accel/tcg: Improve can_do_io management
[qemu/armbru.git] / block / blkdebug.c
blob9da8c9eddc2fdc49a6bf431a1fe096ab3e40381f
1 /*
2 * Block protocol for I/O error injection
4 * Copyright (C) 2016-2017 Red Hat, Inc.
5 * Copyright (c) 2010 Kevin Wolf <kwolf@redhat.com>
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:
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
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.
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-io.h"
31 #include "block/block_int.h"
32 #include "block/qdict.h"
33 #include "qemu/module.h"
34 #include "qemu/option.h"
35 #include "qapi/qapi-visit-block-core.h"
36 #include "qapi/qmp/qdict.h"
37 #include "qapi/qmp/qlist.h"
38 #include "qapi/qmp/qstring.h"
39 #include "qapi/qobject-input-visitor.h"
40 #include "sysemu/qtest.h"
42 /* All APIs are thread-safe */
44 typedef struct BDRVBlkdebugState {
45 /* IN: initialized in blkdebug_open() and never changed */
46 uint64_t align;
47 uint64_t max_transfer;
48 uint64_t opt_write_zero;
49 uint64_t max_write_zero;
50 uint64_t opt_discard;
51 uint64_t max_discard;
52 char *config_file; /* For blkdebug_refresh_filename() */
53 /* initialized in blkdebug_parse_perms() */
54 uint64_t take_child_perms;
55 uint64_t unshare_child_perms;
57 /* State. Protected by lock */
58 int state;
59 QLIST_HEAD(, BlkdebugRule) rules[BLKDBG__MAX];
60 QSIMPLEQ_HEAD(, BlkdebugRule) active_rules;
61 QLIST_HEAD(, BlkdebugSuspendedReq) suspended_reqs;
62 QemuMutex lock;
63 } BDRVBlkdebugState;
65 typedef struct BlkdebugAIOCB {
66 BlockAIOCB common;
67 int ret;
68 } BlkdebugAIOCB;
70 typedef struct BlkdebugSuspendedReq {
71 /* IN: initialized in suspend_request() */
72 Coroutine *co;
73 char *tag;
75 /* List entry protected BDRVBlkdebugState's lock */
76 QLIST_ENTRY(BlkdebugSuspendedReq) next;
77 } BlkdebugSuspendedReq;
79 enum {
80 ACTION_INJECT_ERROR,
81 ACTION_SET_STATE,
82 ACTION_SUSPEND,
83 ACTION__MAX,
86 typedef struct BlkdebugRule {
87 /* IN: initialized in add_rule() or blkdebug_debug_breakpoint() */
88 BlkdebugEvent event;
89 int action;
90 int state;
91 union {
92 struct {
93 uint64_t iotype_mask;
94 int error;
95 int immediately;
96 int once;
97 int64_t offset;
98 } inject;
99 struct {
100 int new_state;
101 } set_state;
102 struct {
103 char *tag;
104 } suspend;
105 } options;
107 /* List entries protected BDRVBlkdebugState's lock */
108 QLIST_ENTRY(BlkdebugRule) next;
109 QSIMPLEQ_ENTRY(BlkdebugRule) active_next;
110 } BlkdebugRule;
112 QEMU_BUILD_BUG_MSG(BLKDEBUG_IO_TYPE__MAX > 64,
113 "BlkdebugIOType mask does not fit into an uint64_t");
115 static QemuOptsList inject_error_opts = {
116 .name = "inject-error",
117 .head = QTAILQ_HEAD_INITIALIZER(inject_error_opts.head),
118 .desc = {
120 .name = "event",
121 .type = QEMU_OPT_STRING,
124 .name = "state",
125 .type = QEMU_OPT_NUMBER,
128 .name = "iotype",
129 .type = QEMU_OPT_STRING,
132 .name = "errno",
133 .type = QEMU_OPT_NUMBER,
136 .name = "sector",
137 .type = QEMU_OPT_NUMBER,
140 .name = "once",
141 .type = QEMU_OPT_BOOL,
144 .name = "immediately",
145 .type = QEMU_OPT_BOOL,
147 { /* end of list */ }
151 static QemuOptsList set_state_opts = {
152 .name = "set-state",
153 .head = QTAILQ_HEAD_INITIALIZER(set_state_opts.head),
154 .desc = {
156 .name = "event",
157 .type = QEMU_OPT_STRING,
160 .name = "state",
161 .type = QEMU_OPT_NUMBER,
164 .name = "new_state",
165 .type = QEMU_OPT_NUMBER,
167 { /* end of list */ }
