hw/arm/exynos4210: Put combiners into state struct
[qemu/ar7.git] / block / copy-before-write.c
bloba8a06fdc09b6a9e2dcdfe82f54c74179fb367ced
1 /*
2 * copy-before-write filter driver
4 * The driver performs Copy-Before-Write (CBW) operation: it is injected above
5 * some node, and before each write it copies _old_ data to the target node.
7 * Copyright (c) 2018-2021 Virtuozzo International GmbH.
9 * Author:
10 * Sementsov-Ogievskiy Vladimir <vsementsov@virtuozzo.com>
12 * This program is free software; you can redistribute it and/or modify
13 * it under the terms of the GNU General Public License as published by
14 * the Free Software Foundation; either version 2 of the License, or
15 * (at your option) any later version.
17 * This program is distributed in the hope that it will be useful,
18 * but WITHOUT ANY WARRANTY; without even the implied warranty of
19 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 * GNU General Public License for more details.
22 * You should have received a copy of the GNU General Public License
23 * along with this program. If not, see <http://www.gnu.org/licenses/>.
26 #include "qemu/osdep.h"
28 #include "sysemu/block-backend.h"
29 #include "qemu/cutils.h"
30 #include "qapi/error.h"
31 #include "block/block_int.h"
32 #include "block/qdict.h"
33 #include "block/block-copy.h"
35 #include "block/copy-before-write.h"
36 #include "block/reqlist.h"
38 #include "qapi/qapi-visit-block-core.h"
40 typedef struct BDRVCopyBeforeWriteState {
41 BlockCopyState *bcs;
42 BdrvChild *target;
45 * @lock: protects access to @access_bitmap, @done_bitmap and
46 * @frozen_read_reqs
48 CoMutex lock;
51 * @access_bitmap: represents areas allowed for reading by fleecing user.
52 * Reading from non-dirty areas leads to -EACCES.
54 BdrvDirtyBitmap *access_bitmap;
57 * @done_bitmap: represents areas that was successfully copied to @target by
58 * copy-before-write operations.
60 BdrvDirtyBitmap *done_bitmap;
63 * @frozen_read_reqs: current read requests for fleecing user in bs->file
64 * node. These areas must not be rewritten by guest.
66 BlockReqList frozen_read_reqs;
67 } BDRVCopyBeforeWriteState;
69 static coroutine_fn int cbw_co_preadv(
70 BlockDriverState *bs, int64_t offset, int64_t bytes,
71 QEMUIOVector *qiov, BdrvRequestFlags flags)
73 return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
77 * Do copy-before-write operation.
79 * On failure guest request must be failed too.
81 * On success, we also wait for all in-flight fleecing read requests in source
82 * node, and it's guaranteed that after cbw_do_copy_before_write() successful
83 * return there are no such requests and they will never appear.
85 static coroutine_fn int cbw_do_copy_before_write(BlockDriverState *bs,
86 uint64_t offset, uint64_t bytes, BdrvRequestFlags flags)
88 BDRVCopyBeforeWriteState *s = bs->opaque;
89 int ret;
90 uint64_t off, end;
91 int64_t cluster_size = block_copy_cluster_size(s->bcs);
93 if (flags & BDRV_REQ_WRITE_UNCHANGED) {
94 return 0;
97 off = QEMU_ALIGN_DOWN(offset, cluster_size);
98 end = QEMU_ALIGN_UP(offset + bytes, cluster_size);
100 ret = block_copy(s->bcs, off, end - off, true);
101 if (ret < 0) {
102 return ret;
105 WITH_QEMU_LOCK_GUARD(&s->lock) {
106 bdrv_set_dirty_bitmap(s->done_bitmap, off, end - off);
107 reqlist_wait_all(&s->frozen_read_reqs, off, end - off, &s->lock);
110 return 0;
113 static int coroutine_fn cbw_co_pdiscard(BlockDriverState *bs,
114 int64_t offset, int64_t bytes)
116 int ret = cbw_do_copy_before_write(bs, offset, bytes, 0);
117 if (ret < 0) {
118 return ret;
121 return bdrv_co_pdiscard(bs->file, offset, bytes);
124 static int coroutine_fn cbw_co_pwrite_zeroes(BlockDriverState *bs,
125 int64_t offset, int64_t bytes, BdrvRequestFlags flags)
127 int ret = cbw_do_copy_before_write(bs, offset, bytes, flags);
128 if (ret < 0) {
129 return ret;
132 return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags);
135 static coroutine_fn int cbw_co_pwritev(BlockDriverState *bs,
136 int64_t offset,
137 int64_t bytes,
138 QEMUIOVector *qiov,
139 BdrvRequestFlags flags)
141 int ret = cbw_do_copy_before_write(bs, offset, bytes, flags);
142 if (ret < 0) {
143 return ret;
146 return bdrv_co_pwritev(bs->file, offset, bytes, qiov, flags);
149 static int coroutine_fn cbw_co_flush(BlockDriverState *bs)
151 if (!bs->file) {
152 return 0;
155 return bdrv_co_flush(bs->file->bs);
159 * If @offset not accessible - return NULL.
