubd_devs[unit].pdev.dev;
 897         }
 898
 899         disk->private_data = &ubd_devs[unit];
 900         disk->queue = ubd_devs[unit].queue;
 901         device_add_disk(parent, disk, NULL);
 902
 903         *disk_out = disk;
 904         return 0;
 905 }
 906
 907 #define ROUND_BLOCK(n) ((n + (SECTOR_SIZE - 1)) & (-SECTOR_SIZE))
 908
 909 static const struct blk_mq_ops ubd_mq_ops = {
 910         .queue_rq = ubd_queue_rq,
 911 };
 912
 913 static int ubd_add(int n, char **error_out)
 914 {
 915         struct ubd *ubd_dev = &ubd_devs[n];
 916         int err = 0;
 917
 918         if(ubd_dev->file == NULL)
 919                 goto out;
 920
 921         err = ubd_file_size(ubd_dev, &ubd_dev->size);
 922         if(err < 0){
 923                 *error_out = "Couldn't determine size of device's file";
 924                 goto out;
 925         }
 926
 927         ubd_dev->size = ROUND_BLOCK(ubd_dev->size);
 928
 929         ubd_dev->tag_set.ops = &ubd_mq_ops;
 930         ubd_dev->tag_set.queue_depth = 64;
 931         ubd_dev->tag_set.numa_node = NUMA_NO_NODE;
 932         ubd_dev->tag_set.flags = BLK_MQ_F_SHOULD_MERGE;
 933         ubd_dev->tag_set.driver_data = ubd_dev;
 934         ubd_dev->tag_set.nr_hw_queues = 1;
 935
 936         err = blk_mq_alloc_tag_set(&ubd_dev->tag_set);
 937         if (err)
 938                 goto out;
 939
 940         ubd_dev->queue = blk_mq_init_queue(&ubd_dev->tag_set);
 941         if (IS_ERR(ubd_dev->queue)) {
 942                 err = PTR_ERR(ubd_dev->queue);
 943                 goto out_cleanup_tags;
 944         }
 945
 946         ubd_dev->queue->queuedata = ubd_dev;
 947         blk_queue_write_cache(ubd_dev->queue, true, false);
 948
 949         blk_queue_max_segments(ubd_dev->queue, MAX_SG);
 950         blk_queue_segment_boundary(ubd_dev->queue, PAGE_SIZE - 1);
 951         err = ubd_disk_register(UBD_MAJOR, ubd_dev->size, n, &ubd_gendisk[n]);
 952         if(err){
 953                 *error_out = "Failed to register device";
 954                 goto out_cleanup_tags;
 955         }
 956
 957         if (fake_major != UBD_MAJOR)
 958                 ubd_disk_register(fake_major, ubd_dev->size, n,
 959                                   &fake_gendisk[n]);
 960
 961         /*
 962          * Perhaps this should also be under the "if (fake_major)" above
 963          * using the fake_disk->disk_name
 964          */
 965         if (fake_ide)
 966                 make_ide_entries(ubd_gendisk[n]->disk_name);
 967
 968         err = 0;
 969 out:
 970         return err;
 971
 972 out_cleanup_tags:
 973         blk_mq_free_tag_set(&ubd_dev->tag_set);
 974         if (!(IS_ERR(ubd_dev->queue)))
 975                 blk_cleanup_queue(ubd_dev->queue);
 976         goto out;
 977 }
 978
 979 static int ubd_config(char *str, char **error_out)
 980 {
 981         int n, ret;
 982
 983         /* This string is possibly broken up and stored, so it's only
 984          * freed if ubd_setup_common fails, or if only general options
 985          * were set.
