block/stream.c

   1 /*
   2  * Image streaming
   3  *
   4  * Copyright IBM, Corp. 2011
   5  *
   6  * Authors:
   7  *  Stefan Hajnoczi   <stefanha@linux.vnet.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU LGPL, version 2 or later.
  10  * See the COPYING.LIB file in the top-level directory.
  11  *
  12  */
  13
  14 #include "trace.h"
  15 #include "block/block_int.h"
  16 #include "block/blockjob.h"
  17 #include "qapi/qmp/qerror.h"
  18 #include "qemu/ratelimit.h"
  19
  20 enum {
  21     /*
  22      * Size of data buffer for populating the image file.  This should be large
  23      * enough to process multiple clusters in a single call, so that populating
  24      * contiguous regions of the image is efficient.
  25      */
  26     STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
  27 };
  28
  29 #define SLICE_TIME 100000000ULL /* ns */
  30
  31 typedef struct StreamBlockJob {
  32     BlockJob common;
  33     RateLimit limit;
  34     BlockDriverState *base;
  35     BlockdevOnError on_error;
  36     char *backing_file_str;
  37 } StreamBlockJob;
  38
  39 static int coroutine_fn stream_populate(BlockDriverState *bs,
  40                                         int64_t sector_num, int nb_sectors,
  41                                         void *buf)
  42 {
  43     struct iovec iov = {
  44         .iov_base = buf,
  45         .iov_len  = nb_sectors * BDRV_SECTOR_SIZE,
  46     };
  47     QEMUIOVector qiov;
  48
  49     qemu_iovec_init_external(&qiov, &iov, 1);
  50
  51     /* Copy-on-read the unallocated clusters */
  52     return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov);
  53 }
  54
  55 typedef struct {
  56     int ret;
  57     bool reached_end;
  58 } StreamCompleteData;
  59
  60 static void stream_complete(BlockJob *job, void *opaque)
  61 {
  62     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
  63     StreamCompleteData *data = opaque;
  64     BlockDriverState *base = s->base;
  65
  66     if (!block_job_is_cancelled(&s->common) && data->reached_end &&
  67         data->ret == 0) {
  68         const char *base_id = NULL, *base_fmt = NULL;
  69         if (base) {
  70             base_id = s->backing_file_str;
  71             if (base->drv) {
  72                 base_fmt = base->drv->format_name;
  73             }
  74         }
  75         data->ret = bdrv_change_backing_file(job->bs, base_id, base_fmt);
  76         bdrv_set_backing_hd(job->bs, base);
  77     }
  78
  79     g_free(s->backing_file_str);
  80     block_job_completed(&s->common, data->ret);
  81     g_free(data);
  82 }
  83
  84 static void coroutine_fn stream_run(void *opaque)
  85 {
  86     StreamBlockJob *s = opaque;
  87     StreamCompleteData *data;
  88     BlockDriverState *bs = s->common.bs;
  89     BlockDriverState *base = s->base;
  90     int64_t sector_num, end;
  91     int error = 0;
  92     int ret = 0;
  93     int n = 0;
  94     void *buf;
  95
  96     if (!bs->backing) {
  97         block_job_completed(&s->common, 0);
  98         return;
  99     }
 100
 101     s->common.len = bdrv_getlength(bs);
 102     if (s->common.len < 0) {
 103         block_job_completed(&s->common, s->common.len);
 104         return;
 105     }
 106
 107     end = s->common.len >> BDRV_SECTOR_BITS;
 108     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 109
 110     /* Turn on copy-on-read for the whole block device so that guest read
 111      * requests help us make progress.  Only do this when copying the entire
 112      * backing chain since the copy-on-read operation does not take base into
 113      * account.
 114      */
 115     if (!base) {
 116         bdrv_enable_copy_on_read(bs);
 117     }
 118
 119     for (sector_num = 0; sector_num < end; sector_num += n) {
 120         uint64_t delay_ns = 0;
 121         bool copy;
 122
 123 wait:
 124         /* Note that even when no rate limit is applied we need to yield
 125          * with no pending I/O here so that bdrv_drain_all() returns.
