| 1 | /* |
| 2 | * Common code for block device models |
| 3 | * |
| 4 | * Copyright (C) 2012 Red Hat, Inc. |
| 5 | * |
| 6 | * This work is licensed under the terms of the GNU GPL, version 2 or |
| 7 | * later. See the COPYING file in the top-level directory. |
| 8 | */ |
| 9 | |
| 10 | #include "qemu/osdep.h" |
| 11 | #include "block/block_int-common.h" |
| 12 | #include "system/blockdev.h" |
| 13 | #include "system/block-backend.h" |
| 14 | #include "hw/block/block.h" |
| 15 | #include "migration/cpr.h" |
| 16 | #include "qapi/error.h" |
| 17 | #include "qapi/qapi-types-block.h" |
| 18 | |
| 19 | /* |
| 20 | * Read the non-zeroes parts of @blk into @buf |
| 21 | * Reading all of the @blk is expensive if the zeroes parts of @blk |
| 22 | * is large enough. Therefore check the block status and only write |
| 23 | * the non-zeroes block into @buf. |
| 24 | * |
| 25 | * Return 0 on success, non-zero on error. |
| 26 | */ |
| 27 | static int blk_pread_nonzeroes(BlockBackend *blk, hwaddr size, void *buf) |
| 28 | { |
| 29 | int ret; |
| 30 | int64_t bytes, offset = 0; |
| 31 | BlockDriverState *bs = blk_bs(blk); |
| 32 | |
| 33 | for (;;) { |
| 34 | bytes = MIN(size - offset, BDRV_REQUEST_MAX_BYTES); |
| 35 | if (bytes <= 0) { |
| 36 | return 0; |
| 37 | } |
| 38 | ret = bdrv_block_status(bs, offset, bytes, &bytes, NULL, NULL); |
| 39 | if (ret < 0) { |
| 40 | return ret; |
| 41 | } |
| 42 | if (!(ret & BDRV_BLOCK_ZERO)) { |
| 43 | ret = blk_pread(blk, offset, bytes, (uint8_t *) buf + offset, 0); |
| 44 | if (ret < 0) { |
| 45 | return ret; |
| 46 | } |
| 47 | } |
| 48 | offset += bytes; |
| 49 | } |
| 50 | } |
| 51 | |
| 52 | /* |
| 53 | * Read the entire contents of @blk into @buf. |
| 54 | * @blk's contents must be @size bytes, and @size must be at most |
| 55 | * BDRV_REQUEST_MAX_BYTES. |
| 56 | * On success, return true. |
| 57 | * On failure, store an error through @errp and return false. |
| 58 | * |
| 59 | * This function not intended for actual block devices, which read on |
| 60 | * demand. It's for things like memory devices that (ab)use a block |
| 61 | * backend to provide persistence. |
| 62 | */ |
| 63 | bool blk_check_size_and_read_all(BlockBackend *blk, DeviceState *dev, |
| 64 | void *buf, hwaddr size, Error **errp) |
| 65 | { |
| 66 | int64_t blk_len; |
| 67 | int ret; |
| 68 | g_autofree char *dev_id = NULL; |
| 69 | |
| 70 | if (cpr_is_incoming()) { |
| 71 | return true; |
| 72 | } |
| 73 | |
| 74 | blk_len = blk_getlength(blk); |
| 75 | if (blk_len < 0) { |
| 76 | error_setg_errno(errp, -blk_len, |
| 77 | "can't get size of %s block backend", blk_name(blk)); |
| 78 | return false; |
| 79 | } |
| 80 | if (blk_len != size) { |
| 81 | dev_id = qdev_get_human_name(dev); |
| 82 | error_setg(errp, "%s device '%s' requires %" HWADDR_PRIu |
| 83 | " bytes, %s block backend provides %" PRIu64 " bytes", |
| 84 | object_get_typename(OBJECT(dev)), dev_id, size, |
| 85 | blk_name(blk), blk_len); |
| 86 | return false; |
| 87 | } |
| 88 | |
| 89 | /* |
| 90 | * We could loop for @size > BDRV_REQUEST_MAX_BYTES, but if we |
| 91 | * ever get to the point we want to read *gigabytes* here, we |
| 92 | * should probably rework the device to be more like an actual |
