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