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/qdict.h"
  20 #include "qemu/ratelimit.h"
  21 #include "sysemu/block-backend.h"
  22 #include "block/copy-on-read.h"
  23
  24 enum {
  25     /*
  26      * Maximum chunk size to feed to copy-on-read.  This should be
  27      * large enough to process multiple clusters in a single call, so
  28      * that populating contiguous regions of the image is efficient.
  29      */
  30     STREAM_CHUNK = 512 * 1024, /* in bytes */
  31 };
  32
  33 typedef struct StreamBlockJob {
  34     BlockJob common;
  35     BlockBackend *blk;
  36     BlockDriverState *base_overlay; /* COW overlay (stream from this) */
  37     BlockDriverState *above_base;   /* Node directly above the base */
  38     BlockDriverState *cor_filter_bs;
  39     BlockDriverState *target_bs;
  40     BlockdevOnError on_error;
  41     char *backing_file_str;
  42     bool bs_read_only;
  43 } StreamBlockJob;
  44
  45 static int coroutine_fn stream_populate(BlockBackend *blk,
  46                                         int64_t offset, uint64_t bytes)
  47 {
  48     assert(bytes < SIZE_MAX);
  49
  50     return blk_co_preadv(blk, offset, bytes, NULL, BDRV_REQ_PREFETCH);
  51 }
  52
  53 static int stream_prepare(Job *job)
  54 {
  55     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  56     BlockDriverState *unfiltered_bs = bdrv_skip_filters(s->target_bs);
  57     BlockDriverState *base;
  58     BlockDriverState *unfiltered_base;
  59     Error *local_err = NULL;
  60     int ret = 0;
  61
  62     /* We should drop filter at this point, as filter hold the backing chain */
  63     bdrv_cor_filter_drop(s->cor_filter_bs);
  64     s->cor_filter_bs = NULL;
  65
  66     /*
  67      * bdrv_set_backing_hd() requires that unfiltered_bs is drained. Drain
  68      * already here and use bdrv_set_backing_hd_drained() instead because
  69      * the polling during drained_begin() might change the graph, and if we do
  70      * this only later, we may end up working with the wrong base node (or it
  71      * might even have gone away by the time we want to use it).
  72      */
  73     bdrv_drained_begin(unfiltered_bs);
  74
  75     base = bdrv_filter_or_cow_bs(s->above_base);
  76     unfiltered_base = bdrv_skip_filters(base);
  77
  78     if (bdrv_cow_child(unfiltered_bs)) {
  79         const char *base_id = NULL, *base_fmt = NULL;
  80         if (unfiltered_base) {
  81             base_id = s->backing_file_str ?: unfiltered_base->filename;
  82             if (unfiltered_base->drv) {
  83                 base_fmt = unfiltered_base->drv->format_name;
  84             }
  85         }
  86
  87         bdrv_set_backing_hd_drained(unfiltered_bs, base, &local_err);
  88
  89         /*
  90          * This call will do I/O, so the graph can change again from here on.
  91          * We have already completed the graph change, so we are not in danger
  92          * of operating on the wrong node any more if this happens.
  93          */
  94         ret = bdrv_change_backing_file(unfiltered_bs, base_id, base_fmt, false);
  95         if (local_err) {
  96             error_report_err(local_err);
  97             ret = -EPERM;
  98             goto out;
  99         }
 100     }
 101
 102 out:
 103     bdrv_drained_end(unfiltered_bs);
 104     return ret;
 105 }
 106
 107 static void stream_clean(Job *job)
 108 {
 109     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
 110
 111     if (s->cor_filter_bs) {
 112         bdrv_cor_filter_drop(s->cor_filter_bs);
 113         s->cor_filter_bs = NULL;
 114     }
 115
 116     blk_unref(s->blk);
 117     s->blk = NULL;
 118
 119     /* Reopen the image back in read-only mode if necessary */
 120     if (s->bs_read_only) {
 121         /* Give up write permissions before making it read-only */
 122         bdrv_reopen_set_read_only(s->target_bs, true, NULL);
 123     }
 124
 125     g_free(s->backing_file_str);
 126 }
 127
 128 static int coroutine_fn stream_run(Job *job, Error **errp)
 129 {
 130     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
 131     BlockDriverState *unfiltered_bs = bdrv_skip_filters(s->target_bs);
 132     int64_t len;
 133     int64_t offset = 0;
 134     uint64_t delay_ns = 0;
 135     int error = 0;
 136     int64_t n = 0; /* bytes */
 137
 138     if (unfiltered_bs == s->base_overlay) {
 139         /* Nothing to stream */
 140         return 0;
 141     }
 142
 143     WITH_GRAPH_RDLOCK_GUARD() {
 144         len = bdrv_co_getlength(s->target_bs);
 145         if (len < 0) {
 146             return len;
 147         }
 148     }
 149     job_progress_set_remaining(&s->common.job, len);
 150
 151     for ( ; offset < len; offset += n) {
 152         bool copy;
 153         int ret;
 154
 155         /* Note that even when no rate limit is applied we need to yield
 156          * with no pending I/O here so that bdrv_drain_all() returns.
