| 1 | /* |
| 2 | * Copy-on-read filter block driver |
| 3 | * |
| 4 | * Copyright (c) 2018 Red Hat, Inc. |
| 5 | * |
| 6 | * Author: |
| 7 | * Max Reitz <mreitz@redhat.com> |
| 8 | * |
| 9 | * This program is free software; you can redistribute it and/or |
| 10 | * modify it under the terms of the GNU General Public License as |
| 11 | * published by the Free Software Foundation; either version 2 or |
| 12 | * (at your option) version 3 of the License. |
| 13 | * |
| 14 | * This program is distributed in the hope that it will be useful, |
| 15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 17 | * GNU General Public License for more details. |
| 18 | * |
| 19 | * You should have received a copy of the GNU General Public License |
| 20 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
| 21 | */ |
| 22 | |
| 23 | #include "qemu/osdep.h" |
| 24 | #include "block/block-io.h" |
| 25 | #include "block/block_int.h" |
| 26 | #include "qemu/module.h" |
| 27 | #include "qapi/error.h" |
| 28 | #include "qobject/qdict.h" |
| 29 | #include "block/copy-on-read.h" |
| 30 | |
| 31 | |
| 32 | typedef struct BDRVStateCOR { |
| 33 | BlockDriverState *bottom_bs; |
| 34 | bool chain_frozen; |
| 35 | } BDRVStateCOR; |
| 36 | |
| 37 | |
| 38 | static int GRAPH_UNLOCKED |
| 39 | cor_open(BlockDriverState *bs, QDict *options, int flags, Error **errp) |
| 40 | { |
| 41 | BlockDriverState *bottom_bs = NULL; |
| 42 | BDRVStateCOR *state = bs->opaque; |
| 43 | /* Find a bottom node name, if any */ |
| 44 | const char *bottom_node = qdict_get_try_str(options, "bottom"); |
| 45 | int ret; |
| 46 | |
| 47 | GLOBAL_STATE_CODE(); |
| 48 | |
| 49 | ret = bdrv_open_file_child(NULL, options, "file", bs, errp); |
| 50 | if (ret < 0) { |
| 51 | return ret; |
| 52 | } |
| 53 | |
| 54 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
| 55 | |
| 56 | bs->supported_read_flags = BDRV_REQ_PREFETCH; |
| 57 | |
| 58 | bs->supported_write_flags = BDRV_REQ_WRITE_UNCHANGED | |
| 59 | (BDRV_REQ_FUA & bs->file->bs->supported_write_flags); |
| 60 | |
| 61 | bs->supported_zero_flags = BDRV_REQ_WRITE_UNCHANGED | |
| 62 | ((BDRV_REQ_FUA | BDRV_REQ_MAY_UNMAP | BDRV_REQ_NO_FALLBACK) & |
| 63 | bs->file->bs->supported_zero_flags); |
| 64 | |
| 65 | if (bottom_node) { |
| 66 | bottom_bs = bdrv_find_node(bottom_node); |
| 67 | if (!bottom_bs) { |
| 68 | error_setg(errp, "Bottom node '%s' not found", bottom_node); |
| 69 | qdict_del(options, "bottom"); |
| 70 | return -EINVAL; |
| 71 | } |
| 72 | qdict_del(options, "bottom"); |
| 73 | |
| 74 | if (!bottom_bs->drv) { |
| 75 | error_setg(errp, "Bottom node '%s' not opened", bottom_node); |
| 76 | return -EINVAL; |
| 77 | } |
| 78 | |
| 79 | if (bottom_bs->drv->is_filter) { |
| 80 | error_setg(errp, "Bottom node '%s' is a filter", bottom_node); |
| 81 | return -EINVAL; |
| 82 | } |
| 83 | |
| 84 | if (bdrv_freeze_backing_chain(bs, bottom_bs, errp) < 0) { |
| 85 | return -EINVAL; |
| 86 | } |
| 87 | state->chain_frozen = true; |
| 88 | |
| 89 | /* |
| 90 | * We do freeze the chain, so it shouldn't be removed. Still, storing a |
| 91 | * pointer worth bdrv_ref(). |
| 92 | */ |
| 93 | bdrv_ref(bottom_bs); |
| 94 | } |
| 95 | state->bottom_bs = bottom_bs; |
| 96 | |
| 97 | /* |
