seccomp: add some basic shared memory syscalls to the whitelist
[qemu.git] / block / mirror.c
blob2932bab27a40e2ebb1bf8dc41a389d85db5708ba
1 /*
2 * Image mirroring
4 * Copyright Red Hat, Inc. 2012
6 * Authors:
7 * Paolo Bonzini <pbonzini@redhat.com>
9 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
10 * See the COPYING.LIB file in the top-level directory.
14 #include "trace.h"
15 #include "block/blockjob.h"
16 #include "block/block_int.h"
17 #include "qemu/ratelimit.h"
18 #include "qemu/bitmap.h"
20 #define SLICE_TIME 100000000ULL /* ns */
21 #define MAX_IN_FLIGHT 16
23 /* The mirroring buffer is a list of granularity-sized chunks.
24 * Free chunks are organized in a list.
26 typedef struct MirrorBuffer {
27 QSIMPLEQ_ENTRY(MirrorBuffer) next;
28 } MirrorBuffer;
30 typedef struct MirrorBlockJob {
31 BlockJob common;
32 RateLimit limit;
33 BlockDriverState *target;
34 BlockDriverState *base;
35 bool is_none_mode;
36 BlockdevOnError on_source_error, on_target_error;
37 bool synced;
38 bool should_complete;
39 int64_t sector_num;
40 int64_t granularity;
41 size_t buf_size;
42 unsigned long *cow_bitmap;
43 BdrvDirtyBitmap *dirty_bitmap;
44 HBitmapIter hbi;
45 uint8_t *buf;
46 QSIMPLEQ_HEAD(, MirrorBuffer) buf_free;
47 int buf_free_count;
49 unsigned long *in_flight_bitmap;
50 int in_flight;
51 int ret;
52 } MirrorBlockJob;
54 typedef struct MirrorOp {
55 MirrorBlockJob *s;
56 QEMUIOVector qiov;
57 int64_t sector_num;
58 int nb_sectors;
59 } MirrorOp;
61 static BlockErrorAction mirror_error_action(MirrorBlockJob *s, bool read,
62 int error)
64 s->synced = false;
65 if (read) {
66 return block_job_error_action(&s->common, s->common.bs,
67 s->on_source_error, true, error);
68 } else {
69 return block_job_error_action(&s->common, s->target,
70 s->on_target_error, false, error);
74 static void mirror_iteration_done(MirrorOp *op, int ret)
76 MirrorBlockJob *s = op->s;
77 struct iovec *iov;
78 int64_t chunk_num;
79 int i, nb_chunks, sectors_per_chunk;
81 trace_mirror_iteration_done(s, op->sector_num, op->nb_sectors, ret);
83 s->in_flight--;
84 iov = op->qiov.iov;
85 for (i = 0; i < op->qiov.niov; i++) {
86 MirrorBuffer *buf = (MirrorBuffer *) iov[i].iov_base;
87 QSIMPLEQ_INSERT_TAIL(&s->buf_free, buf, next);
88 s->buf_free_count++;
91 sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
92 chunk_num = op->sector_num / sectors_per_chunk;
93 nb_chunks = op->nb_sectors / sectors_per_chunk;
94 bitmap_clear(s->in_flight_bitmap, chunk_num, nb_chunks);
95 if (s->cow_bitmap && ret >= 0) {
96 bitmap_set(s->cow_bitmap, chunk_num, nb_chunks);
99 g_slice_free(MirrorOp, op);
100 qemu_coroutine_enter(s->common.co, NULL);
103 static void mirror_write_complete(void *opaque, int ret)
105 MirrorOp *op = opaque;
106 MirrorBlockJob *s = op->s;
107 if (ret < 0) {
108 BlockDriverState *source = s->common.bs;
109 BlockErrorAction action;
111 bdrv_set_dirty(source, op->sector_num, op->nb_sectors);
112 action = mirror_error_action(s, false, -ret);
113 if (action == BDRV_ACTION_REPORT && s->ret >= 0) {
114 s->ret = ret;
117 mirror_iteration_done(op, ret);
120 static void mirror_read_complete(void *opaque, int ret)
122 MirrorOp *op = opaque;
123 MirrorBlockJob *s = op->s;
124 if (ret < 0) {
125 BlockDriverState *source = s->common.bs;
126 BlockErrorAction action;
128 bdrv_set_dirty(source, op->sector_num, op->nb_sectors);
129 action = mirror_error_action(s, true, -ret);
130 if (action == BDRV_ACTION_REPORT && s->ret >= 0) {
131 s->ret = ret;
134 mirror_iteration_done(op, ret);
135 return;
137 bdrv_aio_writev(s->target, op->sector_num, &op->qiov, op->nb_sectors,
138 mirror_write_complete, op);
141 static void coroutine_fn mirror_iteration(MirrorBlockJob *s)
143 BlockDriverState *source = s->common.bs;
144 int nb_sectors, sectors_per_chunk, nb_chunks;
145 int64_t end, sector_num, next_chunk, next_sector, hbitmap_next_sector;
146 MirrorOp *op;
148 s->sector_num = hbitmap_iter_next(&s->hbi);
149 if (s->sector_num < 0) {
150 bdrv_dirty_iter_init(source, s->dirty_bitmap, &s->hbi);
151 s->sector_num = hbitmap_iter_next(&s->hbi);
152 trace_mirror_restart_iter(s,
153 bdrv_get_dirty_count(source, s->dirty_bitmap));
154 assert(s->sector_num >= 0);
157 hbitmap_next_sector = s->sector_num;
158 sector_num = s->sector_num;
159 sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
160 end = s->common.len >> BDRV_SECTOR_BITS;
162 /* Extend the QEMUIOVector to include all adjacent blocks that will
163 * be copied in this operation.
