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 typedef struct QEMU_PACKED QCowSnapshotExtraData {
  50     uint64_t vm_state_size_large;
  51 } QCowSnapshotExtraData;
  52
  53 void qcow2_free_snapshots(BlockDriverState *bs)
  54 {
  55     BDRVQcowState *s = bs->opaque;
  56     int i;
  57
  58     for(i = 0; i < s->nb_snapshots; i++) {
  59         g_free(s->snapshots[i].name);
  60         g_free(s->snapshots[i].id_str);
  61     }
  62     g_free(s->snapshots);
  63     s->snapshots = NULL;
  64     s->nb_snapshots = 0;
  65 }
  66
  67 int qcow2_read_snapshots(BlockDriverState *bs)
  68 {
  69     BDRVQcowState *s = bs->opaque;
  70     QCowSnapshotHeader h;
  71     QCowSnapshotExtraData extra;
  72     QCowSnapshot *sn;
  73     int i, id_str_size, name_size;
  74     int64_t offset;
  75     uint32_t extra_data_size;
  76     int ret;
  77
  78     if (!s->nb_snapshots) {
  79         s->snapshots = NULL;
  80         s->snapshots_size = 0;
  81         return 0;
  82     }
  83
  84     offset = s->snapshots_offset;
  85     s->snapshots = g_malloc0(s->nb_snapshots * sizeof(QCowSnapshot));
  86
  87     for(i = 0; i < s->nb_snapshots; i++) {
  88         /* Read statically sized part of the snapshot header */
  89         offset = align_offset(offset, 8);
  90         ret = bdrv_pread(bs->file, offset, &h, sizeof(h));
  91         if (ret < 0) {
  92             goto fail;
  93         }
  94
  95         offset += sizeof(h);
  96         sn = s->snapshots + i;
  97         sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
  98         sn->l1_size = be32_to_cpu(h.l1_size);
  99         sn->vm_state_size = be32_to_cpu(h.vm_state_size);
 100         sn->date_sec = be32_to_cpu(h.date_sec);
 101         sn->date_nsec = be32_to_cpu(h.date_nsec);
 102         sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
 103         extra_data_size = be32_to_cpu(h.extra_data_size);
 104
 105         id_str_size = be16_to_cpu(h.id_str_size);
 106         name_size = be16_to_cpu(h.name_size);
 107
 108         /* Read extra data */
 109         ret = bdrv_pread(bs->file, offset, &extra,
 110                          MIN(sizeof(extra), extra_data_size));
 111         if (ret < 0) {
 112             goto fail;
 113         }
 114         offset += extra_data_size;
 115
 116         if (extra_data_size >= 8) {
 117             sn->vm_state_size = be64_to_cpu(extra.vm_state_size_large);
 118         }
 119
 120         /* Read snapshot ID */
 121         sn->id_str = g_malloc(id_str_size + 1);
 122         ret = bdrv_pread(bs->file, offset, sn->id_str, id_str_size);
 123         if (ret < 0) {
 124             goto fail;
 125         }
 126         offset += id_str_size;
 127         sn->id_str[id_str_size] = '\0';
 128
 129         /* Read snapshot name */
 130         sn->name = g_malloc(name_size + 1);
 131         ret = bdrv_pread(bs->file, offset, sn->name, name_size);
 132         if (ret < 0) {
 133             goto fail;
 134         }
 135         offset += name_size;
 136         sn->name[name_size] = '\0';
 137     }
 138
 139     s->snapshots_size = offset - s->snapshots_offset;
 140     return 0;
 141
 142 fail:
 143     qcow2_free_snapshots(bs);
 144     return ret;
 145 }
 146
 147 /* add at the end of the file a new list of snapshots */
 148 static int qcow2_write_snapshots(BlockDriverState *bs)
 149 {
 150     BDRVQcowState *s = bs->opaque;
 151     QCowSnapshot *sn;
 152     QCowSnapshotHeader h;
 153     QCowSnapshotExtraData extra;
 154     int i, name_size, id_str_size, snapshots_size;
 155     struct {