| 98 | * We don't need to call bdrv_child_refresh_perms() now as the permissions |
| 99 | * will be updated later when the filter node gets its parent. |
| 100 | */ |
| 101 | |
| 102 | return 0; |
| 103 | } |
| 104 | |
| 105 | |
| 106 | #define PERM_PASSTHROUGH (BLK_PERM_CONSISTENT_READ \ |
| 107 | | BLK_PERM_WRITE \ |
| 108 | | BLK_PERM_RESIZE) |
| 109 | #define PERM_UNCHANGED (BLK_PERM_ALL & ~PERM_PASSTHROUGH) |
| 110 | |
| 111 | static void cor_child_perm(BlockDriverState *bs, BdrvChild *c, |
| 112 | BdrvChildRole role, |
| 113 | BlockReopenQueue *reopen_queue, |
| 114 | uint64_t perm, uint64_t shared, |
| 115 | uint64_t *nperm, uint64_t *nshared) |
| 116 | { |
| 117 | *nperm = perm & PERM_PASSTHROUGH; |
| 118 | *nshared = (shared & PERM_PASSTHROUGH) | PERM_UNCHANGED; |
| 119 | |
| 120 | /* We must not request write permissions for an inactive node, the child |
| 121 | * cannot provide it. */ |
| 122 | if (!(bs->open_flags & BDRV_O_INACTIVE)) { |
| 123 | *nperm |= BLK_PERM_WRITE_UNCHANGED; |
| 124 | } |
| 125 | } |
| 126 | |
| 127 | |
| 128 | static int64_t coroutine_fn GRAPH_RDLOCK cor_co_getlength(BlockDriverState *bs) |
| 129 | { |
| 130 | return bdrv_co_getlength(bs->file->bs); |
| 131 | } |
| 132 | |
| 133 | |
| 134 | static int coroutine_fn GRAPH_RDLOCK |
| 135 | cor_co_preadv_part(BlockDriverState *bs, int64_t offset, int64_t bytes, |
| 136 | QEMUIOVector *qiov, size_t qiov_offset, |
| 137 | BdrvRequestFlags flags) |
| 138 | { |
| 139 | int64_t n; |
| 140 | int local_flags; |
| 141 | int ret; |
| 142 | BDRVStateCOR *state = bs->opaque; |
| 143 | |
| 144 | if (!state->bottom_bs) { |
| 145 | return bdrv_co_preadv_part(bs->file, offset, bytes, qiov, qiov_offset, |
| 146 | flags | BDRV_REQ_COPY_ON_READ); |
| 147 | } |
| 148 | |
| 149 | while (bytes) { |
| 150 | local_flags = flags; |
| 151 | |
| 152 | /* In case of failure, try to copy-on-read anyway */ |
| 153 | ret = bdrv_co_is_allocated(bs->file->bs, offset, bytes, &n); |
| 154 | if (ret <= 0) { |
| 155 | ret = bdrv_co_is_allocated_above(bdrv_backing_chain_next(bs->file->bs), |
| 156 | state->bottom_bs, true, offset, |
| 157 | n, &n); |
| 158 | if (ret > 0 || ret < 0) { |
| 159 | local_flags |= BDRV_REQ_COPY_ON_READ; |
| 160 | } |
| 161 | /* Finish earlier if the end of a backing file has been reached */ |
| 162 | if (n == 0) { |
| 163 | break; |
| 164 | } |
| 165 | } |
| 166 | |
| 167 | /* Skip if neither read nor write are needed */ |
| 168 | if ((local_flags & (BDRV_REQ_PREFETCH | BDRV_REQ_COPY_ON_READ)) != |
| 169 | BDRV_REQ_PREFETCH) { |
| 170 | ret = bdrv_co_preadv_part(bs->file, offset, n, qiov, qiov_offset, |
| 171 | local_flags); |
| 172 | if (ret < 0) { |
| 173 | return ret; |
| 174 | } |
| 175 | } |
| 176 | |
| 177 | offset += n; |
| 178 | qiov_offset += n; |
| 179 | bytes -= n; |
| 180 | } |
| 181 | |
| 182 | return 0; |
| 183 | } |
| 184 | |
| 185 | |
| 186 | static int coroutine_fn GRAPH_RDLOCK |
| 187 | cor_co_pwritev_part(BlockDriverState *bs, int64_t offset, int64_t bytes, |
| 188 | QEMUIOVector *qiov, size_t qiov_offset, |
| 189 | BdrvRequestFlags flags) |
| 190 | { |
| 191 | return bdrv_co_pwritev_part(bs->file, offset, bytes, qiov, qiov_offset, |
| 192 | flags); |
| 193 | } |
| 194 | |
| 195 | |
