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 "qemu/osdep.h"
  15 #include "trace.h"
  16 #include "block/block_int.h"
  17 #include "block/blockjob.h"
  18 #include "qapi/qmp/qerror.h"
  19 #include "qemu/ratelimit.h"
  20 #include "sysemu/block-backend.h"
  21
  22 enum {
  23     /*
  24      * Size of data buffer for populating the image file.  This should be large
  25      * enough to process multiple clusters in a single call, so that populating
  26      * contiguous regions of the image is efficient.
  27      */
  28     STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
  29 };
  30
  31 #define SLICE_TIME 100000000ULL /* ns */
  32
  33 typedef struct StreamBlockJob {
  34     BlockJob common;
  35     RateLimit limit;
  36     BlockDriverState *base;
  37     BlockdevOnError on_error;
  38     char *backing_file_str;
  39 } StreamBlockJob;
  40
  41 static int coroutine_fn stream_populate(BlockDriverState *bs,
  42                                         int64_t sector_num, int nb_sectors,
  43                                         void *buf)
  44 {
  45     struct iovec iov = {
  46         .iov_base = buf,
  47         .iov_len  = nb_sectors * BDRV_SECTOR_SIZE,
  48     };
  49     QEMUIOVector qiov;
  50
  51     qemu_iovec_init_external(&qiov, &iov, 1);
  52
  53     /* Copy-on-read the unallocated clusters */
  54     return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov);
  55 }
  56
  57 typedef struct {
  58     int ret;
  59     bool reached_end;
  60 } StreamCompleteData;
  61
  62 static void stream_complete(BlockJob *job, void *opaque)
  63 {
  64     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
  65     StreamCompleteData *data = opaque;
  66     BlockDriverState *base = s->base;
  67
  68     if (!block_job_is_cancelled(&s->common) && data->reached_end &&
  69         data->ret == 0) {
  70         const char *base_id = NULL, *base_fmt = NULL;
  71         if (base) {
  72             base_id = s->backing_file_str;
  73             if (base->drv) {
  74                 base_fmt = base->drv->format_name;
  75             }
  76         }
  77         data->ret = bdrv_change_backing_file(job->bs, base_id, base_fmt);
  78         bdrv_set_backing_hd(job->bs, base);
  79     }
  80
  81     g_free(s->backing_file_str);
  82     block_job_completed(&s->common, data->ret);
  83     g_free(data);
  84 }
  85
  86 static void coroutine_fn stream_run(void *opaque)
  87 {
  88     StreamBlockJob *s = opaque;
  89     StreamCompleteData *data;
  90     BlockDriverState *bs = s->common.bs;
  91     BlockDriverState *base = s->base;
  92     int64_t sector_num, end;
  93     int error = 0;
  94     int ret = 0;
  95     int n = 0;
  96     void *buf;
  97
  98     if (!bs->backing) {
  99         block_job_completed(&s->common, 0);
 100         return;
 101     }
 102
 103     s->common.len = bdrv_getlength(bs);
 104     if (s->common.len < 0) {
 105         block_job_completed(&s->common, s->common.len);
 106         return;
 107     }
 108
 109     end = s->common.len >> BDRV_SECTOR_BITS;
 110     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 111
 112     /* Turn on copy-on-read for the whole block device so that guest read
 113      * requests help us make progress.  Only do this when copying the entire
 114      * backing chain since the copy-on-read operation does not take base into
 115      * account.
 116      */
 117     if (!base) {
 118         bdrv_enable_copy_on_read(bs);
 119     }
 120
 121     for (sector_num = 0; sector_num < end; sector_num += n) {
 122         uint64_t delay_ns = 0;
 123         bool copy;
 124
 125 wait:
 126         /* Note that even when no rate limit is applied we need to yield
 127          * with no pending I/O here so that bdrv_drain_all() returns.
 128          */
 129         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 130         if (block_job_is_cancelled(&s->common)) {
 131             break;
 132         }
 133
 134         copy = false;
 135
 136         ret = bdrv_is_allocated(bs, sector_num,
 137                                 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
 138         if (ret == 1) {
 139             /* Allocated in the top, no need to copy.  */
 140         } else if (ret >= 0) {
 141             /* Copy if allocated in the intermediate images.  Limit to the
 142              * known-unallocated area [sector_num, sector_num+n).  */
 143             ret = bdrv_is_allocated_above(backing_bs(bs), base,
 144                                           sector_num, n, &n);
 145
 146             /* Finish early if end of backing file has been reached */
 147             if (ret == 0 && n == 0) {
 148                 n = end - sector_num;
 149             }
 150
 151             copy = (ret == 1);
 152         }
 153         trace_stream_one_iteration(s, sector_num, n, ret);
 154         if (copy) {
 155             if (s->common.speed) {
 156                 delay_ns = ratelimit_calculate_delay(&s->limit, n);
 157                 if (delay_ns > 0) {
 158                     goto wait;
 159                 }
 160             }
 161             ret = stream_populate(bs, sector_num, n, buf);
 162         }
 163         if (ret < 0) {
 164             BlockErrorAction action =
 165                 block_job_error_action(&s->common, s->common.bs, s->on_error,
 166                                        true, -ret);
 167             if (action == BLOCK_ERROR_ACTION_STOP) {
 168                 n = 0;
 169                 continue;
 170             }
 171             if (error == 0) {
 172                 error = ret;
 173             }
 174             if (action == BLOCK_ERROR_ACTION_REPORT) {
 175                 break;
 176             }
 177         }
 178         ret = 0;
 179
 180         /* Publish progress */
 181         s->common.offset += n * BDRV_SECTOR_SIZE;
 182     }
 183
 184     if (!base) {
 185         bdrv_disable_copy_on_read(bs);
 186     }
 187
 188     /* Do not remove the backing file if an error was there but ignored.  */
 189     ret = error;
 190
 191     qemu_vfree(buf);
 192
 193     /* Modify backing chain and close BDSes in main loop */
 194     data = g_malloc(sizeof(*data));
 195     data->ret = ret;
 196     data->reached_end = sector_num == end;
 197     block_job_defer_to_main_loop(&s->common, stream_complete, data);
 198 }
 199
 200 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
 201 {
 202     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
 203
 204     if (speed < 0) {
 205         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 206         return;
 207     }
 208     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 209 }
 210
 211 static const BlockJobDriver stream_job_driver = {
 212     .instance_size = sizeof(StreamBlockJob),
 213     .job_type      = BLOCK_JOB_TYPE_STREAM,
 214     .set_speed     = stream_set_speed,
 215 };
 216
 217 void stream_start(BlockDriverState *bs, BlockDriverState *base,
 218                   const char *backing_file_str, int64_t speed,
 219                   BlockdevOnError on_error,
 220                   BlockCompletionFunc *cb,
 221                   void *opaque, Error **errp)
 222 {
 223     StreamBlockJob *s;
 224
 225     if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
 226          on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
 227         (!bs->blk || !blk_iostatus_is_enabled(bs->blk))) {
 228         error_setg(errp, QERR_INVALID_PARAMETER, "on-error");
 229         return;
 230     }
 231
 232     s = block_job_create(&stream_job_driver, bs, speed, cb, opaque, errp);
 233     if (!s) {
 234         return;
 235     }
 236
 237     s->base = base;
 238     s->backing_file_str = g_strdup(backing_file_str);
 239
 240     s->on_error = on_error;
 241     s->common.co = qemu_coroutine_create(stream_run);
 242     trace_stream_start(bs, base, s, s->common.co, opaque);
 243     qemu_coroutine_enter(s->common.co, s);
 244 }