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 "qemu/ratelimit.h"
  18
  19 enum {
  20     /*
  21      * Size of data buffer for populating the image file.  This should be large
  22      * enough to process multiple clusters in a single call, so that populating
  23      * contiguous regions of the image is efficient.
  24      */
  25     STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
  26 };
  27
  28 #define SLICE_TIME 100000000ULL /* ns */
  29
  30 typedef struct StreamBlockJob {
  31     BlockJob common;
  32     RateLimit limit;
  33     BlockDriverState *base;
  34     BlockdevOnError on_error;
  35     char *backing_file_str;
  36 } StreamBlockJob;
  37
  38 static int coroutine_fn stream_populate(BlockDriverState *bs,
  39                                         int64_t sector_num, int nb_sectors,
  40                                         void *buf)
  41 {
  42     struct iovec iov = {
  43         .iov_base = buf,
  44         .iov_len  = nb_sectors * BDRV_SECTOR_SIZE,
  45     };
  46     QEMUIOVector qiov;
  47
  48     qemu_iovec_init_external(&qiov, &iov, 1);
  49
  50     /* Copy-on-read the unallocated clusters */
  51     return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov);
  52 }
  53
  54 static void close_unused_images(BlockDriverState *top, BlockDriverState *base,
  55                                 const char *base_id)
  56 {
  57     BlockDriverState *intermediate;
  58     intermediate = top->backing_hd;
  59
  60     /* Must assign before bdrv_delete() to prevent traversing dangling pointer
  61      * while we delete backing image instances.
  62      */
  63     bdrv_set_backing_hd(top, base);
  64
  65     while (intermediate) {
  66         BlockDriverState *unused;
  67
  68         /* reached base */
  69         if (intermediate == base) {
  70             break;
  71         }
  72
  73         unused = intermediate;
  74         intermediate = intermediate->backing_hd;
  75         bdrv_set_backing_hd(unused, NULL);
  76         bdrv_unref(unused);
  77     }
  78
  79     bdrv_refresh_limits(top, NULL);
  80 }
  81
  82 static void coroutine_fn stream_run(void *opaque)
  83 {
  84     StreamBlockJob *s = opaque;
  85     BlockDriverState *bs = s->common.bs;
  86     BlockDriverState *base = s->base;
  87     int64_t sector_num, end;
  88     int error = 0;
  89     int ret = 0;
  90     int n = 0;
  91     void *buf;
  92
  93     if (!bs->backing_hd) {
  94         block_job_completed(&s->common, 0);
  95         return;
  96     }
  97
  98     s->common.len = bdrv_getlength(bs);
  99     if (s->common.len < 0) {
 100         block_job_completed(&s->common, s->common.len);
 101         return;
 102     }
 103
 104     end = s->common.len >> BDRV_SECTOR_BITS;
 105     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 106
 107     /* Turn on copy-on-read for the whole block device so that guest read
 108      * requests help us make progress.  Only do this when copying the entire
 109      * backing chain since the copy-on-read operation does not take base into
 110      * account.
 111      */
 112     if (!base) {
 113         bdrv_enable_copy_on_read(bs);
 114     }
 115
 116     for (sector_num = 0; sector_num < end; sector_num += n) {
 117         uint64_t delay_ns = 0;
 118         bool copy;
 119
 120 wait:
 121         /* Note that even when no rate limit is applied we need to yield
 122          * with no pending I/O here so that bdrv_drain_all() returns.
 123          */
 124         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 125         if (block_job_is_cancelled(&s->common)) {
 126             break;
 127         }
 128
 129         copy = false;
 130
 131         ret = bdrv_is_allocated(bs, sector_num,
 132                                 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
 133         if (ret == 1) {
 134             /* Allocated in the top, no need to copy.  */
 135         } else if (ret >= 0) {
 136             /* Copy if allocated in the intermediate images.  Limit to the
 137              * known-unallocated area [sector_num, sector_num+n).  */
 138             ret = bdrv_is_allocated_above(bs->backing_hd, base,
 139                                           sector_num, n, &n);
 140
 141             /* Finish early if end of backing file has been reached */
 142             if (ret == 0 && n == 0) {
 143                 n = end - sector_num;
 144             }
 145
 146             copy = (ret == 1);
 147         }
 148         trace_stream_one_iteration(s, sector_num, n, ret);
 149         if (copy) {
 150             if (s->common.speed) {
 151                 delay_ns = ratelimit_calculate_delay(&s->limit, n);
 152                 if (delay_ns > 0) {
 153                     goto wait;
 154                 }
 155             }
 156             ret = stream_populate(bs, sector_num, n, buf);
 157         }
 158         if (ret < 0) {
 159             BlockErrorAction action =
 160                 block_job_error_action(&s->common, s->common.bs, s->on_error,
 161                                        true, -ret);
 162             if (action == BLOCK_ERROR_ACTION_STOP) {
 163                 n = 0;
 164                 continue;
 165             }
 166             if (error == 0) {
 167                 error = ret;
 168             }
 169             if (action == BLOCK_ERROR_ACTION_REPORT) {
 170                 break;
 171             }
 172         }
 173         ret = 0;
 174
 175         /* Publish progress */
 176         s->common.offset += n * BDRV_SECTOR_SIZE;
 177     }
 178
 179     if (!base) {
 180         bdrv_disable_copy_on_read(bs);
 181     }
 182
 183     /* Do not remove the backing file if an error was there but ignored.  */
 184     ret = error;
 185
 186     if (!block_job_is_cancelled(&s->common) && sector_num == end && ret == 0) {
 187         const char *base_id = NULL, *base_fmt = NULL;
 188         if (base) {
 189             base_id = s->backing_file_str;
 190             if (base->drv) {
 191                 base_fmt = base->drv->format_name;
 192             }
 193         }
 194         ret = bdrv_change_backing_file(bs, base_id, base_fmt);
 195         close_unused_images(bs, base, base_id);
 196     }
 197
 198     qemu_vfree(buf);
 199     g_free(s->backing_file_str);
 200     block_job_completed(&s->common, ret);
 201 }
 202
 203 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
 204 {
 205     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
 206
 207     if (speed < 0) {
 208         error_set(errp, QERR_INVALID_PARAMETER, "speed");
 209         return;
 210     }
 211     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 212 }
 213
 214 static const BlockJobDriver stream_job_driver = {
 215     .instance_size = sizeof(StreamBlockJob),
 216     .job_type      = BLOCK_JOB_TYPE_STREAM,
 217     .set_speed     = stream_set_speed,
 218 };
 219
 220 void stream_start(BlockDriverState *bs, BlockDriverState *base,
 221                   const char *backing_file_str, int64_t speed,
 222                   BlockdevOnError on_error,
 223                   BlockDriverCompletionFunc *cb,
 224                   void *opaque, Error **errp)
 225 {
 226     StreamBlockJob *s;
 227
 228     if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
 229          on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
 230         !bdrv_iostatus_is_enabled(bs)) {
 231         error_set(errp, QERR_INVALID_PARAMETER, "on-error");
 232         return;
 233     }
 234
 235     s = block_job_create(&stream_job_driver, bs, speed, cb, opaque, errp);
 236     if (!s) {
 237         return;
 238     }
 239
 240     s->base = base;
 241     s->backing_file_str = g_strdup(backing_file_str);
 242
 243     s->on_error = on_error;
 244     s->common.co = qemu_coroutine_create(stream_run);
 245     trace_stream_start(bs, base, s, s->common.co, opaque);
 246     qemu_coroutine_enter(s->common.co, s);
 247 }