165 * We have to do this if we have no backing file yet in the destination,
166 * and the cluster size is very large. Then we need to do COW ourselves.
167 * The first time a cluster is copied, copy it entirely. Note that,
168 * because both the granularity and the cluster size are powers of two,
169 * the number of sectors to copy cannot exceed one cluster.
171 * We also want to extend the QEMUIOVector to include more adjacent
172 * dirty blocks if possible, to limit the number of I/O operations and
173 * run efficiently even with a small granularity.
175 nb_chunks = 0;
176 nb_sectors = 0;
177 next_sector = sector_num;
178 next_chunk = sector_num / sectors_per_chunk;
180 /* Wait for I/O to this cluster (from a previous iteration) to be done. */
181 while (test_bit(next_chunk, s->in_flight_bitmap)) {
182 trace_mirror_yield_in_flight(s, sector_num, s->in_flight);
183 qemu_coroutine_yield();
186 do {
187 int added_sectors, added_chunks;
189 if (!bdrv_get_dirty(source, s->dirty_bitmap, next_sector) ||
190 test_bit(next_chunk, s->in_flight_bitmap)) {
191 assert(nb_sectors > 0);
192 break;
195 added_sectors = sectors_per_chunk;
196 if (s->cow_bitmap && !test_bit(next_chunk, s->cow_bitmap)) {
197 bdrv_round_to_clusters(s->target,
198 next_sector, added_sectors,
199 &next_sector, &added_sectors);
201 /* On the first iteration, the rounding may make us copy
202 * sectors before the first dirty one.
204 if (next_sector < sector_num) {
205 assert(nb_sectors == 0);
206 sector_num = next_sector;
207 next_chunk = next_sector / sectors_per_chunk;
211 added_sectors = MIN(added_sectors, end - (sector_num + nb_sectors));
212 added_chunks = (added_sectors + sectors_per_chunk - 1) / sectors_per_chunk;
214 /* When doing COW, it may happen that there is not enough space for
215 * a full cluster. Wait if that is the case.
217 while (nb_chunks == 0 && s->buf_free_count < added_chunks) {
218 trace_mirror_yield_buf_busy(s, nb_chunks, s->in_flight);
219 qemu_coroutine_yield();
221 if (s->buf_free_count < nb_chunks + added_chunks) {
222 trace_mirror_break_buf_busy(s, nb_chunks, s->in_flight);
223 break;
226 /* We have enough free space to copy these sectors. */
227 bitmap_set(s->in_flight_bitmap, next_chunk, added_chunks);
229 nb_sectors += added_sectors;
230 nb_chunks += added_chunks;
231 next_sector += added_sectors;
232 next_chunk += added_chunks;
233 } while (next_sector < end);
235 /* Allocate a MirrorOp that is used as an AIO callback. */
236 op = g_slice_new(MirrorOp);
237 op->s = s;
238 op->sector_num = sector_num;
239 op->nb_sectors = nb_sectors;
241 /* Now make a QEMUIOVector taking enough granularity-sized chunks
242 * from s->buf_free.
244 qemu_iovec_init(&op->qiov, nb_chunks);
245 next_sector = sector_num;
246 while (nb_chunks-- > 0) {
247 MirrorBuffer *buf = QSIMPLEQ_FIRST(&s->buf_free);
248 QSIMPLEQ_REMOVE_HEAD(&s->buf_free, next);
249 s->buf_free_count--;
250 qemu_iovec_add(&op->qiov, buf, s->granularity);
252 /* Advance the HBitmapIter in parallel, so that we do not examine
253 * the same sector twice.
