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_int.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     int bs_flags;
  41 } StreamBlockJob;
  42
  43 static int coroutine_fn stream_populate(BlockBackend *blk,
  44                                         int64_t sector_num, int nb_sectors,
  45                                         void *buf)
  46 {
  47     struct iovec iov = {
  48         .iov_base = buf,
  49         .iov_len  = nb_sectors * BDRV_SECTOR_SIZE,
  50     };
  51     QEMUIOVector qiov;
  52
  53     qemu_iovec_init_external(&qiov, &iov, 1);
  54
  55     /* Copy-on-read the unallocated clusters */
  56     return blk_co_preadv(blk, sector_num * BDRV_SECTOR_SIZE, qiov.size, &qiov,
  57                          BDRV_REQ_COPY_ON_READ);
  58 }
  59
  60 typedef struct {
  61     int ret;
  62     bool reached_end;
  63 } StreamCompleteData;
  64
  65 static void stream_complete(BlockJob *job, void *opaque)
  66 {
  67     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
  68     StreamCompleteData *data = opaque;
  69     BlockDriverState *bs = blk_bs(job->blk);
  70     BlockDriverState *base = s->base;
  71
  72     if (!block_job_is_cancelled(&s->common) && data->reached_end &&
  73         data->ret == 0) {
  74         const char *base_id = NULL, *base_fmt = NULL;
  75         if (base) {
  76             base_id = s->backing_file_str;
  77             if (base->drv) {
  78                 base_fmt = base->drv->format_name;
  79             }
  80         }
  81         data->ret = bdrv_change_backing_file(bs, base_id, base_fmt);
  82         bdrv_set_backing_hd(bs, base);
  83     }
  84
  85     /* Reopen the image back in read-only mode if necessary */
  86     if (s->bs_flags != bdrv_get_flags(bs)) {
  87         bdrv_reopen(bs, s->bs_flags, NULL);
  88     }
  89
  90     g_free(s->backing_file_str);
  91     block_job_completed(&s->common, data->ret);
  92     g_free(data);
  93 }
  94
  95 static void coroutine_fn stream_run(void *opaque)
  96 {
  97     StreamBlockJob *s = opaque;
  98     StreamCompleteData *data;
  99     BlockBackend *blk = s->common.blk;
 100     BlockDriverState *bs = blk_bs(blk);
 101     BlockDriverState *base = s->base;
 102     int64_t sector_num = 0;
 103     int64_t end = -1;
 104     uint64_t delay_ns = 0;
 105     int error = 0;
 106     int ret = 0;
 107     int n = 0;
 108     void *buf;
 109
 110     if (!bs->backing) {
 111         goto out;
 112     }
 113
 114     s->common.len = bdrv_getlength(bs);
 115     if (s->common.len < 0) {
 116         ret = s->common.len;
 117         goto out;
 118     }
 119
 120     end = s->common.len >> BDRV_SECTOR_BITS;
 121     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 122
 123     /* Turn on copy-on-read for the whole block device so that guest read
 124      * requests help us make progress.  Only do this when copying the entire
 125      * backing chain since the copy-on-read operation does not take base into
 126      * account.
 127      */
 128     if (!base) {
 129         bdrv_enable_copy_on_read(bs);
 130     }
 131
 132     for (sector_num = 0; sector_num < end; sector_num += n) {
 133         bool copy;
 134
 135         /* Note that even when no rate limit is applied we need to yield
 136          * with no pending I/O here so that bdrv_drain_all() returns.
 137          */
 138         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 139         if (block_job_is_cancelled(&s->common)) {
 140             break;
 141         }
 142
 143         copy = false;
 144
 145         ret = bdrv_is_allocated(bs, sector_num,
 146                                 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
 147         if (ret == 1) {
 148             /* Allocated in the top, no need to copy.  */
 149         } else if (ret >= 0) {
 150             /* Copy if allocated in the intermediate images.  Limit to the
 151              * known-unallocated area [sector_num, sector_num+n).  */
 152             ret = bdrv_is_allocated_above(backing_bs(bs), base,
 153                                           sector_num, n, &n);
 154
 155             /* Finish early if end of backing file has been reached */
 156             if (ret == 0 && n == 0) {
 157                 n = end - sector_num;
 158             }
 159
 160             copy = (ret == 1);
 161         }
 162         trace_stream_one_iteration(s, sector_num, n, ret);
 163         if (copy) {
 164             ret = stream_populate(blk, sector_num, n, buf);
 165         }
 166         if (ret < 0) {
 167             BlockErrorAction action =
 168                 block_job_error_action(&s->common, s->on_error, true, -ret);
 169             if (action == BLOCK_ERROR_ACTION_STOP) {
 170                 n = 0;
 171                 continue;
 172             }
 173             if (error == 0) {
 174                 error = ret;
 175             }
 176             if (action == BLOCK_ERROR_ACTION_REPORT) {
 177                 break;
 178             }
 179         }
 180         ret = 0;
 181
 182         /* Publish progress */
 183         s->common.offset += n * BDRV_SECTOR_SIZE;
 184         if (copy && s->common.speed) {
 185             delay_ns = ratelimit_calculate_delay(&s->limit, n);
 186         }
 187     }
 188
 189     if (!base) {
 190         bdrv_disable_copy_on_read(bs);
 191     }
 192
 193     /* Do not remove the backing file if an error was there but ignored.  */
 194     ret = error;
 195
 196     qemu_vfree(buf);
 197
 198 out:
 199     /* Modify backing chain and close BDSes in main loop */
 200     data = g_malloc(sizeof(*data));
 201     data->ret = ret;
 202     data->reached_end = sector_num == end;
 203     block_job_defer_to_main_loop(&s->common, stream_complete, data);
 204 }
 205
 206 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
 207 {
 208     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
 209
 210     if (speed < 0) {
 211         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 212         return;
 213     }
 214     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 215 }
 216
 217 static const BlockJobDriver stream_job_driver = {
 218     .instance_size = sizeof(StreamBlockJob),
 219     .job_type      = BLOCK_JOB_TYPE_STREAM,
 220     .set_speed     = stream_set_speed,
 221 };
 222
 223 void stream_start(const char *job_id, BlockDriverState *bs,
 224                   BlockDriverState *base, const char *backing_file_str,
 225                   int64_t speed, BlockdevOnError on_error, Error **errp)
 226 {
 227     StreamBlockJob *s;
 228     BlockDriverState *iter;
 229     int orig_bs_flags;
 230
 231     s = block_job_create(job_id, &stream_job_driver, bs, speed,
 232                          BLOCK_JOB_DEFAULT, NULL, NULL, errp);
 233     if (!s) {
 234         return;
 235     }
 236
 237     /* Make sure that the image is opened in read-write mode */
 238     orig_bs_flags = bdrv_get_flags(bs);
 239     if (!(orig_bs_flags & BDRV_O_RDWR)) {
 240         if (bdrv_reopen(bs, orig_bs_flags | BDRV_O_RDWR, errp) != 0) {
 241             block_job_unref(&s->common);
 242             return;
 243         }
 244     }
 245
 246     /* Block all intermediate nodes between bs and base, because they
 247      * will disappear from the chain after this operation */
 248     for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
 249         block_job_add_bdrv(&s->common, iter);
 250     }
 251
 252     s->base = base;
 253     s->backing_file_str = g_strdup(backing_file_str);
 254     s->bs_flags = orig_bs_flags;
 255
 256     s->on_error = on_error;
 257     s->common.co = qemu_coroutine_create(stream_run, s);
 258     trace_stream_start(bs, base, s, s->common.co);
 259     qemu_coroutine_enter(s->common.co);
 260 }