 156         uint32_t nb_snapshots;
 157         uint64_t snapshots_offset;
 158     } QEMU_PACKED header_data;
 159     int64_t offset, snapshots_offset;
 160     int ret;
 161
 162     /* compute the size of the snapshots */
 163     offset = 0;
 164     for(i = 0; i < s->nb_snapshots; i++) {
 165         sn = s->snapshots + i;
 166         offset = align_offset(offset, 8);
 167         offset += sizeof(h);
 168         offset += sizeof(extra);
 169         offset += strlen(sn->id_str);
 170         offset += strlen(sn->name);
 171     }
 172     snapshots_size = offset;
 173
 174     /* Allocate space for the new snapshot list */
 175     snapshots_offset = qcow2_alloc_clusters(bs, snapshots_size);
 176     bdrv_flush(bs->file);
 177     offset = snapshots_offset;
 178     if (offset < 0) {
 179         return offset;
 180     }
 181
 182     /* Write all snapshots to the new list */
 183     for(i = 0; i < s->nb_snapshots; i++) {
 184         sn = s->snapshots + i;
 185         memset(&h, 0, sizeof(h));
 186         h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
 187         h.l1_size = cpu_to_be32(sn->l1_size);
 188         /* If it doesn't fit in 32 bit, older implementations should treat it
 189          * as a disk-only snapshot rather than truncate the VM state */
 190         if (sn->vm_state_size <= 0xffffffff) {
 191             h.vm_state_size = cpu_to_be32(sn->vm_state_size);
 192         }
 193         h.date_sec = cpu_to_be32(sn->date_sec);
 194         h.date_nsec = cpu_to_be32(sn->date_nsec);
 195         h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
 196         h.extra_data_size = cpu_to_be32(sizeof(extra));
 197
 198         memset(&extra, 0, sizeof(extra));
 199         extra.vm_state_size_large = cpu_to_be64(sn->vm_state_size);
 200
 201         id_str_size = strlen(sn->id_str);
 202         name_size = strlen(sn->name);
 203         h.id_str_size = cpu_to_be16(id_str_size);
 204         h.name_size = cpu_to_be16(name_size);
 205         offset = align_offset(offset, 8);
 206
 207         ret = bdrv_pwrite(bs->file, offset, &h, sizeof(h));
 208         if (ret < 0) {
 209             goto fail;
 210         }
 211         offset += sizeof(h);
 212
 213         ret = bdrv_pwrite(bs->file, offset, &extra, sizeof(extra));
 214         if (ret < 0) {
 215             goto fail;
 216         }
 217         offset += sizeof(extra);
 218
 219         ret = bdrv_pwrite(bs->file, offset, sn->id_str, id_str_size);
 220         if (ret < 0) {
 221             goto fail;
 222         }
 223         offset += id_str_size;
 224
 225         ret = bdrv_pwrite(bs->file, offset, sn->name, name_size);
 226         if (ret < 0) {
 227             goto fail;
 228         }
 229         offset += name_size;
 230     }
 231
 232     /*
 233      * Update the header to point to the new snapshot table. This requires the
 234      * new table and its refcounts to be stable on disk.
 235      */
 236     ret = bdrv_flush(bs);
 237     if (ret < 0) {
 238         goto fail;
 239     }
 240
 241     QEMU_BUILD_BUG_ON(offsetof(QCowHeader, snapshots_offset) !=
 242         offsetof(QCowHeader, nb_snapshots) + sizeof(header_data.nb_snapshots));
 243
 244     header_data.nb_snapshots        = cpu_to_be32(s->nb_snapshots);
 245     header_data.snapshots_offset    = cpu_to_be64(snapshots_offset);
 246
 247     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, nb_snapshots),
 248                            &header_data, sizeof(header_data));
 249     if (ret < 0) {
 250         goto fail;
 251     }
 252
