block/blkdebug.c

   1 /*
   2  * Block protocol for I/O error injection
   3  *
   4  * Copyright (C) 2016-2017 Red Hat, Inc.
   5  * Copyright (c) 2010 Kevin Wolf <kwolf@redhat.com>
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a copy
   8  * of this software and associated documentation files (the "Software"), to deal
   9  * in the Software without restriction, including without limitation the rights
  10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  11  * copies of the Software, and to permit persons to whom the Software is
  12  * furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice shall be included in
  15  * all copies or substantial portions of the Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  23  * THE SOFTWARE.
  24  */
  25
  26 #include "qemu/osdep.h"
  27 #include "qapi/error.h"
  28 #include "qemu/cutils.h"
  29 #include "qemu/config-file.h"
  30 #include "block/block_int.h"
  31 #include "block/qdict.h"
  32 #include "qemu/module.h"
  33 #include "qemu/option.h"
  34 #include "qapi/qapi-visit-block-core.h"
  35 #include "qapi/qmp/qdict.h"
  36 #include "qapi/qmp/qlist.h"
  37 #include "qapi/qmp/qstring.h"
  38 #include "qapi/qobject-input-visitor.h"
  39 #include "sysemu/qtest.h"
  40
  41 typedef struct BDRVBlkdebugState {
  42     int state;
  43     int new_state;
  44     uint64_t align;
  45     uint64_t max_transfer;
  46     uint64_t opt_write_zero;
  47     uint64_t max_write_zero;
  48     uint64_t opt_discard;
  49     uint64_t max_discard;
  50
  51     uint64_t take_child_perms;
  52     uint64_t unshare_child_perms;
  53
  54     /* For blkdebug_refresh_filename() */
  55     char *config_file;
  56
  57     QLIST_HEAD(, BlkdebugRule) rules[BLKDBG__MAX];
  58     QSIMPLEQ_HEAD(, BlkdebugRule) active_rules;
  59     QLIST_HEAD(, BlkdebugSuspendedReq) suspended_reqs;
  60 } BDRVBlkdebugState;
  61
  62 typedef struct BlkdebugAIOCB {
  63     BlockAIOCB common;
  64     int ret;
  65 } BlkdebugAIOCB;
  66
  67 typedef struct BlkdebugSuspendedReq {
  68     Coroutine *co;
  69     char *tag;
  70     QLIST_ENTRY(BlkdebugSuspendedReq) next;
  71 } BlkdebugSuspendedReq;
  72
  73 enum {
  74     ACTION_INJECT_ERROR,
  75     ACTION_SET_STATE,
  76     ACTION_SUSPEND,
  77 };
  78
  79 typedef struct BlkdebugRule {
  80     BlkdebugEvent event;
  81     int action;
  82     int state;
  83     union {
  84         struct {
  85             uint64_t iotype_mask;
  86             int error;
  87             int immediately;
  88             int once;
  89             int64_t offset;
  90         } inject;
  91         struct {
  92             int new_state;
  93         } set_state;
  94         struct {
  95             char *tag;
  96         } suspend;
  97     } options;
  98     QLIST_ENTRY(BlkdebugRule) next;
  99     QSIMPLEQ_ENTRY(BlkdebugRule) active_next;
 100 } BlkdebugRule;
 101
 102 QEMU_BUILD_BUG_MSG(BLKDEBUG_IO_TYPE__MAX > 64,
 103                    "BlkdebugIOType mask does not fit into an uint64_t");
 104
 105 static QemuOptsList inject_error_opts = {
 106     .name = "inject-error",
 107     .head = QTAILQ_HEAD_INITIALIZER(inject_error_opts.head),
 108     .desc = {
 109         {
 110             .name = "event",
 111             .type = QEMU_OPT_STRING,
 112         },
 113         {
 114             .name = "state",
 115             .type = QEMU_OPT_NUMBER,
 116         },
 117         {
 118             .name = "iotype",
 119             .type = QEMU_OPT_STRING,
 120         },
 121         {
 122             .name = "errno",
 123             .type = QEMU_OPT_NUMBER,
 124         },
 125         {
 126             .name = "sector",
 127             .type = QEMU_OPT_NUMBER,
 128         },
 129         {
 130             .name = "once",
 131             .type = QEMU_OPT_BOOL,
 132         },
 133         {
 134             .name = "immediately",
 135             .type = QEMU_OPT_BOOL,
 136         },
 137         { /* end of list */ }
 138     },
 139 };
 140
 141 static QemuOptsList set_state_opts = {
 142     .name = "set-state",
 143     .head = QTAILQ_HEAD_INITIALIZER(set_state_opts.head),
 144     .desc = {
 145         {
 146             .name = "event",
 147             .type = QEMU_OPT_STRING,
 148         },
 149         {
 150             .name = "state",
 151             .type = QEMU_OPT_NUMBER,
 152         },
