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      * Maximum chunk size to feed to copy-on-read.  This should be
  26      * large enough to process multiple clusters in a single call, so
  27      * that populating contiguous regions of the image is efficient.
  28      */
  29     STREAM_CHUNK = 512 * 1024, /* in bytes */
  30 };
  31
  32 typedef struct StreamBlockJob {
  33     BlockJob common;
  34     BlockDriverState *base_overlay; /* COW overlay (stream from this) */
  35     BlockDriverState *above_base;   /* Node directly above the base */
  36     BlockdevOnError on_error;
  37     char *backing_file_str;
  38     bool bs_read_only;
  39     bool chain_frozen;
  40 } StreamBlockJob;
  41
  42 static int coroutine_fn stream_populate(BlockBackend *blk,
  43                                         int64_t offset, uint64_t bytes)
  44 {
  45     assert(bytes < SIZE_MAX);
  46
  47     return blk_co_preadv(blk, offset, bytes, NULL,
  48                          BDRV_REQ_COPY_ON_READ | BDRV_REQ_PREFETCH);
  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->above_base);
  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 *unfiltered_bs = bdrv_skip_filters(bs);
  67     BlockDriverState *base = bdrv_filter_or_cow_bs(s->above_base);
  68     Error *local_err = NULL;
  69     int ret = 0;
  70
  71     bdrv_unfreeze_backing_chain(bs, s->above_base);
  72     s->chain_frozen = false;
  73
  74     if (bdrv_cow_child(unfiltered_bs)) {
  75         const char *base_id = NULL, *base_fmt = NULL;
  76         if (base) {
  77             base_id = s->backing_file_str;
  78             if (base->drv) {
  79                 base_fmt = base->drv->format_name;
  80             }
  81         }
  82         bdrv_set_backing_hd(unfiltered_bs, base, &local_err);
  83         ret = bdrv_change_backing_file(unfiltered_bs, base_id, base_fmt, false);
  84         if (local_err) {
  85             error_report_err(local_err);
  86             return -EPERM;
  87         }
  88     }
  89
  90     return ret;
  91 }
  92
  93 static void stream_clean(Job *job)
  94 {
  95     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  96     BlockJob *bjob = &s->common;
  97     BlockDriverState *bs = blk_bs(bjob->blk);
  98
  99     /* Reopen the image back in read-only mode if necessary */
 100     if (s->bs_read_only) {
 101         /* Give up write permissions before making it read-only */
 102         blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
 103         bdrv_reopen_set_read_only(bs, true, NULL);
 104     }
 105
 106     g_free(s->backing_file_str);
 107 }
 108
 109 static int coroutine_fn stream_run(Job *job, Error **errp)
 110 {
 111     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
 112     BlockBackend *blk = s->common.blk;
 113     BlockDriverState *bs = blk_bs(blk);
 114     BlockDriverState *unfiltered_bs = bdrv_skip_filters(bs);
 115     bool enable_cor = !bdrv_cow_child(s->base_overlay);
 116     int64_t len;
 117     int64_t offset = 0;
 118     uint64_t delay_ns = 0;
 119     int error = 0;
 120     int64_t n = 0; /* bytes */
 121
 122     if (unfiltered_bs == s->base_overlay) {
 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     /* Turn on copy-on-read for the whole block device so that guest read
 134      * requests help us make progress.  Only do this when copying the entire
 135      * backing chain since the copy-on-read operation does not take base into
 136      * account.
 137      */
 138     if (enable_cor) {
 139         bdrv_enable_copy_on_read(bs);
 140     }
 141
 142     for ( ; offset < len; offset += n) {
 143         bool copy;
 144         int ret;
 145
 146         /* Note that even when no rate limit is applied we need to yield
 147          * with no pending I/O here so that bdrv_drain_all() returns.