| 93 | * block device and read only on demand. |
| 94 | */ |
| 95 | assert(size <= BDRV_REQUEST_MAX_BYTES); |
| 96 | ret = blk_pread_nonzeroes(blk, size, buf); |
| 97 | if (ret < 0) { |
| 98 | dev_id = qdev_get_human_name(dev); |
| 99 | error_setg_errno(errp, -ret, "can't read %s block backend" |
| 100 | " for %s device '%s'", |
| 101 | blk_name(blk), object_get_typename(OBJECT(dev)), |
| 102 | dev_id); |
| 103 | return false; |
| 104 | } |
| 105 | return true; |
| 106 | } |
| 107 | |
| 108 | bool blkconf_blocksizes(BlockConf *conf, Error **errp) |
| 109 | { |
| 110 | BlockBackend *blk = conf->blk; |
| 111 | BlockSizes blocksizes; |
| 112 | BlockDriverState *bs; |
| 113 | bool use_blocksizes; |
| 114 | bool use_bs; |
| 115 | |
| 116 | switch (conf->backend_defaults) { |
| 117 | case ON_OFF_AUTO_AUTO: |
| 118 | use_blocksizes = !blk_probe_blocksizes(blk, &blocksizes); |
| 119 | use_bs = false; |
| 120 | break; |
| 121 | |
| 122 | case ON_OFF_AUTO_ON: |
| 123 | use_blocksizes = !blk_probe_blocksizes(blk, &blocksizes); |
| 124 | bs = blk_bs(blk); |
| 125 | use_bs = bs; |
| 126 | break; |
| 127 | |
| 128 | case ON_OFF_AUTO_OFF: |
| 129 | use_blocksizes = false; |
| 130 | use_bs = false; |
| 131 | break; |
| 132 | |
| 133 | default: |
| 134 | abort(); |
| 135 | } |
| 136 | |
| 137 | /* fill in detected values if they are not defined via qemu command line */ |
| 138 | if (!conf->physical_block_size) { |
| 139 | if (use_blocksizes) { |
| 140 | conf->physical_block_size = blocksizes.phys; |
| 141 | } else { |
| 142 | conf->physical_block_size = BDRV_SECTOR_SIZE; |
| 143 | } |
| 144 | } |
| 145 | if (!conf->logical_block_size) { |
| 146 | if (use_blocksizes) { |
| 147 | conf->logical_block_size = blocksizes.log; |
| 148 | } else { |
| 149 | conf->logical_block_size = BDRV_SECTOR_SIZE; |
| 150 | } |
| 151 | } |
| 152 | if (use_bs) { |
| 153 | if (!conf->opt_io_size) { |
| 154 | conf->opt_io_size = bs->bl.opt_transfer; |
| 155 | } |
| 156 | if (conf->discard_granularity == -1) { |
| 157 | if (bs->bl.pdiscard_alignment) { |
| 158 | conf->discard_granularity = bs->bl.pdiscard_alignment; |
| 159 | } else if (bs->bl.request_alignment != 1) { |
| 160 | conf->discard_granularity = bs->bl.request_alignment; |
| 161 | } |
| 162 | } |
| 163 | } |
| 164 | |
| 165 | if (conf->logical_block_size > conf->physical_block_size) { |
| 166 | error_setg(errp, |
| 167 | "logical_block_size > physical_block_size not supported"); |
| 168 | return false; |
| 169 | } |
| 170 | |
| 171 | if (!QEMU_IS_ALIGNED(conf->min_io_size, conf->logical_block_size)) { |
| 172 | error_setg(errp, |
| 173 | "min_io_size must be a multiple of logical_block_size"); |
| 174 | return false; |
| 175 | } |
| 176 | |
| 177 | /* |
| 178 | * all devices which support min_io_size (scsi and virtio-blk) expose it to |
| 179 | * the guest as a uint16_t in units of logical blocks |
| 180 | */ |
| 181 | if (conf->min_io_size / conf->logical_block_size > UINT16_MAX) { |
| 182 | error_setg(errp, "min_io_size must not exceed %u logical blocks", |
| 183 | UINT16_MAX); |
| 184 | return false; |
| 185 | } |
| 186 | |
| 187 | if (!QEMU_IS_ALIGNED(conf->opt_io_size, conf->logical_block_size)) { |
| 188 | error_setg(errp, |
| 189 | "opt_io_size must be a multiple of logical_block_size"); |
