xref: /openbmc/qemu/block/copy-before-write.c (revision f44fd7399cf35b2d4e8f264de508e8119c76a14a)
1d003e0aeSVladimir Sementsov-Ogievskiy /*
2d003e0aeSVladimir Sementsov-Ogievskiy  * copy-before-write filter driver
3d003e0aeSVladimir Sementsov-Ogievskiy  *
4d003e0aeSVladimir Sementsov-Ogievskiy  * The driver performs Copy-Before-Write (CBW) operation: it is injected above
5d003e0aeSVladimir Sementsov-Ogievskiy  * some node, and before each write it copies _old_ data to the target node.
6d003e0aeSVladimir Sementsov-Ogievskiy  *
7d003e0aeSVladimir Sementsov-Ogievskiy  * Copyright (c) 2018-2021 Virtuozzo International GmbH.
8d003e0aeSVladimir Sementsov-Ogievskiy  *
9d003e0aeSVladimir Sementsov-Ogievskiy  * Author:
10d003e0aeSVladimir Sementsov-Ogievskiy  *  Sementsov-Ogievskiy Vladimir <vsementsov@virtuozzo.com>
11d003e0aeSVladimir Sementsov-Ogievskiy  *
12d003e0aeSVladimir Sementsov-Ogievskiy  * This program is free software; you can redistribute it and/or modify
13d003e0aeSVladimir Sementsov-Ogievskiy  * it under the terms of the GNU General Public License as published by
14d003e0aeSVladimir Sementsov-Ogievskiy  * the Free Software Foundation; either version 2 of the License, or
15d003e0aeSVladimir Sementsov-Ogievskiy  * (at your option) any later version.
16d003e0aeSVladimir Sementsov-Ogievskiy  *
17d003e0aeSVladimir Sementsov-Ogievskiy  * This program is distributed in the hope that it will be useful,
18d003e0aeSVladimir Sementsov-Ogievskiy  * but WITHOUT ANY WARRANTY; without even the implied warranty of
19d003e0aeSVladimir Sementsov-Ogievskiy  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20d003e0aeSVladimir Sementsov-Ogievskiy  * GNU General Public License for more details.
21d003e0aeSVladimir Sementsov-Ogievskiy  *
22d003e0aeSVladimir Sementsov-Ogievskiy  * You should have received a copy of the GNU General Public License
23d003e0aeSVladimir Sementsov-Ogievskiy  * along with this program. If not, see <http://www.gnu.org/licenses/>.
24d003e0aeSVladimir Sementsov-Ogievskiy  */
25d003e0aeSVladimir Sementsov-Ogievskiy 
26d003e0aeSVladimir Sementsov-Ogievskiy #include "qemu/osdep.h"
27d003e0aeSVladimir Sementsov-Ogievskiy 
28d003e0aeSVladimir Sementsov-Ogievskiy #include "sysemu/block-backend.h"
29d003e0aeSVladimir Sementsov-Ogievskiy #include "qemu/cutils.h"
30d003e0aeSVladimir Sementsov-Ogievskiy #include "qapi/error.h"
31d003e0aeSVladimir Sementsov-Ogievskiy #include "block/block_int.h"
32d003e0aeSVladimir Sementsov-Ogievskiy #include "block/qdict.h"
33d003e0aeSVladimir Sementsov-Ogievskiy #include "block/block-copy.h"
34d003e0aeSVladimir Sementsov-Ogievskiy 
35d003e0aeSVladimir Sementsov-Ogievskiy #include "block/copy-before-write.h"
36d003e0aeSVladimir Sementsov-Ogievskiy 
37d003e0aeSVladimir Sementsov-Ogievskiy typedef struct BDRVCopyBeforeWriteState {
38d003e0aeSVladimir Sementsov-Ogievskiy     BlockCopyState *bcs;
39d003e0aeSVladimir Sementsov-Ogievskiy     BdrvChild *target;
40d003e0aeSVladimir Sementsov-Ogievskiy } BDRVCopyBeforeWriteState;