 986          */
 987         str = kstrdup(str, GFP_KERNEL);
 988         if (str == NULL) {
 989                 *error_out = "Failed to allocate memory";
 990                 return -ENOMEM;
 991         }
 992
 993         ret = ubd_setup_common(str, &n, error_out);
 994         if (ret)
 995                 goto err_free;
 996
 997         if (n == -1) {
 998                 ret = 0;
 999                 goto err_free;
1000         }
1001
1002         mutex_lock(&ubd_lock);
1003         ret = ubd_add(n, error_out);
1004         if (ret)
1005                 ubd_devs[n].file = NULL;
1006         mutex_unlock(&ubd_lock);
1007
1008 out:
1009         return ret;
1010
1011 err_free:
1012         kfree(str);
1013         goto out;
1014 }
1015
1016 static int ubd_get_config(char *name, char *str, int size, char **error_out)
1017 {
1018         struct ubd *ubd_dev;
1019         int n, len = 0;
1020
1021         n = parse_unit(&name);
1022         if((n >= MAX_DEV) || (n < 0)){
1023                 *error_out = "ubd_get_config : device number out of range";
1024                 return -1;
1025         }
1026
1027         ubd_dev = &ubd_devs[n];
1028         mutex_lock(&ubd_lock);
1029
1030         if(ubd_dev->file == NULL){
1031                 CONFIG_CHUNK(str, size, len, "", 1);
1032                 goto out;
1033         }
1034
1035         CONFIG_CHUNK(str, size, len, ubd_dev->file, 0);
1036
1037         if(ubd_dev->cow.file != NULL){
1038                 CONFIG_CHUNK(str, size, len, ",", 0);
1039                 CONFIG_CHUNK(str, size, len, ubd_dev->cow.file, 1);
1040         }
1041         else CONFIG_CHUNK(str, size, len, "", 1);
1042
1043  out:
1044         mutex_unlock(&ubd_lock);
1045         return len;
1046 }
1047
1048 static int ubd_id(char **str, int *start_out, int *end_out)
1049 {
1050         int n;
1051
1052         n = parse_unit(str);
1053         *start_out = 0;
1054         *end_out = MAX_DEV - 1;
1055         return n;
1056 }
1057
1058 static int ubd_remove(int n, char **error_out)
1059 {
1060         struct gendisk *disk = ubd_gendisk[n];
1061         struct ubd *ubd_dev;
1062         int err = -ENODEV;
1063
1064         mutex_lock(&ubd_lock);
1065
1066         ubd_dev = &ubd_devs[n];
1067
1068         if(ubd_dev->file == NULL)
1069                 goto out;
1070
1071         /* you cannot remove a open disk */
1072         err = -EBUSY;
1073         if(ubd_dev->count > 0)
1074                 goto out;
1075
1076         ubd_gendisk[n] = NULL;
1077         if(disk != NULL){
1078                 del_gendisk(disk);
1079                 put_disk(disk);
1080         }
1081
1082         if(fake_gendisk[n] != NULL){
1083                 del_gendisk(fake_gendisk[n]);
1084                 put_disk(fake_gendisk[n]);
1085                 fake_gendisk[n] = NULL;
1086         }
1087
1088         err = 0;
1089         platform_device_unregister(&ubd_dev->pdev);
1090 out:
1091         mutex_unlock(&ubd_lock);
1092         return err;
1093 }
1094
1095 /* All these are called by mconsole in process context and without
1096  * ubd-specific locks.  The structure itself is const except for .list.
