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 *bottom;
  35     BlockdevOnError on_error;
  36     char *backing_file_str;
  37     bool bs_read_only;
  38     bool chain_frozen;
  39 } StreamBlockJob;
  40
  41 static int coroutine_fn stream_populate(BlockBackend *blk,
  42                                         int64_t offset, uint64_t bytes,
  43                                         void *buf)
  44 {
  45     assert(bytes < SIZE_MAX);
  46
  47     /* Copy-on-read the unallocated clusters */
  48     return blk_co_pread(blk, offset, bytes, buf, BDRV_REQ_COPY_ON_READ);
  49 }
  50
  51 static void stream_abort(Job *job)
  52 {
  53     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  54
  55     if (s->chain_frozen) {
  56         BlockJob *bjob = &s->common;
  57         bdrv_unfreeze_backing_chain(blk_bs(bjob->blk), s->bottom);
  58     }
  59 }
  60
  61 static int stream_prepare(Job *job)
  62 {
  63     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  64     BlockJob *bjob = &s->common;
  65     BlockDriverState *bs = blk_bs(bjob->blk);
  66     BlockDriverState *base = backing_bs(s->bottom);
  67     Error *local_err = NULL;
  68     int ret = 0;
  69
  70     bdrv_unfreeze_backing_chain(bs, s->bottom);
  71     s->chain_frozen = false;
  72
  73     if (bs->backing) {
  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         bdrv_set_backing_hd(bs, base, &local_err);
  82         ret = bdrv_change_backing_file(bs, base_id, base_fmt);
  83         if (local_err) {
  84             error_report_err(local_err);
  85             return -EPERM;
  86         }
  87     }
  88
  89     return ret;
  90 }
  91
  92 static void stream_clean(Job *job)
  93 {
  94     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  95     BlockJob *bjob = &s->common;
  96     BlockDriverState *bs = blk_bs(bjob->blk);
  97
  98     /* Reopen the image back in read-only mode if necessary */
  99     if (s->bs_read_only) {
 100         /* Give up write permissions before making it read-only */
 101         blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
 102         bdrv_reopen_set_read_only(bs, true, NULL);
 103     }
 104
 105     g_free(s->backing_file_str);
 106 }
 107
 108 static int coroutine_fn stream_run(Job *job, Error **errp)
 109 {
 110     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
 111     BlockBackend *blk = s->common.blk;
 112     BlockDriverState *bs = blk_bs(blk);
 113     bool enable_cor = !backing_bs(s->bottom);
 114     int64_t len;
 115     int64_t offset = 0;
 116     uint64_t delay_ns = 0;
 117     int error = 0;
 118     int ret = 0;
 119     int64_t n = 0; /* bytes */
 120     void *buf;
 121
 122     if (bs == s->bottom) {
 123         /* Nothing to stream */
 124         return 0;
 125     }
 126
 127     len = bdrv_getlength(bs);
 128     if (len < 0) {
 129         return len;
 130     }
 131     job_progress_set_remaining(&s->common.job, len);
 132
 133     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 134
 135     /* Turn on copy-on-read for the whole block device so that guest read
 136      * requests help us make progress.  Only do this when copying the entire
 137      * backing chain since the copy-on-read operation does not take base into
 138      * account.
 139      */
 140     if (enable_cor) {
 141         bdrv_enable_copy_on_read(bs);
 142     }
 143
 144     for ( ; offset < len; offset += n) {
 145         bool copy;
 146
 147         /* Note that even when no rate limit is applied we need to yield
 148          * with no pending I/O here so that bdrv_drain_all() returns.
