block/copy-before-write.c

   1 /*
   2  * copy-before-write filter driver
   3  *
   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.
   6  *
   7  * Copyright (c) 2018-2021 Virtuozzo International GmbH.
   8  *
   9  * Author:
  10  *  Sementsov-Ogievskiy Vladimir <vsementsov@virtuozzo.com>
  11  *
  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.
  16  *
  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.
  21  *
  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/>.
  24  */
  25
  26 #include "qemu/osdep.h"
  27
  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"
  34
  35 #include "block/copy-before-write.h"
  36
  37 typedef struct BDRVCopyBeforeWriteState {
  38     BlockCopyState *bcs;
  39     BdrvChild *target;
  40 } BDRVCopyBeforeWriteState;
  41
  42 static coroutine_fn int cbw_co_preadv(
  43         BlockDriverState *bs, uint64_t offset, uint64_t bytes,
  44         QEMUIOVector *qiov, int flags)
  45 {
  46     return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
  47 }
  48
  49 static coroutine_fn int cbw_do_copy_before_write(BlockDriverState *bs,
  50         uint64_t offset, uint64_t bytes, BdrvRequestFlags flags)
  51 {
  52     BDRVCopyBeforeWriteState *s = bs->opaque;
  53     uint64_t off, end;
  54     int64_t cluster_size = block_copy_cluster_size(s->bcs);
  55
  56     if (flags & BDRV_REQ_WRITE_UNCHANGED) {
  57         return 0;
  58     }
  59
  60     off = QEMU_ALIGN_DOWN(offset, cluster_size);
  61     end = QEMU_ALIGN_UP(offset + bytes, cluster_size);
  62
  63     return block_copy(s->bcs, off, end - off, true);
  64 }
  65
  66 static int coroutine_fn cbw_co_pdiscard(BlockDriverState *bs,
  67                                         int64_t offset, int bytes)
  68 {
  69     int ret = cbw_do_copy_before_write(bs, offset, bytes, 0);
  70     if (ret < 0) {
  71         return ret;
  72     }
  73
  74     return bdrv_co_pdiscard(bs->file, offset, bytes);
  75 }
  76
  77 static int coroutine_fn cbw_co_pwrite_zeroes(BlockDriverState *bs,
  78         int64_t offset, int bytes, BdrvRequestFlags flags)
  79 {
  80     int ret = cbw_do_copy_before_write(bs, offset, bytes, flags);
  81     if (ret < 0) {
  82         return ret;
  83     }
  84
  85     return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags);
  86 }
  87
  88 static coroutine_fn int cbw_co_pwritev(BlockDriverState *bs,
  89                                        uint64_t offset,
  90                                        uint64_t bytes,
  91                                        QEMUIOVector *qiov, int flags)
  92 {
  93     int ret = cbw_do_copy_before_write(bs, offset, bytes, flags);
  94     if (ret < 0) {
  95         return ret;
  96     }
  97
  98     return bdrv_co_pwritev(bs->file, offset, bytes, qiov, flags);
  99 }
 100
 101 static int coroutine_fn cbw_co_flush(BlockDriverState *bs)
 102 {
 103     if (!bs->file) {
 104         return 0;
 105     }
 106
 107     return bdrv_co_flush(bs->file->bs);
 108 }
 109
 110 static void cbw_refresh_filename(BlockDriverState *bs)
 111 {
 112     pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
 113             bs->file->bs->filename);
 114 }
 115
 116 static void cbw_child_perm(BlockDriverState *bs, BdrvChild *c,
 117                            BdrvChildRole role,
 118                            BlockReopenQueue *reopen_queue,
 119                            uint64_t perm, uint64_t shared,
 120                            uint64_t *nperm, uint64_t *nshared)
 121 {
 122     if (!(role & BDRV_CHILD_FILTERED)) {
 123         /*
 124          * Target child
 125          *
 126          * Share write to target (child_file), to not interfere
 127          * with guest writes to its disk which may be in target backing chain.
 128          * Can't resize during a backup block job because we check the size
 129          * only upfront.