1097  */
1098 static struct mc_device ubd_mc = {
1099         .list           = LIST_HEAD_INIT(ubd_mc.list),
1100         .name           = "ubd",
1101         .config         = ubd_config,
1102         .get_config     = ubd_get_config,
1103         .id             = ubd_id,
1104         .remove         = ubd_remove,
1105 };
1106
1107 static int __init ubd_mc_init(void)
1108 {
1109         mconsole_register_dev(&ubd_mc);
1110         return 0;
1111 }
1112
1113 __initcall(ubd_mc_init);
1114
1115 static int __init ubd0_init(void)
1116 {
1117         struct ubd *ubd_dev = &ubd_devs[0];
1118
1119         mutex_lock(&ubd_lock);
1120         if(ubd_dev->file == NULL)
1121                 ubd_dev->file = "root_fs";
1122         mutex_unlock(&ubd_lock);
1123
1124         return 0;
1125 }
1126
1127 __initcall(ubd0_init);
1128
1129 /* Used in ubd_init, which is an initcall */
1130 static struct platform_driver ubd_driver = {
1131         .driver = {
1132                 .name  = DRIVER_NAME,
1133         },
1134 };
1135
1136 static int __init ubd_init(void)
1137 {
1138         char *error;
1139         int i, err;
1140
1141         if (register_blkdev(UBD_MAJOR, "ubd"))
1142                 return -1;
1143
1144         if (fake_major != UBD_MAJOR) {
1145                 char name[sizeof("ubd_nnn\0")];
1146
1147                 snprintf(name, sizeof(name), "ubd_%d", fake_major);
1148                 if (register_blkdev(fake_major, "ubd"))
1149                         return -1;
1150         }
1151
1152         irq_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1153                                        sizeof(struct io_thread_req *),
1154                                        GFP_KERNEL
1155                 );
1156         irq_remainder = 0;
1157
1158         if (irq_req_buffer == NULL) {
1159                 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1160                 return -1;
1161         }
1162         io_req_buffer = kmalloc_array(UBD_REQ_BUFFER_SIZE,
1163                                       sizeof(struct io_thread_req *),
1164                                       GFP_KERNEL
1165                 );
1166
1167         io_remainder = 0;
1168
1169         if (io_req_buffer == NULL) {
1170                 printk(KERN_ERR "Failed to initialize ubd buffering\n");
1171                 return -1;
1172         }
1173         platform_driver_register(&ubd_driver);
1174         mutex_lock(&ubd_lock);
1175         for (i = 0; i < MAX_DEV; i++){
1176                 err = ubd_add(i, &error);
1177                 if(err)
1178                         printk(KERN_ERR "Failed to initialize ubd device %d :"
1179                                "%s\n", i, error);
1180         }
1181         mutex_unlock(&ubd_lock);
1182         return 0;
1183 }
1184
1185 late_initcall(ubd_init);
1186
1187 static int __init ubd_driver_init(void){
1188         unsigned long stack;
1189         int err;
1190
1191         /* Set by CONFIG_BLK_DEV_UBD_SYNC or ubd=sync.*/
1192         if(global_openflags.s){
1193                 printk(KERN_INFO "ubd: Synchronous mode\n");
1194                 /* Letting ubd=sync be like using ubd#s= instead of ubd#= is
1195                  * enough. So use anyway the io thread. */
1196         }
1197         stack = alloc_stack(0, 0);
1198         io_pid = start_io_thread(stack + PAGE_SIZE - sizeof(void *),
1199                                  &thread_fd);
1200         if(io_pid < 0){
1201                 printk(KERN_ERR
1202                        "ubd : Failed to start I/O thread (errno = %d) - "
1203                        "falling back to synchronous I/O\n", -io_pid);
1204                 io_pid = -1;
1205                 return 0;
1206         }
1207         err = um_request_irq(UBD_IRQ, thread_fd, IRQ_READ, ubd_intr,
1208                              0, "ubd", ubd_devs);
