xref: /openbmc/qemu/block/blkdebug.c (revision a8e2ab5d)
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, errp);
283         if (ret < 0) {
284             goto fail;
285         }
286     }
287 
288     qemu_config_parse_qdict(options, config_groups, &local_err);
289     if (local_err) {
290         error_propagate(errp, local_err);
291         ret = -EINVAL;
292         goto fail;
293     }
294 
295     d.s = s;
296     d.action = ACTION_INJECT_ERROR;
297     qemu_opts_foreach(&inject_error_opts, add_rule, &d, &local_err);
298     if (local_err) {
299         error_propagate(errp, local_err);
300         ret = -EINVAL;
301         goto fail;
302     }
303 
304     d.action = ACTION_SET_STATE;
305     qemu_opts_foreach(&set_state_opts, add_rule, &d, &local_err);
306     if (local_err) {
307         error_propagate(errp, local_err);
308         ret = -EINVAL;
309         goto fail;
310     }
311 
312     ret = 0;
313 fail:
314     qemu_opts_reset(&inject_error_opts);
315     qemu_opts_reset(&set_state_opts);
316     if (f) {
317         fclose(f);
318     }
319     return ret;
320 }
321 
322 /* Valid blkdebug filenames look like blkdebug:path/to/config:path/to/image */
323 static void blkdebug_parse_filename(const char *filename, QDict *options,
324                                     Error **errp)
325 {
326     const char *c;
327 
328     /* Parse the blkdebug: prefix */
329     if (!strstart(filename, "blkdebug:", &filename)) {
330         /* There was no prefix; therefore, all options have to be already
331            present in the QDict (except for the filename) */
332         qdict_put_str(options, "x-image", filename);
333         return;
334     }
335 
336     /* Parse config file path */
337     c = strchr(filename, ':');
338     if (c == NULL) {
339         error_setg(errp, "blkdebug requires both config file and image path");
340         return;
341     }
342 
343     if (c != filename) {
344         QString *config_path;
345         config_path = qstring_from_substr(filename, 0, c - filename);
346         qdict_put(options, "config", config_path);
347     }
348 
349     /* TODO Allow multi-level nesting and set file.filename here */
350     filename = c + 1;
351     qdict_put_str(options, "x-image", filename);
352 }
353 
354 static int blkdebug_parse_perm_list(uint64_t *dest, QDict *options,
355                                     const char *prefix, Error **errp)
356 {
357     int ret = 0;
358     QDict *subqdict = NULL;
359     QObject *crumpled_subqdict = NULL;
360     Visitor *v = NULL;
361     BlockPermissionList *perm_list = NULL, *element;
362 
363     *dest = 0;
364 
365     qdict_extract_subqdict(options, &subqdict, prefix);
366     if (!qdict_size(subqdict)) {
367         goto out;
368     }
369 
370     crumpled_subqdict = qdict_crumple(subqdict, errp);
371     if (!crumpled_subqdict) {
372         ret = -EINVAL;
373         goto out;
374     }
375 
376     v = qobject_input_visitor_new(crumpled_subqdict);
377     if (!visit_type_BlockPermissionList(v, NULL, &perm_list, errp)) {
378         ret = -EINVAL;
379         goto out;
380     }
381 
382     for (element = perm_list; element; element = element->next) {
383         *dest |= bdrv_qapi_perm_to_blk_perm(element->value);
384     }
385 
386 out:
387     qapi_free_BlockPermissionList(perm_list);
388     visit_free(v);
389     qobject_unref(subqdict);
390     qobject_unref(crumpled_subqdict);
391     return ret;
392 }
393 
394 static int blkdebug_parse_perms(BDRVBlkdebugState *s, QDict *options,
395                                 Error **errp)
396 {
397     int ret;
398 
399     ret = blkdebug_parse_perm_list(&s->take_child_perms, options,
400                                    "take-child-perms.", errp);
401     if (ret < 0) {
402         return ret;
403     }
404 
405     ret = blkdebug_parse_perm_list(&s->unshare_child_perms, options,
406                                    "unshare-child-perms.", errp);
407     if (ret < 0) {
408         return ret;
409     }
410 
411     return 0;
412 }
413 
414 static QemuOptsList runtime_opts = {
415     .name = "blkdebug",
416     .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
417     .desc = {
418         {
419             .name = "config",
420             .type = QEMU_OPT_STRING,