 153         {
 154             .name = "new_state",
 155             .type = QEMU_OPT_NUMBER,
 156         },
 157         { /* end of list */ }
 158     },
 159 };
 160
 161 static QemuOptsList *config_groups[] = {
 162     &inject_error_opts,
 163     &set_state_opts,
 164     NULL
 165 };
 166
 167 struct add_rule_data {
 168     BDRVBlkdebugState *s;
 169     int action;
 170 };
 171
 172 static int add_rule(void *opaque, QemuOpts *opts, Error **errp)
 173 {
 174     struct add_rule_data *d = opaque;
 175     BDRVBlkdebugState *s = d->s;
 176     const char* event_name;
 177     int event;
 178     struct BlkdebugRule *rule;
 179     int64_t sector;
 180     BlkdebugIOType iotype;
 181     Error *local_error = NULL;
 182
 183     /* Find the right event for the rule */
 184     event_name = qemu_opt_get(opts, "event");
 185     if (!event_name) {
 186         error_setg(errp, "Missing event name for rule");
 187         return -1;
 188     }
 189     event = qapi_enum_parse(&BlkdebugEvent_lookup, event_name, -1, errp);
 190     if (event < 0) {
 191         return -1;
 192     }
 193
 194     /* Set attributes common for all actions */
 195     rule = g_malloc0(sizeof(*rule));
 196     *rule = (struct BlkdebugRule) {
 197         .event  = event,
 198         .action = d->action,
 199         .state  = qemu_opt_get_number(opts, "state", 0),
 200     };
 201
 202     /* Parse action-specific options */
 203     switch (d->action) {
 204     case ACTION_INJECT_ERROR:
 205         rule->options.inject.error = qemu_opt_get_number(opts, "errno", EIO);
 206         rule->options.inject.once  = qemu_opt_get_bool(opts, "once", 0);
 207         rule->options.inject.immediately =
 208             qemu_opt_get_bool(opts, "immediately", 0);
 209         sector = qemu_opt_get_number(opts, "sector", -1);
 210         rule->options.inject.offset =
 211             sector == -1 ? -1 : sector * BDRV_SECTOR_SIZE;
 212
 213         iotype = qapi_enum_parse(&BlkdebugIOType_lookup,
 214                                  qemu_opt_get(opts, "iotype"),
 215                                  BLKDEBUG_IO_TYPE__MAX, &local_error);
 216         if (local_error) {
 217             error_propagate(errp, local_error);
 218             g_free(rule);
 219             return -1;
 220         }
 221         if (iotype != BLKDEBUG_IO_TYPE__MAX) {
 222             rule->options.inject.iotype_mask = (1ull << iotype);
 223         } else {
 224             /* Apply the default */
 225             rule->options.inject.iotype_mask =
 226                 (1ull << BLKDEBUG_IO_TYPE_READ)
 227                 | (1ull << BLKDEBUG_IO_TYPE_WRITE)
 228                 | (1ull << BLKDEBUG_IO_TYPE_WRITE_ZEROES)
 229                 | (1ull << BLKDEBUG_IO_TYPE_DISCARD)
 230                 | (1ull << BLKDEBUG_IO_TYPE_FLUSH);
 231         }
 232
 233         break;
 234
 235     case ACTION_SET_STATE:
 236         rule->options.set_state.new_state =
 237             qemu_opt_get_number(opts, "new_state", 0);
 238         break;
 239
 240     case ACTION_SUSPEND:
 241         rule->options.suspend.tag =
 242             g_strdup(qemu_opt_get(opts, "tag"));
 243         break;
 244     };
 245
 246     /* Add the rule */
 247     QLIST_INSERT_HEAD(&s->rules[event], rule, next);
 248
 249     return 0;
 250 }
 251
 252 static void remove_rule(BlkdebugRule *rule)
 253 {
 254     switch (rule->action) {
 255     case ACTION_INJECT_ERROR:
 256     case ACTION_SET_STATE:
 257         break;
 258     case ACTION_SUSPEND:
 259         g_free(rule->options.suspend.tag);
 260         break;
 261     }
 262
 263     QLIST_REMOVE(rule, next);
 264     g_free(rule);
 265 }
 266
 267 static int read_config(BDRVBlkdebugState *s, const char *filename,
 268                        QDict *options, Error **errp)
 269 {
 270     FILE *f = NULL;
 271     int ret;
 272     struct add_rule_data d;
 273     Error *local_err = NULL;
 274
 275     if (filename) {
 276         f = fopen(filename, "r");
 277         if (f == NULL) {
 278             error_setg_errno(errp, errno, "Could not read blkdebug config file");
 279             return -errno;
 280         }
 281
 282         ret = qemu_config_parse(f, config_groups, filename);
 283         if (ret < 0) {
 284             error_setg(errp, "Could not parse blkdebug config file");
 285             goto fail;
 286         }
