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     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     QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, buf, bytes);
  46
  47     assert(bytes < SIZE_MAX);
  48
  49     /* Copy-on-read the unallocated clusters */
  50     return blk_co_preadv(blk, offset, qiov.size, &qiov, BDRV_REQ_COPY_ON_READ);
  51 }
  52
  53 static void stream_abort(Job *job)
  54 {
  55     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  56
  57     if (s->chain_frozen) {
  58         BlockJob *bjob = &s->common;
  59         bdrv_unfreeze_backing_chain(blk_bs(bjob->blk), s->base);
  60     }
  61 }
  62
  63 static int stream_prepare(Job *job)
  64 {
  65     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  66     BlockJob *bjob = &s->common;
  67     BlockDriverState *bs = blk_bs(bjob->blk);
  68     BlockDriverState *base = s->base;
  69     Error *local_err = NULL;
  70     int ret = 0;
  71
  72     bdrv_unfreeze_backing_chain(bs, base);
  73     s->chain_frozen = false;
  74
  75     if (bs->backing) {
  76         const char *base_id = NULL, *base_fmt = NULL;
  77         if (base) {
  78             base_id = s->backing_file_str;
  79             if (base->drv) {
  80                 base_fmt = base->drv->format_name;
  81             }
  82         }
  83         ret = bdrv_change_backing_file(bs, base_id, base_fmt);
  84         bdrv_set_backing_hd(bs, base, &local_err);
  85         if (local_err) {
  86             error_report_err(local_err);
  87             return -EPERM;
  88         }
  89     }
  90
  91     return ret;
  92 }
  93
  94 static void stream_clean(Job *job)
  95 {
  96     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  97     BlockJob *bjob = &s->common;
  98     BlockDriverState *bs = blk_bs(bjob->blk);
  99
 100     /* Reopen the image back in read-only mode if necessary */
 101     if (s->bs_read_only) {
 102         /* Give up write permissions before making it read-only */
 103         blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
 104         bdrv_reopen_set_read_only(bs, true, NULL);
 105     }
 106
 107     g_free(s->backing_file_str);
 108 }
 109
 110 static int coroutine_fn stream_run(Job *job, Error **errp)
 111 {
 112     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
 113     BlockBackend *blk = s->common.blk;
 114     BlockDriverState *bs = blk_bs(blk);
 115     BlockDriverState *base = s->base;
 116     int64_t len;
 117     int64_t offset = 0;
 118     uint64_t delay_ns = 0;
 119     int error = 0;
 120     int ret = 0;
 121     int64_t n = 0; /* bytes */
 122     void *buf;
 123
 124     if (!bs->backing) {
 125         goto out;
 126     }
 127
 128     len = bdrv_getlength(bs);
 129     if (len < 0) {
 130         ret = len;
 131         goto out;
 132     }
 133     job_progress_set_remaining(&s->common.job, len);
 134
 135     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 136
 137     /* Turn on copy-on-read for the whole block device so that guest read
 138      * requests help us make progress.  Only do this when copying the entire
 139      * backing chain since the copy-on-read operation does not take base into
 140      * account.
 141      */
 142     if (!base) {
 143         bdrv_enable_copy_on_read(bs);
 144     }
 145
 146     for ( ; offset < len; offset += n) {
 147         bool copy;
 148
 149         /* Note that even when no rate limit is applied we need to yield
 150          * with no pending I/O here so that bdrv_drain_all() returns.