421             .help = "Path to the configuration file",
422         },
423         {
424             .name = "x-image",
425             .type = QEMU_OPT_STRING,
426             .help = "[internal use only, will be removed]",
427         },
428         {
429             .name = "align",
430             .type = QEMU_OPT_SIZE,
431             .help = "Required alignment in bytes",
432         },
433         {
434             .name = "max-transfer",
435             .type = QEMU_OPT_SIZE,
436             .help = "Maximum transfer size in bytes",
437         },
438         {
439             .name = "opt-write-zero",
440             .type = QEMU_OPT_SIZE,
441             .help = "Optimum write zero alignment in bytes",
442         },
443         {
444             .name = "max-write-zero",
445             .type = QEMU_OPT_SIZE,
446             .help = "Maximum write zero size in bytes",
447         },
448         {
449             .name = "opt-discard",
450             .type = QEMU_OPT_SIZE,
451             .help = "Optimum discard alignment in bytes",
452         },
453         {
454             .name = "max-discard",
455             .type = QEMU_OPT_SIZE,
456             .help = "Maximum discard size in bytes",
457         },
458         { /* end of list */ }
459     },
460 };
461 
462 static int blkdebug_open(BlockDriverState *bs, QDict *options, int flags,
463                          Error **errp)
464 {
465     BDRVBlkdebugState *s = bs->opaque;
466     QemuOpts *opts;
467     int ret;
468     uint64_t align;
469 
470     opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
471     if (!qemu_opts_absorb_qdict(opts, options, errp)) {
472         ret = -EINVAL;
473         goto out;
474     }
475 
476     /* Read rules from config file or command line options */
477     s->config_file = g_strdup(qemu_opt_get(opts, "config"));
478     ret = read_config(s, s->config_file, options, errp);
479     if (ret) {
480         goto out;
481     }
482 
483     /* Set initial state */
484     s->state = 1;
485 
486     /* Parse permissions modifiers before opening the image file */
487     ret = blkdebug_parse_perms(s, options, errp);
488     if (ret < 0) {
489         goto out;
490     }
491 
492     /* Open the image file */
493     bs->file = bdrv_open_child(qemu_opt_get(opts, "x-image"), options, "image",
494                                bs, &child_of_bds,
495                                BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
496                                false, errp);
497     if (!bs->file) {
498         ret = -EINVAL;
499         goto out;
500     }
501 
502     bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
503         (BDRV_REQ_FUA & bs->file->bs->supported_write_flags);
504     bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
505         ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) &
506             bs->file->bs->supported_zero_flags);
507     ret = -EINVAL;
508 
509     /* Set alignment overrides */
510     s->align = qemu_opt_get_size(opts, "align", 0);
511     if (s->align && (s->align >= INT_MAX || !is_power_of_2(s->align))) {
512         error_setg(errp, "Cannot meet constraints with align %" PRIu64,
513                    s->align);
514         goto out;
515     }
516     align = MAX(s->align, bs->file->bs->bl.request_alignment);
517 
518     s->max_transfer = qemu_opt_get_size(opts, "max-transfer", 0);
519     if (s->max_transfer &&
520         (s->max_transfer >= INT_MAX ||
521          !QEMU_IS_ALIGNED(s->max_transfer, align))) {
522         error_setg(errp, "Cannot meet constraints with max-transfer %" PRIu64,
523                    s->max_transfer);
524         goto out;
525     }
526 
527     s->opt_write_zero = qemu_opt_get_size(opts, "opt-write-zero", 0);
528     if (s->opt_write_zero &&
529         (s->opt_write_zero >= INT_MAX ||
530          !QEMU_IS_ALIGNED(s->opt_write_zero, align))) {
531         error_setg(errp, "Cannot meet constraints with opt-write-zero %" PRIu64,
532                    s->opt_write_zero);
533         goto out;
534     }
535 
536     s->max_write_zero = qemu_opt_get_size(opts, "max-write-zero", 0);
537     if (s->max_write_zero &&
538         (s->max_write_zero >= INT_MAX ||
539          !QEMU_IS_ALIGNED(s->max_write_zero,