255 if (next_sector > hbitmap_next_sector
256 && bdrv_get_dirty(source, s->dirty_bitmap, next_sector)) {
257 hbitmap_next_sector = hbitmap_iter_next(&s->hbi);
260 next_sector += sectors_per_chunk;
263 bdrv_reset_dirty(source, sector_num, nb_sectors);
265 /* Copy the dirty cluster. */
266 s->in_flight++;
267 trace_mirror_one_iteration(s, sector_num, nb_sectors);
268 bdrv_aio_readv(source, sector_num, &op->qiov, nb_sectors,
269 mirror_read_complete, op);
272 static void mirror_free_init(MirrorBlockJob *s)
274 int granularity = s->granularity;
275 size_t buf_size = s->buf_size;
276 uint8_t *buf = s->buf;
278 assert(s->buf_free_count == 0);
279 QSIMPLEQ_INIT(&s->buf_free);
280 while (buf_size != 0) {
281 MirrorBuffer *cur = (MirrorBuffer *)buf;
282 QSIMPLEQ_INSERT_TAIL(&s->buf_free, cur, next);
283 s->buf_free_count++;
284 buf_size -= granularity;
285 buf += granularity;
289 static void mirror_drain(MirrorBlockJob *s)
291 while (s->in_flight > 0) {
292 qemu_coroutine_yield();
296 static void coroutine_fn mirror_run(void *opaque)
298 MirrorBlockJob *s = opaque;
299 BlockDriverState *bs = s->common.bs;
300 int64_t sector_num, end, sectors_per_chunk, length;
301 uint64_t last_pause_ns;
302 BlockDriverInfo bdi;
303 char backing_filename[1024];
304 int ret = 0;
305 int n;
307 if (block_job_is_cancelled(&s->common)) {
308 goto immediate_exit;
311 s->common.len = bdrv_getlength(bs);
312 if (s->common.len <= 0) {
313 block_job_completed(&s->common, s->common.len);
314 return;
317 length = (bdrv_getlength(bs) + s->granularity - 1) / s->granularity;
318 s->in_flight_bitmap = bitmap_new(length);
320 /* If we have no backing file yet in the destination, we cannot let
321 * the destination do COW. Instead, we copy sectors around the
322 * dirty data if needed. We need a bitmap to do that.
324 bdrv_get_backing_filename(s->target, backing_filename,
325 sizeof(backing_filename));
326 if (backing_filename[0] && !s->target->backing_hd) {
327 bdrv_get_info(s->target, &bdi);
328 if (s->granularity < bdi.cluster_size) {
329 s->buf_size = MAX(s->buf_size, bdi.cluster_size);
330 s->cow_bitmap = bitmap_new(length);
334 end = s->common.len >> BDRV_SECTOR_BITS;
335 s->buf = qemu_blockalign(bs, s->buf_size);
336 sectors_per_chunk = s->granularity >> BDRV_SECTOR_BITS;
337 mirror_free_init(s);
339 if (!s->is_none_mode) {
340 /* First part, loop on the sectors and initialize the dirty bitmap. */
341 BlockDriverState *base = s->base;
342 for (sector_num = 0; sector_num < end; ) {
343 int64_t next = (sector_num | (sectors_per_chunk - 1)) + 1;
344 ret = bdrv_is_allocated_above(bs, base,
345 sector_num, next - sector_num, &n);
347 if (ret < 0) {
348 goto immediate_exit;
351 assert(n > 0);
352 if (ret == 1) {
353 bdrv_set_dirty(bs, sector_num, n);
354 sector_num = next;
355 } else {
356 sector_num += n;
361 bdrv_dirty_iter_init(bs, s->dirty_bitmap, &s->hbi);
362 last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
363 for (;;) {
364 uint64_t delay_ns;
365 int64_t cnt;
366 bool should_complete;
368 if (s->ret < 0) {
369 ret = s->ret;
370 goto immediate_exit;
373 cnt = bdrv_get_dirty_count(bs, s->dirty_bitmap);
375 /* Note that even when no rate limit is applied we need to yield
376 * periodically with no pending I/O so that qemu_aio_flush() returns.
377 * We do so every SLICE_TIME nanoseconds, or when there is an error,
378 * or when the source is clean, whichever comes first.