 157          */
 158         job_sleep_ns(&s->common.job, delay_ns);
 159         if (job_is_cancelled(&s->common.job)) {
 160             break;
 161         }
 162
 163         copy = false;
 164
 165         WITH_GRAPH_RDLOCK_GUARD() {
 166             ret = bdrv_is_allocated(unfiltered_bs, offset, STREAM_CHUNK, &n);
 167             if (ret == 1) {
 168                 /* Allocated in the top, no need to copy.  */
 169             } else if (ret >= 0) {
 170                 /*
 171                  * Copy if allocated in the intermediate images.  Limit to the
 172                  * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE).
 173                  */
 174                 ret = bdrv_is_allocated_above(bdrv_cow_bs(unfiltered_bs),
 175                                             s->base_overlay, true,
 176                                             offset, n, &n);
 177                 /* Finish early if end of backing file has been reached */
 178                 if (ret == 0 && n == 0) {
 179                     n = len - offset;
 180                 }
 181
 182                 copy = (ret > 0);
 183             }
 184         }
 185         trace_stream_one_iteration(s, offset, n, ret);
 186         if (copy) {
 187             ret = stream_populate(s->blk, offset, n);
 188         }
 189         if (ret < 0) {
 190             BlockErrorAction action =
 191                 block_job_error_action(&s->common, s->on_error, true, -ret);
 192             if (action == BLOCK_ERROR_ACTION_STOP) {
 193                 n = 0;
 194                 continue;
 195             }
 196             if (error == 0) {
 197                 error = ret;
 198             }
 199             if (action == BLOCK_ERROR_ACTION_REPORT) {
 200                 break;
 201             }
 202         }
 203
 204         /* Publish progress */
 205         job_progress_update(&s->common.job, n);
 206         if (copy) {
 207             delay_ns = block_job_ratelimit_get_delay(&s->common, n);
 208         } else {
 209             delay_ns = 0;
 210         }
 211     }
 212
 213     /* Do not remove the backing file if an error was there but ignored. */
 214     return error;
 215 }
 216
 217 static const BlockJobDriver stream_job_driver = {
 218     .job_driver = {
 219         .instance_size = sizeof(StreamBlockJob),
 220         .job_type      = JOB_TYPE_STREAM,
 221         .free          = block_job_free,
 222         .run           = stream_run,
 223         .prepare       = stream_prepare,
 224         .clean         = stream_clean,
 225         .user_resume   = block_job_user_resume,
 226     },
 227 };
 228
 229 void stream_start(const char *job_id, BlockDriverState *bs,
 230                   BlockDriverState *base, const char *backing_file_str,
 231                   BlockDriverState *bottom,
 232                   int creation_flags, int64_t speed,
 233                   BlockdevOnError on_error,
 234                   const char *filter_node_name,
 235                   Error **errp)
 236 {
 237     StreamBlockJob *s = NULL;
 238     BlockDriverState *iter;
 239     bool bs_read_only;
 240     int basic_flags = BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED;
 241     BlockDriverState *base_overlay;
 242     BlockDriverState *cor_filter_bs = NULL;
 243     BlockDriverState *above_base;
 244     QDict *opts;
 245     int ret;
 246
 247     GLOBAL_STATE_CODE();
 248
 249     assert(!(base && bottom));
 250     assert(!(backing_file_str && bottom));
 251
 252     if (bottom) {
 253         /*
 254          * New simple interface. The code is written in terms of old interface
 255          * with @base parameter (still, it doesn't freeze link to base, so in
 256          * this mean old code is correct for new interface). So, for now, just
 257          * emulate base_overlay and above_base. Still, when old interface
 258          * finally removed, we should refactor code to use only "bottom", but
 259          * not "*base*" things.
 260          */
 261         assert(!bottom->drv->is_filter);
 262         base_overlay = above_base = bottom;
 263     } else {
 264         base_overlay = bdrv_find_overlay(bs, base);
 265         if (!base_overlay) {
 266             error_setg(errp, "'%s' is not in the backing chain of '%s'",
 267                        base->node_name, bs->node_name);
 268             return;
 269         }
 270
 271         /*
 272          * Find the node directly above @base.  @base_overlay is a COW overlay,
 273          * so it must have a bdrv_cow_child(), but it is the immediate overlay
 274          * of @base, so between the two there can only be filters.
