block/qcow2-snapshot.c

   1 /*
   2  * Block driver for the QCOW version 2 format
   3  *
   4  * Copyright (c) 2004-2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24
  25 #include "qemu-common.h"
  26 #include "block_int.h"
  27 #include "block/qcow2.h"
  28
  29 typedef struct QEMU_PACKED QCowSnapshotHeader {
  30     /* header is 8 byte aligned */
  31     uint64_t l1_table_offset;
  32
  33     uint32_t l1_size;
  34     uint16_t id_str_size;
  35     uint16_t name_size;
  36
  37     uint32_t date_sec;
  38     uint32_t date_nsec;
  39
  40     uint64_t vm_clock_nsec;
  41
  42     uint32_t vm_state_size;
  43     uint32_t extra_data_size; /* for extension */
  44     /* extra data follows */
  45     /* id_str follows */
  46     /* name follows  */
  47 } QCowSnapshotHeader;
  48
  49 void qcow2_free_snapshots(BlockDriverState *bs)
  50 {
  51     BDRVQcowState *s = bs->opaque;
  52     int i;
  53
  54     for(i = 0; i < s->nb_snapshots; i++) {
  55         g_free(s->snapshots[i].name);
  56         g_free(s->snapshots[i].id_str);
  57     }
  58     g_free(s->snapshots);
  59     s->snapshots = NULL;
  60     s->nb_snapshots = 0;
  61 }
  62
  63 int qcow2_read_snapshots(BlockDriverState *bs)
  64 {
  65     BDRVQcowState *s = bs->opaque;
  66     QCowSnapshotHeader h;
  67     QCowSnapshot *sn;
  68     int i, id_str_size, name_size;
  69     int64_t offset;
  70     uint32_t extra_data_size;
  71     int ret;
  72
  73     if (!s->nb_snapshots) {
  74         s->snapshots = NULL;
  75         s->snapshots_size = 0;
  76         return 0;
  77     }
  78
  79     offset = s->snapshots_offset;
  80     s->snapshots = g_malloc0(s->nb_snapshots * sizeof(QCowSnapshot));
  81
  82     for(i = 0; i < s->nb_snapshots; i++) {
  83         /* Read statically sized part of the snapshot header */
  84         offset = align_offset(offset, 8);
  85         ret = bdrv_pread(bs->file, offset, &h, sizeof(h));
  86         if (ret < 0) {
  87             goto fail;
  88         }
  89
  90         offset += sizeof(h);
  91         sn = s->snapshots + i;
  92         sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
  93         sn->l1_size = be32_to_cpu(h.l1_size);
  94         sn->vm_state_size = be32_to_cpu(h.vm_state_size);
  95         sn->date_sec = be32_to_cpu(h.date_sec);
  96         sn->date_nsec = be32_to_cpu(h.date_nsec);
  97         sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
  98         extra_data_size = be32_to_cpu(h.extra_data_size);
  99
 100         id_str_size = be16_to_cpu(h.id_str_size);
 101         name_size = be16_to_cpu(h.name_size);
 102
 103         /* Skip extra data */
 104         offset += extra_data_size;
 105
 106         /* Read snapshot ID */
 107         sn->id_str = g_malloc(id_str_size + 1);
 108         ret = bdrv_pread(bs->file, offset, sn->id_str, id_str_size);
 109         if (ret < 0) {
 110             goto fail;
 111         }
 112         offset += id_str_size;
 113         sn->id_str[id_str_size] = '\0';
 114
 115         /* Read snapshot name */
 116         sn->name = g_malloc(name_size + 1);
 117         ret = bdrv_pread(bs->file, offset, sn->name, name_size);
 118         if (ret < 0) {
 119             goto fail;
 120         }
 121         offset += name_size;
 122         sn->name[name_size] = '\0';
 123     }
 124
 125     s->snapshots_size = offset - s->snapshots_offset;
 126     return 0;
 127
 128 fail:
 129     qcow2_free_snapshots(bs);
 130     return ret;
 131 }
 132
 133 /* add at the end of the file a new list of snapshots */
 134 static int qcow2_write_snapshots(BlockDriverState *bs)
 135 {
 136     BDRVQcowState *s = bs->opaque;
 137     QCowSnapshot *sn;
 138     QCowSnapshotHeader h;
 139     int i, name_size, id_str_size, snapshots_size;
 140     struct {
 141         uint32_t nb_snapshots;
 142         uint64_t snapshots_offset;
 143     } QEMU_PACKED header_data;
 144     int64_t offset, snapshots_offset;
 145     int ret;
 146
 147     /* compute the size of the snapshots */
 148     offset = 0;
 149     for(i = 0; i < s->nb_snapshots; i++) {
 150         sn = s->snapshots + i;
 151         offset = align_offset(offset, 8);
 152         offset += sizeof(h);
 153         offset += strlen(sn->id_str);
 154         offset += strlen(sn->name);
 155     }
 156     snapshots_size = offset;
 157
 158     /* Allocate space for the new snapshot list */
 159     snapshots_offset = qcow2_alloc_clusters(bs, snapshots_size);
 160     bdrv_flush(bs->file);
 161     offset = snapshots_offset;
 162     if (offset < 0) {
 163         return offset;
 164     }
 165
 166     /* Write all snapshots to the new list */
 167     for(i = 0; i < s->nb_snapshots; i++) {
 168         sn = s->snapshots + i;
 169         memset(&h, 0, sizeof(h));
 170         h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
 171         h.l1_size = cpu_to_be32(sn->l1_size);
 172         h.vm_state_size = cpu_to_be32(sn->vm_state_size);
 173         h.date_sec = cpu_to_be32(sn->date_sec);
 174         h.date_nsec = cpu_to_be32(sn->date_nsec);
 175         h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
 176
 177         id_str_size = strlen(sn->id_str);
 178         name_size = strlen(sn->name);
 179         h.id_str_size = cpu_to_be16(id_str_size);
 180         h.name_size = cpu_to_be16(name_size);
 181         offset = align_offset(offset, 8);
 182
 183         ret = bdrv_pwrite(bs->file, offset, &h, sizeof(h));
 184         if (ret < 0) {
 185             goto fail;
 186         }
 187         offset += sizeof(h);
 188
 189         ret = bdrv_pwrite(bs->file, offset, sn->id_str, id_str_size);
 190         if (ret < 0) {
 191             goto fail;
 192         }
 193         offset += id_str_size;
 194
 195         ret = bdrv_pwrite(bs->file, offset, sn->name, name_size);
 196         if (ret < 0) {
 197             goto fail;
 198         }
 199         offset += name_size;
 200     }
 201
 202     /*
 203      * Update the header to point to the new snapshot table. This requires the
 204      * new table and its refcounts to be stable on disk.
 205      */
 206     ret = bdrv_flush(bs);
 207     if (ret < 0) {
 208         goto fail;
 209     }
 210
 211     QEMU_BUILD_BUG_ON(offsetof(QCowHeader, snapshots_offset) !=
 212         offsetof(QCowHeader, nb_snapshots) + sizeof(header_data.nb_snapshots));
 213
 214     header_data.nb_snapshots        = cpu_to_be32(s->nb_snapshots);
 215     header_data.snapshots_offset    = cpu_to_be64(snapshots_offset);
 216
 217     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, nb_snapshots),
 218                            &header_data, sizeof(header_data));
 219     if (ret < 0) {
 220         goto fail;
 221     }
 222
 223     /* free the old snapshot table */
 224     qcow2_free_clusters(bs, s->snapshots_offset, s->snapshots_size);
 225     s->snapshots_offset = snapshots_offset;
