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 #define SLICE_TIME 100000000ULL /* ns */
  33
  34 typedef struct StreamBlockJob {
  35     BlockJob common;
  36     RateLimit limit;
  37     BlockDriverState *base;
  38     BlockdevOnError on_error;
  39     char *backing_file_str;
  40     int bs_flags;
  41 } StreamBlockJob;
  42
  43 static int coroutine_fn stream_populate(BlockBackend *blk,
  44                                         int64_t offset, uint64_t bytes,
  45                                         void *buf)
  46 {
  47     struct iovec iov = {
  48         .iov_base = buf,
  49         .iov_len  = bytes,
  50     };
  51     QEMUIOVector qiov;
  52
  53     assert(bytes < SIZE_MAX);
  54     qemu_iovec_init_external(&qiov, &iov, 1);
  55
  56     /* Copy-on-read the unallocated clusters */
  57     return blk_co_preadv(blk, offset, qiov.size, &qiov, BDRV_REQ_COPY_ON_READ);
  58 }
  59
  60 typedef struct {
  61     int ret;
  62 } StreamCompleteData;
  63
  64 static void stream_complete(BlockJob *job, void *opaque)
  65 {
  66     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
  67     StreamCompleteData *data = opaque;
  68     BlockDriverState *bs = blk_bs(job->blk);
  69     BlockDriverState *base = s->base;
  70     Error *local_err = NULL;
  71
  72     if (!block_job_is_cancelled(&s->common) && bs->backing &&
  73         data->ret == 0) {
  74         const char *base_id = NULL, *base_fmt = NULL;
  75         if (base) {
  76             base_id = s->backing_file_str;
  77             if (base->drv) {
  78                 base_fmt = base->drv->format_name;
  79             }
  80         }
  81         data->ret = bdrv_change_backing_file(bs, base_id, base_fmt);
  82         bdrv_set_backing_hd(bs, base, &local_err);
  83         if (local_err) {
  84             error_report_err(local_err);
  85             data->ret = -EPERM;
  86             goto out;
  87         }
  88     }
  89
  90 out:
  91     /* Reopen the image back in read-only mode if necessary */
  92     if (s->bs_flags != bdrv_get_flags(bs)) {
  93         /* Give up write permissions before making it read-only */
  94         blk_set_perm(job->blk, 0, BLK_PERM_ALL, &error_abort);
  95         bdrv_reopen(bs, s->bs_flags, NULL);
  96     }
  97
  98     g_free(s->backing_file_str);
  99     block_job_completed(&s->common, data->ret);
 100     g_free(data);
 101 }
 102
 103 static void coroutine_fn stream_run(void *opaque)
 104 {
 105     StreamBlockJob *s = opaque;
 106     StreamCompleteData *data;
 107     BlockBackend *blk = s->common.blk;
 108     BlockDriverState *bs = blk_bs(blk);
 109     BlockDriverState *base = s->base;
 110     int64_t sector_num = 0;
 111     int64_t end = -1;
 112     uint64_t delay_ns = 0;
 113     int error = 0;
 114     int ret = 0;
 115     int n = 0;
 116     void *buf;
 117
 118     if (!bs->backing) {
 119         goto out;
 120     }
 121
 122     s->common.len = bdrv_getlength(bs);
 123     if (s->common.len < 0) {
 124         ret = s->common.len;
 125         goto out;
 126     }
 127
 128     end = s->common.len >> BDRV_SECTOR_BITS;
 129     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 130
 131     /* Turn on copy-on-read for the whole block device so that guest read
 132      * requests help us make progress.  Only do this when copying the entire
 133      * backing chain since the copy-on-read operation does not take base into
 134      * account.
 135      */
 136     if (!base) {
 137         bdrv_enable_copy_on_read(bs);
 138     }
 139
 140     for (sector_num = 0; sector_num < end; sector_num += n) {
 141         bool copy;
 142
 143         /* Note that even when no rate limit is applied we need to yield
 144          * with no pending I/O here so that bdrv_drain_all() returns.