161 * Otherwise, set @pnum to some bytes that accessible from @file (@file is set
162 * to bs->file or to s->target). Return newly allocated BlockReq object that
163 * should be than passed to cbw_snapshot_read_unlock().
165 * It's guaranteed that guest writes will not interact in the region until
166 * cbw_snapshot_read_unlock() called.
168 static BlockReq *cbw_snapshot_read_lock(BlockDriverState *bs,
169 int64_t offset, int64_t bytes,
170 int64_t *pnum, BdrvChild **file)
172 BDRVCopyBeforeWriteState *s = bs->opaque;
173 BlockReq *req = g_new(BlockReq, 1);
174 bool done;
176 QEMU_LOCK_GUARD(&s->lock);
178 if (bdrv_dirty_bitmap_next_zero(s->access_bitmap, offset, bytes) != -1) {
179 g_free(req);
180 return NULL;
183 done = bdrv_dirty_bitmap_status(s->done_bitmap, offset, bytes, pnum);
184 if (done) {
186 * Special invalid BlockReq, that is handled in
187 * cbw_snapshot_read_unlock(). We don't need to lock something to read
188 * from s->target.
190 *req = (BlockReq) {.offset = -1, .bytes = -1};
191 *file = s->target;
192 } else {
193 reqlist_init_req(&s->frozen_read_reqs, req, offset, bytes);
194 *file = bs->file;
197 return req;
200 static void cbw_snapshot_read_unlock(BlockDriverState *bs, BlockReq *req)
202 BDRVCopyBeforeWriteState *s = bs->opaque;
204 if (req->offset == -1 && req->bytes == -1) {
205 g_free(req);
206 return;
209 QEMU_LOCK_GUARD(&s->lock);
211 reqlist_remove_req(req);
212 g_free(req);
215 static coroutine_fn int
216 cbw_co_preadv_snapshot(BlockDriverState *bs, int64_t offset, int64_t bytes,
217 QEMUIOVector *qiov, size_t qiov_offset)
219 BlockReq *req;
220 BdrvChild *file;
221 int ret;
223 /* TODO: upgrade to async loop using AioTask */
224 while (bytes) {
225 int64_t cur_bytes;
227 req = cbw_snapshot_read_lock(bs, offset, bytes, &cur_bytes, &file);
228 if (!req) {
229 return -EACCES;
232 ret = bdrv_co_preadv_part(file, offset, cur_bytes,
233 qiov, qiov_offset, 0);
234 cbw_snapshot_read_unlock(bs, req);
235 if (ret < 0) {
236 return ret;
239 bytes -= cur_bytes;
240 offset += cur_bytes;
241 qiov_offset += cur_bytes;
244 return 0;
247 static int coroutine_fn
248 cbw_co_snapshot_block_status(BlockDriverState *bs,
249 bool want_zero, int64_t offset, int64_t bytes,
250 int64_t *pnum, int64_t *map,
251 BlockDriverState **file)
253 BDRVCopyBeforeWriteState *s = bs->opaque;
254 BlockReq *req;
255 int ret;
256 int64_t cur_bytes;
257 BdrvChild *child;
259 req = cbw_snapshot_read_lock(bs, offset, bytes, &cur_bytes, &child);
260 if (!req) {
261 return -EACCES;
264 ret = bdrv_block_status(child->bs, offset, cur_bytes, pnum, map, file);
265 if (child == s->target) {
267 * We refer to s->target only for areas that we've written to it.
