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/block_int.h"
  17 #include "block/blockjob.h"
  18 #include "qapi/qmp/qerror.h"
  19 #include "qemu/ratelimit.h"
  20 #include "sysemu/block-backend.h"
  21
  22 enum {
  23     /*
  24      * Size of data buffer for populating the image file.  This should be large
  25      * enough to process multiple clusters in a single call, so that populating
  26      * contiguous regions of the image is efficient.
  27      */
  28     COMMIT_BUFFER_SIZE = 512 * 1024, /* in bytes */
  29 };
  30
  31 #define SLICE_TIME 100000000ULL /* ns */
  32
  33 typedef struct CommitBlockJob {
  34     BlockJob common;
  35     RateLimit limit;
  36     BlockDriverState *active;
  37     BlockDriverState *top;
  38     BlockDriverState *base;
  39     BlockdevOnError on_error;
  40     int base_flags;
  41     int orig_overlay_flags;
  42     char *backing_file_str;
  43 } CommitBlockJob;
  44
  45 static int coroutine_fn commit_populate(BlockDriverState *bs,
  46                                         BlockDriverState *base,
  47                                         int64_t sector_num, int nb_sectors,
  48                                         void *buf)
  49 {
  50     int ret = 0;
  51
  52     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
  53     if (ret) {
  54         return ret;
  55     }
  56
  57     ret = bdrv_write(base, sector_num, buf, nb_sectors);
  58     if (ret) {
  59         return ret;
  60     }
  61
  62     return 0;
  63 }
  64
  65 typedef struct {
  66     int ret;
  67 } CommitCompleteData;
  68
  69 static void commit_complete(BlockJob *job, void *opaque)
  70 {
  71     CommitBlockJob *s = container_of(job, CommitBlockJob, common);
  72     CommitCompleteData *data = opaque;
  73     BlockDriverState *active = s->active;
  74     BlockDriverState *top = s->top;
  75     BlockDriverState *base = s->base;
  76     BlockDriverState *overlay_bs;
  77     int ret = data->ret;
  78
  79     if (!block_job_is_cancelled(&s->common) && ret == 0) {
  80         /* success */
  81         ret = bdrv_drop_intermediate(active, top, base, s->backing_file_str);
  82     }
  83
  84     /* restore base open flags here if appropriate (e.g., change the base back
  85      * to r/o). These reopens do not need to be atomic, since we won't abort
  86      * even on failure here */
  87     if (s->base_flags != bdrv_get_flags(base)) {
  88         bdrv_reopen(base, s->base_flags, NULL);
  89     }
  90     overlay_bs = bdrv_find_overlay(active, top);
  91     if (overlay_bs && s->orig_overlay_flags != bdrv_get_flags(overlay_bs)) {
  92         bdrv_reopen(overlay_bs, s->orig_overlay_flags, NULL);
  93     }
  94     g_free(s->backing_file_str);
  95     block_job_completed(&s->common, ret);
  96     g_free(data);
  97 }
  98
  99 static void coroutine_fn commit_run(void *opaque)
 100 {
 101     CommitBlockJob *s = opaque;
 102     CommitCompleteData *data;
 103     BlockDriverState *top = s->top;
 104     BlockDriverState *base = s->base;
 105     int64_t sector_num, end;
 106     int ret = 0;
 107     int n = 0;
 108     void *buf = NULL;
 109     int bytes_written = 0;
 110     int64_t base_len;
 111
 112     ret = s->common.len = bdrv_getlength(top);
 113
 114
 115     if (s->common.len < 0) {
 116         goto out;
 117     }
 118
 119     ret = base_len = bdrv_getlength(base);
 120     if (base_len < 0) {
 121         goto out;
 122     }
 123
 124     if (base_len < s->common.len) {
 125         ret = bdrv_truncate(base, s->common.len);
 126         if (ret) {
 127             goto out;
 128         }
 129     }
 130
 131     end = s->common.len >> BDRV_SECTOR_BITS;
 132     buf = qemu_blockalign(top, COMMIT_BUFFER_SIZE);
 133
 134     for (sector_num = 0; sector_num < end; sector_num += n) {
 135         uint64_t delay_ns = 0;
 136         bool copy;
 137
 138 wait:
 139         /* Note that even when no rate limit is applied we need to yield
 140          * with no pending I/O here so that bdrv_drain_all() returns.