 275          */
 276         above_base = base_overlay;
 277         if (bdrv_cow_bs(above_base) != base) {
 278             above_base = bdrv_cow_bs(above_base);
 279             while (bdrv_filter_bs(above_base) != base) {
 280                 above_base = bdrv_filter_bs(above_base);
 281             }
 282         }
 283     }
 284
 285     /* Make sure that the image is opened in read-write mode */
 286     bs_read_only = bdrv_is_read_only(bs);
 287     if (bs_read_only) {
 288         int ret;
 289         /* Hold the chain during reopen */
 290         if (bdrv_freeze_backing_chain(bs, above_base, errp) < 0) {
 291             return;
 292         }
 293
 294         ret = bdrv_reopen_set_read_only(bs, false, errp);
 295
 296         /* failure, or cor-filter will hold the chain */
 297         bdrv_unfreeze_backing_chain(bs, above_base);
 298
 299         if (ret < 0) {
 300             return;
 301         }
 302     }
 303
 304     opts = qdict_new();
 305
 306     qdict_put_str(opts, "driver", "copy-on-read");
 307     qdict_put_str(opts, "file", bdrv_get_node_name(bs));
 308     /* Pass the base_overlay node name as 'bottom' to COR driver */
 309     qdict_put_str(opts, "bottom", base_overlay->node_name);
 310     if (filter_node_name) {
 311         qdict_put_str(opts, "node-name", filter_node_name);
 312     }
 313
 314     cor_filter_bs = bdrv_insert_node(bs, opts, BDRV_O_RDWR, errp);
 315     if (!cor_filter_bs) {
 316         goto fail;
 317     }
 318
 319     if (!filter_node_name) {
 320         cor_filter_bs->implicit = true;
 321     }
 322
 323     s = block_job_create(job_id, &stream_job_driver, NULL, cor_filter_bs,
 324                          0, BLK_PERM_ALL,
 325                          speed, creation_flags, NULL, NULL, errp);
 326     if (!s) {
 327         goto fail;
 328     }
 329
 330     s->blk = blk_new_with_bs(cor_filter_bs, BLK_PERM_CONSISTENT_READ,
 331                              basic_flags | BLK_PERM_WRITE, errp);
 332     if (!s->blk) {
 333         goto fail;
 334     }
 335     /*
 336      * Disable request queuing in the BlockBackend to avoid deadlocks on drain:
 337      * The job reports that it's busy until it reaches a pause point.
 338      */
 339     blk_set_disable_request_queuing(s->blk, true);
 340     blk_set_allow_aio_context_change(s->blk, true);
 341
 342     /*
 343      * Prevent concurrent jobs trying to modify the graph structure here, we
 344      * already have our own plans. Also don't allow resize as the image size is
 345      * queried only at the job start and then cached.
 346      */
 347     if (block_job_add_bdrv(&s->common, "active node", bs, 0,
 348                            basic_flags | BLK_PERM_WRITE, errp)) {
 349         goto fail;
 350     }
 351
 352     /* Block all intermediate nodes between bs and base, because they will
 353      * disappear from the chain after this operation. The streaming job reads
 354      * every block only once, assuming that it doesn't change, so forbid writes
 355      * and resizes. Reassign the base node pointer because the backing BS of the
 356      * bottom node might change after the call to bdrv_reopen_set_read_only()
 357      * due to parallel block jobs running.
 358      * above_base node might change after the call to
 359      * bdrv_reopen_set_read_only() due to parallel block jobs running.
 360      */
 361     base = bdrv_filter_or_cow_bs(above_base);
 362     for (iter = bdrv_filter_or_cow_bs(bs); iter != base;
 363          iter = bdrv_filter_or_cow_bs(iter))
 364     {
 365         ret = block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
 366                                  basic_flags, errp);
 367         if (ret < 0) {
 368             goto fail;
 369         }
 370     }
 371
 372     s->base_overlay = base_overlay;
 373     s->above_base = above_base;
 374     s->backing_file_str = g_strdup(backing_file_str);
 375     s->cor_filter_bs = cor_filter_bs;
 376     s->target_bs = bs;
 377     s->bs_read_only = bs_read_only;
 378
 379     s->on_error = on_error;
 380     trace_stream_start(bs, base, s);
 381     job_start(&s->common.job);
 382     return;
 383
 384 fail:
 385     if (s) {
 386         job_early_fail(&s->common.job);
 387     }
 388     if (cor_filter_bs) {
 389         bdrv_cor_filter_drop(cor_filter_bs);
 390     }
 391     if (bs_read_only) {
 392         bdrv_reopen_set_read_only(bs, true, NULL);
 393     }
 394 }