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 /* All APIs are thread-safe */
  42
  43 typedef struct BDRVBlkdebugState {
  44     /* IN: initialized in blkdebug_open() and never changed */
  45     uint64_t align;
  46     uint64_t max_transfer;
  47     uint64_t opt_write_zero;
  48     uint64_t max_write_zero;
  49     uint64_t opt_discard;
  50     uint64_t max_discard;
  51     char *config_file; /* For blkdebug_refresh_filename() */
  52     /* initialized in blkdebug_parse_perms() */
  53     uint64_t take_child_perms;
  54     uint64_t unshare_child_perms;
  55
  56     /* State. Protected by lock */
  57     int state;
  58     QLIST_HEAD(, BlkdebugRule) rules[BLKDBG__MAX];
  59     QSIMPLEQ_HEAD(, BlkdebugRule) active_rules;
  60     QLIST_HEAD(, BlkdebugSuspendedReq) suspended_reqs;
  61     QemuMutex lock;
  62 } BDRVBlkdebugState;
  63
  64 typedef struct BlkdebugAIOCB {
  65     BlockAIOCB common;
  66     int ret;
  67 } BlkdebugAIOCB;
  68
  69 typedef struct BlkdebugSuspendedReq {
  70     /* IN: initialized in suspend_request() */
  71     Coroutine *co;
  72     char *tag;
  73
  74     /* List entry protected BDRVBlkdebugState's lock */
  75     QLIST_ENTRY(BlkdebugSuspendedReq) next;
  76 } BlkdebugSuspendedReq;
  77
  78 enum {
  79     ACTION_INJECT_ERROR,
  80     ACTION_SET_STATE,
  81     ACTION_SUSPEND,
  82     ACTION__MAX,
  83 };
  84
  85 typedef struct BlkdebugRule {
  86     /* IN: initialized in add_rule() or blkdebug_debug_breakpoint() */
  87     BlkdebugEvent event;
  88     int action;
  89     int state;
  90     union {
  91         struct {
  92             uint64_t iotype_mask;
  93             int error;
  94             int immediately;
  95             int once;
  96             int64_t offset;
  97         } inject;
  98         struct {
  99             int new_state;
 100         } set_state;
 101         struct {
 102             char *tag;
 103         } suspend;
 104     } options;
 105
 106     /* List entries protected BDRVBlkdebugState's lock */
 107     QLIST_ENTRY(BlkdebugRule) next;
 108     QSIMPLEQ_ENTRY(BlkdebugRule) active_next;
 109 } BlkdebugRule;
 110
 111 QEMU_BUILD_BUG_MSG(BLKDEBUG_IO_TYPE__MAX > 64,
 112                    "BlkdebugIOType mask does not fit into an uint64_t");
 113
 114 static QemuOptsList inject_error_opts = {
 115     .name = "inject-error",
 116     .head = QTAILQ_HEAD_INITIALIZER(inject_error_opts.head),
 117     .desc = {
 118         {
 119             .name = "event",
 120             .type = QEMU_OPT_STRING,
 121         },
 122         {
 123             .name = "state",
 124             .type = QEMU_OPT_NUMBER,
 125         },
 126         {
 127             .name = "iotype",
 128             .type = QEMU_OPT_STRING,
 129         },
 130         {
 131             .name = "errno",
 132             .type = QEMU_OPT_NUMBER,
 133         },
 134         {
 135             .name = "sector",
 136             .type = QEMU_OPT_NUMBER,
 137         },
 138         {
 139             .name = "once",
 140             .type = QEMU_OPT_BOOL,
 141         },
 142         {
 143             .name = "immediately",
 144             .type = QEMU_OPT_BOOL,
 145         },
 146         { /* end of list */ }
 147     },
 148 };
 149
 150 static QemuOptsList set_state_opts = {
 151     .name = "set-state",
 152     .head = QTAILQ_HEAD_INITIALIZER(set_state_opts.head),
 153     .desc = {
 154         {
 155             .name = "event",
 156             .type = QEMU_OPT_STRING,
 157         },
 158         {
 159             .name = "state",
 160             .type = QEMU_OPT_NUMBER,
 161         },
 162         {
 163             .name = "new_state",
 164             .type = QEMU_OPT_NUMBER,
 165         },
 166         { /* end of list */ }
 167     },
 168 };
 169
 170 static QemuOptsList *config_groups[] = {
 171     &inject_error_opts,
 172     &set_state_opts,
 173     NULL
 174 };
 175
 176 struct add_rule_data {
 177     BDRVBlkdebugState *s;
 178     int action;
 179 };
 180
 181 static int add_rule(void *opaque, QemuOpts *opts, Error **errp)
 182 {
 183     struct add_rule_data *d = opaque;
 184     BDRVBlkdebugState *s = d->s;
 185     const char *event_name;
 186     int event;
 187     struct BlkdebugRule *rule;
 188     int64_t sector;
 189     BlkdebugIOType iotype;
 190     Error *local_error = NULL;
 191
 192     /* Find the right event for the rule */
 193     event_name = qemu_opt_get(opts, "event");