268 * And we can not report unallocated blocks in s->target: this will
269 * break generic block-status-above logic, that will go to
270 * copy-before-write filtered child in this case.
272 assert(ret & BDRV_BLOCK_ALLOCATED);
275 cbw_snapshot_read_unlock(bs, req);
277 return ret;
280 static int coroutine_fn cbw_co_pdiscard_snapshot(BlockDriverState *bs,
281 int64_t offset, int64_t bytes)
283 BDRVCopyBeforeWriteState *s = bs->opaque;
285 WITH_QEMU_LOCK_GUARD(&s->lock) {
286 bdrv_reset_dirty_bitmap(s->access_bitmap, offset, bytes);
289 block_copy_reset(s->bcs, offset, bytes);
291 return bdrv_co_pdiscard(s->target, offset, bytes);
294 static void cbw_refresh_filename(BlockDriverState *bs)
296 pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
297 bs->file->bs->filename);
300 static void cbw_child_perm(BlockDriverState *bs, BdrvChild *c,
301 BdrvChildRole role,
302 BlockReopenQueue *reopen_queue,
303 uint64_t perm, uint64_t shared,
304 uint64_t *nperm, uint64_t *nshared)
306 if (!(role & BDRV_CHILD_FILTERED)) {
308 * Target child
310 * Share write to target (child_file), to not interfere
311 * with guest writes to its disk which may be in target backing chain.
312 * Can't resize during a backup block job because we check the size
313 * only upfront.
315 *nshared = BLK_PERM_ALL & ~BLK_PERM_RESIZE;
316 *nperm = BLK_PERM_WRITE;
317 } else {
318 /* Source child */
319 bdrv_default_perms(bs, c, role, reopen_queue,
320 perm, shared, nperm, nshared);
322 if (!QLIST_EMPTY(&bs->parents)) {
323 if (perm & BLK_PERM_WRITE) {
324 *nperm = *nperm | BLK_PERM_CONSISTENT_READ;
326 *nshared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
331 static bool cbw_parse_bitmap_option(QDict *options, BdrvDirtyBitmap **bitmap,
332 Error **errp)
334 QDict *bitmap_qdict = NULL;
335 BlockDirtyBitmap *bmp_param = NULL;
336 Visitor *v = NULL;
337 bool ret = false;
339 *bitmap = NULL;
341 qdict_extract_subqdict(options, &bitmap_qdict, "bitmap.");
342 if (!qdict_size(bitmap_qdict)) {
343 ret = true;
344 goto out;
347 v = qobject_input_visitor_new_flat_confused(bitmap_qdict, errp);
348 if (!v) {
349 goto out;
352 visit_type_BlockDirtyBitmap(v, NULL, &bmp_param, errp);
353 if (!bmp_param) {
354 goto out;
357 *bitmap = block_dirty_bitmap_lookup(bmp_param->node, bmp_param->name, NULL,
358 errp);
359 if (!*bitmap) {
360 goto out;
363 ret = true;
365 out:
366 qapi_free_BlockDirtyBitmap(bmp_param);
367 visit_free(v);
368 qobject_unref(bitmap_qdict);
370 return ret;
373 static int cbw_open(BlockDriverState *bs, QDict *options, int flags,
374 Error **errp)
376 BDRVCopyBeforeWriteState *s = bs->opaque;
377 BdrvDirtyBitmap *bitmap = NULL;
378 int64_t cluster_size;
380 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
381 BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
382 false, errp);
383 if (!bs->file) {
384 return -EINVAL;
387 s->target = bdrv_open_child(NULL, options, "target", bs, &child_of_bds,
388 BDRV_CHILD_DATA, false, errp);
389 if (!s->target) {
390 return -EINVAL;
393 if (!cbw_parse_bitmap_option(options, &bitmap, errp)) {
394 return -EINVAL;
