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     int bs_flags;
  38 } StreamBlockJob;
  39
  40 static int coroutine_fn stream_populate(BlockBackend *blk,
  41                                         int64_t offset, uint64_t bytes,
  42                                         void *buf)
  43 {
  44     struct iovec iov = {
  45         .iov_base = buf,
  46         .iov_len  = bytes,
  47     };
  48     QEMUIOVector qiov;
  49
  50     assert(bytes < SIZE_MAX);
  51     qemu_iovec_init_external(&qiov, &iov, 1);
  52
  53     /* Copy-on-read the unallocated clusters */
  54     return blk_co_preadv(blk, offset, qiov.size, &qiov, BDRV_REQ_COPY_ON_READ);
  55 }
  56
  57 typedef struct {
  58     int ret;
  59 } StreamCompleteData;
  60
  61 static void stream_complete(Job *job, void *opaque)
  62 {
  63     StreamBlockJob *s = container_of(job, StreamBlockJob, common.job);
  64     BlockJob *bjob = &s->common;
  65     StreamCompleteData *data = opaque;
  66     BlockDriverState *bs = blk_bs(bjob->blk);
  67     BlockDriverState *base = s->base;
  68     Error *local_err = NULL;
  69
  70     if (!job_is_cancelled(job) && bs->backing && data->ret == 0) {
  71         const char *base_id = NULL, *base_fmt = NULL;
  72         if (base) {
  73             base_id = s->backing_file_str;
  74             if (base->drv) {
  75                 base_fmt = base->drv->format_name;
  76             }
  77         }
  78         data->ret = bdrv_change_backing_file(bs, base_id, base_fmt);
  79         bdrv_set_backing_hd(bs, base, &local_err);
  80         if (local_err) {
  81             error_report_err(local_err);
  82             data->ret = -EPERM;
  83             goto out;
  84         }
  85     }
  86
  87 out:
  88     /* Reopen the image back in read-only mode if necessary */
  89     if (s->bs_flags != bdrv_get_flags(bs)) {
  90         /* Give up write permissions before making it read-only */
  91         blk_set_perm(bjob->blk, 0, BLK_PERM_ALL, &error_abort);
  92         bdrv_reopen(bs, s->bs_flags, NULL);
  93     }
  94
  95     g_free(s->backing_file_str);
  96     job_completed(job, data->ret, NULL);
  97     g_free(data);
  98 }
  99
 100 static void coroutine_fn stream_run(void *opaque)
 101 {
 102     StreamBlockJob *s = opaque;
 103     StreamCompleteData *data;
 104     BlockBackend *blk = s->common.blk;
 105     BlockDriverState *bs = blk_bs(blk);
 106     BlockDriverState *base = s->base;
 107     int64_t len;
 108     int64_t offset = 0;
 109     uint64_t delay_ns = 0;
 110     int error = 0;
 111     int ret = 0;
 112     int64_t n = 0; /* bytes */
 113     void *buf;
 114
 115     if (!bs->backing) {
 116         goto out;
 117     }
 118
 119     len = bdrv_getlength(bs);
 120     if (len < 0) {
 121         ret = len;
 122         goto out;
 123     }
 124     job_progress_set_remaining(&s->common.job, len);
 125
 126     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 127
 128     /* Turn on copy-on-read for the whole block device so that guest read
 129      * requests help us make progress.  Only do this when copying the entire
 130      * backing chain since the copy-on-read operation does not take base into
 131      * account.
 132      */
 133     if (!base) {
 134         bdrv_enable_copy_on_read(bs);
 135     }
 136
 137     for ( ; offset < len; offset += n) {
 138         bool copy;
 139
 140         /* Note that even when no rate limit is applied we need to yield
 141          * with no pending I/O here so that bdrv_drain_all() returns.