380 if (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - last_pause_ns < SLICE_TIME &&
381 s->common.iostatus == BLOCK_DEVICE_IO_STATUS_OK) {
382 if (s->in_flight == MAX_IN_FLIGHT || s->buf_free_count == 0 ||
383 (cnt == 0 && s->in_flight > 0)) {
384 trace_mirror_yield(s, s->in_flight, s->buf_free_count, cnt);
385 qemu_coroutine_yield();
386 continue;
387 } else if (cnt != 0) {
388 mirror_iteration(s);
389 continue;
393 should_complete = false;
394 if (s->in_flight == 0 && cnt == 0) {
395 trace_mirror_before_flush(s);
396 ret = bdrv_flush(s->target);
397 if (ret < 0) {
398 if (mirror_error_action(s, false, -ret) == BDRV_ACTION_REPORT) {
399 goto immediate_exit;
401 } else {
402 /* We're out of the streaming phase. From now on, if the job
403 * is cancelled we will actually complete all pending I/O and
404 * report completion. This way, block-job-cancel will leave
405 * the target in a consistent state.
407 s->common.offset = end * BDRV_SECTOR_SIZE;
408 if (!s->synced) {
409 block_job_ready(&s->common);
410 s->synced = true;
413 should_complete = s->should_complete ||
414 block_job_is_cancelled(&s->common);
415 cnt = bdrv_get_dirty_count(bs, s->dirty_bitmap);
419 if (cnt == 0 && should_complete) {
420 /* The dirty bitmap is not updated while operations are pending.
421 * If we're about to exit, wait for pending operations before
422 * calling bdrv_get_dirty_count(bs), or we may exit while the
423 * source has dirty data to copy!
425 * Note that I/O can be submitted by the guest while
426 * mirror_populate runs.
428 trace_mirror_before_drain(s, cnt);
429 bdrv_drain_all();
430 cnt = bdrv_get_dirty_count(bs, s->dirty_bitmap);
433 ret = 0;
434 trace_mirror_before_sleep(s, cnt, s->synced);
435 if (!s->synced) {
436 /* Publish progress */
437 s->common.offset = (end - cnt) * BDRV_SECTOR_SIZE;
439 if (s->common.speed) {
440 delay_ns = ratelimit_calculate_delay(&s->limit, sectors_per_chunk);
441 } else {
442 delay_ns = 0;
445 block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
446 if (block_job_is_cancelled(&s->common)) {
447 break;
449 } else if (!should_complete) {
450 delay_ns = (s->in_flight == 0 && cnt == 0 ? SLICE_TIME : 0);
451 block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
452 } else if (cnt == 0) {
453 /* The two disks are in sync. Exit and report successful
454 * completion.
456 assert(QLIST_EMPTY(&bs->tracked_requests));
457 s->common.cancelled = false;
458 break;
460 last_pause_ns = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
463 immediate_exit:
464 if (s->in_flight > 0) {
465 /* We get here only if something went wrong. Either the job failed,
466 * or it was cancelled prematurely so that we do not guarantee that
467 * the target is a copy of the source.
469 assert(ret < 0 || (!s->synced && block_job_is_cancelled(&s->common)));
470 mirror_drain(s);
473 assert(s->in_flight == 0);
474 qemu_vfree(s->buf);
475 g_free(s->cow_bitmap);
476 g_free(s->in_flight_bitmap);
477 bdrv_release_dirty_bitmap(bs, s->dirty_bitmap);
478 bdrv_iostatus_disable(s->target);
479 if (s->should_complete && ret == 0) {
480 if (bdrv_get_flags(s->target) != bdrv_get_flags(s->common.bs)) {
481 bdrv_reopen(s->target, bdrv_get_flags(s->common.bs), NULL);
483 bdrv_swap(s->target, s->common.bs);
484 if (s->common.driver->job_type == BLOCK_JOB_TYPE_COMMIT) {
485 /* drop the bs loop chain formed by the swap: break the loop then
486 * trigger the unref from the top one */
487 BlockDriverState *p = s->base->backing_hd;
488 s->base->backing_hd = NULL;
489 bdrv_unref(p);
492 bdrv_unref(s->target);
493 block_job_completed(&s->common, ret);
496 static void mirror_set_speed(BlockJob *job, int64_t speed, Error **errp)
498 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
500 if (speed < 0) {
501 error_set(errp, QERR_INVALID_PARAMETER, "speed");
502 return;
504 ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
507 static void mirror_iostatus_reset(BlockJob *job)