 287     }
 288
 289     qemu_config_parse_qdict(options, config_groups, &local_err);
 290     if (local_err) {
 291         error_propagate(errp, local_err);
 292         ret = -EINVAL;
 293         goto fail;
 294     }
 295
 296     d.s = s;
 297     d.action = ACTION_INJECT_ERROR;
 298     qemu_opts_foreach(&inject_error_opts, add_rule, &d, &local_err);
 299     if (local_err) {
 300         error_propagate(errp, local_err);
 301         ret = -EINVAL;
 302         goto fail;
 303     }
 304
 305     d.action = ACTION_SET_STATE;
 306     qemu_opts_foreach(&set_state_opts, add_rule, &d, &local_err);
 307     if (local_err) {
 308         error_propagate(errp, local_err);
 309         ret = -EINVAL;
 310         goto fail;
 311     }
 312
 313     ret = 0;
 314 fail:
 315     qemu_opts_reset(&inject_error_opts);
 316     qemu_opts_reset(&set_state_opts);
 317     if (f) {
 318         fclose(f);
 319     }
 320     return ret;
 321 }
 322
 323 /* Valid blkdebug filenames look like blkdebug:path/to/config:path/to/image */
 324 static void blkdebug_parse_filename(const char *filename, QDict *options,
 325                                     Error **errp)
 326 {
 327     const char *c;
 328
 329     /* Parse the blkdebug: prefix */
 330     if (!strstart(filename, "blkdebug:", &filename)) {
 331         /* There was no prefix; therefore, all options have to be already
 332            present in the QDict (except for the filename) */
 333         qdict_put_str(options, "x-image", filename);
 334         return;
 335     }
 336
 337     /* Parse config file path */
 338     c = strchr(filename, ':');
 339     if (c == NULL) {
 340         error_setg(errp, "blkdebug requires both config file and image path");
 341         return;
 342     }
 343
 344     if (c != filename) {
 345         QString *config_path;
 346         config_path = qstring_from_substr(filename, 0, c - filename);
 347         qdict_put(options, "config", config_path);
 348     }
 349
 350     /* TODO Allow multi-level nesting and set file.filename here */
 351     filename = c + 1;
 352     qdict_put_str(options, "x-image", filename);
 353 }
 354
 355 static int blkdebug_parse_perm_list(uint64_t *dest, QDict *options,
 356                                     const char *prefix, Error **errp)
 357 {
 358     int ret = 0;
 359     QDict *subqdict = NULL;
 360     QObject *crumpled_subqdict = NULL;
 361     Visitor *v = NULL;
 362     BlockPermissionList *perm_list = NULL, *element;
 363
 364     *dest = 0;
 365
 366     qdict_extract_subqdict(options, &subqdict, prefix);
 367     if (!qdict_size(subqdict)) {
 368         goto out;
 369     }
 370
 371     crumpled_subqdict = qdict_crumple(subqdict, errp);
 372     if (!crumpled_subqdict) {
 373         ret = -EINVAL;
 374         goto out;
 375     }
 376
 377     v = qobject_input_visitor_new(crumpled_subqdict);
 378     if (!visit_type_BlockPermissionList(v, NULL, &perm_list, errp)) {
 379         ret = -EINVAL;
 380         goto out;
 381     }
 382
 383     for (element = perm_list; element; element = element->next) {
 384         *dest |= bdrv_qapi_perm_to_blk_perm(element->value);
 385     }
 386
 387 out:
 388     qapi_free_BlockPermissionList(perm_list);
 389     visit_free(v);
 390     qobject_unref(subqdict);
 391     qobject_unref(crumpled_subqdict);
 392     return ret;
 393 }
 394
 395 static int blkdebug_parse_perms(BDRVBlkdebugState *s, QDict *options,
 396                                 Error **errp)
 397 {
 398     int ret;
 399
 400     ret = blkdebug_parse_perm_list(&s->take_child_perms, options,
 401                                    "take-child-perms.", errp);
 402     if (ret < 0) {
 403         return ret;
 404     }
 405
 406     ret = blkdebug_parse_perm_list(&s->unshare_child_perms, options,
 407                                    "unshare-child-perms.", errp);
 408     if (ret < 0) {
 409         return ret;
 410     }
 411
 412     return 0;
 413 }
 414
 415 static QemuOptsList runtime_opts = {
 416     .name = "blkdebug",
 417     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 418     .desc = {
 419         {
 420             .name = "config",
 421             .type = QEMU_OPT_STRING,
 422             .help = "Path to the configuration file",
 423         },
 424         {
 425             .name = "x-image",
 426             .type = QEMU_OPT_STRING,