 142          */
 143         job_sleep_ns(&s->common.job, delay_ns);
 144         if (job_is_cancelled(&s->common.job)) {
 145             break;
 146         }
 147
 148         copy = false;
 149
 150         ret = bdrv_is_allocated(bs, offset, STREAM_BUFFER_SIZE, &n);
 151         if (ret == 1) {
 152             /* Allocated in the top, no need to copy.  */
 153         } else if (ret >= 0) {
 154             /* Copy if allocated in the intermediate images.  Limit to the
 155              * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE).  */
 156             ret = bdrv_is_allocated_above(backing_bs(bs), base,
 157                                           offset, n, &n);
 158
 159             /* Finish early if end of backing file has been reached */
 160             if (ret == 0 && n == 0) {
 161                 n = len - offset;
 162             }
 163
 164             copy = (ret == 1);
 165         }
 166         trace_stream_one_iteration(s, offset, n, ret);
 167         if (copy) {
 168             ret = stream_populate(blk, offset, n, buf);
 169         }
 170         if (ret < 0) {
 171             BlockErrorAction action =
 172                 block_job_error_action(&s->common, s->on_error, true, -ret);
 173             if (action == BLOCK_ERROR_ACTION_STOP) {
 174                 n = 0;
 175                 continue;
 176             }
 177             if (error == 0) {
 178                 error = ret;
 179             }
 180             if (action == BLOCK_ERROR_ACTION_REPORT) {
 181                 break;
 182             }
 183         }
 184         ret = 0;
 185
 186         /* Publish progress */
 187         job_progress_update(&s->common.job, n);
 188         if (copy) {
 189             delay_ns = block_job_ratelimit_get_delay(&s->common, n);
 190         } else {
 191             delay_ns = 0;
 192         }
 193     }
 194
 195     if (!base) {
 196         bdrv_disable_copy_on_read(bs);
 197     }
 198
 199     /* Do not remove the backing file if an error was there but ignored.  */
 200     ret = error;
 201
 202     qemu_vfree(buf);
 203
 204 out:
 205     /* Modify backing chain and close BDSes in main loop */
 206     data = g_malloc(sizeof(*data));
 207     data->ret = ret;
 208     job_defer_to_main_loop(&s->common.job, stream_complete, data);
 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         .start         = stream_run,
 217         .user_resume   = block_job_user_resume,
 218         .drain         = block_job_drain,
 219     },
 220 };
 221
 222 void stream_start(const char *job_id, BlockDriverState *bs,
 223                   BlockDriverState *base, const char *backing_file_str,
 224                   int64_t speed, BlockdevOnError on_error, Error **errp)
 225 {
 226     StreamBlockJob *s;
 227     BlockDriverState *iter;
 228     int orig_bs_flags;
 229
 230     /* Make sure that the image is opened in read-write mode */
 231     orig_bs_flags = bdrv_get_flags(bs);
 232     if (!(orig_bs_flags & BDRV_O_RDWR)) {
 233         if (bdrv_reopen(bs, orig_bs_flags | BDRV_O_RDWR, errp) != 0) {
 234             return;
 235         }
 236     }
 237
 238     /* Prevent concurrent jobs trying to modify the graph structure here, we
 239      * already have our own plans. Also don't allow resize as the image size is
 240      * queried only at the job start and then cached. */
 241     s = block_job_create(job_id, &stream_job_driver, NULL, bs,
 242                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
 243                          BLK_PERM_GRAPH_MOD,
 244                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
 245                          BLK_PERM_WRITE,
 246                          speed, JOB_DEFAULT, NULL, NULL, errp);
 247     if (!s) {
 248         goto fail;
 249     }
 250
 251     /* Block all intermediate nodes between bs and base, because they will
 252      * disappear from the chain after this operation. The streaming job reads
 253      * every block only once, assuming that it doesn't change, so block writes
 254      * and resizes. */
 255     for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
 256         block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
 257                            BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED,
 258                            &error_abort);
 259     }
 260
 261     s->base = base;
 262     s->backing_file_str = g_strdup(backing_file_str);
 263     s->bs_flags = orig_bs_flags;
 264
 265     s->on_error = on_error;
 266     trace_stream_start(bs, base, s);
 267     job_start(&s->common.job);
 268     return;
 269
 270 fail:
 271     if (orig_bs_flags != bdrv_get_flags(bs)) {
 272         bdrv_reopen(bs, orig_bs_flags, NULL);
 273     }
 274 }