171 static QemuOptsList *config_groups[] = {
172 &inject_error_opts,
173 &set_state_opts,
174 NULL
177 struct add_rule_data {
178 BDRVBlkdebugState *s;
179 int action;
182 static int add_rule(void *opaque, QemuOpts *opts, Error **errp)
184 struct add_rule_data *d = opaque;
185 BDRVBlkdebugState *s = d->s;
186 const char *event_name;
187 int event;
188 struct BlkdebugRule *rule;
189 int64_t sector;
190 BlkdebugIOType iotype;
191 Error *local_error = NULL;
193 /* Find the right event for the rule */
194 event_name = qemu_opt_get(opts, "event");
195 if (!event_name) {
196 error_setg(errp, "Missing event name for rule");
197 return -1;
199 event = qapi_enum_parse(&BlkdebugEvent_lookup, event_name, -1, errp);
200 if (event < 0) {
201 return -1;
204 /* Set attributes common for all actions */
205 rule = g_malloc0(sizeof(*rule));
206 *rule = (struct BlkdebugRule) {
207 .event = event,
208 .action = d->action,
209 .state = qemu_opt_get_number(opts, "state", 0),
212 /* Parse action-specific options */
213 switch (d->action) {
214 case ACTION_INJECT_ERROR:
215 rule->options.inject.error = qemu_opt_get_number(opts, "errno", EIO);
216 rule->options.inject.once = qemu_opt_get_bool(opts, "once", 0);
217 rule->options.inject.immediately =
218 qemu_opt_get_bool(opts, "immediately", 0);
219 sector = qemu_opt_get_number(opts, "sector", -1);
220 rule->options.inject.offset =
221 sector == -1 ? -1 : sector * BDRV_SECTOR_SIZE;
223 iotype = qapi_enum_parse(&BlkdebugIOType_lookup,
224 qemu_opt_get(opts, "iotype"),
225 BLKDEBUG_IO_TYPE__MAX, &local_error);
226 if (local_error) {
227 error_propagate(errp, local_error);
228 g_free(rule);
229 return -1;
231 if (iotype != BLKDEBUG_IO_TYPE__MAX) {
232 rule->options.inject.iotype_mask = (1ull << iotype);
233 } else {
234 /* Apply the default */
235 rule->options.inject.iotype_mask =
236 (1ull << BLKDEBUG_IO_TYPE_READ)
237 | (1ull << BLKDEBUG_IO_TYPE_WRITE)
238 | (1ull << BLKDEBUG_IO_TYPE_WRITE_ZEROES)
239 | (1ull << BLKDEBUG_IO_TYPE_DISCARD)
240 | (1ull << BLKDEBUG_IO_TYPE_FLUSH);
243 break;
245 case ACTION_SET_STATE:
246 rule->options.set_state.new_state =
247 qemu_opt_get_number(opts, "new_state", 0);
248 break;
250 case ACTION_SUSPEND:
251 rule->options.suspend.tag =
252 g_strdup(qemu_opt_get(opts, "tag"));
253 break;
256 /* Add the rule */
257 qemu_mutex_lock(&s->lock);
258 QLIST_INSERT_HEAD(&s->rules[event], rule, next);
259 qemu_mutex_unlock(&s->lock);
261 return 0;
264 /* Called with lock held or from .bdrv_close */
265 static void remove_rule(BlkdebugRule *rule)
267 switch (rule->action) {
268 case ACTION_INJECT_ERROR:
269 case ACTION_SET_STATE:
270 break;
271 case ACTION_SUSPEND:
272 g_free(rule->options.suspend.tag);
273 break;
276 QLIST_REMOVE(rule, next);
277 g_free(rule);
280 static int read_config(BDRVBlkdebugState *s, const char *filename,
281 QDict *options, Error **errp)
283 FILE *f = NULL;
284 int ret;
285 struct add_rule_data d;
286 Error *local_err = NULL;
288 if (filename) {
289 f = fopen(filename, "r");
290 if (f == NULL) {
291 error_setg_errno(errp, errno, "Could not read blkdebug config file");
292 return -errno;
295 ret = qemu_config_parse(f, config_groups, filename, errp);
296 if (ret < 0) {
297 goto fail;
301 if (!qemu_config_parse_qdict(options, config_groups, errp)) {
302 ret = -EINVAL;
303 goto fail;
306 d.s = s;
307 d.action = ACTION_INJECT_ERROR;
308 qemu_opts_foreach(&inject_error_opts, add_rule, &d, &local_err);