 194     if (!event_name) {
 195         error_setg(errp, "Missing event name for rule");
 196         return -1;
 197     }
 198     event = qapi_enum_parse(&BlkdebugEvent_lookup, event_name, -1, errp);
 199     if (event < 0) {
 200         return -1;
 201     }
 202
 203     /* Set attributes common for all actions */
 204     rule = g_malloc0(sizeof(*rule));
 205     *rule = (struct BlkdebugRule) {
 206         .event  = event,
 207         .action = d->action,
 208         .state  = qemu_opt_get_number(opts, "state", 0),
 209     };
 210
 211     /* Parse action-specific options */
 212     switch (d->action) {
 213     case ACTION_INJECT_ERROR:
 214         rule->options.inject.error = qemu_opt_get_number(opts, "errno", EIO);
 215         rule->options.inject.once  = qemu_opt_get_bool(opts, "once", 0);
 216         rule->options.inject.immediately =
 217             qemu_opt_get_bool(opts, "immediately", 0);
 218         sector = qemu_opt_get_number(opts, "sector", -1);
 219         rule->options.inject.offset =
 220             sector == -1 ? -1 : sector * BDRV_SECTOR_SIZE;
 221
 222         iotype = qapi_enum_parse(&BlkdebugIOType_lookup,
 223                                  qemu_opt_get(opts, "iotype"),
 224                                  BLKDEBUG_IO_TYPE__MAX, &local_error);
 225         if (local_error) {
 226             error_propagate(errp, local_error);
 227             g_free(rule);
 228             return -1;
 229         }
 230         if (iotype != BLKDEBUG_IO_TYPE__MAX) {
 231             rule->options.inject.iotype_mask = (1ull << iotype);
 232         } else {
 233             /* Apply the default */
 234             rule->options.inject.iotype_mask =
 235                 (1ull << BLKDEBUG_IO_TYPE_READ)
 236                 | (1ull << BLKDEBUG_IO_TYPE_WRITE)
 237                 | (1ull << BLKDEBUG_IO_TYPE_WRITE_ZEROES)
 238                 | (1ull << BLKDEBUG_IO_TYPE_DISCARD)
 239                 | (1ull << BLKDEBUG_IO_TYPE_FLUSH);
 240         }
 241
 242         break;
 243
 244     case ACTION_SET_STATE:
 245         rule->options.set_state.new_state =
 246             qemu_opt_get_number(opts, "new_state", 0);
 247         break;
 248
 249     case ACTION_SUSPEND:
 250         rule->options.suspend.tag =
 251             g_strdup(qemu_opt_get(opts, "tag"));
 252         break;
 253     };
 254
 255     /* Add the rule */
 256     qemu_mutex_lock(&s->lock);
 257     QLIST_INSERT_HEAD(&s->rules[event], rule, next);
 258     qemu_mutex_unlock(&s->lock);
 259
 260     return 0;
 261 }
 262
 263 /* Called with lock held or from .bdrv_close */
 264 static void remove_rule(BlkdebugRule *rule)
 265 {
 266     switch (rule->action) {
 267     case ACTION_INJECT_ERROR:
 268     case ACTION_SET_STATE:
 269         break;
 270     case ACTION_SUSPEND:
 271         g_free(rule->options.suspend.tag);
 272         break;
 273     }
 274
 275     QLIST_REMOVE(rule, next);
 276     g_free(rule);
 277 }
 278
 279 static int read_config(BDRVBlkdebugState *s, const char *filename,
 280                        QDict *options, Error **errp)
 281 {
 282     FILE *f = NULL;
 283     int ret;
 284     struct add_rule_data d;
 285     Error *local_err = NULL;
 286
 287     if (filename) {
 288         f = fopen(filename, "r");
 289         if (f == NULL) {
 290             error_setg_errno(errp, errno, "Could not read blkdebug config file");
 291             return -errno;
 292         }
 293
 294         ret = qemu_config_parse(f, config_groups, filename, errp);
 295         if (ret < 0) {
 296             goto fail;
 297         }
 298     }
 299
 300     qemu_config_parse_qdict(options, config_groups, &local_err);
 301     if (local_err) {
 302         error_propagate(errp, local_err);
 303         ret = -EINVAL;
 304         goto fail;
 305     }
 306
 307     d.s = s;
 308     d.action = ACTION_INJECT_ERROR;
 309     qemu_opts_foreach(&inject_error_opts, add_rule, &d, &local_err);
 310     if (local_err) {
 311         error_propagate(errp, local_err);
 312         ret = -EINVAL;
 313         goto fail;
 314     }
 315
 316     d.action = ACTION_SET_STATE;
 317     qemu_opts_foreach(&set_state_opts, add_rule, &d, &local_err);
 318     if (local_err) {
 319         error_propagate(errp, local_err);
 320         ret = -EINVAL;
 321         goto fail;
 322     }
 323
 324     ret = 0;
 325 fail:
 326     qemu_opts_reset(&inject_error_opts);