 226     s->snapshots_size = snapshots_size;
 227     return 0;
 228
 229 fail:
 230     return ret;
 231 }
 232
 233 static void find_new_snapshot_id(BlockDriverState *bs,
 234                                  char *id_str, int id_str_size)
 235 {
 236     BDRVQcowState *s = bs->opaque;
 237     QCowSnapshot *sn;
 238     int i, id, id_max = 0;
 239
 240     for(i = 0; i < s->nb_snapshots; i++) {
 241         sn = s->snapshots + i;
 242         id = strtoul(sn->id_str, NULL, 10);
 243         if (id > id_max)
 244             id_max = id;
 245     }
 246     snprintf(id_str, id_str_size, "%d", id_max + 1);
 247 }
 248
 249 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
 250 {
 251     BDRVQcowState *s = bs->opaque;
 252     int i;
 253
 254     for(i = 0; i < s->nb_snapshots; i++) {
 255         if (!strcmp(s->snapshots[i].id_str, id_str))
 256             return i;
 257     }
 258     return -1;
 259 }
 260
 261 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
 262 {
 263     BDRVQcowState *s = bs->opaque;
 264     int i, ret;
 265
 266     ret = find_snapshot_by_id(bs, name);
 267     if (ret >= 0)
 268         return ret;
 269     for(i = 0; i < s->nb_snapshots; i++) {
 270         if (!strcmp(s->snapshots[i].name, name))
 271             return i;
 272     }
 273     return -1;
 274 }
 275
 276 /* if no id is provided, a new one is constructed */
 277 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
 278 {
 279     BDRVQcowState *s = bs->opaque;
 280     QCowSnapshot *new_snapshot_list = NULL;
 281     QCowSnapshot *old_snapshot_list = NULL;
 282     QCowSnapshot sn1, *sn = &sn1;
 283     int i, ret;
 284     uint64_t *l1_table = NULL;
 285     int64_t l1_table_offset;
 286
 287     memset(sn, 0, sizeof(*sn));
 288
 289     /* Generate an ID if it wasn't passed */
 290     if (sn_info->id_str[0] == '\0') {
 291         find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
 292     }
 293
 294     /* Check that the ID is unique */
 295     if (find_snapshot_by_id(bs, sn_info->id_str) >= 0) {
 296         return -ENOENT;
 297     }
 298
 299     /* Populate sn with passed data */
 300     sn->id_str = g_strdup(sn_info->id_str);
 301     sn->name = g_strdup(sn_info->name);
 302
 303     sn->vm_state_size = sn_info->vm_state_size;
 304     sn->date_sec = sn_info->date_sec;
 305     sn->date_nsec = sn_info->date_nsec;
 306     sn->vm_clock_nsec = sn_info->vm_clock_nsec;
 307
 308     /* Allocate the L1 table of the snapshot and copy the current one there. */
 309     l1_table_offset = qcow2_alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
 310     if (l1_table_offset < 0) {
 311         ret = l1_table_offset;
 312         goto fail;
 313     }
 314
 315     sn->l1_table_offset = l1_table_offset;
 316     sn->l1_size = s->l1_size;
 317
 318     l1_table = g_malloc(s->l1_size * sizeof(uint64_t));
 319     for(i = 0; i < s->l1_size; i++) {
 320         l1_table[i] = cpu_to_be64(s->l1_table[i]);
 321     }
 322
 323     ret = bdrv_pwrite(bs->file, sn->l1_table_offset, l1_table,
 324                       s->l1_size * sizeof(uint64_t));
 325     if (ret < 0) {
 326         goto fail;
 327     }
 328
 329     g_free(l1_table);
 330     l1_table = NULL;
 331
 332     /*
 333      * Increase the refcounts of all clusters and make sure everything is
 334      * stable on disk before updating the snapshot table to contain a pointer
 335      * to the new L1 table.