41d003e0aeSVladimir Sementsov-Ogievskiy 
42d003e0aeSVladimir Sementsov-Ogievskiy static coroutine_fn int cbw_co_preadv(
43d003e0aeSVladimir Sementsov-Ogievskiy         BlockDriverState *bs, uint64_t offset, uint64_t bytes,
44d003e0aeSVladimir Sementsov-Ogievskiy         QEMUIOVector *qiov, int flags)
45d003e0aeSVladimir Sementsov-Ogievskiy {
463c1e6327SVladimir Sementsov-Ogievskiy     return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
47d003e0aeSVladimir Sementsov-Ogievskiy }
48d003e0aeSVladimir Sementsov-Ogievskiy 
49d003e0aeSVladimir Sementsov-Ogievskiy static coroutine_fn int cbw_do_copy_before_write(BlockDriverState *bs,
50d003e0aeSVladimir Sementsov-Ogievskiy         uint64_t offset, uint64_t bytes, BdrvRequestFlags flags)
51d003e0aeSVladimir Sementsov-Ogievskiy {
52d003e0aeSVladimir Sementsov-Ogievskiy     BDRVCopyBeforeWriteState *s = bs->opaque;
53d003e0aeSVladimir Sementsov-Ogievskiy     uint64_t off, end;
54b518e9e9SVladimir Sementsov-Ogievskiy     int64_t cluster_size = block_copy_cluster_size(s->bcs);
55d003e0aeSVladimir Sementsov-Ogievskiy 
56d003e0aeSVladimir Sementsov-Ogievskiy     if (flags & BDRV_REQ_WRITE_UNCHANGED) {
57d003e0aeSVladimir Sementsov-Ogievskiy         return 0;
58d003e0aeSVladimir Sementsov-Ogievskiy     }
59d003e0aeSVladimir Sementsov-Ogievskiy 
60b518e9e9SVladimir Sementsov-Ogievskiy     off = QEMU_ALIGN_DOWN(offset, cluster_size);
61b518e9e9SVladimir Sementsov-Ogievskiy     end = QEMU_ALIGN_UP(offset + bytes, cluster_size);
62d003e0aeSVladimir Sementsov-Ogievskiy 
63d003e0aeSVladimir Sementsov-Ogievskiy     return block_copy(s->bcs, off, end - off, true);
64d003e0aeSVladimir Sementsov-Ogievskiy }
65d003e0aeSVladimir Sementsov-Ogievskiy 
66d003e0aeSVladimir Sementsov-Ogievskiy static int coroutine_fn cbw_co_pdiscard(BlockDriverState *bs,
67d003e0aeSVladimir Sementsov-Ogievskiy                                         int64_t offset, int bytes)
68d003e0aeSVladimir Sementsov-Ogievskiy {
69d003e0aeSVladimir Sementsov-Ogievskiy     int ret = cbw_do_copy_before_write(bs, offset, bytes, 0);
70d003e0aeSVladimir Sementsov-Ogievskiy     if (ret < 0) {
71d003e0aeSVladimir Sementsov-Ogievskiy         return ret;
72d003e0aeSVladimir Sementsov-Ogievskiy     }
73d003e0aeSVladimir Sementsov-Ogievskiy 
743c1e6327SVladimir Sementsov-Ogievskiy     return bdrv_co_pdiscard(bs->file, offset, bytes);
75d003e0aeSVladimir Sementsov-Ogievskiy }
76d003e0aeSVladimir Sementsov-Ogievskiy 
77d003e0aeSVladimir Sementsov-Ogievskiy static int coroutine_fn cbw_co_pwrite_zeroes(BlockDriverState *bs,
78d003e0aeSVladimir Sementsov-Ogievskiy         int64_t offset, int bytes, BdrvRequestFlags flags)
79d003e0aeSVladimir Sementsov-Ogievskiy {
80d003e0aeSVladimir Sementsov-Ogievskiy     int ret = cbw_do_copy_before_write(bs, offset, bytes, flags);
81d003e0aeSVladimir Sementsov-Ogievskiy     if (ret < 0) {
82d003e0aeSVladimir Sementsov-Ogievskiy         return ret;
83d003e0aeSVladimir Sementsov-Ogievskiy     }
84d003e0aeSVladimir Sementsov-Ogievskiy 