540                           MAX(s->opt_write_zero, align)))) {
541         error_setg(errp, "Cannot meet constraints with max-write-zero %" PRIu64,
542                    s->max_write_zero);
543         goto out;
544     }
545 
546     s->opt_discard = qemu_opt_get_size(opts, "opt-discard", 0);
547     if (s->opt_discard &&
548         (s->opt_discard >= INT_MAX ||
549          !QEMU_IS_ALIGNED(s->opt_discard, align))) {
550         error_setg(errp, "Cannot meet constraints with opt-discard %" PRIu64,
551                    s->opt_discard);
552         goto out;
553     }
554 
555     s->max_discard = qemu_opt_get_size(opts, "max-discard", 0);
556     if (s->max_discard &&
557         (s->max_discard >= INT_MAX ||
558          !QEMU_IS_ALIGNED(s->max_discard,
559                           MAX(s->opt_discard, align)))) {
560         error_setg(errp, "Cannot meet constraints with max-discard %" PRIu64,
561                    s->max_discard);
562         goto out;
563     }
564 
565     bdrv_debug_event(bs, BLKDBG_NONE);
566 
567     ret = 0;
568 out:
569     if (ret < 0) {
570         g_free(s->config_file);
571     }
572     qemu_opts_del(opts);
573     return ret;
574 }
575 
576 static int rule_check(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
577                       BlkdebugIOType iotype)
578 {
579     BDRVBlkdebugState *s = bs->opaque;
580     BlkdebugRule *rule = NULL;
581     int error;
582     bool immediately;
583 
584     QSIMPLEQ_FOREACH(rule, &s->active_rules, active_next) {
585         uint64_t inject_offset = rule->options.inject.offset;
586 
587         if ((inject_offset == -1 ||
588              (bytes && inject_offset >= offset &&
589               inject_offset < offset + bytes)) &&
590             (rule->options.inject.iotype_mask & (1ull << iotype)))
591         {
592             break;
593         }
594     }
595 
596     if (!rule || !rule->options.inject.error) {
597         return 0;
598     }
599 
600     immediately = rule->options.inject.immediately;
601     error = rule->options.inject.error;
602 
603     if (rule->options.inject.once) {
604         QSIMPLEQ_REMOVE(&s->active_rules, rule, BlkdebugRule, active_next);
605         remove_rule(rule);
606     }
607 
608     if (!immediately) {
609         aio_co_schedule(qemu_get_current_aio_context(), qemu_coroutine_self());
610         qemu_coroutine_yield();
611     }
612 
613     return -error;
614 }
615 
616 static int coroutine_fn
617 blkdebug_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
618                    QEMUIOVector *qiov, int flags)
619 {
620     int err;
621 
622     /* Sanity check block layer guarantees */
623     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
624     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
625     if (bs->bl.max_transfer) {
626         assert(bytes <= bs->bl.max_transfer);
627     }
628 
629     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_READ);
630     if (err) {
631         return err;
632     }
633 
634     return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
635 }
636 
637 static int coroutine_fn
638 blkdebug_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
639                     QEMUIOVector *qiov, int flags)
640 {
641     int err;
642 
643     /* Sanity check block layer guarantees */
644     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
645     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
646     if (bs->bl.max_transfer) {
647         assert(bytes <= bs->bl.max_transfer);
648     }
649 
650     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_WRITE);
651     if (err) {
652         return err;
653     }
654 
655     return bdrv_co_pwritev(bs->file, offset, bytes, qiov, flags);
656 }
657 
658 static int blkdebug_co_flush(BlockDriverState *bs)
659 {
660     int err = rule_check(bs, 0, 0, BLKDEBUG_IO_TYPE_FLUSH);
661 
662     if (err) {
663         return err;
664     }
665 
666     return bdrv_co_flush(bs->file->bs);
667 }
668 
669 static int coroutine_fn blkdebug_co_pwrite_zeroes(BlockDriverState *bs,
670                                                   int64_t offset, int bytes,
671                                                   BdrvRequestFlags flags)
672 {
673     uint32_t align = MAX(bs->bl.request_alignment,