 336      */
 337     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
 338     if (ret < 0) {
 339         goto fail;
 340     }
 341
 342     ret = bdrv_flush(bs);
 343     if (ret < 0) {
 344         goto fail;
 345     }
 346
 347     /* Append the new snapshot to the snapshot list */
 348     new_snapshot_list = g_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
 349     if (s->snapshots) {
 350         memcpy(new_snapshot_list, s->snapshots,
 351                s->nb_snapshots * sizeof(QCowSnapshot));
 352         old_snapshot_list = s->snapshots;
 353     }
 354     s->snapshots = new_snapshot_list;
 355     s->snapshots[s->nb_snapshots++] = *sn;
 356
 357     ret = qcow2_write_snapshots(bs);
 358     if (ret < 0) {
 359         g_free(s->snapshots);
 360         s->snapshots = old_snapshot_list;
 361         goto fail;
 362     }
 363
 364     g_free(old_snapshot_list);
 365
 366 #ifdef DEBUG_ALLOC
 367     {
 368       BdrvCheckResult result = {0};
 369       qcow2_check_refcounts(bs, &result);
 370     }
 371 #endif
 372     return 0;
 373
 374 fail:
 375     g_free(sn->id_str);
 376     g_free(sn->name);
 377     g_free(l1_table);
 378
 379     return ret;
 380 }
 381
 382 /* copy the snapshot 'snapshot_name' into the current disk image */
 383 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
 384 {
 385     BDRVQcowState *s = bs->opaque;
 386     QCowSnapshot *sn;
 387     int i, snapshot_index;
 388     int cur_l1_bytes, sn_l1_bytes;
 389     int ret;
 390     uint64_t *sn_l1_table = NULL;
 391
 392     /* Search the snapshot */
 393     snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
 394     if (snapshot_index < 0) {
 395         return -ENOENT;
 396     }
 397     sn = &s->snapshots[snapshot_index];
 398
 399     /*
 400      * Make sure that the current L1 table is big enough to contain the whole
 401      * L1 table of the snapshot. If the snapshot L1 table is smaller, the
 402      * current one must be padded with zeros.
 403      */
 404     ret = qcow2_grow_l1_table(bs, sn->l1_size, true);
 405     if (ret < 0) {
 406         goto fail;
 407     }
 408
 409     cur_l1_bytes = s->l1_size * sizeof(uint64_t);
 410     sn_l1_bytes = sn->l1_size * sizeof(uint64_t);
 411
 412     /*
 413      * Copy the snapshot L1 table to the current L1 table.
 414      *
 415      * Before overwriting the old current L1 table on disk, make sure to
 416      * increase all refcounts for the clusters referenced by the new one.
 417      * Decrease the refcount referenced by the old one only when the L1
 418      * table is overwritten.
 419      */
 420     sn_l1_table = g_malloc0(cur_l1_bytes);
 421
 422     ret = bdrv_pread(bs->file, sn->l1_table_offset, sn_l1_table, sn_l1_bytes);
 423     if (ret < 0) {
 424         goto fail;
 425     }
 426
 427     ret = qcow2_update_snapshot_refcount(bs, sn->l1_table_offset,
 428                                          sn->l1_size, 1);
 429     if (ret < 0) {
 430         goto fail;
 431     }
 432
 433     ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset, sn_l1_table,
 434                            cur_l1_bytes);
 435     if (ret < 0) {
 436         goto fail;
 437     }
 438
 439     /*
 440      * Decrease refcount of clusters of current L1 table.
 441      *
 442      * At this point, the in-memory s->l1_table points to the old L1 table,
 443      * whereas on disk we already have the new one.
 444      *
 445      * qcow2_update_snapshot_refcount special cases the current L1 table to use
 446      * the in-memory data instead of really using the offset to load a new one,
 447      * which is why this works.
 448      */
 449     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset,
 450                                          s->l1_size, -1);
 451
 452     /*
 453      * Now update the in-memory L1 table to be in sync with the on-disk one. We
 454      * need to do this even if updating refcounts failed.
 455      */
 456     for(i = 0;i < s->l1_size; i++) {
 457         s->l1_table[i] = be64_to_cpu(sn_l1_table[i]);
 458     }
 459
 460     if (ret < 0) {
 461         goto fail;
 462     }
 463
 464     g_free(sn_l1_table);
 465     sn_l1_table = NULL;
 466
 467     /*
 468      * Update QCOW_OFLAG_COPIED in the active L1 table (it may have changed
 469      * when we decreased the refcount of the old snapshot.