1209         if(err != 0)
1210                 printk(KERN_ERR "um_request_irq failed - errno = %d\n", -err);
1211         return 0;
1212 }
1213
1214 device_initcall(ubd_driver_init);
1215
1216 static int ubd_open(struct block_device *bdev, fmode_t mode)
1217 {
1218         struct gendisk *disk = bdev->bd_disk;
1219         struct ubd *ubd_dev = disk->private_data;
1220         int err = 0;
1221
1222         mutex_lock(&ubd_mutex);
1223         if(ubd_dev->count == 0){
1224                 err = ubd_open_dev(ubd_dev);
1225                 if(err){
1226                         printk(KERN_ERR "%s: Can't open \"%s\": errno = %d\n",
1227                                disk->disk_name, ubd_dev->file, -err);
1228                         goto out;
1229                 }
1230         }
1231         ubd_dev->count++;
1232         set_disk_ro(disk, !ubd_dev->openflags.w);
1233
1234         /* This should no more be needed. And it didn't work anyway to exclude
1235          * read-write remounting of filesystems.*/
1236         /*if((mode & FMODE_WRITE) && !ubd_dev->openflags.w){
1237                 if(--ubd_dev->count == 0) ubd_close_dev(ubd_dev);
1238                 err = -EROFS;
1239         }*/
1240 out:
1241         mutex_unlock(&ubd_mutex);
1242         return err;
1243 }
1244
1245 static void ubd_release(struct gendisk *disk, fmode_t mode)
1246 {
1247         struct ubd *ubd_dev = disk->private_data;
1248
1249         mutex_lock(&ubd_mutex);
1250         if(--ubd_dev->count == 0)
1251                 ubd_close_dev(ubd_dev);
1252         mutex_unlock(&ubd_mutex);
1253 }
1254
1255 static void cowify_bitmap(__u64 io_offset, int length, unsigned long *cow_mask,
1256                           __u64 *cow_offset, unsigned long *bitmap,
1257                           __u64 bitmap_offset, unsigned long *bitmap_words,
1258                           __u64 bitmap_len)
1259 {
1260         __u64 sector = io_offset >> SECTOR_SHIFT;
1261         int i, update_bitmap = 0;
1262
1263         for (i = 0; i < length >> SECTOR_SHIFT; i++) {
1264                 if(cow_mask != NULL)
1265                         ubd_set_bit(i, (unsigned char *) cow_mask);
1266                 if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1267                         continue;
1268
1269                 update_bitmap = 1;
1270                 ubd_set_bit(sector + i, (unsigned char *) bitmap);
1271         }
1272
1273         if(!update_bitmap)
1274                 return;
1275
1276         *cow_offset = sector / (sizeof(unsigned long) * 8);
1277
1278         /* This takes care of the case where we're exactly at the end of the
1279          * device, and *cow_offset + 1 is off the end.  So, just back it up
1280          * by one word.  Thanks to Lynn Kerby for the fix and James McMechan
1281          * for the original diagnosis.
1282          */
1283         if (*cow_offset == (DIV_ROUND_UP(bitmap_len,
1284                                          sizeof(unsigned long)) - 1))
1285                 (*cow_offset)--;
1286
1287         bitmap_words[0] = bitmap[*cow_offset];
1288         bitmap_words[1] = bitmap[*cow_offset + 1];
1289
1290         *cow_offset *= sizeof(unsigned long);
1291         *cow_offset += bitmap_offset;
1292 }
1293
1294 static void cowify_req(struct io_thread_req *req, struct io_desc *segment,
1295                        unsigned long offset, unsigned long *bitmap,
1296                        __u64 bitmap_offset, __u64 bitmap_len)
1297 {
1298         __u64 sector = offset >> SECTOR_SHIFT;
1299         int i;
1300
1301         if (segment->length > (sizeof(segment->sector_mask) * 8) << SECTOR_SHIFT)
1302                 panic("Operation too long");
1303
1304         if (req_op(req->req) == REQ_OP_READ) {
1305                 for (i = 0; i < segment->length >> SECTOR_SHIFT; i++) {