 130          */
 131         *nshared = BLK_PERM_ALL & ~BLK_PERM_RESIZE;
 132         *nperm = BLK_PERM_WRITE;
 133     } else {
 134         /* Source child */
 135         bdrv_default_perms(bs, c, role, reopen_queue,
 136                            perm, shared, nperm, nshared);
 137
 138         if (!QLIST_EMPTY(&bs->parents)) {
 139             if (perm & BLK_PERM_WRITE) {
 140                 *nperm = *nperm | BLK_PERM_CONSISTENT_READ;
 141             }
 142             *nshared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
 143         }
 144     }
 145 }
 146
 147 static int cbw_open(BlockDriverState *bs, QDict *options, int flags,
 148                     Error **errp)
 149 {
 150     BDRVCopyBeforeWriteState *s = bs->opaque;
 151     BdrvDirtyBitmap *copy_bitmap;
 152
 153     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds,
 154                                BDRV_CHILD_FILTERED | BDRV_CHILD_PRIMARY,
 155                                false, errp);
 156     if (!bs->file) {
 157         return -EINVAL;
 158     }
 159
 160     s->target = bdrv_open_child(NULL, options, "target", bs, &child_of_bds,
 161                                 BDRV_CHILD_DATA, false, errp);
 162     if (!s->target) {
 163         return -EINVAL;
 164     }
 165
 166     bs->total_sectors = bs->file->bs->total_sectors;
 167     bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
 168             (BDRV_REQ_FUA & bs->file->bs->supported_write_flags);
 169     bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
 170             ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) &
 171              bs->file->bs->supported_zero_flags);
 172
 173     s->bcs = block_copy_state_new(bs->file, s->target, errp);
 174     if (!s->bcs) {
 175         error_prepend(errp, "Cannot create block-copy-state: ");
 176         return -EINVAL;
 177     }
 178
 179     copy_bitmap = block_copy_dirty_bitmap(s->bcs);
 180     bdrv_set_dirty_bitmap(copy_bitmap, 0, bdrv_dirty_bitmap_size(copy_bitmap));
 181
 182     return 0;
 183 }
 184
 185 static void cbw_close(BlockDriverState *bs)
 186 {
 187     BDRVCopyBeforeWriteState *s = bs->opaque;
 188
 189     block_copy_state_free(s->bcs);
 190     s->bcs = NULL;
 191 }
 192
 193 BlockDriver bdrv_cbw_filter = {
 194     .format_name = "copy-before-write",
 195     .instance_size = sizeof(BDRVCopyBeforeWriteState),
 196
 197     .bdrv_open                  = cbw_open,
 198     .bdrv_close                 = cbw_close,
 199
 200     .bdrv_co_preadv             = cbw_co_preadv,
 201     .bdrv_co_pwritev            = cbw_co_pwritev,
 202     .bdrv_co_pwrite_zeroes      = cbw_co_pwrite_zeroes,
 203     .bdrv_co_pdiscard           = cbw_co_pdiscard,
 204     .bdrv_co_flush              = cbw_co_flush,
 205
 206     .bdrv_refresh_filename      = cbw_refresh_filename,
 207
 208     .bdrv_child_perm            = cbw_child_perm,
 209
 210     .is_filter = true,
 211 };
 212
 213 BlockDriverState *bdrv_cbw_append(BlockDriverState *source,
 214                                   BlockDriverState *target,
 215                                   const char *filter_node_name,
 216                                   BlockCopyState **bcs,
 217                                   Error **errp)
 218 {
 219     ERRP_GUARD();
 220     BDRVCopyBeforeWriteState *state;
 221     BlockDriverState *top;
 222     QDict *opts;
 223
 224     assert(source->total_sectors == target->total_sectors);
 225
 226     opts = qdict_new();
 227     qdict_put_str(opts, "driver", "copy-before-write");
 228     if (filter_node_name) {
 229         qdict_put_str(opts, "node-name", filter_node_name);
 230     }
 231     qdict_put_str(opts, "file", bdrv_get_node_name(source));
 232     qdict_put_str(opts, "target", bdrv_get_node_name(target));
 233
 234     top = bdrv_insert_node(source, opts, BDRV_O_RDWR, errp);
 235     if (!top) {
 236         return NULL;
 237     }
 238
 239     state = top->opaque;
 240     *bcs = state->bcs;
 241
 242     return top;
 243 }
 244
 245 void bdrv_cbw_drop(BlockDriverState *bs)
 246 {
 247     bdrv_drop_filter(bs, &error_abort);
 248     bdrv_unref(bs);
 249 }
 250
 251 static void cbw_init(void)
 252 {
 253     bdrv_register(&bdrv_cbw_filter);
 254 }
 255
 256 block_init(cbw_init);