853c1e6327SVladimir Sementsov-Ogievskiy     return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags);
86d003e0aeSVladimir Sementsov-Ogievskiy }
87d003e0aeSVladimir Sementsov-Ogievskiy 
88d003e0aeSVladimir Sementsov-Ogievskiy static coroutine_fn int cbw_co_pwritev(BlockDriverState *bs,
89d003e0aeSVladimir Sementsov-Ogievskiy                                        uint64_t offset,
90d003e0aeSVladimir Sementsov-Ogievskiy                                        uint64_t bytes,
91d003e0aeSVladimir Sementsov-Ogievskiy                                        QEMUIOVector *qiov, int flags)
92d003e0aeSVladimir Sementsov-Ogievskiy {
93d003e0aeSVladimir Sementsov-Ogievskiy     int ret = cbw_do_copy_before_write(bs, offset, bytes, flags);
94d003e0aeSVladimir Sementsov-Ogievskiy     if (ret < 0) {
95d003e0aeSVladimir Sementsov-Ogievskiy         return ret;
96d003e0aeSVladimir Sementsov-Ogievskiy     }
97d003e0aeSVladimir Sementsov-Ogievskiy 
983c1e6327SVladimir Sementsov-Ogievskiy     return bdrv_co_pwritev(bs->file, offset, bytes, qiov, flags);
99d003e0aeSVladimir Sementsov-Ogievskiy }
100d003e0aeSVladimir Sementsov-Ogievskiy 
101d003e0aeSVladimir Sementsov-Ogievskiy static int coroutine_fn cbw_co_flush(BlockDriverState *bs)
102d003e0aeSVladimir Sementsov-Ogievskiy {
1033c1e6327SVladimir Sementsov-Ogievskiy     if (!bs->file) {
104d003e0aeSVladimir Sementsov-Ogievskiy         return 0;
105d003e0aeSVladimir Sementsov-Ogievskiy     }
106d003e0aeSVladimir Sementsov-Ogievskiy 
1073c1e6327SVladimir Sementsov-Ogievskiy     return bdrv_co_flush(bs->file->bs);
108d003e0aeSVladimir Sementsov-Ogievskiy }
109d003e0aeSVladimir Sementsov-Ogievskiy 
110d003e0aeSVladimir Sementsov-Ogievskiy static void cbw_refresh_filename(BlockDriverState *bs)
111d003e0aeSVladimir Sementsov-Ogievskiy {
112d003e0aeSVladimir Sementsov-Ogievskiy     pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
1133c1e6327SVladimir Sementsov-Ogievskiy             bs->file->bs->filename);
114d003e0aeSVladimir Sementsov-Ogievskiy }
115d003e0aeSVladimir Sementsov-Ogievskiy 
116d003e0aeSVladimir Sementsov-Ogievskiy static void cbw_child_perm(BlockDriverState *bs, BdrvChild *c,
117d003e0aeSVladimir Sementsov-Ogievskiy                            BdrvChildRole role,
118d003e0aeSVladimir Sementsov-Ogievskiy                            BlockReopenQueue *reopen_queue,
119d003e0aeSVladimir Sementsov-Ogievskiy                            uint64_t perm, uint64_t shared,
120d003e0aeSVladimir Sementsov-Ogievskiy                            uint64_t *nperm, uint64_t *nshared)
121d003e0aeSVladimir Sementsov-Ogievskiy {
122d003e0aeSVladimir Sementsov-Ogievskiy     if (!(role & BDRV_CHILD_FILTERED)) {
123d003e0aeSVladimir Sementsov-Ogievskiy         /*
124d003e0aeSVladimir Sementsov-Ogievskiy          * Target child
125d003e0aeSVladimir Sementsov-Ogievskiy          *
126d003e0aeSVladimir Sementsov-Ogievskiy          * Share write to target (child_file), to not interfere
127d003e0aeSVladimir Sementsov-Ogievskiy          * with guest writes to its disk which may be in target backing chain.