 145          */
 146         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 147         if (block_job_is_cancelled(&s->common)) {
 148             break;
 149         }
 150
 151         copy = false;
 152
 153         ret = bdrv_is_allocated(bs, sector_num,
 154                                 STREAM_BUFFER_SIZE / BDRV_SECTOR_SIZE, &n);
 155         if (ret == 1) {
 156             /* Allocated in the top, no need to copy.  */
 157         } else if (ret >= 0) {
 158             /* Copy if allocated in the intermediate images.  Limit to the
 159              * known-unallocated area [sector_num, sector_num+n).  */
 160             ret = bdrv_is_allocated_above(backing_bs(bs), base,
 161                                           sector_num, n, &n);
 162
 163             /* Finish early if end of backing file has been reached */
 164             if (ret == 0 && n == 0) {
 165                 n = end - sector_num;
 166             }
 167
 168             copy = (ret == 1);
 169         }
 170         trace_stream_one_iteration(s, sector_num * BDRV_SECTOR_SIZE,
 171                                    n * BDRV_SECTOR_SIZE, ret);
 172         if (copy) {
 173             ret = stream_populate(blk, sector_num * BDRV_SECTOR_SIZE,
 174                                   n * BDRV_SECTOR_SIZE, buf);
 175         }
 176         if (ret < 0) {
 177             BlockErrorAction action =
 178                 block_job_error_action(&s->common, s->on_error, true, -ret);
 179             if (action == BLOCK_ERROR_ACTION_STOP) {
 180                 n = 0;
 181                 continue;
 182             }
 183             if (error == 0) {
 184                 error = ret;
 185             }
 186             if (action == BLOCK_ERROR_ACTION_REPORT) {
 187                 break;
 188             }
 189         }
 190         ret = 0;
 191
 192         /* Publish progress */
 193         s->common.offset += n * BDRV_SECTOR_SIZE;
 194         if (copy && s->common.speed) {
 195             delay_ns = ratelimit_calculate_delay(&s->limit,
 196                                                  n * BDRV_SECTOR_SIZE);
 197         }
 198     }
 199
 200     if (!base) {
 201         bdrv_disable_copy_on_read(bs);
 202     }
 203
 204     /* Do not remove the backing file if an error was there but ignored.  */
 205     ret = error;
 206
 207     qemu_vfree(buf);
 208
 209 out:
 210     /* Modify backing chain and close BDSes in main loop */
 211     data = g_malloc(sizeof(*data));
 212     data->ret = ret;
 213     block_job_defer_to_main_loop(&s->common, stream_complete, data);
 214 }
 215
 216 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
 217 {
 218     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
 219
 220     if (speed < 0) {
 221         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 222         return;
 223     }
 224     ratelimit_set_speed(&s->limit, speed, SLICE_TIME);
 225 }
 226
 227 static const BlockJobDriver stream_job_driver = {
 228     .instance_size = sizeof(StreamBlockJob),
 229     .job_type      = BLOCK_JOB_TYPE_STREAM,
 230     .set_speed     = stream_set_speed,
 231     .start         = stream_run,
 232 };
 233
 234 void stream_start(const char *job_id, BlockDriverState *bs,
 235                   BlockDriverState *base, const char *backing_file_str,
 236                   int64_t speed, BlockdevOnError on_error, Error **errp)
 237 {
 238     StreamBlockJob *s;
 239     BlockDriverState *iter;
 240     int orig_bs_flags;
 241
 242     /* Make sure that the image is opened in read-write mode */
 243     orig_bs_flags = bdrv_get_flags(bs);
 244     if (!(orig_bs_flags & BDRV_O_RDWR)) {
 245         if (bdrv_reopen(bs, orig_bs_flags | BDRV_O_RDWR, errp) != 0) {
 246             return;
 247         }
 248     }
 249
 250     /* Prevent concurrent jobs trying to modify the graph structure here, we
 251      * already have our own plans. Also don't allow resize as the image size is
 252      * queried only at the job start and then cached. */
 253     s = block_job_create(job_id, &stream_job_driver, bs,
 254                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
 255                          BLK_PERM_GRAPH_MOD,
 256                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
 257                          BLK_PERM_WRITE,
 258                          speed, BLOCK_JOB_DEFAULT, NULL, NULL, errp);
 259     if (!s) {
 260         goto fail;
 261     }
 262
 263     /* Block all intermediate nodes between bs and base, because they will
 264      * disappear from the chain after this operation. The streaming job reads
 265      * every block only once, assuming that it doesn't change, so block writes
 266      * and resizes. */
 267     for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
 268         block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
 269                            BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED,
 270                            &error_abort);
 271     }
 272
 273     s->base = base;
 274     s->backing_file_str = g_strdup(backing_file_str);
 275     s->bs_flags = orig_bs_flags;
 276
 277     s->on_error = on_error;
 278     trace_stream_start(bs, base, s);
 279     block_job_start(&s->common);
 280     return;
 281
 282 fail:
 283     if (orig_bs_flags != bdrv_get_flags(bs)) {
 284         bdrv_reopen(bs, orig_bs_flags, NULL);
 285     }
 286 }