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 "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     char *backing_file_str;
  41 } CommitBlockJob;
  42
  43 static int coroutine_fn commit_populate(BlockDriverState *bs,
  44                                         BlockDriverState *base,
  45                                         int64_t sector_num, int nb_sectors,
  46                                         void *buf)
  47 {
  48     int ret = 0;
  49
  50     ret = bdrv_read(bs, sector_num, buf, nb_sectors);
  51     if (ret) {
  52         return ret;
  53     }
  54
  55     ret = bdrv_write(base, sector_num, buf, nb_sectors);
  56     if (ret) {
  57         return ret;
  58     }
  59
  60     return 0;
  61 }
  62
  63 static void coroutine_fn commit_run(void *opaque)
  64 {
  65     CommitBlockJob *s = opaque;
  66     BlockDriverState *active = s->active;
  67     BlockDriverState *top = s->top;
  68     BlockDriverState *base = s->base;
  69     BlockDriverState *overlay_bs;
  70     int64_t sector_num, end;
  71     int ret = 0;
  72     int n = 0;
  73     void *buf;
  74     int bytes_written = 0;
  75     int64_t base_len;
  76
  77     ret = s->common.len = bdrv_getlength(top);
  78
  79
  80     if (s->common.len < 0) {
  81         goto exit_restore_reopen;
  82     }
  83
  84     ret = base_len = bdrv_getlength(base);
  85     if (base_len < 0) {
  86         goto exit_restore_reopen;
  87     }
  88
  89     if (base_len < s->common.len) {
  90         ret = bdrv_truncate(base, s->common.len);
  91         if (ret) {
  92             goto exit_restore_reopen;
  93         }
  94     }
  95
  96     end = s->common.len >> BDRV_SECTOR_BITS;
  97     buf = qemu_blockalign(top, COMMIT_BUFFER_SIZE);
  98
  99     for (sector_num = 0; sector_num < end; sector_num += n) {
 100         uint64_t delay_ns = 0;
 101         bool copy;
 102
 103 wait:
 104         /* Note that even when no rate limit is applied we need to yield
 105          * with no pending I/O here so that bdrv_drain_all() returns.
 106          */
 107         block_job_sleep_ns(&s->common, QEMU_CLOCK_REALTIME, delay_ns);
 108         if (block_job_is_cancelled(&s->common)) {
 109             break;
 110         }
 111         /* Copy if allocated above the base */
 112         ret = bdrv_is_allocated_above(top, base, sector_num,
 113                                       COMMIT_BUFFER_SIZE / BDRV_SECTOR_SIZE,
 114                                       &n);
 115         copy = (ret == 1);
 116         trace_commit_one_iteration(s, sector_num, n, ret);
 117         if (copy) {
 118             if (s->common.speed) {
 119                 delay_ns = ratelimit_calculate_delay(&s->limit, n);
 120                 if (delay_ns > 0) {
 121                     goto wait;
 122                 }
 123             }
 124             ret = commit_populate(top, base, sector_num, n, buf);
 125             bytes_written += n * BDRV_SECTOR_SIZE;
 126         }
 127         if (ret < 0) {
 128             if (s->on_error == BLOCKDEV_ON_ERROR_STOP ||
 129                 s->on_error == BLOCKDEV_ON_ERROR_REPORT||
 130                 (s->on_error == BLOCKDEV_ON_ERROR_ENOSPC && ret == -ENOSPC)) {
 131                 goto exit_free_buf;
 132             } else {
 133                 n = 0;
 134                 continue;
 135             }
 136         }
 137         /* Publish progress */
 138         s->common.offset += n * BDRV_SECTOR_SIZE;
 139     }
 140
 141     ret = 0;
 142
 143     if (!block_job_is_cancelled(&s->common) && sector_num == end) {
 144         /* success */
 145         ret = bdrv_drop_intermediate(active, top, base, s->backing_file_str);
 146     }
 147
 148 exit_free_buf:
 149     qemu_vfree(buf);
 150
 151 exit_restore_reopen:
 152     /* restore base open flags here if appropriate (e.g., change the base back
 153      * to r/o). These reopens do not need to be atomic, since we won't abort
 154      * even on failure here */
 155     if (s->base_flags != bdrv_get_flags(base)) {
 156         bdrv_reopen(base, s->base_flags, NULL);
 157     }
 158     overlay_bs = bdrv_find_overlay(active, top);
 159     if (overlay_bs && s->orig_overlay_flags != bdrv_get_flags(overlay_bs)) {
 160         bdrv_reopen(overlay_bs, s->orig_overlay_flags, NULL);
 161     }
 162     g_free(s->backing_file_str);
 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 const BlockJobDriver commit_job_driver = {
 178     .instance_size = sizeof(CommitBlockJob),
 179     .job_type      = BLOCK_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, const char *backing_file_str, 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_setg(errp, "Invalid parameter combination");
 199         return;
 200     }
 201
 202     assert(top != bs);
 203     if (top == base) {
 204         error_setg(errp, "Invalid files for merge: top and base are the same");
 205         return;
 206     }
 207
 208     overlay_bs = bdrv_find_overlay(bs, top);
 209
 210     if (overlay_bs == NULL) {
 211         error_setg(errp, "Could not find overlay image for %s:", top->filename);
 212         return;
 213     }
 214
 215     orig_base_flags    = bdrv_get_flags(base);
 216     orig_overlay_flags = bdrv_get_flags(overlay_bs);
 217
 218     /* convert base & overlay_bs to r/w, if necessary */
 219     if (!(orig_base_flags & BDRV_O_RDWR)) {
 220         reopen_queue = bdrv_reopen_queue(reopen_queue, base,
 221                                          orig_base_flags | BDRV_O_RDWR);
 222     }
 223     if (!(orig_overlay_flags & BDRV_O_RDWR)) {
 224         reopen_queue = bdrv_reopen_queue(reopen_queue, overlay_bs,
 225                                          orig_overlay_flags | BDRV_O_RDWR);
 226     }
 227     if (reopen_queue) {
 228         bdrv_reopen_multiple(reopen_queue, &local_err);
 229         if (local_err != NULL) {
 230             error_propagate(errp, local_err);
 231             return;
 232         }
 233     }
 234
 235
 236     s = block_job_create(&commit_job_driver, bs, speed, cb, opaque, errp);
 237     if (!s) {
 238         return;
 239     }
 240
 241     s->base   = base;
 242     s->top    = top;
 243     s->active = bs;
 244
 245     s->base_flags          = orig_base_flags;
 246     s->orig_overlay_flags  = orig_overlay_flags;
 247
 248     s->backing_file_str = g_strdup(backing_file_str);
 249
 250     s->on_error = on_error;
 251     s->common.co = qemu_coroutine_create(commit_run);
 252
 253     trace_commit_start(bs, base, top, s, s->common.co, opaque);
 254     qemu_coroutine_enter(s->common.co, s);
 255 }