 327     qemu_opts_reset(&set_state_opts);
 328     if (f) {
 329         fclose(f);
 330     }
 331     return ret;
 332 }
 333
 334 /* Valid blkdebug filenames look like blkdebug:path/to/config:path/to/image */
 335 static void blkdebug_parse_filename(const char *filename, QDict *options,
 336                                     Error **errp)
 337 {
 338     const char *c;
 339
 340     /* Parse the blkdebug: prefix */
 341     if (!strstart(filename, "blkdebug:", &filename)) {
 342         /* There was no prefix; therefore, all options have to be already
 343            present in the QDict (except for the filename) */
 344         qdict_put_str(options, "x-image", filename);
 345         return;
 346     }
 347
 348     /* Parse config file path */
 349     c = strchr(filename, ':');
 350     if (c == NULL) {
 351         error_setg(errp, "blkdebug requires both config file and image path");
 352         return;
 353     }
 354
 355     if (c != filename) {
 356         QString *config_path;
 357         config_path = qstring_from_substr(filename, 0, c - filename);
 358         qdict_put(options, "config", config_path);
 359     }
 360
 361     /* TODO Allow multi-level nesting and set file.filename here */
 362     filename = c + 1;
 363     qdict_put_str(options, "x-image", filename);
 364 }
 365
 366 static int blkdebug_parse_perm_list(uint64_t *dest, QDict *options,
 367                                     const char *prefix, Error **errp)
 368 {
 369     int ret = 0;
 370     QDict *subqdict = NULL;
 371     QObject *crumpled_subqdict = NULL;
 372     Visitor *v = NULL;
 373     BlockPermissionList *perm_list = NULL, *element;
 374
 375     *dest = 0;
 376
 377     qdict_extract_subqdict(options, &subqdict, prefix);
 378     if (!qdict_size(subqdict)) {
 379         goto out;
 380     }
 381
 382     crumpled_subqdict = qdict_crumple(subqdict, errp);
 383     if (!crumpled_subqdict) {
 384         ret = -EINVAL;
 385         goto out;
 386     }
 387
 388     v = qobject_input_visitor_new(crumpled_subqdict);
 389     if (!visit_type_BlockPermissionList(v, NULL, &perm_list, errp)) {
 390         ret = -EINVAL;
 391         goto out;
 392     }
 393
 394     for (element = perm_list; element; element = element->next) {
 395         *dest |= bdrv_qapi_perm_to_blk_perm(element->value);
 396     }
 397
 398 out:
 399     qapi_free_BlockPermissionList(perm_list);
 400     visit_free(v);
 401     qobject_unref(subqdict);
 402     qobject_unref(crumpled_subqdict);
 403     return ret;
 404 }
 405
 406 static int blkdebug_parse_perms(BDRVBlkdebugState *s, QDict *options,
 407                                 Error **errp)
 408 {
 409     int ret;
 410
 411     ret = blkdebug_parse_perm_list(&s->take_child_perms, options,
 412                                    "take-child-perms.", errp);
 413     if (ret < 0) {
 414         return ret;
 415     }
 416
 417     ret = blkdebug_parse_perm_list(&s->unshare_child_perms, options,
 418                                    "unshare-child-perms.", errp);
 419     if (ret < 0) {
 420         return ret;
 421     }
 422
 423     return 0;
 424 }
 425
 426 static QemuOptsList runtime_opts = {
 427     .name = "blkdebug",
 428     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
 429     .desc = {
 430         {
 431             .name = "config",
 432             .type = QEMU_OPT_STRING,
 433             .help = "Path to the configuration file",
 434         },
 435         {
 436             .name = "x-image",
 437             .type = QEMU_OPT_STRING,
 438             .help = "[internal use only, will be removed]",
 439         },
 440         {
 441             .name = "align",
 442             .type = QEMU_OPT_SIZE,
 443             .help = "Required alignment in bytes",
 444         },
 445         {
 446             .name = "max-transfer",
 447             .type = QEMU_OPT_SIZE,
 448             .help = "Maximum transfer size in bytes",
 449         },
 450         {
 451             .name = "opt-write-zero",
 452             .type = QEMU_OPT_SIZE,
 453             .help = "Optimum write zero alignment in bytes",
 454         },
 455         {
 456             .name = "max-write-zero",
 457             .type = QEMU_OPT_SIZE,
 458             .help = "Maximum write zero size in bytes",
 459         },
 460         {
 461             .name = "opt-discard",
 462             .type = QEMU_OPT_SIZE,
 463             .help = "Optimum discard alignment in bytes",
