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 typedef struct StreamBlockJob {
  33     BlockJob common;
  34     BlockDriverState *base;
  35     BlockdevOnError on_error;
  36     char *backing_file_str;
  37     bool bs_read_only;
  38 } StreamBlockJob;
  39
  40 static int coroutine_fn stream_populate(BlockBackend *blk,
  41                                         int64_t offset, uint64_t bytes,
  42                                         void *buf)
  43 {
  44     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
  45
  46     assert(bytes < SIZE_MAX);
  47
  48     /* Copy-on-read the unallocated clusters */
  49     return blk_co_preadv(blk, offset, qiov.size, &qiov, BDRV_REQ_COPY_ON_READ);
  50 }
  51
  52 static int stream_prepare(Job *job)
  53 {
  54     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  55     BlockJob *bjob = &s->common;
  56     BlockDriverState *bs = blk_bs(bjob->blk);
  57     BlockDriverState *base = s->base;
  58     Error *local_err = NULL;
  59     int ret = 0;
  60
  61     if (bs->backing) {
  62         const char *base_id = NULL, *base_fmt = NULL;
  63         if (base) {
  64             base_id = s->backing_file_str;
  65             if (base->drv) {
  66                 base_fmt = base->drv->format_name;
  67             }
  68         }
  69         ret = bdrv_change_backing_file(bs, base_id, base_fmt);
  70         bdrv_set_backing_hd(bs, base, &local_err);
  71         if (local_err) {
  72             error_report_err(local_err);
  73             return -EPERM;
  74         }
  75     }
  76
  77     return ret;
  78 }
  79
  80 static void stream_clean(Job *job)
  81 {
  82     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  83     BlockJob *bjob = &s->common;
  84     BlockDriverState *bs = blk_bs(bjob->blk);
  85
  86     /* Reopen the image back in read-only mode if necessary */
  87     if (s->bs_read_only) {
  88         /* Give up write permissions before making it read-only */
  89         blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
  90         bdrv_reopen_set_read_only(bs, true, NULL);
  91     }
  92
  93     g_free(s->backing_file_str);
  94 }
  95
  96 static int coroutine_fn stream_run(Job *job, Error **errp)
  97 {
  98     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  99     BlockBackend *blk = s->common.blk;
 100     BlockDriverState *bs = blk_bs(blk);
 101     BlockDriverState *base = s->base;
 102     int64_t len;
 103     int64_t offset = 0;
 104     uint64_t delay_ns = 0;
 105     int error = 0;
 106     int ret = 0;
 107     int64_t n = 0; /* bytes */
 108     void *buf;
 109
 110     if (!bs->backing) {
 111         goto out;
 112     }
 113
 114     len = bdrv_getlength(bs);
 115     if (len < 0) {
 116         ret = len;
 117         goto out;
 118     }
 119     job_progress_set_remaining(&s->common.job, len);
 120
 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 ( ; offset < len; offset += 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         job_sleep_ns(&s->common.job, delay_ns);
 139         if (job_is_cancelled(&s->common.job)) {
 140             break;
 141         }
 142
 143         copy = false;
 144
 145         ret = bdrv_is_allocated(bs, offset, STREAM_BUFFER_SIZE, &n);
 146         if (ret == 1) {
 147             /* Allocated in the top, no need to copy.  */
 148         } else if (ret >= 0) {
 149             /* Copy if allocated in the intermediate images.  Limit to the
 150              * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE).  */
 151             ret = bdrv_is_allocated_above(backing_bs(bs), base,
 152                                           offset, n, &n);
 153
 154             /* Finish early if end of backing file has been reached */
 155             if (ret == 0 && n == 0) {
 156                 n = len - offset;
 157             }
 158
 159             copy = (ret == 1);
 160         }
 161         trace_stream_one_iteration(s, offset, n, ret);
 162         if (copy) {
 163             ret = stream_populate(blk, offset, n, buf);
 164         }
 165         if (ret < 0) {
 166             BlockErrorAction action =
 167                 block_job_error_action(&s->common, s->on_error, 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         job_progress_update(&s->common.job, n);
 183         if (copy) {
 184             delay_ns = block_job_ratelimit_get_delay(&s->common, n);
 185         } else {
 186             delay_ns = 0;
 187         }
 188     }
 189
 190     if (!base) {
 191         bdrv_disable_copy_on_read(bs);
 192     }
 193
 194     /* Do not remove the backing file if an error was there but ignored.  */
 195     ret = error;
 196
 197     qemu_vfree(buf);
 198
 199 out:
 200     /* Modify backing chain and close BDSes in main loop */
 201     return ret;
 202 }
 203
 204 static const BlockJobDriver stream_job_driver = {
 205     .job_driver = {
 206         .instance_size = sizeof(StreamBlockJob),
 207         .job_type      = JOB_TYPE_STREAM,
 208         .free          = block_job_free,
 209         .run           = stream_run,
 210         .prepare       = stream_prepare,
 211         .clean         = stream_clean,
 212         .user_resume   = block_job_user_resume,
 213         .drain         = block_job_drain,
 214     },
 215 };
 216
 217 void stream_start(const char *job_id, BlockDriverState *bs,
 218                   BlockDriverState *base, const char *backing_file_str,
 219                   int creation_flags, int64_t speed,
 220                   BlockdevOnError on_error, Error **errp)
 221 {
 222     StreamBlockJob *s;
 223     BlockDriverState *iter;
 224     bool bs_read_only;
 225
 226     /* Make sure that the image is opened in read-write mode */
 227     bs_read_only = bdrv_is_read_only(bs);
 228     if (bs_read_only) {
 229         if (bdrv_reopen_set_read_only(bs, false, errp) != 0) {
 230             return;
 231         }
 232     }
 233
 234     /* Prevent concurrent jobs trying to modify the graph structure here, we
 235      * already have our own plans. Also don't allow resize as the image size is
 236      * queried only at the job start and then cached. */
 237     s = block_job_create(job_id, &stream_job_driver, NULL, bs,
 238                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
 239                          BLK_PERM_GRAPH_MOD,
 240                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
 241                          BLK_PERM_WRITE,
 242                          speed, creation_flags, NULL, NULL, errp);
 243     if (!s) {
 244         goto fail;
 245     }
 246
 247     /* Block all intermediate nodes between bs and base, because they will
 248      * disappear from the chain after this operation. The streaming job reads
 249      * every block only once, assuming that it doesn't change, so block writes
 250      * and resizes. */
 251     for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
 252         block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
 253                            BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED,
 254                            &error_abort);
 255     }
 256
 257     s->base = base;
 258     s->backing_file_str = g_strdup(backing_file_str);
 259     s->bs_read_only = bs_read_only;
 260
 261     s->on_error = on_error;
 262     trace_stream_start(bs, base, s);
 263     job_start(&s->common.job);
 264     return;
 265
 266 fail:
 267     if (bs_read_only) {
 268         bdrv_reopen_set_read_only(bs, true, NULL);
 269     }
 270 }