309 if (local_err) {
310 error_propagate(errp, local_err);
311 ret = -EINVAL;
312 goto fail;
315 d.action = ACTION_SET_STATE;
316 qemu_opts_foreach(&set_state_opts, add_rule, &d, &local_err);
317 if (local_err) {
318 error_propagate(errp, local_err);
319 ret = -EINVAL;
320 goto fail;
323 ret = 0;
324 fail:
325 qemu_opts_reset(&inject_error_opts);
326 qemu_opts_reset(&set_state_opts);
327 if (f) {
328 fclose(f);
330 return ret;
333 /* Valid blkdebug filenames look like blkdebug:path/to/config:path/to/image */
334 static void blkdebug_parse_filename(const char *filename, QDict *options,
335 Error **errp)
337 const char *c;
339 /* Parse the blkdebug: prefix */
340 if (!strstart(filename, "blkdebug:", &filename)) {
341 /* There was no prefix; therefore, all options have to be already
342 present in the QDict (except for the filename) */
343 qdict_put_str(options, "x-image", filename);
344 return;
347 /* Parse config file path */
348 c = strchr(filename, ':');
349 if (c == NULL) {
350 error_setg(errp, "blkdebug requires both config file and image path");
351 return;
354 if (c != filename) {
355 QString *config_path;
356 config_path = qstring_from_substr(filename, 0, c - filename);
357 qdict_put(options, "config", config_path);
360 /* TODO Allow multi-level nesting and set file.filename here */
361 filename = c + 1;
362 qdict_put_str(options, "x-image", filename);
365 static int blkdebug_parse_perm_list(uint64_t *dest, QDict *options,
366 const char *prefix, Error **errp)
368 int ret = 0;
369 QDict *subqdict = NULL;
370 QObject *crumpled_subqdict = NULL;
371 Visitor *v = NULL;
372 BlockPermissionList *perm_list = NULL, *element;
374 *dest = 0;
376 qdict_extract_subqdict(options, &subqdict, prefix);
377 if (!qdict_size(subqdict)) {
378 goto out;
381 crumpled_subqdict = qdict_crumple(subqdict, errp);
382 if (!crumpled_subqdict) {
383 ret = -EINVAL;
384 goto out;
387 v = qobject_input_visitor_new(crumpled_subqdict);
388 if (!visit_type_BlockPermissionList(v, NULL, &perm_list, errp)) {
389 ret = -EINVAL;
390 goto out;
393 for (element = perm_list; element; element = element->next) {
394 *dest |= bdrv_qapi_perm_to_blk_perm(element->value);
397 out:
398 qapi_free_BlockPermissionList(perm_list);
399 visit_free(v);
400 qobject_unref(subqdict);
401 qobject_unref(crumpled_subqdict);
402 return ret;
405 static int blkdebug_parse_perms(BDRVBlkdebugState *s, QDict *options,
406 Error **errp)
408 int ret;
410 ret = blkdebug_parse_perm_list(&s->take_child_perms, options,
411 "take-child-perms.", errp);
412 if (ret < 0) {
413 return ret;
416 ret = blkdebug_parse_perm_list(&s->unshare_child_perms, options,
417 "unshare-child-perms.", errp);
418 if (ret < 0) {
419 return ret;
422 return 0;
425 static QemuOptsList runtime_opts = {
426 .name = "blkdebug",
427 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
428 .desc = {
430 .name = "config",
431 .type = QEMU_OPT_STRING,
432 .help = "Path to the configuration file",
435 .name = "x-image",
436 .type = QEMU_OPT_STRING,
437 .help = "[internal use only, will be removed]",
440 .name = "align",
441 .type = QEMU_OPT_SIZE,
442 .help = "Required alignment in bytes",
445 .name = "max-transfer",
446 .type = QEMU_OPT_SIZE,
447 .help = "Maximum transfer size in bytes",
450 .name = "opt-write-zero",
451 .type = QEMU_OPT_SIZE,
452 .help = "Optimum write zero alignment in bytes",
455 .name = "max-write-zero",
456 .type = QEMU_OPT_SIZE,
457 .help = "Maximum write zero size in bytes",