 470      */
 471     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
 472     if (ret < 0) {
 473         goto fail;
 474     }
 475
 476 #ifdef DEBUG_ALLOC
 477     {
 478         BdrvCheckResult result = {0};
 479         qcow2_check_refcounts(bs, &result);
 480     }
 481 #endif
 482     return 0;
 483
 484 fail:
 485     g_free(sn_l1_table);
 486     return ret;
 487 }
 488
 489 int qcow2_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
 490 {
 491     BDRVQcowState *s = bs->opaque;
 492     QCowSnapshot sn;
 493     int snapshot_index, ret;
 494
 495     /* Search the snapshot */
 496     snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
 497     if (snapshot_index < 0) {
 498         return -ENOENT;
 499     }
 500     sn = s->snapshots[snapshot_index];
 501
 502     /* Remove it from the snapshot list */
 503     memmove(s->snapshots + snapshot_index,
 504             s->snapshots + snapshot_index + 1,
 505             (s->nb_snapshots - snapshot_index - 1) * sizeof(sn));
 506     s->nb_snapshots--;
 507     ret = qcow2_write_snapshots(bs);
 508     if (ret < 0) {
 509         return ret;
 510     }
 511
 512     /*
 513      * The snapshot is now unused, clean up. If we fail after this point, we
 514      * won't recover but just leak clusters.
 515      */
 516     g_free(sn.id_str);
 517     g_free(sn.name);
 518
 519     /*
 520      * Now decrease the refcounts of clusters referenced by the snapshot and
 521      * free the L1 table.
 522      */
 523     ret = qcow2_update_snapshot_refcount(bs, sn.l1_table_offset,
 524                                          sn.l1_size, -1);
 525     if (ret < 0) {
 526         return ret;
 527     }
 528     qcow2_free_clusters(bs, sn.l1_table_offset, sn.l1_size * sizeof(uint64_t));
 529
 530     /* must update the copied flag on the current cluster offsets */
 531     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
 532     if (ret < 0) {
 533         return ret;
 534     }
 535
 536 #ifdef DEBUG_ALLOC
 537     {
 538         BdrvCheckResult result = {0};
 539         qcow2_check_refcounts(bs, &result);
 540     }
 541 #endif
 542     return 0;
 543 }
 544
 545 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
 546 {
 547     BDRVQcowState *s = bs->opaque;
 548     QEMUSnapshotInfo *sn_tab, *sn_info;
 549     QCowSnapshot *sn;
 550     int i;
 551
 552     if (!s->nb_snapshots) {
 553         *psn_tab = NULL;
 554         return s->nb_snapshots;
 555     }
 556
 557     sn_tab = g_malloc0(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
 558     for(i = 0; i < s->nb_snapshots; i++) {
 559         sn_info = sn_tab + i;
 560         sn = s->snapshots + i;
 561         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
 562                 sn->id_str);
 563         pstrcpy(sn_info->name, sizeof(sn_info->name),
 564                 sn->name);
 565         sn_info->vm_state_size = sn->vm_state_size;
 566         sn_info->date_sec = sn->date_sec;
 567         sn_info->date_nsec = sn->date_nsec;
 568         sn_info->vm_clock_nsec = sn->vm_clock_nsec;
 569     }
 570     *psn_tab = sn_tab;
 571     return s->nb_snapshots;
 572 }
 573
 574 int qcow2_snapshot_load_tmp(BlockDriverState *bs, const char *snapshot_name)
 575 {
 576     int i, snapshot_index;
 577     BDRVQcowState *s = bs->opaque;
 578     QCowSnapshot *sn;
 579     uint64_t *new_l1_table;
 580     int new_l1_bytes;
 581     int ret;
 582
 583     assert(bs->read_only);
 584
 585     /* Search the snapshot */
 586     snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_name);
 587     if (snapshot_index < 0) {
 588         return -ENOENT;
 589     }
 590     sn = &s->snapshots[snapshot_index];
 591
 592     /* Allocate and read in the snapshot's L1 table */
 593     new_l1_bytes = s->l1_size * sizeof(uint64_t);
 594     new_l1_table = g_malloc0(align_offset(new_l1_bytes, 512));
 595
 596     ret = bdrv_pread(bs->file, sn->l1_table_offset, new_l1_table, new_l1_bytes);
 597     if (ret < 0) {
 598         g_free(new_l1_table);
 599         return ret;
 600     }
 601
 602     /* Switch the L1 table */
 603     g_free(s->l1_table);
 604
 605     s->l1_size = sn->l1_size;
 606     s->l1_table_offset = sn->l1_table_offset;
 607     s->l1_table = new_l1_table;
 608
 609     for(i = 0;i < s->l1_size; i++) {
 610         be64_to_cpus(&s->l1_table[i]);
 611     }
 612
 613     return 0;
 614 }