| 190 | return false; |
| 191 | } |
| 192 | |
| 193 | if (conf->discard_granularity != -1 && |
| 194 | !QEMU_IS_ALIGNED(conf->discard_granularity, |
| 195 | conf->logical_block_size)) { |
| 196 | error_setg(errp, "discard_granularity must be " |
| 197 | "a multiple of logical_block_size"); |
| 198 | return false; |
| 199 | } |
| 200 | |
| 201 | return true; |
| 202 | } |
| 203 | |
| 204 | bool blkconf_apply_backend_options(BlockConf *conf, bool readonly, |
| 205 | bool resizable, Error **errp) |
| 206 | { |
| 207 | BlockBackend *blk = conf->blk; |
| 208 | BlockdevOnError rerror, werror; |
| 209 | uint64_t perm, shared_perm; |
| 210 | bool wce; |
| 211 | int ret; |
| 212 | |
| 213 | perm = BLK_PERM_CONSISTENT_READ; |
| 214 | if (!readonly) { |
| 215 | perm |= BLK_PERM_WRITE; |
| 216 | } |
| 217 | |
| 218 | shared_perm = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED; |
| 219 | if (resizable) { |
| 220 | shared_perm |= BLK_PERM_RESIZE; |
| 221 | } |
| 222 | if (conf->share_rw) { |
| 223 | shared_perm |= BLK_PERM_WRITE; |
| 224 | } |
| 225 | |
| 226 | ret = blk_set_perm(blk, perm, shared_perm, errp); |
| 227 | if (ret < 0) { |
| 228 | return false; |
| 229 | } |
| 230 | |
| 231 | switch (conf->wce) { |
| 232 | case ON_OFF_AUTO_ON: wce = true; break; |
| 233 | case ON_OFF_AUTO_OFF: wce = false; break; |
| 234 | case ON_OFF_AUTO_AUTO: wce = blk_enable_write_cache(blk); break; |
| 235 | default: |
| 236 | abort(); |
| 237 | } |
| 238 | |
| 239 | rerror = conf->rerror; |
| 240 | if (rerror == BLOCKDEV_ON_ERROR_AUTO) { |
| 241 | rerror = blk_get_on_error(blk, true); |
| 242 | } |
| 243 | |
| 244 | werror = conf->werror; |
| 245 | if (werror == BLOCKDEV_ON_ERROR_AUTO) { |
| 246 | werror = blk_get_on_error(blk, false); |
| 247 | } |
| 248 | |
| 249 | blk_set_enable_write_cache(blk, wce); |
| 250 | blk_set_on_error(blk, rerror, werror); |
| 251 | |
| 252 | if (!block_acct_setup(blk_get_stats(blk), conf->account_invalid, |
| 253 | conf->account_failed, conf->stats_intervals, |
| 254 | conf->num_stats_intervals, errp)) { |
| 255 | return false; |
| 256 | } |
| 257 | return true; |
| 258 | } |
| 259 | |
| 260 | bool blkconf_geometry(BlockConf *conf, int *ptrans, |
| 261 | unsigned cyls_max, unsigned heads_max, unsigned secs_max, |
| 262 | Error **errp) |
| 263 | { |
| 264 | if (!conf->cyls && !conf->heads && !conf->secs) { |
| 265 | hd_geometry_guess(conf->blk, |
| 266 | &conf->cyls, &conf->heads, &conf->secs, |
| 267 | ptrans); |
| 268 | } else if (ptrans && *ptrans == BIOS_ATA_TRANSLATION_AUTO) { |
| 269 | *ptrans = hd_bios_chs_auto_trans(conf->cyls, conf->heads, conf->secs); |
| 270 | } |
| 271 | if (conf->cyls || conf->heads || conf->secs) { |
| 272 | if (conf->cyls < 1 || conf->cyls > cyls_max) { |
| 273 | error_setg(errp, "cyls must be between 1 and %u", cyls_max); |
| 274 | return false; |
| 275 | } |
| 276 | if (conf->heads < 1 || conf->heads > heads_max) { |
| 277 | error_setg(errp, "heads must be between 1 and %u", heads_max); |
| 278 | return false; |
| 279 | } |
| 280 | if (conf->secs < 1 || conf->secs > secs_max) { |
| 281 | error_setg(errp, "secs must be between 1 and %u", secs_max); |
| 282 | return false; |
| 283 | } |
| 284 | } |
| 285 | return true; |
| 286 | } |