 151          */
 152         job_sleep_ns(&s->common.job, delay_ns);
 153         if (job_is_cancelled(&s->common.job)) {
 154             break;
 155         }
 156
 157         copy = false;
 158
 159         ret = bdrv_is_allocated(bs, offset, STREAM_BUFFER_SIZE, &n);
 160         if (ret == 1) {
 161             /* Allocated in the top, no need to copy.  */
 162         } else if (ret >= 0) {
 163             /* Copy if allocated in the intermediate images.  Limit to the
 164              * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE).  */
 165             ret = bdrv_is_allocated_above(backing_bs(bs), base,
 166                                           offset, n, &n);
 167
 168             /* Finish early if end of backing file has been reached */
 169             if (ret == 0 && n == 0) {
 170                 n = len - offset;
 171             }
 172
 173             copy = (ret == 1);
 174         }
 175         trace_stream_one_iteration(s, offset, n, ret);
 176         if (copy) {
 177             ret = stream_populate(blk, offset, n, buf);
 178         }
 179         if (ret < 0) {
 180             BlockErrorAction action =
 181                 block_job_error_action(&s->common, s->on_error, true, -ret);
 182             if (action == BLOCK_ERROR_ACTION_STOP) {
 183                 n = 0;
 184                 continue;
 185             }
 186             if (error == 0) {
 187                 error = ret;
 188             }
 189             if (action == BLOCK_ERROR_ACTION_REPORT) {
 190                 break;
 191             }
 192         }
 193         ret = 0;
 194
 195         /* Publish progress */
 196         job_progress_update(&s->common.job, n);
 197         if (copy) {
 198             delay_ns = block_job_ratelimit_get_delay(&s->common, n);
 199         } else {
 200             delay_ns = 0;
 201         }
 202     }
 203
 204     if (!base) {
 205         bdrv_disable_copy_on_read(bs);
 206     }
 207
 208     /* Do not remove the backing file if an error was there but ignored.  */
 209     ret = error;
 210
 211     qemu_vfree(buf);
 212
 213 out:
 214     /* Modify backing chain and close BDSes in main loop */
 215     return ret;
 216 }
 217
 218 static const BlockJobDriver stream_job_driver = {
 219     .job_driver = {
 220         .instance_size = sizeof(StreamBlockJob),
 221         .job_type      = JOB_TYPE_STREAM,
 222         .free          = block_job_free,
 223         .run           = stream_run,
 224         .prepare       = stream_prepare,
 225         .abort         = stream_abort,
 226         .clean         = stream_clean,
 227         .user_resume   = block_job_user_resume,
 228         .drain         = block_job_drain,
 229     },
 230 };
 231
 232 void stream_start(const char *job_id, BlockDriverState *bs,
 233                   BlockDriverState *base, const char *backing_file_str,
 234                   int creation_flags, int64_t speed,
 235                   BlockdevOnError on_error, Error **errp)
 236 {
 237     StreamBlockJob *s;
 238     BlockDriverState *iter;
 239     bool bs_read_only;
 240
 241     if (bdrv_freeze_backing_chain(bs, base, errp) < 0) {
 242         return;
 243     }
 244
 245     /* Make sure that the image is opened in read-write mode */
 246     bs_read_only = bdrv_is_read_only(bs);
 247     if (bs_read_only) {
 248         if (bdrv_reopen_set_read_only(bs, false, errp) != 0) {
 249             bs_read_only = false;
 250             goto fail;
 251         }
 252     }
 253
 254     /* Prevent concurrent jobs trying to modify the graph structure here, we
 255      * already have our own plans. Also don't allow resize as the image size is
 256      * queried only at the job start and then cached. */
 257     s = block_job_create(job_id, &stream_job_driver, NULL, bs,
 258                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
 259                          BLK_PERM_GRAPH_MOD,
 260                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
 261                          BLK_PERM_WRITE,
 262                          speed, creation_flags, NULL, NULL, errp);
 263     if (!s) {
 264         goto fail;
 265     }
 266
 267     /* Block all intermediate nodes between bs and base, because they will
 268      * disappear from the chain after this operation. The streaming job reads
 269      * every block only once, assuming that it doesn't change, so block writes
 270      * and resizes. */
 271     for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
 272         block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
 273                            BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED,
 274                            &error_abort);
 275     }
 276
 277     s->base = base;
 278     s->backing_file_str = g_strdup(backing_file_str);
 279     s->bs_read_only = bs_read_only;
 280     s->chain_frozen = true;
 281
 282     s->on_error = on_error;
 283     trace_stream_start(bs, base, s);
 284     job_start(&s->common.job);
 285     return;
 286
 287 fail:
 288     if (bs_read_only) {
 289         bdrv_reopen_set_read_only(bs, true, NULL);
 290     }
 291     bdrv_unfreeze_backing_chain(bs, base);
 292 }