397 bs->total_sectors = bs->file->bs->total_sectors;
398 bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
399 (BDRV_REQ_FUA & bs->file->bs->supported_write_flags);
400 bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
401 ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) &
402 bs->file->bs->supported_zero_flags);
404 s->bcs = block_copy_state_new(bs->file, s->target, bitmap, errp);
405 if (!s->bcs) {
406 error_prepend(errp, "Cannot create block-copy-state: ");
407 return -EINVAL;
410 cluster_size = block_copy_cluster_size(s->bcs);
412 s->done_bitmap = bdrv_create_dirty_bitmap(bs, cluster_size, NULL, errp);
413 if (!s->done_bitmap) {
414 return -EINVAL;
416 bdrv_disable_dirty_bitmap(s->done_bitmap);
418 /* s->access_bitmap starts equal to bcs bitmap */
419 s->access_bitmap = bdrv_create_dirty_bitmap(bs, cluster_size, NULL, errp);
420 if (!s->access_bitmap) {
421 return -EINVAL;
423 bdrv_disable_dirty_bitmap(s->access_bitmap);
424 bdrv_dirty_bitmap_merge_internal(s->access_bitmap,
425 block_copy_dirty_bitmap(s->bcs), NULL,
426 true);
428 qemu_co_mutex_init(&s->lock);
429 QLIST_INIT(&s->frozen_read_reqs);
431 return 0;
434 static void cbw_close(BlockDriverState *bs)
436 BDRVCopyBeforeWriteState *s = bs->opaque;
438 bdrv_release_dirty_bitmap(s->access_bitmap);
439 bdrv_release_dirty_bitmap(s->done_bitmap);
441 block_copy_state_free(s->bcs);
442 s->bcs = NULL;
445 BlockDriver bdrv_cbw_filter = {
446 .format_name = "copy-before-write",
447 .instance_size = sizeof(BDRVCopyBeforeWriteState),
449 .bdrv_open = cbw_open,
450 .bdrv_close = cbw_close,
452 .bdrv_co_preadv = cbw_co_preadv,
453 .bdrv_co_pwritev = cbw_co_pwritev,
454 .bdrv_co_pwrite_zeroes = cbw_co_pwrite_zeroes,
455 .bdrv_co_pdiscard = cbw_co_pdiscard,
456 .bdrv_co_flush = cbw_co_flush,
458 .bdrv_co_preadv_snapshot = cbw_co_preadv_snapshot,
459 .bdrv_co_pdiscard_snapshot = cbw_co_pdiscard_snapshot,
460 .bdrv_co_snapshot_block_status = cbw_co_snapshot_block_status,
462 .bdrv_refresh_filename = cbw_refresh_filename,
464 .bdrv_child_perm = cbw_child_perm,
466 .is_filter = true,
469 BlockDriverState *bdrv_cbw_append(BlockDriverState *source,
470 BlockDriverState *target,
471 const char *filter_node_name,
472 BlockCopyState **bcs,
473 Error **errp)
475 ERRP_GUARD();
476 BDRVCopyBeforeWriteState *state;
477 BlockDriverState *top;
478 QDict *opts;
480 assert(source->total_sectors == target->total_sectors);
481 GLOBAL_STATE_CODE();
483 opts = qdict_new();
484 qdict_put_str(opts, "driver", "copy-before-write");
485 if (filter_node_name) {
486 qdict_put_str(opts, "node-name", filter_node_name);
488 qdict_put_str(opts, "file", bdrv_get_node_name(source));
489 qdict_put_str(opts, "target", bdrv_get_node_name(target));
491 top = bdrv_insert_node(source, opts, BDRV_O_RDWR, errp);
492 if (!top) {
493 return NULL;
496 state = top->opaque;
497 *bcs = state->bcs;
499 return top;
502 void bdrv_cbw_drop(BlockDriverState *bs)
504 GLOBAL_STATE_CODE();
505 bdrv_drop_filter(bs, &error_abort);
506 bdrv_unref(bs);
509 static void cbw_init(void)
511 bdrv_register(&bdrv_cbw_filter);
514 block_init(cbw_init);