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 "trace.h"
  15 #include "block/block_int.h"
  16 #include "block/blockjob.h"
  17 #include "qemu/ratelimit.h"
  18
  19 enum {
  20     /*
  21      * Size of data buffer for populating the image file.  This should be large
  22      * enough to process multiple clusters in a single call, so that populating
  23      * contiguous regions of the image is efficient.
  24      */
  25     STREAM_BUFFER_SIZE = 512 * 1024, /* in bytes */
  26 };
  27
  28 #define SLICE_TIME 100000000ULL /* ns */
  29
  30 typedef struct StreamBlockJob {
  31     BlockJob common;
  32     RateLimit limit;
  33     BlockDriverState *base;
  34     BlockdevOnError on_error;
  35     char backing_file_id[1024];
  36 } StreamBlockJob;
  37
  38 static int coroutine_fn stream_populate(BlockDriverState *bs,
  39                                         int64_t sector_num, int nb_sectors,
  40                                         void *buf)
  41 {
  42     struct iovec iov = {
  43         .iov_base = buf,
  44         .iov_len  = nb_sectors * BDRV_SECTOR_SIZE,
  45     };
  46     QEMUIOVector qiov;
  47
  48     qemu_iovec_init_external(&qiov, &iov, 1);
  49
  50     /* Copy-on-read the unallocated clusters */
  51     return bdrv_co_copy_on_readv(bs, sector_num, nb_sectors, &qiov);
  52 }
  53
  54 static void close_unused_images(BlockDriverState *top, BlockDriverState *base,
  55                                 const char *base_id)
  56 {
  57     BlockDriverState *intermediate;
  58     intermediate = top->backing_hd;
  59
  60     /* Must assign before bdrv_delete() to prevent traversing dangling pointer
  61      * while we delete backing image instances.
  62      */
  63     top->backing_hd = base;
  64
  65     while (intermediate) {
  66         BlockDriverState *unused;
  67
  68         /* reached base */
  69         if (intermediate == base) {
  70             break;
  71         }
  72
  73         unused = intermediate;
  74         intermediate = intermediate->backing_hd;
  75         unused->backing_hd = NULL;
  76         bdrv_unref(unused);
  77     }
  78 }
  79
  80 static void coroutine_fn stream_run(void *opaque)
  81 {
  82     StreamBlockJob *s = opaque;
  83     BlockDriverState *bs = s->common.bs;
  84     BlockDriverState *base = s->base;
  85     int64_t sector_num, end;
  86     int error = 0;
  87     int ret = 0;
  88     int n = 0;
  89     void *buf;
  90
  91     if (!bs->backing_hd) {
  92         block_job_completed(&s->common, 0);
  93         return;
  94     }
  95
  96     s->common.len = bdrv_getlength(bs);
  97     if (s->common.len < 0) {
  98         block_job_completed(&s->common, s->common.len);
  99         return;
 100     }
 101
 102     end = s->common.len >> BDRV_SECTOR_BITS;
 103     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 104
 105     /* Turn on copy-on-read for the whole block device so that guest read
 106      * requests help us make progress.  Only do this when copying the entire
 107      * backing chain since the copy-on-read operation does not take base into
 108      * account.
 109      */
 110     if (!base) {
 111         bdrv_enable_copy_on_read(bs);
 112     }
 113
 114     for (sector_num = 0; sector_num < end; sector_num += n) {
 115         uint64_t delay_ns = 0;
 116         bool copy;
 117
 118 wait:
 119         /* Note that even when no rate limit is applied we need to yield
 120          * with no pending I/O here so that bdrv_drain_all() returns.