 464         },
 465         {
 466             .name = "max-discard",
 467             .type = QEMU_OPT_SIZE,
 468             .help = "Maximum discard size in bytes",
 469         },
 470         { /* end of list */ }
 471     },
 472 };
 473
 474 static int blkdebug_open(BlockDriverState *bs, QDict *options, int flags,
 475                          Error **errp)
 476 {
 477     BDRVBlkdebugState *s = bs->opaque;
 478     QemuOpts *opts;
 479     int ret;
 480     uint64_t align;
 481
 482     qemu_mutex_init(&s->lock);
 483     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
 484     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
 485         ret = -EINVAL;
 486         goto out;
 487     }
 488
 489     /* Read rules from config file or command line options */
 490     s->config_file = g_strdup(qemu_opt_get(opts, "config"));
 491     ret = read_config(s, s->config_file, options, errp);
 492     if (ret) {
 493         goto out;
 494     }
 495
 496     /* Set initial state */
 497     s->state = 1;
 498
 499     /* Parse permissions modifiers before opening the image file */
 500     ret = blkdebug_parse_perms(s, options, errp);
 501     if (ret < 0) {
 502         goto out;
 503     }
 504
 505     /* Open the image file */
 506     bs->file = bdrv_open_child(qemu_opt_get(opts, "x-image"), options, "image",
 507                                bs, &child_of_bds,
 508                                BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
 509                                false, errp);
 510     if (!bs->file) {
 511         ret = -EINVAL;
 512         goto out;
 513     }
 514
 515     bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
 516         (BDRV_REQ_FUA & bs->file->bs->supported_write_flags);
 517     bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
 518         ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) &
 519             bs->file->bs->supported_zero_flags);
 520     ret = -EINVAL;
 521
 522     /* Set alignment overrides */
 523     s->align = qemu_opt_get_size(opts, "align", 0);
 524     if (s->align && (s->align >= INT_MAX || !is_power_of_2(s->align))) {
 525         error_setg(errp, "Cannot meet constraints with align %" PRIu64,
 526                    s->align);
 527         goto out;
 528     }
 529     align = MAX(s->align, bs->file->bs->bl.request_alignment);
 530
 531     s->max_transfer = qemu_opt_get_size(opts, "max-transfer", 0);
 532     if (s->max_transfer &&
 533         (s->max_transfer >= INT_MAX ||
 534          !QEMU_IS_ALIGNED(s->max_transfer, align))) {
 535         error_setg(errp, "Cannot meet constraints with max-transfer %" PRIu64,
 536                    s->max_transfer);
 537         goto out;
 538     }
 539
 540     s->opt_write_zero = qemu_opt_get_size(opts, "opt-write-zero", 0);
 541     if (s->opt_write_zero &&
 542         (s->opt_write_zero >= INT_MAX ||
 543          !QEMU_IS_ALIGNED(s->opt_write_zero, align))) {
 544         error_setg(errp, "Cannot meet constraints with opt-write-zero %" PRIu64,
 545                    s->opt_write_zero);
 546         goto out;
 547     }
 548
 549     s->max_write_zero = qemu_opt_get_size(opts, "max-write-zero", 0);
 550     if (s->max_write_zero &&
 551         (s->max_write_zero >= INT_MAX ||
 552          !QEMU_IS_ALIGNED(s->max_write_zero,
 553                           MAX(s->opt_write_zero, align)))) {
 554         error_setg(errp, "Cannot meet constraints with max-write-zero %" PRIu64,
 555                    s->max_write_zero);
 556         goto out;
 557     }
 558
 559     s->opt_discard = qemu_opt_get_size(opts, "opt-discard", 0);
 560     if (s->opt_discard &&
 561         (s->opt_discard >= INT_MAX ||
 562          !QEMU_IS_ALIGNED(s->opt_discard, align))) {
 563         error_setg(errp, "Cannot meet constraints with opt-discard %" PRIu64,
 564                    s->opt_discard);
 565         goto out;
 566     }
 567
 568     s->max_discard = qemu_opt_get_size(opts, "max-discard", 0);
 569     if (s->max_discard &&
 570         (s->max_discard >= INT_MAX ||
 571          !QEMU_IS_ALIGNED(s->max_discard,
 572                           MAX(s->opt_discard, align)))) {
 573         error_setg(errp, "Cannot meet constraints with max-discard %" PRIu64,
 574                    s->max_discard);
 575         goto out;
 576     }
 577
 578     bdrv_debug_event(bs, BLKDBG_NONE);
 579
 580     ret = 0;