 141          */
 142         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 143         if (block_job_is_cancelled(&s->common)) {
 144             break;
 145         }
 146         /* Copy if allocated above the base */
 147         ret = bdrv_is_allocated_above(top, base, sector_num,
 148                                       COMMIT_BUFFER_SIZE / BDRV_SECTOR_SIZE,
 149                                       &n);
 150         copy = (ret == 1);
 151         trace_commit_one_iteration(s, sector_num, n, ret);
 152         if (copy) {
 153             if (s->common.speed) {
 154                 delay_ns = ratelimit_calculate_delay(&s->limit, n);
 155                 if (delay_ns > 0) {
 156                     goto wait;
 157                 }
 158             }
 159             ret = commit_populate(top, base, sector_num, n, buf);
 160             bytes_written += n * BDRV_SECTOR_SIZE;
 161         }
 162         if (ret < 0) {
 163             if (s->on_error == BLOCKDEV_ON_ERROR_STOP ||
 164                 s->on_error == BLOCKDEV_ON_ERROR_REPORT||
 165                 (s->on_error == BLOCKDEV_ON_ERROR_ENOSPC && ret == -ENOSPC)) {
 166                 goto out;
 167             } else {
 168                 n = 0;
 169                 continue;
 170             }
 171         }
 172         /* Publish progress */
 173         s->common.offset += n * BDRV_SECTOR_SIZE;
 174     }
 175
 176     ret = 0;
 177
 178 out:
 179     qemu_vfree(buf);
 180
 181     data = g_malloc(sizeof(*data));
 182     data->ret = ret;
 183     block_job_defer_to_main_loop(&s->common, commit_complete, data);
 184 }
 185
 186 static void commit_set_speed(BlockJob *job, int64_t speed, Error **errp)
 187 {
 188     CommitBlockJob *s = container_of(job, CommitBlockJob, common);
 189
 190     if (speed < 0) {
 191         error_setg(errp, QERR_INVALID_PARAMETER, "speed");
 192         return;
 193     }
 194     ratelimit_set_speed(&s->limit, speed / BDRV_SECTOR_SIZE, SLICE_TIME);
 195 }
 196
 197 static const BlockJobDriver commit_job_driver = {
 198     .instance_size = sizeof(CommitBlockJob),
 199     .job_type      = BLOCK_JOB_TYPE_COMMIT,
 200     .set_speed     = commit_set_speed,
 201 };
 202
 203 void commit_start(BlockDriverState *bs, BlockDriverState *base,
 204                   BlockDriverState *top, int64_t speed,
 205                   BlockdevOnError on_error, BlockCompletionFunc *cb,
 206                   void *opaque, const char *backing_file_str, Error **errp)
 207 {
 208     CommitBlockJob *s;
 209     BlockReopenQueue *reopen_queue = NULL;
 210     int orig_overlay_flags;
 211     int orig_base_flags;
 212     BlockDriverState *overlay_bs;
 213     Error *local_err = NULL;
 214
 215     if ((on_error == BLOCKDEV_ON_ERROR_STOP ||
 216          on_error == BLOCKDEV_ON_ERROR_ENOSPC) &&
 217         (!bs->blk || !blk_iostatus_is_enabled(bs->blk))) {
 218         error_setg(errp, "Invalid parameter combination");
 219         return;
 220     }
 221
 222     assert(top != bs);
 223     if (top == base) {
 224         error_setg(errp, "Invalid files for merge: top and base are the same");
 225         return;
 226     }
 227
 228     overlay_bs = bdrv_find_overlay(bs, top);
 229
 230     if (overlay_bs == NULL) {
 231         error_setg(errp, "Could not find overlay image for %s:", top->filename);
 232         return;
 233     }
 234
 235     orig_base_flags    = bdrv_get_flags(base);
 236     orig_overlay_flags = bdrv_get_flags(overlay_bs);
 237
 238     /* convert base & overlay_bs to r/w, if necessary */
 239     if (!(orig_overlay_flags & BDRV_O_RDWR)) {
 240         reopen_queue = bdrv_reopen_queue(reopen_queue, overlay_bs, NULL,
 241                                          orig_overlay_flags | BDRV_O_RDWR);
 242     }
 243     if (!(orig_base_flags & BDRV_O_RDWR)) {
 244         reopen_queue = bdrv_reopen_queue(reopen_queue, base, NULL,
 245                                          orig_base_flags | BDRV_O_RDWR);
 246     }
 247     if (reopen_queue) {
 248         bdrv_reopen_multiple(reopen_queue, &local_err);
 249         if (local_err != NULL) {
 250             error_propagate(errp, local_err);
 251             return;
 252         }
 253     }
 254
 255
 256     s = block_job_create(&commit_job_driver, bs, speed, cb, opaque, errp);
 257     if (!s) {
 258         return;
 259     }
 260
 261     s->base   = base;
 262     s->top    = top;
 263     s->active = bs;
 264
 265     s->base_flags          = orig_base_flags;
 266     s->orig_overlay_flags  = orig_overlay_flags;
 267
 268     s->backing_file_str = g_strdup(backing_file_str);
 269
 270     s->on_error = on_error;
 271     s->common.co = qemu_coroutine_create(commit_run);
 272
 273     trace_commit_start(bs, base, top, s, s->common.co, opaque);
 274     qemu_coroutine_enter(s->common.co, s);
 275 }