1306                         if(ubd_test_bit(sector + i, (unsigned char *) bitmap))
1307                                 ubd_set_bit(i, (unsigned char *)
1308                                             &segment->sector_mask);
1309                 }
1310         } else {
1311                 cowify_bitmap(offset, segment->length, &segment->sector_mask,
1312                               &segment->cow_offset, bitmap, bitmap_offset,
1313                               segment->bitmap_words, bitmap_len);
1314         }
1315 }
1316
1317 static void ubd_map_req(struct ubd *dev, struct io_thread_req *io_req,
1318                         struct request *req)
1319 {
1320         struct bio_vec bvec;
1321         struct req_iterator iter;
1322         int i = 0;
1323         unsigned long byte_offset = io_req->offset;
1324         int op = req_op(req);
1325
1326         if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD) {
1327                 io_req->io_desc[0].buffer = NULL;
1328                 io_req->io_desc[0].length = blk_rq_bytes(req);
1329         } else {
1330                 rq_for_each_segment(bvec, req, iter) {
1331                         BUG_ON(i >= io_req->desc_cnt);
1332
1333                         io_req->io_desc[i].buffer =
1334                                 page_address(bvec.bv_page) + bvec.bv_offset;
1335                         io_req->io_desc[i].length = bvec.bv_len;
1336                         i++;
1337                 }
1338         }
1339
1340         if (dev->cow.file) {
1341                 for (i = 0; i < io_req->desc_cnt; i++) {
1342                         cowify_req(io_req, &io_req->io_desc[i], byte_offset,
1343                                    dev->cow.bitmap, dev->cow.bitmap_offset,
1344                                    dev->cow.bitmap_len);
1345                         byte_offset += io_req->io_desc[i].length;
1346                 }
1347
1348         }
1349 }
1350
1351 static struct io_thread_req *ubd_alloc_req(struct ubd *dev, struct request *req,
1352                                            int desc_cnt)
1353 {
1354         struct io_thread_req *io_req;
1355         int i;
1356
1357         io_req = kmalloc(sizeof(*io_req) +
1358                          (desc_cnt * sizeof(struct io_desc)),
1359                          GFP_ATOMIC);
1360         if (!io_req)
1361                 return NULL;
1362
1363         io_req->req = req;
1364         if (dev->cow.file)
1365                 io_req->fds[0] = dev->cow.fd;
1366         else
1367                 io_req->fds[0] = dev->fd;
1368         io_req->error = 0;
1369         io_req->sectorsize = SECTOR_SIZE;
1370         io_req->fds[1] = dev->fd;
1371         io_req->offset = (u64) blk_rq_pos(req) << SECTOR_SHIFT;
1372         io_req->offsets[0] = 0;
1373         io_req->offsets[1] = dev->cow.data_offset;
1374
1375         for (i = 0 ; i < desc_cnt; i++) {
1376                 io_req->io_desc[i].sector_mask = 0;
1377                 io_req->io_desc[i].cow_offset = -1;
1378         }
1379
1380         return io_req;
1381 }
1382
1383 static int ubd_submit_request(struct ubd *dev, struct request *req)
1384 {
1385         int segs = 0;
1386         struct io_thread_req *io_req;
1387         int ret;
1388         int op = req_op(req);
1389
1390         if (op == REQ_OP_FLUSH)
1391                 segs = 0;
1392         else if (op == REQ_OP_WRITE_ZEROES || op == REQ_OP_DISCARD)
1393                 segs = 1;
1394         else
1395                 segs = blk_rq_nr_phys_segments(req);
1396
1397         io_req = ubd_alloc_req(dev, req, segs);
1398         if (!io_req)
1399                 return -ENOMEM;
1400
1401         io_req->desc_cnt = segs;
1402         if (segs)
1403                 ubd_map_req(dev, io_req, req);
1404
1405         ret = os_write_file(thread_fd, &io_req, sizeof(io_req));
1406         if (ret != sizeof(io_req)) {
1407                 if (ret != -EAGAIN)