674                          bs->bl.pwrite_zeroes_alignment);
675     int err;
676 
677     /* Only pass through requests that are larger than requested
678      * preferred alignment (so that we test the fallback to writes on
679      * unaligned portions), and check that the block layer never hands
680      * us anything unaligned that crosses an alignment boundary.  */
681     if (bytes < align) {
682         assert(QEMU_IS_ALIGNED(offset, align) ||
683                QEMU_IS_ALIGNED(offset + bytes, align) ||
684                DIV_ROUND_UP(offset, align) ==
685                DIV_ROUND_UP(offset + bytes, align));
686         return -ENOTSUP;
687     }
688     assert(QEMU_IS_ALIGNED(offset, align));
689     assert(QEMU_IS_ALIGNED(bytes, align));
690     if (bs->bl.max_pwrite_zeroes) {
691         assert(bytes <= bs->bl.max_pwrite_zeroes);
692     }
693 
694     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_WRITE_ZEROES);
695     if (err) {
696         return err;
697     }
698 
699     return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags);
700 }
701 
702 static int coroutine_fn blkdebug_co_pdiscard(BlockDriverState *bs,
703                                              int64_t offset, int bytes)
704 {
705     uint32_t align = bs->bl.pdiscard_alignment;
706     int err;
707 
708     /* Only pass through requests that are larger than requested
709      * minimum alignment, and ensure that unaligned requests do not
710      * cross optimum discard boundaries. */
711     if (bytes < bs->bl.request_alignment) {
712         assert(QEMU_IS_ALIGNED(offset, align) ||
713                QEMU_IS_ALIGNED(offset + bytes, align) ||
714                DIV_ROUND_UP(offset, align) ==
715                DIV_ROUND_UP(offset + bytes, align));
716         return -ENOTSUP;
717     }
718     assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
719     assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
720     if (align && bytes >= align) {
721         assert(QEMU_IS_ALIGNED(offset, align));
722         assert(QEMU_IS_ALIGNED(bytes, align));
723     }
724     if (bs->bl.max_pdiscard) {
725         assert(bytes <= bs->bl.max_pdiscard);
726     }
727 
728     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_DISCARD);
729     if (err) {
730         return err;
731     }
732 
733     return bdrv_co_pdiscard(bs->file, offset, bytes);
734 }
735 
736 static int coroutine_fn blkdebug_co_block_status(BlockDriverState *bs,
737                                                  bool want_zero,
738                                                  int64_t offset,
739                                                  int64_t bytes,
740                                                  int64_t *pnum,
741                                                  int64_t *map,
742                                                  BlockDriverState **file)
743 {
744     int err;
745 
746     assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
747 
748     err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_BLOCK_STATUS);
749     if (err) {
750         return err;
751     }
752 
753     assert(bs->file && bs->file->bs);
754     *pnum = bytes;
755     *map = offset;
756     *file = bs->file->bs;
757     return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID;
758 }
759 
760 static void blkdebug_close(BlockDriverState *bs)
761 {
762     BDRVBlkdebugState *s = bs->opaque;
763     BlkdebugRule *rule, *next;
764     int i;
765 
766     for (i = 0; i < BLKDBG__MAX; i++) {
767         QLIST_FOREACH_SAFE(rule, &s->rules[i], next, next) {
768             remove_rule(rule);
769         }
770     }
771 
772     g_free(s->config_file);
773 }
774 
775 static void suspend_request(BlockDriverState *bs, BlkdebugRule *rule)
776 {
777     BDRVBlkdebugState *s = bs->opaque;
778     BlkdebugSuspendedReq r;
779 
780     r = (BlkdebugSuspendedReq) {
781         .co         = qemu_coroutine_self(),
782         .tag        = g_strdup(rule->options.suspend.tag),
783     };
784 
785     remove_rule(rule);
786     QLIST_INSERT_HEAD(&s->suspended_reqs, &r, next);
787 
788     if (!qtest_enabled()) {
789         printf("blkdebug: Suspended request '%s'\n", r.tag);
790     }
791     qemu_coroutine_yield();
792     if (!qtest_enabled()) {
793         printf("blkdebug: Resuming request '%s'\n", r.tag);
