block/commit.c

   1 /*
   2  * Live block commit
   3  *
   4  * Copyright Red Hat, Inc. 2012
   5  *
   6  * Authors:
   7  *  Jeff Cody   <jcody@redhat.com>
   8  *  Based on stream.c by Stefan Hajnoczi
   9  *
  10  * This work is licensed under the terms of the GNU LGPL, version 2 or later.
  11  * See the COPYING.LIB file in the top-level directory.
  12  *
  13  */
  14
  15 #include "trace.h"
  16 #include "block_int.h"
  17 #include "blockjob.h"
  18 #include "qemu/ratelimit.h"
  19
  20 enum {
  21     /*
  22      * Size of data buffer for populating the image file.  This should be large
  23      * enough to process multiple clusters in a single call, so that populating
  24      * contiguous regions of the image is efficient.
  25      */
  26     COMMIT_BUFFER_SIZE = 512 * 1024, /* in bytes */
  27 };
  28
  29 #define SLICE_TIME 100000000ULL /* ns */
  30
  31 typedef struct CommitBlockJob {
  32     BlockJob common;
  33     RateLimit limit;
  34     BlockDriverState *active;
  35     BlockDriverState *top;
  36     BlockDriverState *base;
  37     BlockdevOnError on_error;
  38     int base_flags;
  39     int orig_overlay_flags;
  40 } CommitBlockJob;
  41
  42 static int coroutine_fn commit_populate(BlockDriverState *bs,
  43                                         BlockDriverState *base,
  44                                         int64_t sector_num, int nb_sectors,
  45                                         void *buf)
  46 {
  47     int ret = 0;
  48
  49     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
  50     if (ret) {
  51         return ret;
  52     }
  53
  54     ret = bdrv_write(base, sector_num, buf, nb_sectors);
  55     if (ret) {
  56         return ret;
  57     }
  58
  59     return 0;
  60 }
  61
  62 static void coroutine_fn commit_run(void *opaque)
  63 {
  64     CommitBlockJob *s = opaque;
  65     BlockDriverState *active = s->active;
  66     BlockDriverState *top = s->top;
  67     BlockDriverState *base = s->base;
  68     BlockDriverState *overlay_bs = NULL;
  69     int64_t sector_num, end;
  70     int ret = 0;
  71     int n = 0;
  72     void *buf;
  73     int bytes_written = 0;
  74     int64_t base_len;
  75
  76     ret = s->common.len = bdrv_getlength(top);
  77
  78
  79     if (s->common.len < 0) {
  80         goto exit_restore_reopen;
  81     }
  82
  83     ret = base_len = bdrv_getlength(base);
  84     if (base_len < 0) {
  85         goto exit_restore_reopen;
  86     }
  87
  88     if (base_len < s->common.len) {
  89         ret = bdrv_truncate(base, s->common.len);
  90         if (ret) {
  91             goto exit_restore_reopen;
  92         }
  93     }
  94
  95     overlay_bs = bdrv_find_overlay(active, top);
  96
  97     end = s->common.len >> BDRV_SECTOR_BITS;
  98     buf = qemu_blockalign(top, COMMIT_BUFFER_SIZE);
  99
 100     for (sector_num = 0; sector_num < end; sector_num += n) {
 101         uint64_t delay_ns = 0;
 102         bool copy;
 103
 104 wait:
 105         /* Note that even when no rate limit is applied we need to yield
 106          * with no pending I/O here so that qemu_aio_flush() returns.