 427             .help = "[internal use only, will be removed]",
 428         },
 429         {
 430             .name = "align",
 431             .type = QEMU_OPT_SIZE,
 432             .help = "Required alignment in bytes",
 433         },
 434         {
 435             .name = "max-transfer",
 436             .type = QEMU_OPT_SIZE,
 437             .help = "Maximum transfer size in bytes",
 438         },
 439         {
 440             .name = "opt-write-zero",
 441             .type = QEMU_OPT_SIZE,
 442             .help = "Optimum write zero alignment in bytes",
 443         },
 444         {
 445             .name = "max-write-zero",
 446             .type = QEMU_OPT_SIZE,
 447             .help = "Maximum write zero size in bytes",
 448         },
 449         {
 450             .name = "opt-discard",
 451             .type = QEMU_OPT_SIZE,
 452             .help = "Optimum discard alignment in bytes",
 453         },
 454         {
 455             .name = "max-discard",
 456             .type = QEMU_OPT_SIZE,
 457             .help = "Maximum discard size in bytes",
 458         },
 459         { /* end of list */ }
 460     },
 461 };
 462
 463 static int blkdebug_open(BlockDriverState *bs, QDict *options, int flags,
 464                          Error **errp)
 465 {
 466     BDRVBlkdebugState *s = bs->opaque;
 467     QemuOpts *opts;
 468     Error *local_err = NULL;
 469     int ret;
 470     uint64_t align;
 471
 472     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 473     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
 474         ret = -EINVAL;
 475         goto out;
 476     }
 477
 478     /* Read rules from config file or command line options */
 479     s->config_file = g_strdup(qemu_opt_get(opts, "config"));
 480     ret = read_config(s, s->config_file, options, errp);
 481     if (ret) {
 482         goto out;
 483     }
 484
 485     /* Set initial state */
 486     s->state = 1;
 487
 488     /* Parse permissions modifiers before opening the image file */
 489     ret = blkdebug_parse_perms(s, options, errp);
 490     if (ret < 0) {
 491         goto out;
 492     }
 493
 494     /* Open the image file */
 495     bs->file = bdrv_open_child(qemu_opt_get(opts, "x-image"), options, "image",
 496                                bs, &child_of_bds,
 497                                BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
 498                                false, &local_err);
 499     if (local_err) {
 500         ret = -EINVAL;
 501         error_propagate(errp, local_err);
 502         goto out;
 503     }
 504
 505     bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
 506         (BDRV_REQ_FUA & bs->file->bs->supported_write_flags);
 507     bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
 508         ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) &
 509             bs->file->bs->supported_zero_flags);
 510     ret = -EINVAL;
 511
 512     /* Set alignment overrides */
 513     s->align = qemu_opt_get_size(opts, "align", 0);
 514     if (s->align && (s->align >= INT_MAX || !is_power_of_2(s->align))) {
 515         error_setg(errp, "Cannot meet constraints with align %" PRIu64,
 516                    s->align);
 517         goto out;
 518     }
 519     align = MAX(s->align, bs->file->bs->bl.request_alignment);
 520
 521     s->max_transfer = qemu_opt_get_size(opts, "max-transfer", 0);
 522     if (s->max_transfer &&
 523         (s->max_transfer >= INT_MAX ||
 524          !QEMU_IS_ALIGNED(s->max_transfer, align))) {
 525         error_setg(errp, "Cannot meet constraints with max-transfer %" PRIu64,
 526                    s->max_transfer);
 527         goto out;
 528     }
 529
 530     s->opt_write_zero = qemu_opt_get_size(opts, "opt-write-zero", 0);
 531     if (s->opt_write_zero &&
 532         (s->opt_write_zero >= INT_MAX ||
 533          !QEMU_IS_ALIGNED(s->opt_write_zero, align))) {
 534         error_setg(errp, "Cannot meet constraints with opt-write-zero %" PRIu64,
 535                    s->opt_write_zero);
 536         goto out;
 537     }
 538
 539     s->max_write_zero = qemu_opt_get_size(opts, "max-write-zero", 0);
 540     if (s->max_write_zero &&
 541         (s->max_write_zero >= INT_MAX ||
 542          !QEMU_IS_ALIGNED(s->max_write_zero,
 543                           MAX(s->opt_write_zero, align)))) {