 253     /* free the old snapshot table */
 254     qcow2_free_clusters(bs, s->snapshots_offset, s->snapshots_size);
 255     s->snapshots_offset = snapshots_offset;
 256     s->snapshots_size = snapshots_size;
 257     return 0;
 258
 259 fail:
 260     return ret;
 261 }
 262
 263 static void find_new_snapshot_id(BlockDriverState *bs,
 264                                  char *id_str, int id_str_size)
 265 {
 266     BDRVQcowState *s = bs->opaque;
 267     QCowSnapshot *sn;
 268     int i, id, id_max = 0;
 269
 270     for(i = 0; i < s->nb_snapshots; i++) {
 271         sn = s->snapshots + i;
 272         id = strtoul(sn->id_str, NULL, 10);
 273         if (id > id_max)
 274             id_max = id;
 275     }
 276     snprintf(id_str, id_str_size, "%d", id_max + 1);
 277 }
 278
 279 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
 280 {
 281     BDRVQcowState *s = bs->opaque;
 282     int i;
 283
 284     for(i = 0; i < s->nb_snapshots; i++) {
 285         if (!strcmp(s->snapshots[i].id_str, id_str))
 286             return i;
 287     }
 288     return -1;
 289 }
 290
 291 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
 292 {
 293     BDRVQcowState *s = bs->opaque;
 294     int i, ret;
 295
 296     ret = find_snapshot_by_id(bs, name);
 297     if (ret >= 0)
 298         return ret;
 299     for(i = 0; i < s->nb_snapshots; i++) {
 300         if (!strcmp(s->snapshots[i].name, name))
 301             return i;
 302     }
 303     return -1;
 304 }
 305
 306 /* if no id is provided, a new one is constructed */
 307 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
 308 {
 309     BDRVQcowState *s = bs->opaque;
 310     QCowSnapshot *new_snapshot_list = NULL;
 311     QCowSnapshot *old_snapshot_list = NULL;
 312     QCowSnapshot sn1, *sn = &sn1;
 313     int i, ret;
 314     uint64_t *l1_table = NULL;
 315     int64_t l1_table_offset;
 316
 317     memset(sn, 0, sizeof(*sn));
 318
 319     /* Generate an ID if it wasn't passed */
 320     if (sn_info->id_str[0] == '\0') {
 321         find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
 322     }
 323
 324     /* Check that the ID is unique */
 325     if (find_snapshot_by_id(bs, sn_info->id_str) >= 0) {
 326         return -ENOENT;
 327     }
 328
 329     /* Populate sn with passed data */
 330     sn->id_str = g_strdup(sn_info->id_str);
 331     sn->name = g_strdup(sn_info->name);
 332
 333     sn->vm_state_size = sn_info->vm_state_size;
 334     sn->date_sec = sn_info->date_sec;
 335     sn->date_nsec = sn_info->date_nsec;
 336     sn->vm_clock_nsec = sn_info->vm_clock_nsec;
 337
 338     /* Allocate the L1 table of the snapshot and copy the current one there. */
 339     l1_table_offset = qcow2_alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
 340     if (l1_table_offset < 0) {
 341         ret = l1_table_offset;
 342         goto fail;
 343     }
 344
 345     sn->l1_table_offset = l1_table_offset;
 346     sn->l1_size = s->l1_size;
 347
 348     l1_table = g_malloc(s->l1_size * sizeof(uint64_t));
 349     for(i = 0; i < s->l1_size; i++) {
 350         l1_table[i] = cpu_to_be64(s->l1_table[i]);
 351     }
 352
 353     ret = bdrv_pwrite(bs->file, sn->l1_table_offset, l1_table,
 354                       s->l1_size * sizeof(uint64_t));
 355     if (ret < 0) {
 356         goto fail;
 357     }
 358
 359     g_free(l1_table);
 360     l1_table = NULL;
 361
 362     /*
 363      * Increase the refcounts of all clusters and make sure everything is
 364      * stable on disk before updating the snapshot table to contain a pointer
 365      * to the new L1 table.