460 .name = "opt-discard",
461 .type = QEMU_OPT_SIZE,
462 .help = "Optimum discard alignment in bytes",
465 .name = "max-discard",
466 .type = QEMU_OPT_SIZE,
467 .help = "Maximum discard size in bytes",
469 { /* end of list */ }
473 static int blkdebug_open(BlockDriverState *bs, QDict *options, int flags,
474 Error **errp)
476 BDRVBlkdebugState *s = bs->opaque;
477 QemuOpts *opts;
478 int ret;
479 uint64_t align;
481 qemu_mutex_init(&s->lock);
482 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
483 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
484 ret = -EINVAL;
485 goto out;
488 /* Read rules from config file or command line options */
489 s->config_file = g_strdup(qemu_opt_get(opts, "config"));
490 ret = read_config(s, s->config_file, options, errp);
491 if (ret) {
492 goto out;
495 /* Set initial state */
496 s->state = 1;
498 /* Parse permissions modifiers before opening the image file */
499 ret = blkdebug_parse_perms(s, options, errp);
500 if (ret < 0) {
501 goto out;
504 /* Open the image file */
505 ret = bdrv_open_file_child(qemu_opt_get(opts, "x-image"), options, "image",
506 bs, errp);
507 if (ret < 0) {
508 goto out;
511 bdrv_graph_rdlock_main_loop();
513 bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
514 (BDRV_REQ_FUA & bs->file->bs->supported_write_flags);
515 bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
516 ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) &
517 bs->file->bs->supported_zero_flags);
518 ret = -EINVAL;
520 /* Set alignment overrides */
521 s->align = qemu_opt_get_size(opts, "align", 0);
522 if (s->align && (s->align >= INT_MAX || !is_power_of_2(s->align))) {
523 error_setg(errp, "Cannot meet constraints with align %" PRIu64,
524 s->align);
525 goto out_rdlock;
527 align = MAX(s->align, bs->file->bs->bl.request_alignment);
529 s->max_transfer = qemu_opt_get_size(opts, "max-transfer", 0);
530 if (s->max_transfer &&
531 (s->max_transfer >= INT_MAX ||
532 !QEMU_IS_ALIGNED(s->max_transfer, align))) {
533 error_setg(errp, "Cannot meet constraints with max-transfer %" PRIu64,
534 s->max_transfer);
535 goto out_rdlock;
538 s->opt_write_zero = qemu_opt_get_size(opts, "opt-write-zero", 0);
539 if (s->opt_write_zero &&
540 (s->opt_write_zero >= INT_MAX ||
541 !QEMU_IS_ALIGNED(s->opt_write_zero, align))) {
542 error_setg(errp, "Cannot meet constraints with opt-write-zero %" PRIu64,
543 s->opt_write_zero);
544 goto out_rdlock;
547 s->max_write_zero = qemu_opt_get_size(opts, "max-write-zero", 0);
548 if (s->max_write_zero &&
549 (s->max_write_zero >= INT_MAX ||
550 !QEMU_IS_ALIGNED(s->max_write_zero,
551 MAX(s->opt_write_zero, align)))) {
552 error_setg(errp, "Cannot meet constraints with max-write-zero %" PRIu64,
553 s->max_write_zero);
554 goto out_rdlock;
557 s->opt_discard = qemu_opt_get_size(opts, "opt-discard", 0);
558 if (s->opt_discard &&
559 (s->opt_discard >= INT_MAX ||
560 !QEMU_IS_ALIGNED(s->opt_discard, align))) {
561 error_setg(errp, "Cannot meet constraints with opt-discard %" PRIu64,
562 s->opt_discard);
563 goto out_rdlock;
566 s->max_discard = qemu_opt_get_size(opts, "max-discard", 0);
567 if (s->max_discard &&
568 (s->max_discard >= INT_MAX ||
569 !QEMU_IS_ALIGNED(s->max_discard,
570 MAX(s->opt_discard, align)))) {
571 error_setg(errp, "Cannot meet constraints with max-discard %" PRIu64,
572 s->max_discard);
573 goto out_rdlock;
576 bdrv_debug_event(bs, BLKDBG_NONE);
578 ret = 0;
579 out_rdlock:
580 bdrv_graph_rdunlock_main_loop();
581 out:
582 if (ret < 0) {
583 qemu_mutex_destroy(&s->lock);
584 g_free(s->config_file);
586 qemu_opts_del(opts);
587 return ret;
590 static int coroutine_fn rule_check(BlockDriverState *bs, uint64_t offset,
591 uint64_t bytes, BlkdebugIOType iotype)
593 BDRVBlkdebugState *s = bs->opaque;
594 BlkdebugRule *rule = NULL;
595 int error;
596 bool immediately;
598 qemu_mutex_lock(&s->lock);
599 QSIMPLEQ_FOREACH(rule, &s->active_rules, active_next) {
600 uint64_t inject_offset = rule->options.inject.offset;
602 if ((inject_offset == -1 ||
603 (bytes && inject_offset >= offset &&
604 inject_offset < offset + bytes)) &&
605 (rule->options.inject.iotype_mask & (1ull << iotype)))
607 break;
611 if (!rule || !rule->options.inject.error) {
612 qemu_mutex_unlock(&s->lock);
613 return 0;
616 immediately = rule->options.inject.immediately;
617 error = rule->options.inject.error;
619 if (rule->options.inject.once) {
620 QSIMPLEQ_REMOVE(&s->active_rules, rule, BlkdebugRule, active_next);
621 remove_rule(rule);
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();
630 return -error;
633 static int coroutine_fn GRAPH_RDLOCK
634 blkdebug_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
635 QEMUIOVector *qiov, BdrvRequestFlags flags)
637 int err;
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);
646 err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_READ);
647 if (err) {
648 return err;
651 return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
654 static int coroutine_fn GRAPH_RDLOCK
655 blkdebug_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
656 QEMUIOVector *qiov, BdrvRequestFlags flags)
658 int err;
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);
667 err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_WRITE);
668 if (err) {
669 return err;
672 return bdrv_co_pwritev(bs->file, offset, bytes, qiov, flags);
675 static int GRAPH_RDLOCK coroutine_fn blkdebug_co_flush(BlockDriverState *bs)
677 int err = rule_check(bs, 0, 0, BLKDEBUG_IO_TYPE_FLUSH);
679 if (err) {
680 return err;
683 return bdrv_co_flush(bs->file->bs);
686 static int coroutine_fn GRAPH_RDLOCK
687 blkdebug_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes,
688 BdrvRequestFlags flags)
690 uint32_t align = MAX(bs->bl.request_alignment,
691 bs->bl.pwrite_zeroes_alignment);
692 int err;
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;
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);
711 err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_WRITE_ZEROES);
712 if (err) {
713 return err;
716 return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags);
719 static int coroutine_fn GRAPH_RDLOCK
720 blkdebug_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
722 uint32_t align = bs->bl.pdiscard_alignment;
723 int err;
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;
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));
741 if (bs->bl.max_pdiscard) {
742 assert(bytes <= bs->bl.max_pdiscard);
745 err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_DISCARD);
746 if (err) {
747 return err;
750 return bdrv_co_pdiscard(bs->file, offset, bytes);
753 static int coroutine_fn GRAPH_RDLOCK
754 blkdebug_co_block_status(BlockDriverState *bs, bool want_zero, int64_t offset,
755 int64_t bytes, int64_t *pnum, int64_t *map,
756 BlockDriverState **file)
758 int err;
760 assert(QEMU_IS_ALIGNED(offset | bytes, bs->bl.request_alignment));