 148          */
 149         job_sleep_ns(&s->common.job, delay_ns);
 150         if (job_is_cancelled(&s->common.job)) {
 151             break;
 152         }
 153
 154         copy = false;
 155
 156         ret = bdrv_is_allocated(unfiltered_bs, offset, STREAM_CHUNK, &n);
 157         if (ret == 1) {
 158             /* Allocated in the top, no need to copy.  */
 159         } else if (ret >= 0) {
 160             /* Copy if allocated in the intermediate images.  Limit to the
 161              * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE).  */
 162             ret = bdrv_is_allocated_above(bdrv_cow_bs(unfiltered_bs),
 163                                           s->base_overlay, 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 > 0);
 171         }
 172         trace_stream_one_iteration(s, offset, n, ret);
 173         if (copy) {
 174             ret = stream_populate(blk, offset, n);
 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
 191         /* Publish progress */
 192         job_progress_update(&s->common.job, n);
 193         if (copy) {
 194             delay_ns = block_job_ratelimit_get_delay(&s->common, n);
 195         } else {
 196             delay_ns = 0;
 197         }
 198     }
 199
 200     if (enable_cor) {
 201         bdrv_disable_copy_on_read(bs);
 202     }
 203
 204     /* Do not remove the backing file if an error was there but ignored. */
 205     return error;
 206 }
 207
 208 static const BlockJobDriver stream_job_driver = {
 209     .job_driver = {
 210         .instance_size = sizeof(StreamBlockJob),
 211         .job_type      = JOB_TYPE_STREAM,
 212         .free          = block_job_free,
 213         .run           = stream_run,
 214         .prepare       = stream_prepare,
 215         .abort         = stream_abort,
 216         .clean         = stream_clean,
 217         .user_resume   = block_job_user_resume,
 218     },
 219 };
 220
 221 void stream_start(const char *job_id, BlockDriverState *bs,
 222                   BlockDriverState *base, const char *backing_file_str,
 223                   int creation_flags, int64_t speed,
 224                   BlockdevOnError on_error,
 225                   const char *filter_node_name,
 226                   Error **errp)
 227 {
 228     StreamBlockJob *s;
 229     BlockDriverState *iter;
 230     bool bs_read_only;
 231     int basic_flags = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
 232     BlockDriverState *base_overlay = bdrv_find_overlay(bs, base);
 233     BlockDriverState *above_base;
 234
 235     if (!base_overlay) {
 236         error_setg(errp, "'%s' is not in the backing chain of '%s'",
 237                    base->node_name, bs->node_name);
 238         return;
 239     }
 240
 241     /*
 242      * Find the node directly above @base.  @base_overlay is a COW overlay, so
 243      * it must have a bdrv_cow_child(), but it is the immediate overlay of
 244      * @base, so between the two there can only be filters.
 245      */
 246     above_base = base_overlay;
 247     if (bdrv_cow_bs(above_base) != base) {
 248         above_base = bdrv_cow_bs(above_base);
 249         while (bdrv_filter_bs(above_base) != base) {
 250             above_base = bdrv_filter_bs(above_base);
 251         }
 252     }
 253
 254     if (bdrv_freeze_backing_chain(bs, above_base, errp) < 0) {
 255         return;
 256     }
 257
 258     /* Make sure that the image is opened in read-write mode */
 259     bs_read_only = bdrv_is_read_only(bs);
 260     if (bs_read_only) {
 261         if (bdrv_reopen_set_read_only(bs, false, errp) != 0) {
 262             bs_read_only = false;
 263             goto fail;
 264         }
 265     }
 266
 267     /* Prevent concurrent jobs trying to modify the graph structure here, we
 268      * already have our own plans. Also don't allow resize as the image size is
 269      * queried only at the job start and then cached. */
 270     s = block_job_create(job_id, &stream_job_driver, NULL, bs,
 271                          basic_flags | BLK_PERM_GRAPH_MOD,
 272                          basic_flags | BLK_PERM_WRITE,
 273                          speed, creation_flags, NULL, NULL, errp);
 274     if (!s) {
 275         goto fail;
 276     }
 277
 278     /* Block all intermediate nodes between bs and base, because they will
 279      * disappear from the chain after this operation. The streaming job reads
 280      * every block only once, assuming that it doesn't change, so forbid writes
 281      * and resizes. Reassign the base node pointer because the backing BS of the
 282      * bottom node might change after the call to bdrv_reopen_set_read_only()
 283      * due to parallel block jobs running.
 284      * above_base node might change after the call to
 285      * bdrv_reopen_set_read_only() due to parallel block jobs running.
 286      */
 287     base = bdrv_filter_or_cow_bs(above_base);
 288     for (iter = bdrv_filter_or_cow_bs(bs); iter != base;
 289          iter = bdrv_filter_or_cow_bs(iter))
 290     {
 291         block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
 292                            basic_flags, &error_abort);
 293     }
 294
 295     s->base_overlay = base_overlay;
 296     s->above_base = above_base;
 297     s->backing_file_str = g_strdup(backing_file_str);
 298     s->bs_read_only = bs_read_only;
 299     s->chain_frozen = true;
 300
 301     s->on_error = on_error;
 302     trace_stream_start(bs, base, s);
 303     job_start(&s->common.job);
 304     return;
 305
 306 fail:
 307     if (bs_read_only) {
 308         bdrv_reopen_set_read_only(bs, true, NULL);
 309     }
 310     bdrv_unfreeze_backing_chain(bs, above_base);
 311 }