 126          */
 127         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 128         if (block_job_is_cancelled(&s->common)) {
 129             break;
 130         }
 131
 132         copy = false;
 133
 134         ret = bdrv_is_allocated(bs, sector_num,
 135                                 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
 136         if (ret == 1) {
 137             /* Allocated in the top, no need to copy.  */
 138         } else if (ret >= 0) {
 139             /* Copy if allocated in the intermediate images.  Limit to the
 140              * known-unallocated area [sector_num, sector_num+n).  */
 141             ret = bdrv_is_allocated_above(backing_bs(bs), base,
 142                                           sector_num, n, &n);
 143
 144             /* Finish early if end of backing file has been reached */
 145             if (ret == 0 && n == 0) {
 146                 n = end - sector_num;
 147             }
 148
 149             copy = (ret == 1);
 150         }
 151         trace_stream_one_iteration(s, sector_num, n, ret);
 152         if (copy) {
 153             if (s->common.speed) {
 154                 delay_ns = ratelimit_calculate_delay(&s->limit, n);
 155                 if (delay_ns > 0) {
 156                     goto wait;
 157                 }
 158             }
 159             ret = stream_populate(bs, sector_num, n, buf);
 160         }
 161         if (ret < 0) {
 162             BlockErrorAction action =
 163                 block_job_error_action(&s->common, s->common.bs, s->on_error,
 164                                        true, -ret);
 165             if (action == BLOCK_ERROR_ACTION_STOP) {
 166                 n = 0;
 167                 continue;
 168             }
 169             if (error == 0) {
 170                 error = ret;
 171             }
 172             if (action == BLOCK_ERROR_ACTION_REPORT) {
 173                 break;
 174             }
 175         }
 176         ret = 0;
 177
 178         /* Publish progress */
 179         s->common.offset += n * BDRV_SECTOR_SIZE;
 180     }
 181
 182     if (!base) {
 183         bdrv_disable_copy_on_read(bs);
 184     }
 185
 186     /* Do not remove the backing file if an error was there but ignored.  */
 187     ret = error;
 188
 189     qemu_vfree(buf);
 190
 191     /* Modify backing chain and close BDSes in main loop */
 192     data = g_malloc(sizeof(*data));
 193     data->ret = ret;
 194     data->reached_end = sector_num == end;
 195     block_job_defer_to_main_loop(&s->common, stream_complete, data);
 196 }
 197
 198 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
 199 {
 200     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
 201
 202     if (speed < 0) {
 203         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 204         return;
 205     }
 206     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 207 }
 208
 209 static const BlockJobDriver stream_job_driver = {
 210     .instance_size = sizeof(StreamBlockJob),
 211     .job_type      = BLOCK_JOB_TYPE_STREAM,
 212     .set_speed     = stream_set_speed,
 213 };
 214
 215 void stream_start(BlockDriverState *bs, BlockDriverState *base,
 216                   const char *backing_file_str, int64_t speed,
 217                   BlockdevOnError on_error,
 218                   BlockCompletionFunc *cb,
 219                   void *opaque, Error **errp)
 220 {
 221     StreamBlockJob *s;
 222
 223     if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
 224          on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
 225         !bdrv_iostatus_is_enabled(bs)) {
 226         error_setg(errp, QERR_INVALID_PARAMETER, "on-error");
 227         return;
 228     }
 229
 230     s = block_job_create(&stream_job_driver, bs, speed, cb, opaque, errp);
 231     if (!s) {
 232         return;
 233     }
 234
 235     s->base = base;
 236     s->backing_file_str = g_strdup(backing_file_str);
 237
 238     s->on_error = on_error;
 239     s->common.co = qemu_coroutine_create(stream_run);
 240     trace_stream_start(bs, base, s, s->common.co, opaque);
 241     qemu_coroutine_enter(s->common.co, s);
 242 }