 149          */
 150         job_sleep_ns(&s->common.job, delay_ns);
 151         if (job_is_cancelled(&s->common.job)) {
 152             break;
 153         }
 154
 155         copy = false;
 156
 157         ret = bdrv_is_allocated(bs, offset, STREAM_BUFFER_SIZE, &n);
 158         if (ret == 1) {
 159             /* Allocated in the top, no need to copy.  */
 160         } else if (ret >= 0) {
 161             /* Copy if allocated in the intermediate images.  Limit to the
 162              * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE).  */
 163             ret = bdrv_is_allocated_above(backing_bs(bs), s->bottom, true,
 164                                           offset, n, &n);
 165             /* Finish early if end of backing file has been reached */
 166             if (ret == 0 && n == 0) {
 167                 n = len - offset;
 168             }
 169
 170             copy = (ret == 1);
 171         }
 172         trace_stream_one_iteration(s, offset, n, ret);
 173         if (copy) {
 174             ret = stream_populate(blk, offset, n, buf);
 175         }
 176         if (ret < 0) {
 177             BlockErrorAction action =
 178                 block_job_error_action(&s->common, s->on_error, true, -ret);
 179             if (action == BLOCK_ERROR_ACTION_STOP) {
 180                 n = 0;
 181                 continue;
 182             }
 183             if (error == 0) {
 184                 error = ret;
 185             }
 186             if (action == BLOCK_ERROR_ACTION_REPORT) {
 187                 break;
 188             }
 189         }
 190         ret = 0;
 191
 192         /* Publish progress */
 193         job_progress_update(&s->common.job, n);
 194         if (copy) {
 195             delay_ns = block_job_ratelimit_get_delay(&s->common, n);
 196         } else {
 197             delay_ns = 0;
 198         }
 199     }
 200
 201     if (enable_cor) {
 202         bdrv_disable_copy_on_read(bs);
 203     }
 204
 205     qemu_vfree(buf);
 206
 207     /* Do not remove the backing file if an error was there but ignored. */
 208     return error;
 209 }
 210
 211 static const BlockJobDriver stream_job_driver = {
 212     .job_driver = {
 213         .instance_size = sizeof(StreamBlockJob),
 214         .job_type      = JOB_TYPE_STREAM,
 215         .free          = block_job_free,
 216         .run           = stream_run,
 217         .prepare       = stream_prepare,
 218         .abort         = stream_abort,
 219         .clean         = stream_clean,
 220         .user_resume   = block_job_user_resume,
 221         .drain         = block_job_drain,
 222     },
 223 };
 224
 225 void stream_start(const char *job_id, BlockDriverState *bs,
 226                   BlockDriverState *base, const char *backing_file_str,
 227                   int creation_flags, int64_t speed,
 228                   BlockdevOnError on_error, Error **errp)
 229 {
 230     StreamBlockJob *s;
 231     BlockDriverState *iter;
 232     bool bs_read_only;
 233     int basic_flags = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
 234     BlockDriverState *bottom = bdrv_find_overlay(bs, base);
 235
 236     if (bdrv_freeze_backing_chain(bs, bottom, errp) < 0) {
 237         return;
 238     }
 239
 240     /* Make sure that the image is opened in read-write mode */
 241     bs_read_only = bdrv_is_read_only(bs);
 242     if (bs_read_only) {
 243         if (bdrv_reopen_set_read_only(bs, false, errp) != 0) {
 244             bs_read_only = false;
 245             goto fail;
 246         }
 247     }
 248
 249     /* Prevent concurrent jobs trying to modify the graph structure here, we
 250      * already have our own plans. Also don't allow resize as the image size is
 251      * queried only at the job start and then cached. */
 252     s = block_job_create(job_id, &stream_job_driver, NULL, bs,
 253                          basic_flags | BLK_PERM_GRAPH_MOD,
 254                          basic_flags | BLK_PERM_WRITE,
 255                          speed, creation_flags, NULL, NULL, errp);
 256     if (!s) {
 257         goto fail;
 258     }
 259
 260     /* Block all intermediate nodes between bs and base, because they will
 261      * disappear from the chain after this operation. The streaming job reads
 262      * every block only once, assuming that it doesn't change, so forbid writes
 263      * and resizes. Reassign the base node pointer because the backing BS of the
 264      * bottom node might change after the call to bdrv_reopen_set_read_only()
 265      * due to parallel block jobs running.
 266      */
 267     base = backing_bs(bottom);
 268     for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
 269         block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
 270                            basic_flags, &error_abort);
 271     }
 272
 273     s->bottom = bottom;
 274     s->backing_file_str = g_strdup(backing_file_str);
 275     s->bs_read_only = bs_read_only;
 276     s->chain_frozen = true;
 277
 278     s->on_error = on_error;
 279     trace_stream_start(bs, base, s);
 280     job_start(&s->common.job);
 281     return;
 282
 283 fail:
 284     if (bs_read_only) {
 285         bdrv_reopen_set_read_only(bs, true, NULL);
 286     }
 287     bdrv_unfreeze_backing_chain(bs, bottom);
 288 }