509 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
511 bdrv_iostatus_reset(s->target);
514 static void mirror_complete(BlockJob *job, Error **errp)
516 MirrorBlockJob *s = container_of(job, MirrorBlockJob, common);
517 Error *local_err = NULL;
518 int ret;
520 ret = bdrv_open_backing_file(s->target, NULL, &local_err);
521 if (ret < 0) {
522 char backing_filename[PATH_MAX];
523 bdrv_get_full_backing_filename(s->target, backing_filename,
524 sizeof(backing_filename));
525 error_propagate(errp, local_err);
526 return;
528 if (!s->synced) {
529 error_set(errp, QERR_BLOCK_JOB_NOT_READY, job->bs->device_name);
530 return;
533 s->should_complete = true;
534 block_job_resume(job);
537 static const BlockJobDriver mirror_job_driver = {
538 .instance_size = sizeof(MirrorBlockJob),
539 .job_type = BLOCK_JOB_TYPE_MIRROR,
540 .set_speed = mirror_set_speed,
541 .iostatus_reset= mirror_iostatus_reset,
542 .complete = mirror_complete,
545 static const BlockJobDriver commit_active_job_driver = {
546 .instance_size = sizeof(MirrorBlockJob),
547 .job_type = BLOCK_JOB_TYPE_COMMIT,
548 .set_speed = mirror_set_speed,
549 .iostatus_reset
550 = mirror_iostatus_reset,
551 .complete = mirror_complete,
554 static void mirror_start_job(BlockDriverState *bs, BlockDriverState *target,
555 int64_t speed, int64_t granularity,
556 int64_t buf_size,
557 BlockdevOnError on_source_error,
558 BlockdevOnError on_target_error,
559 BlockDriverCompletionFunc *cb,
560 void *opaque, Error **errp,
561 const BlockJobDriver *driver,
562 bool is_none_mode, BlockDriverState *base)
564 MirrorBlockJob *s;
566 if (granularity == 0) {
567 /* Choose the default granularity based on the target file's cluster
568 * size, clamped between 4k and 64k. */
569 BlockDriverInfo bdi;
570 if (bdrv_get_info(target, &bdi) >= 0 && bdi.cluster_size != 0) {
571 granularity = MAX(4096, bdi.cluster_size);
572 granularity = MIN(65536, granularity);
573 } else {
574 granularity = 65536;
578 assert ((granularity & (granularity - 1)) == 0);
580 if ((on_source_error == BLOCKDEV_ON_ERROR_STOP ||
581 on_source_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
582 !bdrv_iostatus_is_enabled(bs)) {
583 error_set(errp, QERR_INVALID_PARAMETER, "on-source-error");
584 return;
588 s = block_job_create(driver, bs, speed, cb, opaque, errp);
589 if (!s) {
590 return;
593 s->on_source_error = on_source_error;
594 s->on_target_error = on_target_error;
595 s->target = target;
596 s->is_none_mode = is_none_mode;
597 s->base = base;
598 s->granularity = granularity;
599 s->buf_size = MAX(buf_size, granularity);
601 s->dirty_bitmap = bdrv_create_dirty_bitmap(bs, granularity);
602 bdrv_set_enable_write_cache(s->target, true);
603 bdrv_set_on_error(s->target, on_target_error, on_target_error);
604 bdrv_iostatus_enable(s->target);
605 s->common.co = qemu_coroutine_create(mirror_run);
606 trace_mirror_start(bs, s, s->common.co, opaque);
607 qemu_coroutine_enter(s->common.co, s);
610 void mirror_start(BlockDriverState *bs, BlockDriverState *target,
611 int64_t speed, int64_t granularity, int64_t buf_size,
612 MirrorSyncMode mode, BlockdevOnError on_source_error,
613 BlockdevOnError on_target_error,
614 BlockDriverCompletionFunc *cb,
615 void *opaque, Error **errp)
617 bool is_none_mode;
618 BlockDriverState *base;
620 is_none_mode = mode == MIRROR_SYNC_MODE_NONE;
621 base = mode == MIRROR_SYNC_MODE_TOP ? bs->backing_hd : NULL;
622 mirror_start_job(bs, target, speed, granularity, buf_size,
623 on_source_error, on_target_error, cb, opaque, errp,
624 &mirror_job_driver, is_none_mode, base);
627 void commit_active_start(BlockDriverState *bs, BlockDriverState *base,
628 int64_t speed,
629 BlockdevOnError on_error,
630 BlockDriverCompletionFunc *cb,
631 void *opaque, Error **errp)
633 if (bdrv_reopen(base, bs->open_flags, errp)) {
634 return;
636 bdrv_ref(base);
637 mirror_start_job(bs, base, speed, 0, 0,
638 on_error, on_error, cb, opaque, errp,
639 &commit_active_job_driver, false, base);