 581 out:
 582     if (ret < 0) {
 583         qemu_mutex_destroy(&s->lock);
 584         g_free(s->config_file);
 585     }
 586     qemu_opts_del(opts);
 587     return ret;
 588 }
 589
 590 static int rule_check(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
 591                       BlkdebugIOType iotype)
 592 {
 593     BDRVBlkdebugState *s = bs->opaque;
 594     BlkdebugRule *rule = NULL;
 595     int error;
 596     bool immediately;
 597
 598     qemu_mutex_lock(&s->lock);
 599     QSIMPLEQ_FOREACH(rule, &s->active_rules, active_next) {
 600         uint64_t inject_offset = rule->options.inject.offset;
 601
 602         if ((inject_offset == -1 ||
 603              (bytes && inject_offset >= offset &&
 604               inject_offset < offset + bytes)) &&
 605             (rule->options.inject.iotype_mask & (1ull << iotype)))
 606         {
 607             break;
 608         }
 609     }
 610
 611     if (!rule || !rule->options.inject.error) {
 612         qemu_mutex_unlock(&s->lock);
 613         return 0;
 614     }
 615
 616     immediately = rule->options.inject.immediately;
 617     error = rule->options.inject.error;
 618
 619     if (rule->options.inject.once) {
 620         QSIMPLEQ_REMOVE(&s->active_rules, rule, BlkdebugRule, active_next);
 621         remove_rule(rule);
 622     }
 623
 624     qemu_mutex_unlock(&s->lock);
 625     if (!immediately) {
 626         aio_co_schedule(qemu_get_current_aio_context(), qemu_coroutine_self());
 627         qemu_coroutine_yield();
 628     }
 629
 630     return -error;
 631 }
 632
 633 static int coroutine_fn
 634 blkdebug_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
 635                    QEMUIOVector *qiov, BdrvRequestFlags flags)
 636 {
 637     int err;
 638
 639     /* Sanity check block layer guarantees */
 640     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
 641     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
 642     if (bs->bl.max_transfer) {
 643         assert(bytes <= bs->bl.max_transfer);
 644     }
 645
 646     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_READ);
 647     if (err) {
 648         return err;
 649     }
 650
 651     return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
 652 }
 653
 654 static int coroutine_fn
 655 blkdebug_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
 656                     QEMUIOVector *qiov, BdrvRequestFlags flags)
 657 {
 658     int err;
 659
 660     /* Sanity check block layer guarantees */
 661     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
 662     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
 663     if (bs->bl.max_transfer) {
 664         assert(bytes <= bs->bl.max_transfer);
 665     }
 666
 667     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_WRITE);
 668     if (err) {
 669         return err;
 670     }
 671
 672     return bdrv_co_pwritev(bs->file, offset, bytes, qiov, flags);
 673 }
 674
 675 static int blkdebug_co_flush(BlockDriverState *bs)
 676 {
 677     int err = rule_check(bs, 0, 0, BLKDEBUG_IO_TYPE_FLUSH);
 678
 679     if (err) {
 680         return err;
 681     }
 682
 683     return bdrv_co_flush(bs->file->bs);
 684 }
 685
 686 static int coroutine_fn blkdebug_co_pwrite_zeroes(BlockDriverState *bs,
 687                                                   int64_t offset, int64_t bytes,
 688                                                   BdrvRequestFlags flags)
 689 {
 690     uint32_t align = MAX(bs->bl.request_alignment,
 691                          bs->bl.pwrite_zeroes_alignment);
 692     int err;
 693
 694     /* Only pass through requests that are larger than requested
 695      * preferred alignment (so that we test the fallback to writes on
 696      * unaligned portions), and check that the block layer never hands
 697      * us anything unaligned that crosses an alignment boundary.  */
 698     if (bytes < align) {
 699         assert(QEMU_IS_ALIGNED(offset, align) ||
 700                QEMU_IS_ALIGNED(offset + bytes, align) ||
 701                DIV_ROUND_UP(offset, align) ==
 702                DIV_ROUND_UP(offset + bytes, align));
 703         return -ENOTSUP;
 704     }
 705     assert(QEMU_IS_ALIGNED(offset, align));
 706     assert(QEMU_IS_ALIGNED(bytes, align));
 707     if (bs->bl.max_pwrite_zeroes) {
 708         assert(bytes <= bs->bl.max_pwrite_zeroes);
 709     }
 710
 711     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_WRITE_ZEROES);