794     }
795 
796     QLIST_REMOVE(&r, next);
797     g_free(r.tag);
798 }
799 
800 static bool process_rule(BlockDriverState *bs, struct BlkdebugRule *rule,
801     bool injected)
802 {
803     BDRVBlkdebugState *s = bs->opaque;
804 
805     /* Only process rules for the current state */
806     if (rule->state && rule->state != s->state) {
807         return injected;
808     }
809 
810     /* Take the action */
811     switch (rule->action) {
812     case ACTION_INJECT_ERROR:
813         if (!injected) {
814             QSIMPLEQ_INIT(&s->active_rules);
815             injected = true;
816         }
817         QSIMPLEQ_INSERT_HEAD(&s->active_rules, rule, active_next);
818         break;
819 
820     case ACTION_SET_STATE:
821         s->new_state = rule->options.set_state.new_state;
822         break;
823 
824     case ACTION_SUSPEND:
825         suspend_request(bs, rule);
826         break;
827     }
828     return injected;
829 }
830 
831 static void blkdebug_debug_event(BlockDriverState *bs, BlkdebugEvent event)
832 {
833     BDRVBlkdebugState *s = bs->opaque;
834     struct BlkdebugRule *rule, *next;
835     bool injected;
836 
837     assert((int)event >= 0 && event < BLKDBG__MAX);
838 
839     injected = false;
840     s->new_state = s->state;
841     QLIST_FOREACH_SAFE(rule, &s->rules[event], next, next) {
842         injected = process_rule(bs, rule, injected);
843     }
844     s->state = s->new_state;
845 }
846 
847 static int blkdebug_debug_breakpoint(BlockDriverState *bs, const char *event,
848                                      const char *tag)
849 {
850     BDRVBlkdebugState *s = bs->opaque;
851     struct BlkdebugRule *rule;
852     int blkdebug_event;
853 
854     blkdebug_event = qapi_enum_parse(&BlkdebugEvent_lookup, event, -1, NULL);
855     if (blkdebug_event < 0) {
856         return -ENOENT;
857     }
858 
859     rule = g_malloc(sizeof(*rule));
860     *rule = (struct BlkdebugRule) {
861         .event  = blkdebug_event,
862         .action = ACTION_SUSPEND,
863         .state  = 0,
864         .options.suspend.tag = g_strdup(tag),
865     };
866 
867     QLIST_INSERT_HEAD(&s->rules[blkdebug_event], rule, next);
868 
869     return 0;
870 }
871 
872 static int blkdebug_debug_resume(BlockDriverState *bs, const char *tag)
873 {
874     BDRVBlkdebugState *s = bs->opaque;
875     BlkdebugSuspendedReq *r, *next;
876 
877     QLIST_FOREACH_SAFE(r, &s->suspended_reqs, next, next) {
878         if (!strcmp(r->tag, tag)) {
879             qemu_coroutine_enter(r->co);
880             return 0;
881         }
882     }
883     return -ENOENT;
884 }
885 
886 static int blkdebug_debug_remove_breakpoint(BlockDriverState *bs,
887                                             const char *tag)
888 {
889     BDRVBlkdebugState *s = bs->opaque;
890     BlkdebugSuspendedReq *r, *r_next;
891     BlkdebugRule *rule, *next;
892     int i, ret = -ENOENT;
893 
894     for (i = 0; i < BLKDBG__MAX; i++) {
895         QLIST_FOREACH_SAFE(rule, &s->rules[i], next, next) {
896             if (rule->action == ACTION_SUSPEND &&
897                 !strcmp(rule->options.suspend.tag, tag)) {
898                 remove_rule(rule);
899                 ret = 0;
900             }
901         }
902     }
903     QLIST_FOREACH_SAFE(r, &s->suspended_reqs, next, r_next) {
904         if (!strcmp(r->tag, tag)) {
905             qemu_coroutine_enter(r->co);
906             ret = 0;
907         }
908     }
909     return ret;
910 }
911 
912 static bool blkdebug_debug_is_suspended(BlockDriverState *bs, const char *tag)
913 {
914     BDRVBlkdebugState *s = bs->opaque;
915     BlkdebugSuspendedReq *r;
916 
917     QLIST_FOREACH(r, &s->suspended_reqs, next) {
918         if (!strcmp(r->tag, tag)) {
919             return true;
920         }
921     }
922     return false;
923 }
924 
925 static int64_t blkdebug_getlength(BlockDriverState *bs)
926 {
927     return bdrv_getlength(bs->file->bs);
928 }
929 
930 static void blkdebug_refresh_filename(BlockDriverState *bs)
931 {
932     BDRVBlkdebugState *s = bs->opaque;
933     const QDictEntry *e;
934     int ret;
935 
936     if (!bs->file->bs->exact_filename[0]) {
937         return;
938     }
939 
940     for (e = qdict_first(bs->full_open_options); e;