 121          */
 122         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 123         if (block_job_is_cancelled(&s->common)) {
 124             break;
 125         }
 126
 127         copy = false;
 128
 129         ret = bdrv_is_allocated(bs, sector_num,
 130                                 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
 131         if (ret == 1) {
 132             /* Allocated in the top, no need to copy.  */
 133         } else if (ret >= 0) {
 134             /* Copy if allocated in the intermediate images.  Limit to the
 135              * known-unallocated area [sector_num, sector_num+n).  */
 136             ret = bdrv_is_allocated_above(bs->backing_hd, base,
 137                                           sector_num, n, &n);
 138
 139             /* Finish early if end of backing file has been reached */
 140             if (ret == 0 && n == 0) {
 141                 n = end - sector_num;
 142             }
 143
 144             copy = (ret == 1);
 145         }
 146         trace_stream_one_iteration(s, sector_num, n, ret);
 147         if (copy) {
 148             if (s->common.speed) {
 149                 delay_ns = ratelimit_calculate_delay(&s->limit, n);
 150                 if (delay_ns > 0) {
 151                     goto wait;
 152                 }
 153             }
 154             ret = stream_populate(bs, sector_num, n, buf);
 155         }
 156         if (ret < 0) {
 157             BlockErrorAction action =
 158                 block_job_error_action(&s->common, s->common.bs, s->on_error,
 159                                        true, -ret);
 160             if (action == BDRV_ACTION_STOP) {
 161                 n = 0;
 162                 continue;
 163             }
 164             if (error == 0) {
 165                 error = ret;
 166             }
 167             if (action == BDRV_ACTION_REPORT) {
 168                 break;
 169             }
 170         }
 171         ret = 0;
 172
 173         /* Publish progress */
 174         s->common.offset += n * BDRV_SECTOR_SIZE;
 175     }
 176
 177     if (!base) {
 178         bdrv_disable_copy_on_read(bs);
 179     }
 180
 181     /* Do not remove the backing file if an error was there but ignored.  */
 182     ret = error;
 183
 184     if (!block_job_is_cancelled(&s->common) && sector_num == end && ret == 0) {
 185         const char *base_id = NULL, *base_fmt = NULL;
 186         if (base) {
 187             base_id = s->backing_file_id;
 188             if (base->drv) {
 189                 base_fmt = base->drv->format_name;
 190             }
 191         }
 192         ret = bdrv_change_backing_file(bs, base_id, base_fmt);
 193         close_unused_images(bs, base, base_id);
 194     }
 195
 196     qemu_vfree(buf);
 197     block_job_completed(&s->common, ret);
 198 }
 199
 200 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
 201 {
 202     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
 203
 204     if (speed < 0) {
 205         error_set(errp, QERR_INVALID_PARAMETER, "speed");
 206         return;
 207     }
 208     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 209 }
 210
 211 static const BlockJobDriver stream_job_driver = {
 212     .instance_size = sizeof(StreamBlockJob),
 213     .job_type      = BLOCK_JOB_TYPE_STREAM,
 214     .set_speed     = stream_set_speed,
 215 };
 216
 217 void stream_start(BlockDriverState *bs, BlockDriverState *base,
 218                   const char *base_id, int64_t speed,
 219                   BlockdevOnError on_error,
 220                   BlockDriverCompletionFunc *cb,
 221                   void *opaque, Error **errp)
 222 {
 223     StreamBlockJob *s;
 224
 225     if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
 226          on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
 227         !bdrv_iostatus_is_enabled(bs)) {
 228         error_set(errp, QERR_INVALID_PARAMETER, "on-error");
 229         return;
 230     }
 231
 232     s = block_job_create(&stream_job_driver, bs, speed, cb, opaque, errp);
 233     if (!s) {
 234         return;
 235     }
 236
 237     s->base = base;
 238     if (base_id) {
 239         pstrcpy(s->backing_file_id, sizeof(s->backing_file_id), base_id);
 240     }
 241
 242     s->on_error = on_error;
 243     s->common.co = qemu_coroutine_create(stream_run);
 244     trace_stream_start(bs, base, s, s->common.co, opaque);
 245     qemu_coroutine_enter(s->common.co, s);
 246 }