 712     if (err) {
 713         return err;
 714     }
 715
 716     return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags);
 717 }
 718
 719 static int coroutine_fn blkdebug_co_pdiscard(BlockDriverState *bs,
 720                                              int64_t offset, int64_t bytes)
 721 {
 722     uint32_t align = bs->bl.pdiscard_alignment;
 723     int err;
 724
 725     /* Only pass through requests that are larger than requested
 726      * minimum alignment, and ensure that unaligned requests do not
 727      * cross optimum discard boundaries. */
 728     if (bytes < bs->bl.request_alignment) {
 729         assert(QEMU_IS_ALIGNED(offset, align) ||
 730                QEMU_IS_ALIGNED(offset + bytes, align) ||
 731                DIV_ROUND_UP(offset, align) ==
 732                DIV_ROUND_UP(offset + bytes, align));
 733         return -ENOTSUP;
 734     }
 735     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
 736     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
 737     if (align && bytes >= align) {
 738         assert(QEMU_IS_ALIGNED(offset, align));
 739         assert(QEMU_IS_ALIGNED(bytes, align));
 740     }
 741     if (bs->bl.max_pdiscard) {
 742         assert(bytes <= bs->bl.max_pdiscard);
 743     }
 744
 745     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_DISCARD);
 746     if (err) {
 747         return err;
 748     }
 749
 750     return bdrv_co_pdiscard(bs->file, offset, bytes);
 751 }
 752
 753 static int coroutine_fn blkdebug_co_block_status(BlockDriverState *bs,
 754                                                  bool want_zero,
 755                                                  int64_t offset,
 756                                                  int64_t bytes,
 757                                                  int64_t *pnum,
 758                                                  int64_t *map,
 759                                                  BlockDriverState **file)
 760 {
 761     int err;
 762
 763     assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
 764
 765     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_BLOCK_STATUS);
 766     if (err) {
 767         return err;
 768     }
 769
 770     assert(bs->file && bs->file->bs);
 771     *pnum = bytes;
 772     *map = offset;
 773     *file = bs->file->bs;
 774     return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID;
 775 }
 776
 777 static void blkdebug_close(BlockDriverState *bs)
 778 {
 779     BDRVBlkdebugState *s = bs->opaque;
 780     BlkdebugRule *rule, *next;
 781     int i;
 782
 783     for (i = 0; i < BLKDBG__MAX; i++) {
 784         QLIST_FOREACH_SAFE(rule, &s->rules[i], next, next) {
 785             remove_rule(rule);
 786         }
 787     }
 788
 789     g_free(s->config_file);
 790     qemu_mutex_destroy(&s->lock);
 791 }
 792
 793 /* Called with lock held.  */
 794 static void suspend_request(BlockDriverState *bs, BlkdebugRule *rule)
 795 {
 796     BDRVBlkdebugState *s = bs->opaque;
 797     BlkdebugSuspendedReq *r;
 798
 799     r = g_new(BlkdebugSuspendedReq, 1);
 800
 801     r->co         = qemu_coroutine_self();
 802     r->tag        = g_strdup(rule->options.suspend.tag);
 803
 804     remove_rule(rule);
 805     QLIST_INSERT_HEAD(&s->suspended_reqs, r, next);
 806
 807     if (!qtest_enabled()) {
 808         printf("blkdebug: Suspended request '%s'\n", r->tag);
 809     }
 810 }
 811
 812 /* Called with lock held.  */
 813 static void process_rule(BlockDriverState *bs, struct BlkdebugRule *rule,
 814                          int *action_count, int *new_state)
 815 {
 816     BDRVBlkdebugState *s = bs->opaque;
 817
 818     /* Only process rules for the current state */
 819     if (rule->state && rule->state != s->state) {
 820         return;
 821     }
 822
 823     /* Take the action */
 824     action_count[rule->action]++;
 825     switch (rule->action) {
 826     case ACTION_INJECT_ERROR:
 827         if (action_count[ACTION_INJECT_ERROR] == 1) {
 828             QSIMPLEQ_INIT(&s->active_rules);
 829         }
 830         QSIMPLEQ_INSERT_HEAD(&s->active_rules, rule, active_next);
 831         break;
 832
 833     case ACTION_SET_STATE:
 834         *new_state = rule->options.set_state.new_state;
 835         break;
 836
 837     case ACTION_SUSPEND:
 838         suspend_request(bs, rule);
 839         break;
 840     }
 841 }
 842