 366      */
 367     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
 368     if (ret < 0) {
 369         goto fail;
 370     }
 371
 372     ret = bdrv_flush(bs);
 373     if (ret < 0) {
 374         goto fail;
 375     }
 376
 377     /* Append the new snapshot to the snapshot list */
 378     new_snapshot_list = g_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
 379     if (s->snapshots) {
 380         memcpy(new_snapshot_list, s->snapshots,
 381                s->nb_snapshots * sizeof(QCowSnapshot));
 382         old_snapshot_list = s->snapshots;
 383     }
 384     s->snapshots = new_snapshot_list;
 385     s->snapshots[s->nb_snapshots++] = *sn;
 386
 387     ret = qcow2_write_snapshots(bs);
 388     if (ret < 0) {
 389         g_free(s->snapshots);
 390         s->snapshots = old_snapshot_list;
 391         goto fail;
 392     }
 393
 394     g_free(old_snapshot_list);
 395
 396 #ifdef DEBUG_ALLOC
 397     {
 398       BdrvCheckResult result = {0};
 399       qcow2_check_refcounts(bs, &result);
 400     }
 401 #endif
 402     return 0;
 403
 404 fail:
 405     g_free(sn->id_str);
 406     g_free(sn->name);
 407     g_free(l1_table);
 408
 409     return ret;
 410 }
 411
 412 /* copy the snapshot 'snapshot_name' into the current disk image */
 413 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
 414 {
 415     BDRVQcowState *s = bs->opaque;
 416     QCowSnapshot *sn;
 417     int i, snapshot_index;
 418     int cur_l1_bytes, sn_l1_bytes;
 419     int ret;
 420     uint64_t *sn_l1_table = NULL;
 421
 422     /* Search the snapshot */
 423     snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
 424     if (snapshot_index < 0) {
 425         return -ENOENT;
 426     }
 427     sn = &s->snapshots[snapshot_index];
 428
 429     /*
 430      * Make sure that the current L1 table is big enough to contain the whole
 431      * L1 table of the snapshot. If the snapshot L1 table is smaller, the
 432      * current one must be padded with zeros.
 433      */
 434     ret = qcow2_grow_l1_table(bs, sn->l1_size, true);
 435     if (ret < 0) {
 436         goto fail;
 437     }
 438
 439     cur_l1_bytes = s->l1_size * sizeof(uint64_t);
 440     sn_l1_bytes = sn->l1_size * sizeof(uint64_t);
 441
 442     /*
 443      * Copy the snapshot L1 table to the current L1 table.
 444      *
 445      * Before overwriting the old current L1 table on disk, make sure to
 446      * increase all refcounts for the clusters referenced by the new one.
 447      * Decrease the refcount referenced by the old one only when the L1
 448      * table is overwritten.
 449      */
 450     sn_l1_table = g_malloc0(cur_l1_bytes);
 451
 452     ret = bdrv_pread(bs->file, sn->l1_table_offset, sn_l1_table, sn_l1_bytes);
 453     if (ret < 0) {
 454         goto fail;
 455     }
 456
 457     ret = qcow2_update_snapshot_refcount(bs, sn->l1_table_offset,
 458                                          sn->l1_size, 1);
 459     if (ret < 0) {
 460         goto fail;
 461     }
 462
 463     ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset, sn_l1_table,
 464                            cur_l1_bytes);
 465     if (ret < 0) {
 466         goto fail;
 467     }
 468
 469     /*
 470      * Decrease refcount of clusters of current L1 table.
 471      *
 472      * At this point, the in-memory s->l1_table points to the old L1 table,
 473      * whereas on disk we already have the new one.
 474      *
 475      * qcow2_update_snapshot_refcount special cases the current L1 table to use
 476      * the in-memory data instead of really using the offset to load a new one,
 477      * which is why this works.
 478      */
 479     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset,
 480                                          s->l1_size, -1);
 481
 482     /*
 483      * Now update the in-memory L1 table to be in sync with the on-disk one. We
 484      * need to do this even if updating refcounts failed.
 485      */
 486     for(i = 0;i < s->l1_size; i++) {
 487         s->l1_table[i] = be64_to_cpu(sn_l1_table[i]);
 488     }
 489
 490     if (ret < 0) {
 491         goto fail;
 492     }
 493
 494     g_free(sn_l1_table);
 495     sn_l1_table = NULL;
 496
 497     /*
 498      * Update QCOW_OFLAG_COPIED in the active L1 table (it may have changed
 499      * when we decreased the refcount of the old snapshot.