 544         error_setg(errp, "Cannot meet constraints with max-write-zero %" PRIu64,
 545                    s->max_write_zero);
 546         goto out;
 547     }
 548
 549     s->opt_discard = qemu_opt_get_size(opts, "opt-discard", 0);
 550     if (s->opt_discard &&
 551         (s->opt_discard >= INT_MAX ||
 552          !QEMU_IS_ALIGNED(s->opt_discard, align))) {
 553         error_setg(errp, "Cannot meet constraints with opt-discard %" PRIu64,
 554                    s->opt_discard);
 555         goto out;
 556     }
 557
 558     s->max_discard = qemu_opt_get_size(opts, "max-discard", 0);
 559     if (s->max_discard &&
 560         (s->max_discard >= INT_MAX ||
 561          !QEMU_IS_ALIGNED(s->max_discard,
 562                           MAX(s->opt_discard, align)))) {
 563         error_setg(errp, "Cannot meet constraints with max-discard %" PRIu64,
 564                    s->max_discard);
 565         goto out;
 566     }
 567
 568     bdrv_debug_event(bs, BLKDBG_NONE);
 569
 570     ret = 0;
 571 out:
 572     if (ret < 0) {
 573         g_free(s->config_file);
 574     }
 575     qemu_opts_del(opts);
 576     return ret;
 577 }
 578
 579 static int rule_check(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
 580                       BlkdebugIOType iotype)
 581 {
 582     BDRVBlkdebugState *s = bs->opaque;
 583     BlkdebugRule *rule = NULL;
 584     int error;
 585     bool immediately;
 586
 587     QSIMPLEQ_FOREACH(rule, &s->active_rules, active_next) {
 588         uint64_t inject_offset = rule->options.inject.offset;
 589
 590         if ((inject_offset == -1 ||
 591              (bytes && inject_offset >= offset &&
 592               inject_offset < offset + bytes)) &&
 593             (rule->options.inject.iotype_mask & (1ull << iotype)))
 594         {
 595             break;
 596         }
 597     }
 598
 599     if (!rule || !rule->options.inject.error) {
 600         return 0;
 601     }
 602
 603     immediately = rule->options.inject.immediately;
 604     error = rule->options.inject.error;
 605
 606     if (rule->options.inject.once) {
 607         QSIMPLEQ_REMOVE(&s->active_rules, rule, BlkdebugRule, active_next);
 608         remove_rule(rule);
 609     }
 610
 611     if (!immediately) {
 612         aio_co_schedule(qemu_get_current_aio_context(), qemu_coroutine_self());
 613         qemu_coroutine_yield();
 614     }
 615
 616     return -error;
 617 }
 618
 619 static int coroutine_fn
 620 blkdebug_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
 621                    QEMUIOVector *qiov, int flags)
 622 {
 623     int err;
 624
 625     /* Sanity check block layer guarantees */
 626     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
 627     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
 628     if (bs->bl.max_transfer) {
 629         assert(bytes <= bs->bl.max_transfer);
 630     }
 631
 632     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_READ);
 633     if (err) {
 634         return err;
 635     }
 636
 637     return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
 638 }
 639
 640 static int coroutine_fn
 641 blkdebug_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
 642                     QEMUIOVector *qiov, int flags)
 643 {
 644     int err;
 645
 646     /* Sanity check block layer guarantees */
 647     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
 648     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
 649     if (bs->bl.max_transfer) {
 650         assert(bytes <= bs->bl.max_transfer);
 651     }
 652
 653     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_WRITE);
 654     if (err) {
 655         return err;
 656     }
 657
 658     return bdrv_co_pwritev(bs->file, offset, bytes, qiov, flags);
 659 }
 660
 661 static int blkdebug_co_flush(BlockDriverState *bs)
 662 {
 663     int err = rule_check(bs, 0, 0, BLKDEBUG_IO_TYPE_FLUSH);
 664
 665     if (err) {
 666         return err;
 667     }
 668
 669     return bdrv_co_flush(bs->file->bs);
 670 }
 671
 672 static int coroutine_fn blkdebug_co_pwrite_zeroes(BlockDriverState *bs,
 673                                                   int64_t offset, int bytes,
 674                                                   BdrvRequestFlags flags)
 675 {
 676     uint32_t align = MAX(bs->bl.request_alignment,
 677                          bs->bl.pwrite_zeroes_alignment);