762 err = rule_check(bs, offset, bytes, BLKDEBUG_IO_TYPE_BLOCK_STATUS);
763 if (err) {
764 return err;
767 assert(bs->file && bs->file->bs);
768 *pnum = bytes;
769 *map = offset;
770 *file = bs->file->bs;
771 return BDRV_BLOCK_RAW | BDRV_BLOCK_OFFSET_VALID;
774 static void blkdebug_close(BlockDriverState *bs)
776 BDRVBlkdebugState *s = bs->opaque;
777 BlkdebugRule *rule, *next;
778 int i;
780 for (i = 0; i < BLKDBG__MAX; i++) {
781 QLIST_FOREACH_SAFE(rule, &s->rules[i], next, next) {
782 remove_rule(rule);
786 g_free(s->config_file);
787 qemu_mutex_destroy(&s->lock);
790 /* Called with lock held. */
791 static void suspend_request(BlockDriverState *bs, BlkdebugRule *rule)
793 BDRVBlkdebugState *s = bs->opaque;
794 BlkdebugSuspendedReq *r;
796 r = g_new(BlkdebugSuspendedReq, 1);
798 r->co = qemu_coroutine_self();
799 r->tag = g_strdup(rule->options.suspend.tag);
801 remove_rule(rule);
802 QLIST_INSERT_HEAD(&s->suspended_reqs, r, next);
804 if (!qtest_enabled()) {
805 printf("blkdebug: Suspended request '%s'\n", r->tag);
809 /* Called with lock held. */
810 static void process_rule(BlockDriverState *bs, struct BlkdebugRule *rule,
811 int *action_count, int *new_state)
813 BDRVBlkdebugState *s = bs->opaque;
815 /* Only process rules for the current state */
816 if (rule->state && rule->state != s->state) {
817 return;
820 /* Take the action */
821 action_count[rule->action]++;
822 switch (rule->action) {
823 case ACTION_INJECT_ERROR:
824 if (action_count[ACTION_INJECT_ERROR] == 1) {
825 QSIMPLEQ_INIT(&s->active_rules);
827 QSIMPLEQ_INSERT_HEAD(&s->active_rules, rule, active_next);
828 break;
830 case ACTION_SET_STATE:
831 *new_state = rule->options.set_state.new_state;
832 break;
834 case ACTION_SUSPEND:
835 suspend_request(bs, rule);
836 break;
840 static void coroutine_fn
841 blkdebug_co_debug_event(BlockDriverState *bs, BlkdebugEvent event)
843 BDRVBlkdebugState *s = bs->opaque;
844 struct BlkdebugRule *rule, *next;
845 int new_state;
846 int actions_count[ACTION__MAX] = { 0 };
848 assert((int)event >= 0 && event < BLKDBG__MAX);
850 WITH_QEMU_LOCK_GUARD(&s->lock) {
851 new_state = s->state;
852 QLIST_FOREACH_SAFE(rule, &s->rules[event], next, next) {
853 process_rule(bs, rule, actions_count, &new_state);
855 s->state = new_state;
858 while (actions_count[ACTION_SUSPEND] > 0) {
859 qemu_coroutine_yield();
860 actions_count[ACTION_SUSPEND]--;
864 static int blkdebug_debug_breakpoint(BlockDriverState *bs, const char *event,
865 const char *tag)
867 BDRVBlkdebugState *s = bs->opaque;
868 struct BlkdebugRule *rule;
869 int blkdebug_event;
871 blkdebug_event = qapi_enum_parse(&BlkdebugEvent_lookup, event, -1, NULL);
872 if (blkdebug_event < 0) {
873 return -ENOENT;
876 rule = g_malloc(sizeof(*rule));
877 *rule = (struct BlkdebugRule) {
878 .event = blkdebug_event,
879 .action = ACTION_SUSPEND,
880 .state = 0,
881 .options.suspend.tag = g_strdup(tag),
884 qemu_mutex_lock(&s->lock);
885 QLIST_INSERT_HEAD(&s->rules[blkdebug_event], rule, next);
886 qemu_mutex_unlock(&s->lock);
888 return 0;
891 /* Called with lock held. May temporarily release lock. */
892 static int resume_req_by_tag(BDRVBlkdebugState *s, const char *tag, bool all)
894 BlkdebugSuspendedReq *r;
896 retry:
898 * No need for _SAFE, since a different coroutine can remove another node
899 * (not the current one) in this list, and when the current one is removed
900 * the iteration starts back from beginning anyways.