 500      */
 501     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
 502     if (ret < 0) {
 503         goto fail;
 504     }
 505
 506 #ifdef DEBUG_ALLOC
 507     {
 508         BdrvCheckResult result = {0};
 509         qcow2_check_refcounts(bs, &result);
 510     }
 511 #endif
 512     return 0;
 513
 514 fail:
 515     g_free(sn_l1_table);
 516     return ret;
 517 }
 518
 519 int qcow2_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
 520 {
 521     BDRVQcowState *s = bs->opaque;
 522     QCowSnapshot sn;
 523     int snapshot_index, ret;
 524
 525     /* Search the snapshot */
 526     snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
 527     if (snapshot_index < 0) {
 528         return -ENOENT;
 529     }
 530     sn = s->snapshots[snapshot_index];
 531
 532     /* Remove it from the snapshot list */
 533     memmove(s->snapshots + snapshot_index,
 534             s->snapshots + snapshot_index + 1,
 535             (s->nb_snapshots - snapshot_index - 1) * sizeof(sn));
 536     s->nb_snapshots--;
 537     ret = qcow2_write_snapshots(bs);
 538     if (ret < 0) {
 539         return ret;
 540     }
 541
 542     /*
 543      * The snapshot is now unused, clean up. If we fail after this point, we
 544      * won't recover but just leak clusters.
 545      */
 546     g_free(sn.id_str);
 547     g_free(sn.name);
 548
 549     /*
 550      * Now decrease the refcounts of clusters referenced by the snapshot and
 551      * free the L1 table.
 552      */
 553     ret = qcow2_update_snapshot_refcount(bs, sn.l1_table_offset,
 554                                          sn.l1_size, -1);
 555     if (ret < 0) {
 556         return ret;
 557     }
 558     qcow2_free_clusters(bs, sn.l1_table_offset, sn.l1_size * sizeof(uint64_t));
 559
 560     /* must update the copied flag on the current cluster offsets */
 561     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
 562     if (ret < 0) {
 563         return ret;
 564     }
 565
 566 #ifdef DEBUG_ALLOC
 567     {
 568         BdrvCheckResult result = {0};
 569         qcow2_check_refcounts(bs, &result);
 570     }
 571 #endif
 572     return 0;
 573 }
 574
 575 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
 576 {
 577     BDRVQcowState *s = bs->opaque;
 578     QEMUSnapshotInfo *sn_tab, *sn_info;
 579     QCowSnapshot *sn;
 580     int i;
 581
 582     if (!s->nb_snapshots) {
 583         *psn_tab = NULL;
 584         return s->nb_snapshots;
 585     }
 586
 587     sn_tab = g_malloc0(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
 588     for(i = 0; i < s->nb_snapshots; i++) {
 589         sn_info = sn_tab + i;
 590         sn = s->snapshots + i;
 591         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
 592                 sn->id_str);
 593         pstrcpy(sn_info->name, sizeof(sn_info->name),
 594                 sn->name);
 595         sn_info->vm_state_size = sn->vm_state_size;
 596         sn_info->date_sec = sn->date_sec;
 597         sn_info->date_nsec = sn->date_nsec;
 598         sn_info->vm_clock_nsec = sn->vm_clock_nsec;
 599     }
 600     *psn_tab = sn_tab;
 601     return s->nb_snapshots;
 602 }
 603
 604 int qcow2_snapshot_load_tmp(BlockDriverState *bs, const char *snapshot_name)
 605 {
 606     int i, snapshot_index;
 607     BDRVQcowState *s = bs->opaque;
 608     QCowSnapshot *sn;
 609     uint64_t *new_l1_table;
 610     int new_l1_bytes;
 611     int ret;
 612
 613     assert(bs->read_only);
 614
 615     /* Search the snapshot */
 616     snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_name);
 617     if (snapshot_index < 0) {
 618         return -ENOENT;
 619     }
 620     sn = &s->snapshots[snapshot_index];
 621
 622     /* Allocate and read in the snapshot's L1 table */
 623     new_l1_bytes = s->l1_size * sizeof(uint64_t);
 624     new_l1_table = g_malloc0(align_offset(new_l1_bytes, 512));
 625
 626     ret = bdrv_pread(bs->file, sn->l1_table_offset, new_l1_table, new_l1_bytes);
 627     if (ret < 0) {
 628         g_free(new_l1_table);
 629         return ret;
 630     }
 631
 632     /* Switch the L1 table */
 633     g_free(s->l1_table);
 634
 635     s->l1_size = sn->l1_size;
 636     s->l1_table_offset = sn->l1_table_offset;
 637     s->l1_table = new_l1_table;
 638
 639     for(i = 0;i < s->l1_size; i++) {
 640         be64_to_cpus(&s->l1_table[i]);
 641     }
 642
 643     return 0;
 644 }