 678     int err;
 679
 680     /* Only pass through requests that are larger than requested
 681      * preferred alignment (so that we test the fallback to writes on
 682      * unaligned portions), and check that the block layer never hands
 683      * us anything unaligned that crosses an alignment boundary.  */
 684     if (bytes < align) {
 685         assert(QEMU_IS_ALIGNED(offset, align) ||
 686                QEMU_IS_ALIGNED(offset + bytes, align) ||
 687                DIV_ROUND_UP(offset, align) ==
 688                DIV_ROUND_UP(offset + bytes, align));
 689         return -ENOTSUP;
 690     }
 691     assert(QEMU_IS_ALIGNED(offset, align));
 692     assert(QEMU_IS_ALIGNED(bytes, align));
 693     if (bs->bl.max_pwrite_zeroes) {
 694         assert(bytes <= bs->bl.max_pwrite_zeroes);
 695     }
 696
 697     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_WRITE_ZEROES);
 698     if (err) {
 699         return err;
 700     }
 701
 702     return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags);
 703 }
 704
 705 static int coroutine_fn blkdebug_co_pdiscard(BlockDriverState *bs,
 706                                              int64_t offset, int bytes)
 707 {
 708     uint32_t align = bs->bl.pdiscard_alignment;
 709     int err;
 710
 711     /* Only pass through requests that are larger than requested
 712      * minimum alignment, and ensure that unaligned requests do not
 713      * cross optimum discard boundaries. */
 714     if (bytes < bs->bl.request_alignment) {
 715         assert(QEMU_IS_ALIGNED(offset, align) ||
 716                QEMU_IS_ALIGNED(offset + bytes, align) ||
 717                DIV_ROUND_UP(offset, align) ==
 718                DIV_ROUND_UP(offset + bytes, align));
 719         return -ENOTSUP;
 720     }
 721     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
 722     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
 723     if (align && bytes >= align) {
 724         assert(QEMU_IS_ALIGNED(offset, align));
 725         assert(QEMU_IS_ALIGNED(bytes, align));
 726     }
 727     if (bs->bl.max_pdiscard) {
 728         assert(bytes <= bs->bl.max_pdiscard);
 729     }
 730
 731     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_DISCARD);
 732     if (err) {
 733         return err;
 734     }
 735
 736     return bdrv_co_pdiscard(bs->file, offset, bytes);
 737 }
 738
 739 static int coroutine_fn blkdebug_co_block_status(BlockDriverState *bs,
 740                                                  bool want_zero,
 741                                                  int64_t offset,
 742                                                  int64_t bytes,
 743                                                  int64_t *pnum,
 744                                                  int64_t *map,
 745                                                  BlockDriverState **file)
 746 {
 747     int err;
 748
 749     assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
 750
 751     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_BLOCK_STATUS);
 752     if (err) {
 753         return err;
 754     }
 755
 756     assert(bs->file && bs->file->bs);
 757     *pnum = bytes;
 758     *map = offset;
 759     *file = bs->file->bs;
 760     return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID;
 761 }
 762
 763 static void blkdebug_close(BlockDriverState *bs)
 764 {
 765     BDRVBlkdebugState *s = bs->opaque;
 766     BlkdebugRule *rule, *next;
 767     int i;
 768
 769     for (i = 0; i < BLKDBG__MAX; i++) {
 770         QLIST_FOREACH_SAFE(rule, &s->rules[i], next, next) {
 771             remove_rule(rule);
 772         }
 773     }
 774
 775     g_free(s->config_file);
 776 }
 777
 778 static void suspend_request(BlockDriverState *bs, BlkdebugRule *rule)
 779 {
 780     BDRVBlkdebugState *s = bs->opaque;
 781     BlkdebugSuspendedReq r;
 782
 783     r = (BlkdebugSuspendedReq) {
 784         .co         = qemu_coroutine_self(),
 785         .tag        = g_strdup(rule->options.suspend.tag),
 786     };
 787
 788     remove_rule(rule);
 789     QLIST_INSERT_HEAD(&s->suspended_reqs, &r, next);
 790
 791     if (!qtest_enabled()) {
 792         printf("blkdebug: Suspended request '%s'\n", r.tag);
 793     }
 794     qemu_coroutine_yield();
 795     if (!qtest_enabled()) {
 796         printf("blkdebug: Resuming request '%s'\n", r.tag);