128d003e0aeSVladimir Sementsov-Ogievskiy          * Can't resize during a backup block job because we check the size
129d003e0aeSVladimir Sementsov-Ogievskiy          * only upfront.
130d003e0aeSVladimir Sementsov-Ogievskiy          */
131d003e0aeSVladimir Sementsov-Ogievskiy         *nshared = BLK_PERM_ALL & ~BLK_PERM_RESIZE;
132d003e0aeSVladimir Sementsov-Ogievskiy         *nperm = BLK_PERM_WRITE;
133d003e0aeSVladimir Sementsov-Ogievskiy     } else {
134d003e0aeSVladimir Sementsov-Ogievskiy         /* Source child */
135d003e0aeSVladimir Sementsov-Ogievskiy         bdrv_default_perms(bs, c, role, reopen_queue,
136d003e0aeSVladimir Sementsov-Ogievskiy                            perm, shared, nperm, nshared);
137d003e0aeSVladimir Sementsov-Ogievskiy 
1383860c020SVladimir Sementsov-Ogievskiy         if (!QLIST_EMPTY(&bs->parents)) {
139d003e0aeSVladimir Sementsov-Ogievskiy             if (perm & BLK_PERM_WRITE) {
140d003e0aeSVladimir Sementsov-Ogievskiy                 *nperm = *nperm | BLK_PERM_CONSISTENT_READ;
141d003e0aeSVladimir Sementsov-Ogievskiy             }
142d003e0aeSVladimir Sementsov-Ogievskiy             *nshared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
143d003e0aeSVladimir Sementsov-Ogievskiy         }
144d003e0aeSVladimir Sementsov-Ogievskiy     }
1453860c020SVladimir Sementsov-Ogievskiy }
146d003e0aeSVladimir Sementsov-Ogievskiy 
147*f44fd739SVladimir Sementsov-Ogievskiy static int cbw_init(BlockDriverState *bs, QDict *options, Error **errp)
1481f0cacb9SVladimir Sementsov-Ogievskiy {
149fe7ea40cSVladimir Sementsov-Ogievskiy     BDRVCopyBeforeWriteState *s = bs->opaque;
1501f0cacb9SVladimir Sementsov-Ogievskiy 
151*f44fd739SVladimir Sementsov-Ogievskiy     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
152*f44fd739SVladimir Sementsov-Ogievskiy                                BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
153*f44fd739SVladimir Sementsov-Ogievskiy                                false, errp);
154*f44fd739SVladimir Sementsov-Ogievskiy     if (!bs->file) {
1551f0cacb9SVladimir Sementsov-Ogievskiy         return -EINVAL;
1561f0cacb9SVladimir Sementsov-Ogievskiy     }
1571f0cacb9SVladimir Sementsov-Ogievskiy 
158*f44fd739SVladimir Sementsov-Ogievskiy     s->target = bdrv_open_child(NULL, options, "target", bs, &child_of_bds,
159*f44fd739SVladimir Sementsov-Ogievskiy                                 BDRV_CHILD_DATA, false, errp);
160*f44fd739SVladimir Sementsov-Ogievskiy     if (!s->target) {
1611f0cacb9SVladimir Sementsov-Ogievskiy         return -EINVAL;
1621f0cacb9SVladimir Sementsov-Ogievskiy     }
1631f0cacb9SVladimir Sementsov-Ogievskiy 
1645a507426SVladimir Sementsov-Ogievskiy     bs->total_sectors = bs->file->bs->total_sectors;
1655a507426SVladimir Sementsov-Ogievskiy     bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
1665a507426SVladimir Sementsov-Ogievskiy             (BDRV_REQ_FUA & bs->file->bs->supported_write_flags);
1675a507426SVladimir Sementsov-Ogievskiy     bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