941          e = qdict_next(bs->full_open_options, e))
942     {
943         /* Real child options are under "image", but "x-image" may
944          * contain a filename */
945         if (strcmp(qdict_entry_key(e), "config") &&
946             strcmp(qdict_entry_key(e), "image") &&
947             strcmp(qdict_entry_key(e), "x-image") &&
948             strcmp(qdict_entry_key(e), "driver"))
949         {
950             return;
951         }
952     }
953 
954     ret = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
955                    "blkdebug:%s:%s",
956                    s->config_file ?: "", bs->file->bs->exact_filename);
957     if (ret >= sizeof(bs->exact_filename)) {
958         /* An overflow makes the filename unusable, so do not report any */
959         bs->exact_filename[0] = 0;
960     }
961 }
962 
963 static void blkdebug_refresh_limits(BlockDriverState *bs, Error **errp)
964 {
965     BDRVBlkdebugState *s = bs->opaque;
966 
967     if (s->align) {
968         bs->bl.request_alignment = s->align;
969     }
970     if (s->max_transfer) {
971         bs->bl.max_transfer = s->max_transfer;
972     }
973     if (s->opt_write_zero) {
974         bs->bl.pwrite_zeroes_alignment = s->opt_write_zero;
975     }
976     if (s->max_write_zero) {
977         bs->bl.max_pwrite_zeroes = s->max_write_zero;
978     }
979     if (s->opt_discard) {
980         bs->bl.pdiscard_alignment = s->opt_discard;
981     }
982     if (s->max_discard) {
983         bs->bl.max_pdiscard = s->max_discard;
984     }
985 }
986 
987 static int blkdebug_reopen_prepare(BDRVReopenState *reopen_state,
988                                    BlockReopenQueue *queue, Error **errp)
989 {
990     return 0;
991 }
992 
993 static void blkdebug_child_perm(BlockDriverState *bs, BdrvChild *c,
994                                 BdrvChildRole role,
995                                 BlockReopenQueue *reopen_queue,
996                                 uint64_t perm, uint64_t shared,
997                                 uint64_t *nperm, uint64_t *nshared)
998 {
999     BDRVBlkdebugState *s = bs->opaque;
1000 
1001     bdrv_default_perms(bs, c, role, reopen_queue,
1002                        perm, shared, nperm, nshared);
1003 
1004     *nperm |= s->take_child_perms;
1005     *nshared &= ~s->unshare_child_perms;
1006 }
1007 
1008 static const char *const blkdebug_strong_runtime_opts[] = {
1009     "config",
1010     "inject-error.",
1011     "set-state.",
1012     "align",
1013     "max-transfer",
1014     "opt-write-zero",
1015     "max-write-zero",
1016     "opt-discard",
1017     "max-discard",
1018 
1019     NULL
1020 };
1021 
1022 static BlockDriver bdrv_blkdebug = {
1023     .format_name            = "blkdebug",
1024     .protocol_name          = "blkdebug",
1025     .instance_size          = sizeof(BDRVBlkdebugState),
1026     .is_filter              = true,
1027 
1028     .bdrv_parse_filename    = blkdebug_parse_filename,
1029     .bdrv_file_open         = blkdebug_open,
1030     .bdrv_close             = blkdebug_close,
1031     .bdrv_reopen_prepare    = blkdebug_reopen_prepare,
1032     .bdrv_child_perm        = blkdebug_child_perm,
1033 
1034     .bdrv_getlength         = blkdebug_getlength,
1035     .bdrv_refresh_filename  = blkdebug_refresh_filename,
1036     .bdrv_refresh_limits    = blkdebug_refresh_limits,
1037 
1038     .bdrv_co_preadv         = blkdebug_co_preadv,
1039     .bdrv_co_pwritev        = blkdebug_co_pwritev,
1040     .bdrv_co_flush_to_disk  = blkdebug_co_flush,
1041     .bdrv_co_pwrite_zeroes  = blkdebug_co_pwrite_zeroes,
1042     .bdrv_co_pdiscard       = blkdebug_co_pdiscard,
1043     .bdrv_co_block_status   = blkdebug_co_block_status,
1044 
1045     .bdrv_debug_event           = blkdebug_debug_event,
1046     .bdrv_debug_breakpoint      = blkdebug_debug_breakpoint,
1047     .bdrv_debug_remove_breakpoint
1048                                 = blkdebug_debug_remove_breakpoint,
1049     .bdrv_debug_resume          = blkdebug_debug_resume,
1050     .bdrv_debug_is_suspended    = blkdebug_debug_is_suspended,
1051 
1052     .strong_runtime_opts        = blkdebug_strong_runtime_opts,
1053 };
1054 
1055 static void bdrv_blkdebug_init(void)
1056 {
1057     bdrv_register(&bdrv_blkdebug);
1058 }
1059 
1060 block_init(bdrv_blkdebug_init);
1061