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 offset = 0;
 111     uint64_t delay_ns = 0;
 112     int error = 0;
 113     int ret = 0;
 114     int n = 0; /* sectors */
 115     void *buf;
 116
 117     if (!bs->backing) {
 118         goto out;
 119     }
 120
 121     s->common.len = bdrv_getlength(bs);
 122     if (s->common.len < 0) {
 123         ret = s->common.len;
 124         goto out;
 125     }
 126
 127     buf = qemu_blockalign(bs, STREAM_BUFFER_SIZE);
 128
 129     /* Turn on copy-on-read for the whole block device so that guest read
 130      * requests help us make progress.  Only do this when copying the entire
 131      * backing chain since the copy-on-read operation does not take base into
 132      * account.
 133      */
 134     if (!base) {
 135         bdrv_enable_copy_on_read(bs);
 136     }
 137
 138     for ( ; offset < s->common.len; offset += n * BDRV_SECTOR_SIZE) {
 139         bool copy;
 140         int64_t count = 0;
 141
 142         /* Note that even when no rate limit is applied we need to yield
 143          * with no pending I/O here so that bdrv_drain_all() returns.
 144          */
 145         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 146         if (block_job_is_cancelled(&s->common)) {
 147             break;
 148         }
 149
 150         copy = false;
 151
 152         ret = bdrv_is_allocated(bs, offset, STREAM_BUFFER_SIZE, &count);
 153         /* TODO relax this once bdrv_is_allocated does not enforce sectors */
 154         assert(QEMU_IS_ALIGNED(count, BDRV_SECTOR_SIZE));
 155         n = count >> BDRV_SECTOR_BITS;
 156         if (ret == 1) {
 157             /* Allocated in the top, no need to copy.  */
 158         } else if (ret >= 0) {
 159             /* Copy if allocated in the intermediate images.  Limit to the
 160              * known-unallocated area [offset, offset+n*BDRV_SECTOR_SIZE).  */
 161             ret = bdrv_is_allocated_above(backing_bs(bs), base,
 162                                           offset / BDRV_SECTOR_SIZE, n, &n);
 163
 164             /* Finish early if end of backing file has been reached */
 165             if (ret == 0 && n == 0) {
 166                 n = (s->common.len - offset) / BDRV_SECTOR_SIZE;
 167             }
 168
 169             copy = (ret == 1);
 170         }
 171         trace_stream_one_iteration(s, offset, n * BDRV_SECTOR_SIZE, ret);
 172         if (copy) {
 173             ret = stream_populate(blk, offset, n * BDRV_SECTOR_SIZE, buf);
 174         }
 175         if (ret < 0) {
 176             BlockErrorAction action =
 177                 block_job_error_action(&s->common, s->on_error, true, -ret);
 178             if (action == BLOCK_ERROR_ACTION_STOP) {
 179                 n = 0;
 180                 continue;
 181             }
 182             if (error == 0) {
 183                 error = ret;
 184             }
 185             if (action == BLOCK_ERROR_ACTION_REPORT) {
 186                 break;
 187             }
 188         }
 189         ret = 0;
 190
 191         /* Publish progress */
 192         s->common.offset += n * BDRV_SECTOR_SIZE;
 193         if (copy && s->common.speed) {
 194             delay_ns = ratelimit_calculate_delay(&s->limit,
 195                                                  n * BDRV_SECTOR_SIZE);
 196         }
 197     }
 198
 199     if (!base) {
 200         bdrv_disable_copy_on_read(bs);
 201     }
 202
 203     /* Do not remove the backing file if an error was there but ignored.  */
 204     ret = error;
 205
 206     qemu_vfree(buf);
 207
 208 out:
 209     /* Modify backing chain and close BDSes in main loop */
 210     data = g_malloc(sizeof(*data));
 211     data->ret = ret;
 212     block_job_defer_to_main_loop(&s->common, stream_complete, data);
 213 }
 214
 215 static void stream_set_speed(BlockJob *job, int64_t speed, Error **errp)
 216 {
 217     StreamBlockJob *s = container_of(job, StreamBlockJob, common);
 218
 219     if (speed < 0) {
 220         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 221         return;
 222     }
 223     ratelimit_set_speed(&s->limit, speed, SLICE_TIME);
 224 }
 225
 226 static const BlockJobDriver stream_job_driver = {
 227     .instance_size = sizeof(StreamBlockJob),
 228     .job_type      = BLOCK_JOB_TYPE_STREAM,
 229     .set_speed     = stream_set_speed,
 230     .start         = stream_run,
 231 };
 232
 233 void stream_start(const char *job_id, BlockDriverState *bs,
 234                   BlockDriverState *base, const char *backing_file_str,
 235                   int64_t speed, BlockdevOnError on_error, Error **errp)
 236 {
 237     StreamBlockJob *s;
 238     BlockDriverState *iter;
 239     int orig_bs_flags;
 240
 241     /* Make sure that the image is opened in read-write mode */
 242     orig_bs_flags = bdrv_get_flags(bs);
 243     if (!(orig_bs_flags & BDRV_O_RDWR)) {
 244         if (bdrv_reopen(bs, orig_bs_flags | BDRV_O_RDWR, errp) != 0) {
 245             return;
 246         }
 247     }
 248
 249     /* Prevent concurrent jobs trying to modify the graph structure here, we
 250      * already have our own plans. Also don't allow resize as the image size is
 251      * queried only at the job start and then cached. */
 252     s = block_job_create(job_id, &stream_job_driver, bs,
 253                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
 254                          BLK_PERM_GRAPH_MOD,
 255                          BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED |
 256                          BLK_PERM_WRITE,
 257                          speed, BLOCK_JOB_DEFAULT, NULL, NULL, errp);
 258     if (!s) {
 259         goto fail;
 260     }
 261
 262     /* Block all intermediate nodes between bs and base, because they will
 263      * disappear from the chain after this operation. The streaming job reads
 264      * every block only once, assuming that it doesn't change, so block writes
 265      * and resizes. */
 266     for (iter = backing_bs(bs); iter && iter != base; iter = backing_bs(iter)) {
 267         block_job_add_bdrv(&s->common, "intermediate node", iter, 0,
 268                            BLK_PERM_CONSISTENT_READ | BLK_PERM_WRITE_UNCHANGED,
 269                            &error_abort);
 270     }
 271
 272     s->base = base;
 273     s->backing_file_str = g_strdup(backing_file_str);
 274     s->bs_flags = orig_bs_flags;
 275
 276     s->on_error = on_error;
 277     trace_stream_start(bs, base, s);
 278     block_job_start(&s->common);
 279     return;
 280
 281 fail:
 282     if (orig_bs_flags != bdrv_get_flags(bs)) {
 283         bdrv_reopen(bs, orig_bs_flags, NULL);
 284     }
 285 }