1685a507426SVladimir Sementsov-Ogievskiy             ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) &
1695a507426SVladimir Sementsov-Ogievskiy              bs->file->bs->supported_zero_flags);
1705a507426SVladimir Sementsov-Ogievskiy 
1714c1e992bSVladimir Sementsov-Ogievskiy     s->bcs = block_copy_state_new(bs->file, s->target, false, false, errp);
172fe7ea40cSVladimir Sementsov-Ogievskiy     if (!s->bcs) {
1731f0cacb9SVladimir Sementsov-Ogievskiy         error_prepend(errp, "Cannot create block-copy-state: ");
1741f0cacb9SVladimir Sementsov-Ogievskiy         return -EINVAL;
1751f0cacb9SVladimir Sementsov-Ogievskiy     }
1761f0cacb9SVladimir Sementsov-Ogievskiy 
1771f0cacb9SVladimir Sementsov-Ogievskiy     return 0;
1781f0cacb9SVladimir Sementsov-Ogievskiy }
1791f0cacb9SVladimir Sementsov-Ogievskiy 
180d003e0aeSVladimir Sementsov-Ogievskiy BlockDriver bdrv_cbw_filter = {
181d003e0aeSVladimir Sementsov-Ogievskiy     .format_name = "copy-before-write",
182d003e0aeSVladimir Sementsov-Ogievskiy     .instance_size = sizeof(BDRVCopyBeforeWriteState),
183d003e0aeSVladimir Sementsov-Ogievskiy 
184d003e0aeSVladimir Sementsov-Ogievskiy     .bdrv_co_preadv             = cbw_co_preadv,
185d003e0aeSVladimir Sementsov-Ogievskiy     .bdrv_co_pwritev            = cbw_co_pwritev,
186d003e0aeSVladimir Sementsov-Ogievskiy     .bdrv_co_pwrite_zeroes      = cbw_co_pwrite_zeroes,
187d003e0aeSVladimir Sementsov-Ogievskiy     .bdrv_co_pdiscard           = cbw_co_pdiscard,
188d003e0aeSVladimir Sementsov-Ogievskiy     .bdrv_co_flush              = cbw_co_flush,
189d003e0aeSVladimir Sementsov-Ogievskiy 
190d003e0aeSVladimir Sementsov-Ogievskiy     .bdrv_refresh_filename      = cbw_refresh_filename,
191d003e0aeSVladimir Sementsov-Ogievskiy 
192d003e0aeSVladimir Sementsov-Ogievskiy     .bdrv_child_perm            = cbw_child_perm,
193d003e0aeSVladimir Sementsov-Ogievskiy 
194d003e0aeSVladimir Sementsov-Ogievskiy     .is_filter = true,
195d003e0aeSVladimir Sementsov-Ogievskiy };
196d003e0aeSVladimir Sementsov-Ogievskiy 
197d003e0aeSVladimir Sementsov-Ogievskiy BlockDriverState *bdrv_cbw_append(BlockDriverState *source,
198d003e0aeSVladimir Sementsov-Ogievskiy                                   BlockDriverState *target,
199d003e0aeSVladimir Sementsov-Ogievskiy                                   const char *filter_node_name,
200d003e0aeSVladimir Sementsov-Ogievskiy                                   BlockCopyState **bcs,
201d003e0aeSVladimir Sementsov-Ogievskiy                                   Error **errp)
202d003e0aeSVladimir Sementsov-Ogievskiy {
203d003e0aeSVladimir Sementsov-Ogievskiy     ERRP_GUARD();
204d003e0aeSVladimir Sementsov-Ogievskiy     int ret;
205d003e0aeSVladimir Sementsov-Ogievskiy     BDRVCopyBeforeWriteState *state;
206d003e0aeSVladimir Sementsov-Ogievskiy     BlockDriverState *top;
207*f44fd739SVladimir Sementsov-Ogievskiy     QDict *opts;
208d003e0aeSVladimir Sementsov-Ogievskiy 
209d003e0aeSVladimir Sementsov-Ogievskiy     assert(source->total_sectors == target->total_sectors);