 843 static void blkdebug_debug_event(BlockDriverState *bs, BlkdebugEvent event)
 844 {
 845     BDRVBlkdebugState *s = bs->opaque;
 846     struct BlkdebugRule *rule, *next;
 847     int new_state;
 848     int actions_count[ACTION__MAX] = { 0 };
 849
 850     assert((int)event >= 0 && event < BLKDBG__MAX);
 851
 852     WITH_QEMU_LOCK_GUARD(&s->lock) {
 853         new_state = s->state;
 854         QLIST_FOREACH_SAFE(rule, &s->rules[event], next, next) {
 855             process_rule(bs, rule, actions_count, &new_state);
 856         }
 857         s->state = new_state;
 858     }
 859
 860     while (actions_count[ACTION_SUSPEND] > 0) {
 861         qemu_coroutine_yield();
 862         actions_count[ACTION_SUSPEND]--;
 863     }
 864 }
 865
 866 static int blkdebug_debug_breakpoint(BlockDriverState *bs, const char *event,
 867                                      const char *tag)
 868 {
 869     BDRVBlkdebugState *s = bs->opaque;
 870     struct BlkdebugRule *rule;
 871     int blkdebug_event;
 872
 873     blkdebug_event = qapi_enum_parse(&BlkdebugEvent_lookup, event, -1, NULL);
 874     if (blkdebug_event < 0) {
 875         return -ENOENT;
 876     }
 877
 878     rule = g_malloc(sizeof(*rule));
 879     *rule = (struct BlkdebugRule) {
 880         .event  = blkdebug_event,
 881         .action = ACTION_SUSPEND,
 882         .state  = 0,
 883         .options.suspend.tag = g_strdup(tag),
 884     };
 885
 886     qemu_mutex_lock(&s->lock);
 887     QLIST_INSERT_HEAD(&s->rules[blkdebug_event], rule, next);
 888     qemu_mutex_unlock(&s->lock);
 889
 890     return 0;
 891 }
 892
 893 /* Called with lock held. May temporarily release lock. */
 894 static int resume_req_by_tag(BDRVBlkdebugState *s, const char *tag, bool all)
 895 {
 896     BlkdebugSuspendedReq *r;
 897
 898 retry:
 899     /*
 900      * No need for _SAFE, since a different coroutine can remove another node
 901      * (not the current one) in this list, and when the current one is removed
 902      * the iteration starts back from beginning anyways.
 903      */
 904     QLIST_FOREACH(r, &s->suspended_reqs, next) {
 905         if (!strcmp(r->tag, tag)) {
 906             Coroutine *co = r->co;
 907
 908             if (!qtest_enabled()) {
 909                 printf("blkdebug: Resuming request '%s'\n", r->tag);
 910             }
 911
 912             QLIST_REMOVE(r, next);
 913             g_free(r->tag);
 914             g_free(r);
 915
 916             qemu_mutex_unlock(&s->lock);
 917             qemu_coroutine_enter(co);
 918             qemu_mutex_lock(&s->lock);
 919
 920             if (all) {
 921                 goto retry;
 922             }
 923             return 0;
 924         }
 925     }
 926     return -ENOENT;
 927 }
 928
 929 static int blkdebug_debug_resume(BlockDriverState *bs, const char *tag)
 930 {
 931     BDRVBlkdebugState *s = bs->opaque;
 932     QEMU_LOCK_GUARD(&s->lock);
 933     return resume_req_by_tag(s, tag, false);
 934 }
 935
 936 static int blkdebug_debug_remove_breakpoint(BlockDriverState *bs,
 937                                             const char *tag)
 938 {
 939     BDRVBlkdebugState *s = bs->opaque;
 940     BlkdebugRule *rule, *next;
 941     int i, ret = -ENOENT;
 942
 943     QEMU_LOCK_GUARD(&s->lock);
 944     for (i = 0; i < BLKDBG__MAX; i++) {
 945         QLIST_FOREACH_SAFE(rule, &s->rules[i], next, next) {
 946             if (rule->action == ACTION_SUSPEND &&
 947                 !strcmp(rule->options.suspend.tag, tag)) {
 948                 remove_rule(rule);
 949                 ret = 0;
 950             }
 951         }
 952     }
 953     if (resume_req_by_tag(s, tag, true) == 0) {
 954         ret = 0;
 955     }
 956     return ret;
 957 }
 958
 959 static bool blkdebug_debug_is_suspended(BlockDriverState *bs, const char *tag)
 960 {
 961     BDRVBlkdebugState *s = bs->opaque;
 962     BlkdebugSuspendedReq *r;
 963
 964     QEMU_LOCK_GUARD(&s->lock);
 965     QLIST_FOREACH(r, &s->suspended_reqs, next) {
 966         if (!strcmp(r->tag, tag)) {
 967             return true;
 968         }
 969     }
 970     return false;
 971 }
 972
 973 static int64_t blkdebug_getlength(BlockDriverState *bs)
 974 {
 975     return bdrv_getlength(bs->file->bs);
 976 }
 977
 978 static void blkdebug_refresh_filename(BlockDriverState *bs)
 979 {
 980     BDRVBlkdebugState *s = bs->opaque;
 981     const QDictEntry *e;
 982     int ret;
 983
 984     if (!bs->file->bs->exact_filename[0]) {