 797     }
 798
 799     QLIST_REMOVE(&r, next);
 800     g_free(r.tag);
 801 }
 802
 803 static bool process_rule(BlockDriverState *bs, struct BlkdebugRule *rule,
 804     bool injected)
 805 {
 806     BDRVBlkdebugState *s = bs->opaque;
 807
 808     /* Only process rules for the current state */
 809     if (rule->state && rule->state != s->state) {
 810         return injected;
 811     }
 812
 813     /* Take the action */
 814     switch (rule->action) {
 815     case ACTION_INJECT_ERROR:
 816         if (!injected) {
 817             QSIMPLEQ_INIT(&s->active_rules);
 818             injected = true;
 819         }
 820         QSIMPLEQ_INSERT_HEAD(&s->active_rules, rule, active_next);
 821         break;
 822
 823     case ACTION_SET_STATE:
 824         s->new_state = rule->options.set_state.new_state;
 825         break;
 826
 827     case ACTION_SUSPEND:
 828         suspend_request(bs, rule);
 829         break;
 830     }
 831     return injected;
 832 }
 833
 834 static void blkdebug_debug_event(BlockDriverState *bs, BlkdebugEvent event)
 835 {
 836     BDRVBlkdebugState *s = bs->opaque;
 837     struct BlkdebugRule *rule, *next;
 838     bool injected;
 839
 840     assert((int)event >= 0 && event < BLKDBG__MAX);
 841
 842     injected = false;
 843     s->new_state = s->state;
 844     QLIST_FOREACH_SAFE(rule, &s->rules[event], next, next) {
 845         injected = process_rule(bs, rule, injected);
 846     }
 847     s->state = s->new_state;
 848 }
 849
 850 static int blkdebug_debug_breakpoint(BlockDriverState *bs, const char *event,
 851                                      const char *tag)
 852 {
 853     BDRVBlkdebugState *s = bs->opaque;
 854     struct BlkdebugRule *rule;
 855     int blkdebug_event;
 856
 857     blkdebug_event = qapi_enum_parse(&BlkdebugEvent_lookup, event, -1, NULL);
 858     if (blkdebug_event < 0) {
 859         return -ENOENT;
 860     }
 861
 862     rule = g_malloc(sizeof(*rule));
 863     *rule = (struct BlkdebugRule) {
 864         .event  = blkdebug_event,
 865         .action = ACTION_SUSPEND,
 866         .state  = 0,
 867         .options.suspend.tag = g_strdup(tag),
 868     };
 869
 870     QLIST_INSERT_HEAD(&s->rules[blkdebug_event], rule, next);
 871
 872     return 0;
 873 }
 874
 875 static int blkdebug_debug_resume(BlockDriverState *bs, const char *tag)
 876 {
 877     BDRVBlkdebugState *s = bs->opaque;
 878     BlkdebugSuspendedReq *r, *next;
 879
 880     QLIST_FOREACH_SAFE(r, &s->suspended_reqs, next, next) {
 881         if (!strcmp(r->tag, tag)) {
 882             qemu_coroutine_enter(r->co);
 883             return 0;
 884         }
 885     }
 886     return -ENOENT;
 887 }
 888
 889 static int blkdebug_debug_remove_breakpoint(BlockDriverState *bs,
 890                                             const char *tag)
 891 {
 892     BDRVBlkdebugState *s = bs->opaque;
 893     BlkdebugSuspendedReq *r, *r_next;
 894     BlkdebugRule *rule, *next;
 895     int i, ret = -ENOENT;
 896
 897     for (i = 0; i < BLKDBG__MAX; i++) {
 898         QLIST_FOREACH_SAFE(rule, &s->rules[i], next, next) {
 899             if (rule->action == ACTION_SUSPEND &&
 900                 !strcmp(rule->options.suspend.tag, tag)) {
 901                 remove_rule(rule);
 902                 ret = 0;
 903             }
 904         }
 905     }
 906     QLIST_FOREACH_SAFE(r, &s->suspended_reqs, next, r_next) {
 907         if (!strcmp(r->tag, tag)) {
 908             qemu_coroutine_enter(r->co);
 909             ret = 0;
 910         }
 911     }
 912     return ret;
 913 }
 914
 915 static bool blkdebug_debug_is_suspended(BlockDriverState *bs, const char *tag)
 916 {
 917     BDRVBlkdebugState *s = bs->opaque;
 918     BlkdebugSuspendedReq *r;
 919
 920     QLIST_FOREACH(r, &s->suspended_reqs, next) {
 921         if (!strcmp(r->tag, tag)) {
 922             return true;
 923         }
 924     }
 925     return false;
 926 }
 927
 928 static int64_t blkdebug_getlength(BlockDriverState *bs)
 929 {
 930     return bdrv_getlength(bs->file->bs);
 931 }
 932
 933 static void blkdebug_refresh_filename(BlockDriverState *bs)
 934 {
 935     BDRVBlkdebugState *s = bs->opaque;
 936     const QDictEntry *e;
 937     int ret;
 938
 939     if (!bs->file->bs->exact_filename[0]) {
 940         return;
 941     }
 942
 943     for (e = qdict_first(bs->full_open_options); e;