210d003e0aeSVladimir Sementsov-Ogievskiy 
211d003e0aeSVladimir Sementsov-Ogievskiy     top = bdrv_new_open_driver(&bdrv_cbw_filter, filter_node_name,
212d003e0aeSVladimir Sementsov-Ogievskiy                                BDRV_O_RDWR, errp);
213d003e0aeSVladimir Sementsov-Ogievskiy     if (!top) {
2143c1e6327SVladimir Sementsov-Ogievskiy         error_prepend(errp, "Cannot open driver: ");
215d003e0aeSVladimir Sementsov-Ogievskiy         return NULL;
216d003e0aeSVladimir Sementsov-Ogievskiy     }
217d003e0aeSVladimir Sementsov-Ogievskiy     state = top->opaque;
218d003e0aeSVladimir Sementsov-Ogievskiy 
219*f44fd739SVladimir Sementsov-Ogievskiy     opts = qdict_new();
220*f44fd739SVladimir Sementsov-Ogievskiy     qdict_put_str(opts, "file", bdrv_get_node_name(source));
221*f44fd739SVladimir Sementsov-Ogievskiy     qdict_put_str(opts, "target", bdrv_get_node_name(target));
222*f44fd739SVladimir Sementsov-Ogievskiy 
223*f44fd739SVladimir Sementsov-Ogievskiy     ret = cbw_init(top, opts, errp);
224*f44fd739SVladimir Sementsov-Ogievskiy     qobject_unref(opts);
2251f0cacb9SVladimir Sementsov-Ogievskiy     if (ret < 0) {
226d003e0aeSVladimir Sementsov-Ogievskiy         goto fail;
227d003e0aeSVladimir Sementsov-Ogievskiy     }
228d003e0aeSVladimir Sementsov-Ogievskiy 
2297ddbce2dSVladimir Sementsov-Ogievskiy     bdrv_drained_begin(source);
2307ddbce2dSVladimir Sementsov-Ogievskiy     ret = bdrv_replace_node(source, top, errp);
231d003e0aeSVladimir Sementsov-Ogievskiy     bdrv_drained_end(source);
2327ddbce2dSVladimir Sementsov-Ogievskiy     if (ret < 0) {
2337ddbce2dSVladimir Sementsov-Ogievskiy         error_prepend(errp, "Cannot append copy-before-write filter: ");
2347ddbce2dSVladimir Sementsov-Ogievskiy         goto fail;
2357ddbce2dSVladimir Sementsov-Ogievskiy     }
2367ddbce2dSVladimir Sementsov-Ogievskiy 
2377ddbce2dSVladimir Sementsov-Ogievskiy     *bcs = state->bcs;
238d003e0aeSVladimir Sementsov-Ogievskiy 
239d003e0aeSVladimir Sementsov-Ogievskiy     return top;
240d003e0aeSVladimir Sementsov-Ogievskiy 
241d003e0aeSVladimir Sementsov-Ogievskiy fail:
2427ddbce2dSVladimir Sementsov-Ogievskiy     block_copy_state_free(state->bcs);
243d003e0aeSVladimir Sementsov-Ogievskiy     bdrv_unref(top);
244d003e0aeSVladimir Sementsov-Ogievskiy     return NULL;
245d003e0aeSVladimir Sementsov-Ogievskiy }
246d003e0aeSVladimir Sementsov-Ogievskiy 
247d003e0aeSVladimir Sementsov-Ogievskiy void bdrv_cbw_drop(BlockDriverState *bs)
248d003e0aeSVladimir Sementsov-Ogievskiy {
249d003e0aeSVladimir Sementsov-Ogievskiy     BDRVCopyBeforeWriteState *s = bs->opaque;
250d003e0aeSVladimir Sementsov-Ogievskiy 
251d003e0aeSVladimir Sementsov-Ogievskiy     bdrv_drop_filter(bs, &error_abort);
252d003e0aeSVladimir Sementsov-Ogievskiy 
253d003e0aeSVladimir Sementsov-Ogievskiy     block_copy_state_free(s->bcs);
254d003e0aeSVladimir Sementsov-Ogievskiy 
255d003e0aeSVladimir Sementsov-Ogievskiy     bdrv_unref(bs);
256d003e0aeSVladimir Sementsov-Ogievskiy }
257