902 QLIST_FOREACH(r, &s->suspended_reqs, next) {
903 if (!strcmp(r->tag, tag)) {
904 Coroutine *co = r->co;
906 if (!qtest_enabled()) {
907 printf("blkdebug: Resuming request '%s'\n", r->tag);
910 QLIST_REMOVE(r, next);
911 g_free(r->tag);
912 g_free(r);
914 qemu_mutex_unlock(&s->lock);
915 qemu_coroutine_enter(co);
916 qemu_mutex_lock(&s->lock);
918 if (all) {
919 goto retry;
921 return 0;
924 return -ENOENT;
927 static int blkdebug_debug_resume(BlockDriverState *bs, const char *tag)
929 BDRVBlkdebugState *s = bs->opaque;
930 QEMU_LOCK_GUARD(&s->lock);
931 return resume_req_by_tag(s, tag, false);
934 static int blkdebug_debug_remove_breakpoint(BlockDriverState *bs,
935 const char *tag)
937 BDRVBlkdebugState *s = bs->opaque;
938 BlkdebugRule *rule, *next;
939 int i, ret = -ENOENT;
941 QEMU_LOCK_GUARD(&s->lock);
942 for (i = 0; i < BLKDBG__MAX; i++) {
943 QLIST_FOREACH_SAFE(rule, &s->rules[i], next, next) {
944 if (rule->action == ACTION_SUSPEND &&
945 !strcmp(rule->options.suspend.tag, tag)) {
946 remove_rule(rule);
947 ret = 0;
951 if (resume_req_by_tag(s, tag, true) == 0) {
952 ret = 0;
954 return ret;
957 static bool blkdebug_debug_is_suspended(BlockDriverState *bs, const char *tag)
959 BDRVBlkdebugState *s = bs->opaque;
960 BlkdebugSuspendedReq *r;
962 QEMU_LOCK_GUARD(&s->lock);
963 QLIST_FOREACH(r, &s->suspended_reqs, next) {
964 if (!strcmp(r->tag, tag)) {
965 return true;
968 return false;
971 static int64_t coroutine_fn GRAPH_RDLOCK
972 blkdebug_co_getlength(BlockDriverState *bs)
974 return bdrv_co_getlength(bs->file->bs);
977 static void GRAPH_RDLOCK blkdebug_refresh_filename(BlockDriverState *bs)
979 BDRVBlkdebugState *s = bs->opaque;
980 const QDictEntry *e;
981 int ret;
983 if (!bs->file->bs->exact_filename[0]) {
984 return;
987 for (e = qdict_first(bs->full_open_options); e;
988 e = qdict_next(bs->full_open_options, e))
990 /* Real child options are under "image", but "x-image" may
991 * contain a filename */
992 if (strcmp(qdict_entry_key(e), "config") &&
993 strcmp(qdict_entry_key(e), "image") &&
994 strcmp(qdict_entry_key(e), "x-image") &&
995 strcmp(qdict_entry_key(e), "driver"))
997 return;
1001 ret = snprintf(bs->exact_filename, sizeof(bs->exact_filename),
1002 "blkdebug:%s:%s",
1003 s->config_file ?: "", bs->file->bs->exact_filename);
1004 if (ret >= sizeof(bs->exact_filename)) {
1005 /* An overflow makes the filename unusable, so do not report any */
1006 bs->exact_filename[0] = 0;
1010 static void blkdebug_refresh_limits(BlockDriverState *bs, Error **errp)
1012 BDRVBlkdebugState *s = bs->opaque;
