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