hw/block/block.c

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