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     int64_t cluster_size;
  41 } BDRVCopyBeforeWriteState;
  42
  43 static coroutine_fn int cbw_co_preadv(
  44         BlockDriverState *bs, uint64_t offset, uint64_t bytes,
  45         QEMUIOVector *qiov, int flags)
  46 {
  47     return bdrv_co_preadv(bs->backing, offset, bytes, qiov, flags);
  48 }
  49
  50 static coroutine_fn int cbw_do_copy_before_write(BlockDriverState *bs,
  51         uint64_t offset, uint64_t bytes, BdrvRequestFlags flags)
  52 {
  53     BDRVCopyBeforeWriteState *s = bs->opaque;
  54     uint64_t off, end;
  55
  56     if (flags & BDRV_REQ_WRITE_UNCHANGED) {
  57         return 0;
  58     }
  59
  60     off = QEMU_ALIGN_DOWN(offset, s->cluster_size);
  61     end = QEMU_ALIGN_UP(offset + bytes, s->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->backing, 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->backing, 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->backing, offset, bytes, qiov, flags);
  99 }
 100
 101 static int coroutine_fn cbw_co_flush(BlockDriverState *bs)
 102 {
 103     if (!bs->backing) {
 104         return 0;
 105     }
 106
 107     return bdrv_co_flush(bs->backing->bs);
 108 }
 109
 110 static void cbw_refresh_filename(BlockDriverState *bs)
 111 {
 112     if (bs->backing == NULL) {
 113         /*
 114          * we can be here after failed bdrv_attach_child in
 115          * bdrv_set_backing_hd
 116          */
 117         return;
 118     }
 119     pstrcpy(bs->exact_filename, sizeof(bs->exact_filename),
 120             bs->backing->bs->filename);
 121 }
 122
 123 static void cbw_child_perm(BlockDriverState *bs, BdrvChild *c,
 124                            BdrvChildRole role,
 125                            BlockReopenQueue *reopen_queue,
 126                            uint64_t perm, uint64_t shared,
 127                            uint64_t *nperm, uint64_t *nshared)
 128 {
 129     if (!(role & BDRV_CHILD_FILTERED)) {
 130         /*
 131          * Target child
 132          *
 133          * Share write to target (child_file), to not interfere
 134          * with guest writes to its disk which may be in target backing chain.
 135          * Can't resize during a backup block job because we check the size
 136          * only upfront.
 137          */
 138         *nshared = BLK_PERM_ALL & ~BLK_PERM_RESIZE;
 139         *nperm = BLK_PERM_WRITE;
 140     } else {
 141         /* Source child */
 142         bdrv_default_perms(bs, c, role, reopen_queue,
 143                            perm, shared, nperm, nshared);
 144
 145         if (perm & BLK_PERM_WRITE) {
 146             *nperm = *nperm | BLK_PERM_CONSISTENT_READ;
 147         }
 148         *nshared &= ~(BLK_PERM_WRITE | BLK_PERM_RESIZE);
 149     }
 150 }
 151
 152 BlockDriver bdrv_cbw_filter = {
 153     .format_name = "copy-before-write",
 154     .instance_size = sizeof(BDRVCopyBeforeWriteState),
 155
 156     .bdrv_co_preadv             = cbw_co_preadv,
 157     .bdrv_co_pwritev            = cbw_co_pwritev,
 158     .bdrv_co_pwrite_zeroes      = cbw_co_pwrite_zeroes,
 159     .bdrv_co_pdiscard           = cbw_co_pdiscard,
 160     .bdrv_co_flush              = cbw_co_flush,
 161
 162     .bdrv_refresh_filename      = cbw_refresh_filename,
 163
 164     .bdrv_child_perm            = cbw_child_perm,
 165
 166     .is_filter = true,
 167 };
 168
 169 BlockDriverState *bdrv_cbw_append(BlockDriverState *source,
 170                                   BlockDriverState *target,
 171                                   const char *filter_node_name,
 172                                   uint64_t cluster_size,
 173                                   bool compress,
 174                                   BlockCopyState **bcs,
 175                                   Error **errp)
 176 {
 177     ERRP_GUARD();
 178     int ret;
 179     BDRVCopyBeforeWriteState *state;
 180     BlockDriverState *top;
 181     bool appended = false;
 182
 183     assert(source->total_sectors == target->total_sectors);
 184
 185     top = bdrv_new_open_driver(&bdrv_cbw_filter, filter_node_name,
 186                                BDRV_O_RDWR, errp);
 187     if (!top) {
 188         return NULL;
 189     }
 190
 191     state = top->opaque;
 192     top->total_sectors = source->total_sectors;
 193     top->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED |
 194             (BDRV_REQ_FUA & source->supported_write_flags);
 195     top->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED |
 196             ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) &
 197              source->supported_zero_flags);
 198
 199     bdrv_ref(target);
 200     state->target = bdrv_attach_child(top, target, "target", &child_of_bds,
 201                                       BDRV_CHILD_DATA, errp);
 202     if (!state->target) {
 203         bdrv_unref(target);
 204         bdrv_unref(top);
 205         return NULL;
 206     }
 207
 208     bdrv_drained_begin(source);
 209
 210     ret = bdrv_append(top, source, errp);
 211     if (ret < 0) {
 212         error_prepend(errp, "Cannot append copy-before-write filter: ");
 213         goto fail;
 214     }
 215     appended = true;
 216
 217     state->cluster_size = cluster_size;
 218     state->bcs = block_copy_state_new(top->backing, state->target,
 219                                       cluster_size, false, compress, errp);
 220     if (!state->bcs) {
 221         error_prepend(errp, "Cannot create block-copy-state: ");
 222         goto fail;
 223     }
 224     *bcs = state->bcs;
 225
 226     bdrv_drained_end(source);
 227
 228     return top;
 229
 230 fail:
 231     if (appended) {
 232         bdrv_cbw_drop(top);
 233     } else {
 234         bdrv_unref(top);
 235     }
 236
 237     bdrv_drained_end(source);
 238
 239     return NULL;
 240 }
 241
 242 void bdrv_cbw_drop(BlockDriverState *bs)
 243 {
 244     BDRVCopyBeforeWriteState *s = bs->opaque;
 245
 246     bdrv_drop_filter(bs, &error_abort);
 247
 248     block_copy_state_free(s->bcs);
 249
 250     bdrv_unref(bs);
 251 }