 944          e = qdict_next(bs->full_open_options, e))
 945     {
 946         /* Real child options are under "image", but "x-image" may
 947          * contain a filename */
 948         if (strcmp(qdict_entry_key(e), "config") &&
 949             strcmp(qdict_entry_key(e), "image") &&
 950             strcmp(qdict_entry_key(e), "x-image") &&
 951             strcmp(qdict_entry_key(e), "driver"))
 952         {
 953             return;
 954         }
 955     }
 956
 957     ret = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
 958                    "blkdebug:%s:%s",
 959                    s->config_file ?: "", bs->file->bs->exact_filename);
 960     if (ret >= sizeof(bs->exact_filename)) {
 961         /* An overflow makes the filename unusable, so do not report any */
 962         bs->exact_filename[0] = 0;
 963     }
 964 }
 965
 966 static void blkdebug_refresh_limits(BlockDriverState *bs, Error **errp)
 967 {
 968     BDRVBlkdebugState *s = bs->opaque;
 969
 970     if (s->align) {
 971         bs->bl.request_alignment = s->align;
 972     }
 973     if (s->max_transfer) {
 974         bs->bl.max_transfer = s->max_transfer;
 975     }
 976     if (s->opt_write_zero) {
 977         bs->bl.pwrite_zeroes_alignment = s->opt_write_zero;
 978     }
 979     if (s->max_write_zero) {
 980         bs->bl.max_pwrite_zeroes = s->max_write_zero;
 981     }
 982     if (s->opt_discard) {
 983         bs->bl.pdiscard_alignment = s->opt_discard;
 984     }
 985     if (s->max_discard) {
 986         bs->bl.max_pdiscard = s->max_discard;
 987     }
 988 }
 989
 990 static int blkdebug_reopen_prepare(BDRVReopenState *reopen_state,
 991                                    BlockReopenQueue *queue, Error **errp)
 992 {
 993     return 0;
 994 }
 995
 996 static void blkdebug_child_perm(BlockDriverState *bs, BdrvChild *c,
 997                                 BdrvChildRole role,
 998                                 BlockReopenQueue *reopen_queue,
 999                                 uint64_t perm, uint64_t shared,
1000                                 uint64_t *nperm, uint64_t *nshared)
1001 {
1002     BDRVBlkdebugState *s = bs->opaque;
1003
1004     bdrv_default_perms(bs, c, role, reopen_queue,
1005                        perm, shared, nperm, nshared);
1006
1007     *nperm |= s->take_child_perms;
1008     *nshared &= ~s->unshare_child_perms;
1009 }
1010
1011 static const char *const blkdebug_strong_runtime_opts[] = {
1012     "config",
1013     "inject-error.",
1014     "set-state.",
1015     "align",
1016     "max-transfer",
1017     "opt-write-zero",
1018     "max-write-zero",
1019     "opt-discard",
1020     "max-discard",
1021
1022     NULL
1023 };
1024
1025 static BlockDriver bdrv_blkdebug = {
1026     .format_name            = "blkdebug",
1027     .protocol_name          = "blkdebug",
1028     .instance_size          = sizeof(BDRVBlkdebugState),
1029     .is_filter              = true,
1030
1031     .bdrv_parse_filename    = blkdebug_parse_filename,
1032     .bdrv_file_open         = blkdebug_open,
1033     .bdrv_close             = blkdebug_close,
1034     .bdrv_reopen_prepare    = blkdebug_reopen_prepare,
1035     .bdrv_child_perm        = blkdebug_child_perm,
1036
1037     .bdrv_getlength         = blkdebug_getlength,
1038     .bdrv_refresh_filename  = blkdebug_refresh_filename,
1039     .bdrv_refresh_limits    = blkdebug_refresh_limits,
1040
1041     .bdrv_co_preadv         = blkdebug_co_preadv,
1042     .bdrv_co_pwritev        = blkdebug_co_pwritev,
1043     .bdrv_co_flush_to_disk  = blkdebug_co_flush,
1044     .bdrv_co_pwrite_zeroes  = blkdebug_co_pwrite_zeroes,
1045     .bdrv_co_pdiscard       = blkdebug_co_pdiscard,
1046     .bdrv_co_block_status   = blkdebug_co_block_status,
1047
1048     .bdrv_debug_event           = blkdebug_debug_event,
1049     .bdrv_debug_breakpoint      = blkdebug_debug_breakpoint,
1050     .bdrv_debug_remove_breakpoint
1051                                 = blkdebug_debug_remove_breakpoint,
1052     .bdrv_debug_resume          = blkdebug_debug_resume,
1053     .bdrv_debug_is_suspended    = blkdebug_debug_is_suspended,
1054
1055     .strong_runtime_opts        = blkdebug_strong_runtime_opts,
1056 };
1057
1058 static void bdrv_blkdebug_init(void)
1059 {
1060     bdrv_register(&bdrv_blkdebug);
1061 }
1062
1063 block_init(bdrv_blkdebug_init);