 985         return;
 986     }
 987
 988     for (e = qdict_first(bs->full_open_options); e;
 989          e = qdict_next(bs->full_open_options, e))
 990     {
 991         /* Real child options are under "image", but "x-image" may
 992          * contain a filename */
 993         if (strcmp(qdict_entry_key(e), "config") &&
 994             strcmp(qdict_entry_key(e), "image") &&
 995             strcmp(qdict_entry_key(e), "x-image") &&
 996             strcmp(qdict_entry_key(e), "driver"))
 997         {
 998             return;
 999         }
1000     }
1001
1002     ret = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
1003                    "blkdebug:%s:%s",
1004                    s->config_file ?: "", bs->file->bs->exact_filename);
1005     if (ret >= sizeof(bs->exact_filename)) {
1006         /* An overflow makes the filename unusable, so do not report any */
1007         bs->exact_filename[0] = 0;
1008     }
1009 }
1010
1011 static void blkdebug_refresh_limits(BlockDriverState *bs, Error **errp)
1012 {
1013     BDRVBlkdebugState *s = bs->opaque;
1014
1015     if (s->align) {
1016         bs->bl.request_alignment = s->align;
1017     }
1018     if (s->max_transfer) {
1019         bs->bl.max_transfer = s->max_transfer;
1020     }
1021     if (s->opt_write_zero) {
1022         bs->bl.pwrite_zeroes_alignment = s->opt_write_zero;
1023     }
1024     if (s->max_write_zero) {
1025         bs->bl.max_pwrite_zeroes = s->max_write_zero;
1026     }
1027     if (s->opt_discard) {
1028         bs->bl.pdiscard_alignment = s->opt_discard;
1029     }
1030     if (s->max_discard) {
1031         bs->bl.max_pdiscard = s->max_discard;
1032     }
1033 }
1034
1035 static int blkdebug_reopen_prepare(BDRVReopenState *reopen_state,
1036                                    BlockReopenQueue *queue, Error **errp)
1037 {
1038     return 0;
1039 }
1040
1041 static void blkdebug_child_perm(BlockDriverState *bs, BdrvChild *c,
1042                                 BdrvChildRole role,
1043                                 BlockReopenQueue *reopen_queue,
1044                                 uint64_t perm, uint64_t shared,
1045                                 uint64_t *nperm, uint64_t *nshared)
1046 {
1047     BDRVBlkdebugState *s = bs->opaque;
1048
1049     bdrv_default_perms(bs, c, role, reopen_queue,
1050                        perm, shared, nperm, nshared);
1051
1052     *nperm |= s->take_child_perms;
1053     *nshared &= ~s->unshare_child_perms;
1054 }
1055
1056 static const char *const blkdebug_strong_runtime_opts[] = {
1057     "config",
1058     "inject-error.",
1059     "set-state.",
1060     "align",
1061     "max-transfer",
1062     "opt-write-zero",
1063     "max-write-zero",
1064     "opt-discard",
1065     "max-discard",
1066
1067     NULL
1068 };
1069
1070 static BlockDriver bdrv_blkdebug = {
1071     .format_name            = "blkdebug",
1072     .protocol_name          = "blkdebug",
1073     .instance_size          = sizeof(BDRVBlkdebugState),
1074     .is_filter              = true,
1075
1076     .bdrv_parse_filename    = blkdebug_parse_filename,
1077     .bdrv_file_open         = blkdebug_open,
1078     .bdrv_close             = blkdebug_close,
1079     .bdrv_reopen_prepare    = blkdebug_reopen_prepare,
1080     .bdrv_child_perm        = blkdebug_child_perm,
1081
1082     .bdrv_getlength         = blkdebug_getlength,
1083     .bdrv_refresh_filename  = blkdebug_refresh_filename,
1084     .bdrv_refresh_limits    = blkdebug_refresh_limits,
1085
1086     .bdrv_co_preadv         = blkdebug_co_preadv,
1087     .bdrv_co_pwritev        = blkdebug_co_pwritev,
1088     .bdrv_co_flush_to_disk  = blkdebug_co_flush,
1089     .bdrv_co_pwrite_zeroes  = blkdebug_co_pwrite_zeroes,
1090     .bdrv_co_pdiscard       = blkdebug_co_pdiscard,
1091     .bdrv_co_block_status   = blkdebug_co_block_status,
1092
1093     .bdrv_debug_event           = blkdebug_debug_event,
1094     .bdrv_debug_breakpoint      = blkdebug_debug_breakpoint,
1095     .bdrv_debug_remove_breakpoint
1096                                 = blkdebug_debug_remove_breakpoint,
1097     .bdrv_debug_resume          = blkdebug_debug_resume,
1098     .bdrv_debug_is_suspended    = blkdebug_debug_is_suspended,
1099
1100     .strong_runtime_opts        = blkdebug_strong_runtime_opts,
1101 };
1102
1103 static void bdrv_blkdebug_init(void)
1104 {
1105     bdrv_register(&bdrv_blkdebug);
1106 }
1107
1108 block_init(bdrv_blkdebug_init);