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 #include "sysemu/block-backend.h"
  20
  21 enum {
  22     /*
  23      * Size of data buffer for populating the image file.  This should be large
  24      * enough to process multiple clusters in a single call, so that populating
  25      * contiguous regions of the image is efficient.
  26      */
  27     STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
  28 };
  29
  30 #define SLICE_TIME 100000000ULL /* ns */
  31
  32 typedef struct StreamBlockJob {
  33     BlockJob common;
  34     RateLimit limit;
  35     BlockDriverState *base;
  36     BlockdevOnError on_error;
  37     char *backing_file_str;
  38 } StreamBlockJob;
  39
  40 static int coroutine_fn stream_populate(BlockDriverState *bs,
  41                                         int64_t sector_num, int nb_sectors,
  42                                         void *buf)
  43 {
  44     struct iovec iov = {
  45         .iov_base = buf,
  46         .iov_len  = nb_sectors * BDRV_SECTOR_SIZE,
  47     };
  48     QEMUIOVector qiov;
  49
  50     qemu_iovec_init_external(&qiov, &iov, 1);
  51
  52     /* Copy-on-read the unallocated clusters */
  53     return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov);
  54 }
  55
  56 typedef struct {
  57     int ret;
  58     bool reached_end;
  59 } StreamCompleteData;
  60
  61 static void stream_complete(BlockJob *job, void *opaque)
  62 {
  63     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
  64     StreamCompleteData *data = opaque;
  65     BlockDriverState *base = s->base;
  66
  67     if (!block_job_is_cancelled(&s->common) && data->reached_end &&
  68         data->ret == 0) {
  69         const char *base_id = NULL, *base_fmt = NULL;
  70         if (base) {
  71             base_id = s->backing_file_str;
  72             if (base->drv) {
  73                 base_fmt = base->drv->format_name;
  74             }
  75         }
  76         data->ret = bdrv_change_backing_file(job->bs, base_id, base_fmt);
  77         bdrv_set_backing_hd(job->bs, base);
  78     }
  79
  80     g_free(s->backing_file_str);
  81     block_job_completed(&s->common, data->ret);
  82     g_free(data);
  83 }
  84
  85 static void coroutine_fn stream_run(void *opaque)
  86 {
  87     StreamBlockJob *s = opaque;
  88     StreamCompleteData *data;
  89     BlockDriverState *bs = s->common.bs;
  90     BlockDriverState *base = s->base;
  91     int64_t sector_num, end;
  92     int error = 0;
  93     int ret = 0;
  94     int n = 0;
  95     void *buf;
  96
  97     if (!bs->backing) {
  98         block_job_completed(&s->common, 0);
  99         return;
 100     }
 101
 102     s->common.len = bdrv_getlength(bs);
 103     if (s->common.len < 0) {
 104         block_job_completed(&s->common, s->common.len);
 105         return;
 106     }
 107
 108     end = s->common.len >> BDRV_SECTOR_BITS;
 109     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 110
 111     /* Turn on copy-on-read for the whole block device so that guest read
 112      * requests help us make progress.  Only do this when copying the entire
 113      * backing chain since the copy-on-read operation does not take base into
 114      * account.
 115      */
 116     if (!base) {
 117         bdrv_enable_copy_on_read(bs);
 118     }
 119
 120     for (sector_num = 0; sector_num < end; sector_num += n) {
 121         uint64_t delay_ns = 0;
 122         bool copy;
 123
 124 wait:
 125         /* Note that even when no rate limit is applied we need to yield
 126          * with no pending I/O here so that bdrv_drain_all() returns.
 127          */
 128         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 129         if (block_job_is_cancelled(&s->common)) {
 130             break;
 131         }
 132
 133         copy = false;
 134
 135         ret = bdrv_is_allocated(bs, sector_num,
 136                                 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
 137         if (ret == 1) {
 138             /* Allocated in the top, no need to copy.  */
 139         } else if (ret >= 0) {
 140             /* Copy if allocated in the intermediate images.  Limit to the
 141              * known-unallocated area [sector_num, sector_num+n).  */
 142             ret = bdrv_is_allocated_above(backing_bs(bs), base,
 143                                           sector_num, n, &n);
 144
 145             /* Finish early if end of backing file has been reached */
 146             if (ret == 0 && n == 0) {
 147                 n = end - sector_num;
 148             }
 149
 150             copy = (ret == 1);
 151         }
 152         trace_stream_one_iteration(s, sector_num, n, ret);
 153         if (copy) {
 154             if (s->common.speed) {
 155                 delay_ns = ratelimit_calculate_delay(&s->limit, n);
 156                 if (delay_ns > 0) {
 157                     goto wait;
 158                 }
 159             }
 160             ret = stream_populate(bs, sector_num, n, buf);
 161         }
 162         if (ret < 0) {
 163             BlockErrorAction action =
 164                 block_job_error_action(&s->common, s->common.bs, s->on_error,
 165                                        true, -ret);
 166             if (action == BLOCK_ERROR_ACTION_STOP) {
 167                 n = 0;
 168                 continue;
 169             }
 170             if (error == 0) {
 171                 error = ret;
 172             }
 173             if (action == BLOCK_ERROR_ACTION_REPORT) {
 174                 break;
 175             }
 176         }
 177         ret = 0;
 178
 179         /* Publish progress */
 180         s->common.offset += n * BDRV_SECTOR_SIZE;
 181     }
 182
 183     if (!base) {
 184         bdrv_disable_copy_on_read(bs);
 185     }
 186
 187     /* Do not remove the backing file if an error was there but ignored.  */
 188     ret = error;
 189
 190     qemu_vfree(buf);
 191
 192     /* Modify backing chain and close BDSes in main loop */
 193     data = g_malloc(sizeof(*data));
 194     data->ret = ret;
 195     data->reached_end = sector_num == end;
 196     block_job_defer_to_main_loop(&s->common, stream_complete, data);
 197 }
 198
 199 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
 200 {
 201     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
 202
 203     if (speed < 0) {
 204         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 205         return;
 206     }
 207     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 208 }
 209
 210 static const BlockJobDriver stream_job_driver = {
 211     .instance_size = sizeof(StreamBlockJob),
 212     .job_type      = BLOCK_JOB_TYPE_STREAM,
 213     .set_speed     = stream_set_speed,
 214 };
 215
 216 void stream_start(BlockDriverState *bs, BlockDriverState *base,
 217                   const char *backing_file_str, int64_t speed,
 218                   BlockdevOnError on_error,
 219                   BlockCompletionFunc *cb,
 220                   void *opaque, Error **errp)
 221 {
 222     StreamBlockJob *s;
 223
 224     if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
 225          on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
 226         (!bs->blk || !blk_iostatus_is_enabled(bs->blk))) {
 227         error_setg(errp, QERR_INVALID_PARAMETER, "on-error");
 228         return;
 229     }
 230
 231     s = block_job_create(&stream_job_driver, bs, speed, cb, opaque, errp);
 232     if (!s) {
 233         return;
 234     }
 235
 236     s->base = base;
 237     s->backing_file_str = g_strdup(backing_file_str);
 238
 239     s->on_error = on_error;
 240     s->common.co = qemu_coroutine_create(stream_run);
 241     trace_stream_start(bs, base, s, s->common.co, opaque);
 242     qemu_coroutine_enter(s->common.co, s);
 243 }