 107          */
 108         block_job_sleep_ns(&s->common, rt_clock, delay_ns);
 109         if (block_job_is_cancelled(&s->common)) {
 110             break;
 111         }
 112         /* Copy if allocated above the base */
 113         ret = bdrv_co_is_allocated_above(top, base, sector_num,
 114                                          COMMIT_BUFFER_SIZE / BDRV_SECTOR_SIZE,
 115                                          &n);
 116         copy = (ret == 1);
 117         trace_commit_one_iteration(s, sector_num, n, ret);
 118         if (copy) {
 119             if (s->common.speed) {
 120                 delay_ns = ratelimit_calculate_delay(&s->limit, n);
 121                 if (delay_ns > 0) {
 122                     goto wait;
 123                 }
 124             }
 125             ret = commit_populate(top, base, sector_num, n, buf);
 126             bytes_written += n * BDRV_SECTOR_SIZE;
 127         }
 128         if (ret < 0) {
 129             if (s->on_error == BLOCKDEV_ON_ERROR_STOP ||
 130                 s->on_error == BLOCKDEV_ON_ERROR_REPORT||
 131                 (s->on_error == BLOCKDEV_ON_ERROR_ENOSPC && ret == -ENOSPC)) {
 132                 goto exit_free_buf;
 133             } else {
 134                 n = 0;
 135                 continue;
 136             }
 137         }
 138         /* Publish progress */
 139         s->common.offset += n * BDRV_SECTOR_SIZE;
 140     }
 141
 142     ret = 0;
 143
 144     if (!block_job_is_cancelled(&s->common) && sector_num == end) {
 145         /* success */
 146         ret = bdrv_drop_intermediate(active, top, base);
 147     }
 148
 149 exit_free_buf:
 150     qemu_vfree(buf);
 151
 152 exit_restore_reopen:
 153     /* restore base open flags here if appropriate (e.g., change the base back
 154      * to r/o). These reopens do not need to be atomic, since we won't abort
 155      * even on failure here */
 156     if (s->base_flags != bdrv_get_flags(base)) {
 157         bdrv_reopen(base, s->base_flags, NULL);
 158     }
 159     if (s->orig_overlay_flags != bdrv_get_flags(overlay_bs)) {
 160         bdrv_reopen(overlay_bs, s->orig_overlay_flags, NULL);
 161     }
 162
 163     block_job_completed(&s->common, ret);
 164 }
 165
 166 static void commit_set_speed(BlockJob *job, int64_t speed, Error **errp)
 167 {
 168     CommitBlockJob *s = container_of(job, CommitBlockJob, common);
 169
 170     if (speed < 0) {
 171         error_set(errp, QERR_INVALID_PARAMETER, "speed");
 172         return;
 173     }
 174     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 175 }
 176
 177 static BlockJobType commit_job_type = {
 178     .instance_size = sizeof(CommitBlockJob),
 179     .job_type      = "commit",
 180     .set_speed     = commit_set_speed,
 181 };
 182
 183 void commit_start(BlockDriverState *bs, BlockDriverState *base,
 184                   BlockDriverState *top, int64_t speed,
 185                   BlockdevOnError on_error, BlockDriverCompletionFunc *cb,
 186                   void *opaque, Error **errp)
 187 {
 188     CommitBlockJob *s;
 189     BlockReopenQueue *reopen_queue = NULL;
 190     int orig_overlay_flags;
 191     int orig_base_flags;
 192     BlockDriverState *overlay_bs;
 193     Error *local_err = NULL;
 194
 195     if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
 196          on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
 197         !bdrv_iostatus_is_enabled(bs)) {
 198         error_set(errp, QERR_INVALID_PARAMETER_COMBINATION);
 199         return;
 200     }
 201
 202     /* Once we support top == active layer, remove this check */
 203     if (top == bs) {
 204         error_setg(errp,
 205                    "Top image as the active layer is currently unsupported");
 206         return;
 207     }
 208
 209     if (top == base) {
 210         error_setg(errp, "Invalid files for merge: top and base are the same");
 211         return;
 212     }
 213
 214     overlay_bs = bdrv_find_overlay(bs, top);
 215
 216     if (overlay_bs == NULL) {
 217         error_setg(errp, "Could not find overlay image for %s:", top->filename);
 218         return;
 219     }
 220
 221     orig_base_flags    = bdrv_get_flags(base);
 222     orig_overlay_flags = bdrv_get_flags(overlay_bs);
 223
 224     /* convert base & overlay_bs to r/w, if necessary */
 225     if (!(orig_base_flags & BDRV_O_RDWR)) {
 226         reopen_queue = bdrv_reopen_queue(reopen_queue, base,
 227                                          orig_base_flags | BDRV_O_RDWR);
 228     }
 229     if (!(orig_overlay_flags & BDRV_O_RDWR)) {
 230         reopen_queue = bdrv_reopen_queue(reopen_queue, overlay_bs,
 231                                          orig_overlay_flags | BDRV_O_RDWR);
 232     }
 233     if (reopen_queue) {
 234         bdrv_reopen_multiple(reopen_queue, &local_err);
 235         if (local_err != NULL) {
 236             error_propagate(errp, local_err);
 237             return;
 238         }
 239     }
 240
 241
 242     s = block_job_create(&commit_job_type, bs, speed, cb, opaque, errp);
 243     if (!s) {
 244         return;
 245     }
 246
 247     s->base   = base;
 248     s->top    = top;
 249     s->active = bs;
 250
 251     s->base_flags          = orig_base_flags;
 252     s->orig_overlay_flags  = orig_overlay_flags;
 253
 254     s->on_error = on_error;
 255     s->common.co = qemu_coroutine_create(commit_run);
 256
 257     trace_commit_start(bs, base, top, s, s->common.co, opaque);
 258     qemu_coroutine_enter(s->common.co, s);
 259 }