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/error.h"
  19 #include "qapi/qmp/qerror.h"
  20 #include "qemu/ratelimit.h"
  21 #include "sysemu/block-backend.h"
  22
  23 enum {
  24     /*
  25      * Size of data buffer for populating the image file.  This should be large
  26      * enough to process multiple clusters in a single call, so that populating
  27      * contiguous regions of the image is efficient.
  28      */
  29     STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
  30 };
  31
  32 #define SLICE_TIME 100000000ULL /* ns */
  33
  34 typedef struct StreamBlockJob {
  35     BlockJob common;
  36     RateLimit limit;
  37     BlockDriverState *base;
  38     BlockdevOnError on_error;
  39     char *backing_file_str;
  40 } StreamBlockJob;
  41
  42 static int coroutine_fn stream_populate(BlockDriverState *bs,
  43                                         int64_t sector_num, int nb_sectors,
  44                                         void *buf)
  45 {
  46     struct iovec iov = {
  47         .iov_base = buf,
  48         .iov_len  = nb_sectors * BDRV_SECTOR_SIZE,
  49     };
  50     QEMUIOVector qiov;
  51
  52     qemu_iovec_init_external(&qiov, &iov, 1);
  53
  54     /* Copy-on-read the unallocated clusters */
  55     return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov);
  56 }
  57
  58 typedef struct {
  59     int ret;
  60     bool reached_end;
  61 } StreamCompleteData;
  62
  63 static void stream_complete(BlockJob *job, void *opaque)
  64 {
  65     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
  66     StreamCompleteData *data = opaque;
  67     BlockDriverState *base = s->base;
  68
  69     if (!block_job_is_cancelled(&s->common) && data->reached_end &&
  70         data->ret == 0) {
  71         const char *base_id = NULL, *base_fmt = NULL;
  72         if (base) {
  73             base_id = s->backing_file_str;
  74             if (base->drv) {
  75                 base_fmt = base->drv->format_name;
  76             }
  77         }
  78         data->ret = bdrv_change_backing_file(job->bs, base_id, base_fmt);
  79         bdrv_set_backing_hd(job->bs, base);
  80     }
  81
  82     g_free(s->backing_file_str);
  83     block_job_completed(&s->common, data->ret);
  84     g_free(data);
  85 }
  86
  87 static void coroutine_fn stream_run(void *opaque)
  88 {
  89     StreamBlockJob *s = opaque;
  90     StreamCompleteData *data;
  91     BlockDriverState *bs = s->common.bs;
  92     BlockDriverState *base = s->base;
  93     int64_t sector_num, end;
  94     int error = 0;
  95     int ret = 0;
  96     int n = 0;
  97     void *buf;
  98
  99     if (!bs->backing) {
 100         block_job_completed(&s->common, 0);
 101         return;
 102     }
 103
 104     s->common.len = bdrv_getlength(bs);
 105     if (s->common.len < 0) {
 106         block_job_completed(&s->common, s->common.len);
 107         return;
 108     }
 109
 110     end = s->common.len >> BDRV_SECTOR_BITS;
 111     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 112
 113     /* Turn on copy-on-read for the whole block device so that guest read
 114      * requests help us make progress.  Only do this when copying the entire
 115      * backing chain since the copy-on-read operation does not take base into
 116      * account.
 117      */
 118     if (!base) {
 119         bdrv_enable_copy_on_read(bs);
 120     }
 121
 122     for (sector_num = 0; sector_num < end; sector_num += n) {
 123         uint64_t delay_ns = 0;
 124         bool copy;
 125
 126 wait:
 127         /* Note that even when no rate limit is applied we need to yield
 128          * with no pending I/O here so that bdrv_drain_all() returns.
 129          */
 130         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 131         if (block_job_is_cancelled(&s->common)) {
 132             break;
 133         }
 134
 135         copy = false;
 136
 137         ret = bdrv_is_allocated(bs, sector_num,
 138                                 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
 139         if (ret == 1) {
 140             /* Allocated in the top, no need to copy.  */
 141         } else if (ret >= 0) {
 142             /* Copy if allocated in the intermediate images.  Limit to the
 143              * known-unallocated area [sector_num, sector_num+n).  */
 144             ret = bdrv_is_allocated_above(backing_bs(bs), base,
 145                                           sector_num, n, &n);
 146
 147             /* Finish early if end of backing file has been reached */
 148             if (ret == 0 && n == 0) {
 149                 n = end - sector_num;
 150             }
 151
 152             copy = (ret == 1);
 153         }
 154         trace_stream_one_iteration(s, sector_num, n, ret);
 155         if (copy) {
 156             if (s->common.speed) {
 157                 delay_ns = ratelimit_calculate_delay(&s->limit, n);
 158                 if (delay_ns > 0) {
 159                     goto wait;
 160                 }
 161             }
 162             ret = stream_populate(bs, sector_num, n, buf);
 163         }
 164         if (ret < 0) {
 165             BlockErrorAction action =
 166                 block_job_error_action(&s->common, s->common.bs, s->on_error,
 167                                        true, -ret);
 168             if (action == BLOCK_ERROR_ACTION_STOP) {
 169                 n = 0;
 170                 continue;
 171             }
 172             if (error == 0) {
 173                 error = ret;
 174             }
 175             if (action == BLOCK_ERROR_ACTION_REPORT) {
 176                 break;
 177             }
 178         }
 179         ret = 0;
 180
 181         /* Publish progress */
 182         s->common.offset += n * BDRV_SECTOR_SIZE;
 183     }
 184
 185     if (!base) {
 186         bdrv_disable_copy_on_read(bs);
 187     }
 188
 189     /* Do not remove the backing file if an error was there but ignored.  */
 190     ret = error;
 191
 192     qemu_vfree(buf);
 193
 194     /* Modify backing chain and close BDSes in main loop */
 195     data = g_malloc(sizeof(*data));
 196     data->ret = ret;
 197     data->reached_end = sector_num == end;
 198     block_job_defer_to_main_loop(&s->common, stream_complete, data);
 199 }
 200
 201 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
 202 {
 203     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
 204
 205     if (speed < 0) {
 206         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 207         return;
 208     }
 209     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 210 }
 211
 212 static const BlockJobDriver stream_job_driver = {
 213     .instance_size = sizeof(StreamBlockJob),
 214     .job_type      = BLOCK_JOB_TYPE_STREAM,
 215     .set_speed     = stream_set_speed,
 216 };
 217
 218 void stream_start(BlockDriverState *bs, BlockDriverState *base,
 219                   const char *backing_file_str, int64_t speed,
 220                   BlockdevOnError on_error,
 221                   BlockCompletionFunc *cb,
 222                   void *opaque, Error **errp)
 223 {
 224     StreamBlockJob *s;
 225
 226     if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
 227          on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
 228         (!bs->blk || !blk_iostatus_is_enabled(bs->blk))) {
 229         error_setg(errp, QERR_INVALID_PARAMETER, "on-error");
 230         return;
 231     }
 232
 233     s = block_job_create(&stream_job_driver, bs, speed, cb, opaque, errp);
 234     if (!s) {
 235         return;
 236     }
 237
 238     s->base = base;
 239     s->backing_file_str = g_strdup(backing_file_str);
 240
 241     s->on_error = on_error;
 242     s->common.co = qemu_coroutine_create(stream_run);
 243     trace_stream_start(bs, base, s, s->common.co, opaque);
 244     qemu_coroutine_enter(s->common.co, s);
 245 }