| 196 | static int coroutine_fn GRAPH_RDLOCK |
| 197 | cor_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, |
| 198 | BdrvRequestFlags flags) |
| 199 | { |
| 200 | return bdrv_co_pwrite_zeroes(bs->file, offset, bytes, flags); |
| 201 | } |
| 202 | |
| 203 | |
| 204 | static int coroutine_fn GRAPH_RDLOCK |
| 205 | cor_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) |
| 206 | { |
| 207 | return bdrv_co_pdiscard(bs->file, offset, bytes); |
| 208 | } |
| 209 | |
| 210 | |
| 211 | static int coroutine_fn GRAPH_RDLOCK |
| 212 | cor_co_pwritev_compressed(BlockDriverState *bs, int64_t offset, int64_t bytes, |
| 213 | QEMUIOVector *qiov) |
| 214 | { |
| 215 | return bdrv_co_pwritev(bs->file, offset, bytes, qiov, |
| 216 | BDRV_REQ_WRITE_COMPRESSED); |
| 217 | } |
| 218 | |
| 219 | |
| 220 | static void coroutine_fn GRAPH_RDLOCK |
| 221 | cor_co_eject(BlockDriverState *bs, bool eject_flag) |
| 222 | { |
| 223 | bdrv_co_eject(bs->file->bs, eject_flag); |
| 224 | } |
| 225 | |
| 226 | |
| 227 | static void coroutine_fn GRAPH_RDLOCK |
| 228 | cor_co_lock_medium(BlockDriverState *bs, bool locked) |
| 229 | { |
| 230 | bdrv_co_lock_medium(bs->file->bs, locked); |
| 231 | } |
| 232 | |
| 233 | |
| 234 | static void GRAPH_UNLOCKED cor_close(BlockDriverState *bs) |
| 235 | { |
| 236 | BDRVStateCOR *s = bs->opaque; |
| 237 | |
| 238 | GLOBAL_STATE_CODE(); |
| 239 | |
| 240 | if (s->chain_frozen) { |
| 241 | bdrv_graph_rdlock_main_loop(); |
| 242 | s->chain_frozen = false; |
| 243 | bdrv_unfreeze_backing_chain(bs, s->bottom_bs); |
| 244 | bdrv_graph_rdunlock_main_loop(); |
| 245 | } |
| 246 | |
| 247 | bdrv_unref(s->bottom_bs); |
| 248 | } |
| 249 | |
| 250 | |
| 251 | static BlockDriver bdrv_copy_on_read = { |
| 252 | .format_name = "copy-on-read", |
| 253 | .instance_size = sizeof(BDRVStateCOR), |
| 254 | |
| 255 | .bdrv_open = cor_open, |
| 256 | .bdrv_close = cor_close, |
| 257 | .bdrv_child_perm = cor_child_perm, |
| 258 | |
| 259 | .bdrv_co_getlength = cor_co_getlength, |
| 260 | |
| 261 | .bdrv_co_preadv_part = cor_co_preadv_part, |
| 262 | .bdrv_co_pwritev_part = cor_co_pwritev_part, |
| 263 | .bdrv_co_pwrite_zeroes = cor_co_pwrite_zeroes, |
| 264 | .bdrv_co_pdiscard = cor_co_pdiscard, |
| 265 | .bdrv_co_pwritev_compressed = cor_co_pwritev_compressed, |
| 266 | |
| 267 | .bdrv_co_eject = cor_co_eject, |
| 268 | .bdrv_co_lock_medium = cor_co_lock_medium, |
| 269 | |
| 270 | .is_filter = true, |
| 271 | }; |
| 272 | |
| 273 | |
| 274 | void no_coroutine_fn bdrv_cor_filter_drop(BlockDriverState *cor_filter_bs) |
| 275 | { |
| 276 | BDRVStateCOR *s = cor_filter_bs->opaque; |
| 277 | |
| 278 | GLOBAL_STATE_CODE(); |
| 279 | |
| 280 | /* unfreeze, as otherwise bdrv_replace_node() will fail */ |
| 281 | if (s->chain_frozen) { |
| 282 | GRAPH_RDLOCK_GUARD_MAINLOOP(); |
| 283 | s->chain_frozen = false; |
| 284 | bdrv_unfreeze_backing_chain(cor_filter_bs, s->bottom_bs); |
| 285 | } |
| 286 | bdrv_drop_filter(cor_filter_bs, &error_abort); |
| 287 | bdrv_unref(cor_filter_bs); |
| 288 | } |
| 289 | |
| 290 | |
| 291 | static void bdrv_copy_on_read_init(void) |
| 292 | { |
| 293 | bdrv_register(&bdrv_copy_on_read); |
| 294 | } |
| 295 | |
| 296 | block_init(bdrv_copy_on_read_init); |