1408                         pr_err("write to io thread failed: %d\n", -ret);
1409                 kfree(io_req);
1410         }
1411         return ret;
1412 }
1413
1414 static blk_status_t ubd_queue_rq(struct blk_mq_hw_ctx *hctx,
1415                                  const struct blk_mq_queue_data *bd)
1416 {
1417         struct ubd *ubd_dev = hctx->queue->queuedata;
1418         struct request *req = bd->rq;
1419         int ret = 0, res = BLK_STS_OK;
1420
1421         blk_mq_start_request(req);
1422
1423         spin_lock_irq(&ubd_dev->lock);
1424
1425         switch (req_op(req)) {
1426         case REQ_OP_FLUSH:
1427         case REQ_OP_READ:
1428         case REQ_OP_WRITE:
1429         case REQ_OP_DISCARD:
1430         case REQ_OP_WRITE_ZEROES:
1431                 ret = ubd_submit_request(ubd_dev, req);
1432                 break;
1433         default:
1434                 WARN_ON_ONCE(1);
1435                 res = BLK_STS_NOTSUPP;
1436         }
1437
1438         spin_unlock_irq(&ubd_dev->lock);
1439
1440         if (ret < 0) {
1441                 if (ret == -ENOMEM)
1442                         res = BLK_STS_RESOURCE;
1443                 else
1444                         res = BLK_STS_DEV_RESOURCE;
1445         }
1446
1447         return res;
1448 }
1449
1450 static int ubd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1451 {
1452         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1453
1454         geo->heads = 128;
1455         geo->sectors = 32;
1456         geo->cylinders = ubd_dev->size / (128 * 32 * 512);
1457         return 0;
1458 }
1459
1460 static int ubd_ioctl(struct block_device *bdev, fmode_t mode,
1461                      unsigned int cmd, unsigned long arg)
1462 {
1463         struct ubd *ubd_dev = bdev->bd_disk->private_data;
1464         u16 ubd_id[ATA_ID_WORDS];
1465
1466         switch (cmd) {
1467                 struct cdrom_volctrl volume;
1468         case HDIO_GET_IDENTITY:
1469                 memset(&ubd_id, 0, ATA_ID_WORDS * 2);
1470                 ubd_id[ATA_ID_CYLS]     = ubd_dev->size / (128 * 32 * 512);
1471                 ubd_id[ATA_ID_HEADS]    = 128;
1472                 ubd_id[ATA_ID_SECTORS]  = 32;
1473                 if(copy_to_user((char __user *) arg, (char *) &ubd_id,
1474                                  sizeof(ubd_id)))
1475                         return -EFAULT;
1476                 return 0;
1477
1478         case CDROMVOLREAD:
1479                 if(copy_from_user(&volume, (char __user *) arg, sizeof(volume)))
1480                         return -EFAULT;
1481                 volume.channel0 = 255;
1482                 volume.channel1 = 255;
1483                 volume.channel2 = 255;
1484                 volume.channel3 = 255;
1485                 if(copy_to_user((char __user *) arg, &volume, sizeof(volume)))
1486                         return -EFAULT;
1487                 return 0;
1488         }
1489         return -EINVAL;
1490 }
1491
1492 static int map_error(int error_code)
1493 {
1494         switch (error_code) {
1495         case 0:
1496                 return BLK_STS_OK;
1497         case ENOSYS:
1498         case EOPNOTSUPP:
1499                 return BLK_STS_NOTSUPP;
1500         case ENOSPC:
1501                 return BLK_STS_NOSPC;
1502         }
1503         return BLK_STS_IOERR;
1504 }
1505
1506 /*
1507  * Everything from here onwards *IS NOT PART OF THE KERNEL*
1508  *
1509  * The following functions are part of UML hypervisor code.
1510  * All functions from here onwards are executed as a helper
1511  * thread and are not allowed to execute any kernel functions.
1512  *
1513  * Any communication must occur strictly via shared memory and IPC.
1514  *
1515  * Do not add printks, locks, kernel memory operations, etc - it
1516  * will result in unpredictable behaviour and/or crashes.