1014 if (s->align) {
1015 bs->bl.request_alignment = s->align;
1017 if (s->max_transfer) {
1018 bs->bl.max_transfer = s->max_transfer;
1020 if (s->opt_write_zero) {
1021 bs->bl.pwrite_zeroes_alignment = s->opt_write_zero;
1023 if (s->max_write_zero) {
1024 bs->bl.max_pwrite_zeroes = s->max_write_zero;
1026 if (s->opt_discard) {
1027 bs->bl.pdiscard_alignment = s->opt_discard;
1029 if (s->max_discard) {
1030 bs->bl.max_pdiscard = s->max_discard;
1034 static int blkdebug_reopen_prepare(BDRVReopenState *reopen_state,
1035 BlockReopenQueue *queue, Error **errp)
1037 return 0;
1040 static void blkdebug_child_perm(BlockDriverState *bs, BdrvChild *c,
1041 BdrvChildRole role,
1042 BlockReopenQueue *reopen_queue,
1043 uint64_t perm, uint64_t shared,
1044 uint64_t *nperm, uint64_t *nshared)
1046 BDRVBlkdebugState *s = bs->opaque;
1048 bdrv_default_perms(bs, c, role, reopen_queue,
1049 perm, shared, nperm, nshared);
1051 *nperm |= s->take_child_perms;
1052 *nshared &= ~s->unshare_child_perms;
1055 static const char *const blkdebug_strong_runtime_opts[] = {
1056 "config",
1057 "inject-error.",
1058 "set-state.",
1059 "align",
1060 "max-transfer",
1061 "opt-write-zero",
1062 "max-write-zero",
1063 "opt-discard",
1064 "max-discard",
1066 NULL
1069 static BlockDriver bdrv_blkdebug = {
1070 .format_name = "blkdebug",
1071 .protocol_name = "blkdebug",
1072 .instance_size = sizeof(BDRVBlkdebugState),
1073 .is_filter = true,
1075 .bdrv_parse_filename = blkdebug_parse_filename,
1076 .bdrv_file_open = blkdebug_open,
1077 .bdrv_close = blkdebug_close,
1078 .bdrv_reopen_prepare = blkdebug_reopen_prepare,
1079 .bdrv_child_perm = blkdebug_child_perm,
1081 .bdrv_co_getlength = blkdebug_co_getlength,
1082 .bdrv_refresh_filename = blkdebug_refresh_filename,
1083 .bdrv_refresh_limits = blkdebug_refresh_limits,
1085 .bdrv_co_preadv = blkdebug_co_preadv,
1086 .bdrv_co_pwritev = blkdebug_co_pwritev,
1087 .bdrv_co_flush_to_disk = blkdebug_co_flush,
1088 .bdrv_co_pwrite_zeroes = blkdebug_co_pwrite_zeroes,
1089 .bdrv_co_pdiscard = blkdebug_co_pdiscard,
1090 .bdrv_co_block_status = blkdebug_co_block_status,
1092 .bdrv_co_debug_event = blkdebug_co_debug_event,
1093 .bdrv_debug_breakpoint = blkdebug_debug_breakpoint,
1094 .bdrv_debug_remove_breakpoint
1095 = blkdebug_debug_remove_breakpoint,
1096 .bdrv_debug_resume = blkdebug_debug_resume,
1097 .bdrv_debug_is_suspended = blkdebug_debug_is_suspended,
1099 .strong_runtime_opts = blkdebug_strong_runtime_opts,
1102 static void bdrv_blkdebug_init(void)
1104 bdrv_register(&bdrv_blkdebug);
1107 block_init(bdrv_blkdebug_init);