1517  */
1518
1519 static int update_bitmap(struct io_thread_req *req, struct io_desc *segment)
1520 {
1521         int n;
1522
1523         if (segment->cow_offset == -1)
1524                 return map_error(0);
1525
1526         n = os_pwrite_file(req->fds[1], &segment->bitmap_words,
1527                           sizeof(segment->bitmap_words), segment->cow_offset);
1528         if (n != sizeof(segment->bitmap_words))
1529                 return map_error(-n);
1530
1531         return map_error(0);
1532 }
1533
1534 static void do_io(struct io_thread_req *req, struct io_desc *desc)
1535 {
1536         char *buf = NULL;
1537         unsigned long len;
1538         int n, nsectors, start, end, bit;
1539         __u64 off;
1540
1541         /* FLUSH is really a special case, we cannot "case" it with others */
1542
1543         if (req_op(req->req) == REQ_OP_FLUSH) {
1544                 /* fds[0] is always either the rw image or our cow file */
1545                 req->error = map_error(-os_sync_file(req->fds[0]));
1546                 return;
1547         }
1548
1549         nsectors = desc->length / req->sectorsize;
1550         start = 0;
1551         do {
1552                 bit = ubd_test_bit(start, (unsigned char *) &desc->sector_mask);
1553                 end = start;
1554                 while((end < nsectors) &&
1555                       (ubd_test_bit(end, (unsigned char *) &desc->sector_mask) == bit))
1556                         end++;
1557
1558                 off = req->offset + req->offsets[bit] +
1559                         start * req->sectorsize;
1560                 len = (end - start) * req->sectorsize;
1561                 if (desc->buffer != NULL)
1562                         buf = &desc->buffer[start * req->sectorsize];
1563
1564                 switch (req_op(req->req)) {
1565                 case REQ_OP_READ:
1566                         n = 0;
1567                         do {
1568                                 buf = &buf[n];
1569                                 len -= n;
1570                                 n = os_pread_file(req->fds[bit], buf, len, off);
1571                                 if (n < 0) {
1572                                         req->error = map_error(-n);
1573                                         return;
1574                                 }
1575                         } while((n < len) && (n != 0));
1576                         if (n < len) memset(&buf[n], 0, len - n);
1577                         break;
1578                 case REQ_OP_WRITE:
1579                         n = os_pwrite_file(req->fds[bit], buf, len, off);
1580                         if(n != len){
1581                                 req->error = map_error(-n);
1582                                 return;
1583                         }
1584                         break;
1585                 case REQ_OP_DISCARD:
1586                 case REQ_OP_WRITE_ZEROES:
1587                         n = os_falloc_punch(req->fds[bit], off, len);
1588                         if (n) {
1589                                 req->error = map_error(-n);
1590                                 return;
1591                         }
1592                         break;
1593                 default:
1594                         WARN_ON_ONCE(1);
1595                         req->error = BLK_STS_NOTSUPP;
1596                         return;
1597                 }
1598
1599                 start = end;
1600         } while(start < nsectors);
1601
1602         req->offset += len;
1603         req->error = update_bitmap(req, desc);
1604 }
1605
1606 /* Changed in start_io_thread, which is serialized by being called only
1607  * from ubd_init, which is an initcall.
1608  */
1609 int kernel_fd = -1;
1610
1611 /* Only changed by the io thread. XXX: currently unused. */
1612 static int io_count = 0;
1613
1614 int io_thread(void *arg)
1615 {
1616         int n, count, written, res;
1617
1618         os_fix_helper_signals();
1619
1620         while(1){
1621                 n = bulk_req_safe_read(
1622                         kernel_fd,
1623                         io_req_buffer,
1624                         &io_remainder,
1625                         &io_remainder_size,
1626                         UBD_REQ_BUFFER_SIZE
1627                 );
1628                 if (n < 0) {
1629                         if (n == -EAGAIN) {
1630                                 ubd_read_poll(-1);
1631                                 continue;
1632                         }
1633                 }
1634
1635                 for (count = 0; count < n/sizeof(struct io_thread_req *); count++) {
1636                         struct io_thread_req *req = (*io_req_buffer)[count];
1637                         int i;
1638
1639                         io_count++;
1640                         for (i = 0; !req->error && i < req->desc_cnt; i++)
1641                                 do_io(req, &(req->io_desc[i]));
1642
1643                 }
1644
1645                 written = 0;
1646
1647                 do {
1648                         res = os_write_file(kernel_fd,
1649                                             ((char *) io_req_buffer) + written,
1650                                             n - written);
1651                         if (res >= 0) {
1652                                 written += res;
1653                         }
1654                         if (written < n) {
1655                                 ubd_write_poll(-1);
1656                         }
1657                 } while (written < n);
1658         }
1659
1660         return 0;
1661 }