xref: /openbmc/qemu/block/qcow2-snapshot.c (revision 5242ef88)
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/osdep.h"
26 #include "sysemu/block-backend.h"
27 #include "qapi/error.h"
28 #include "qcow2.h"
29 #include "qemu/bswap.h"
30 #include "qemu/error-report.h"
31 #include "qemu/cutils.h"
32 #include "qemu/memalign.h"
33 
34 static void qcow2_free_single_snapshot(BlockDriverState *bs, int i)
35 {
36     BDRVQcow2State *s = bs->opaque;
37 
38     assert(i >= 0 && i < s->nb_snapshots);
39     g_free(s->snapshots[i].name);
40     g_free(s->snapshots[i].id_str);
41     g_free(s->snapshots[i].unknown_extra_data);
42     memset(&s->snapshots[i], 0, sizeof(s->snapshots[i]));
43 }
44 
45 void qcow2_free_snapshots(BlockDriverState *bs)
46 {
47     BDRVQcow2State *s = bs->opaque;
48     int i;
49 
50     for(i = 0; i < s->nb_snapshots; i++) {
51         qcow2_free_single_snapshot(bs, i);
52     }
53     g_free(s->snapshots);
54     s->snapshots = NULL;
55     s->nb_snapshots = 0;
56 }
57 
58 /*
59  * If @repair is true, try to repair a broken snapshot table instead
60  * of just returning an error:
61  *
62  * - If the snapshot table was too long, set *nb_clusters_reduced to
63  *   the number of snapshots removed off the end.
64  *   The caller will update the on-disk nb_snapshots accordingly;
65  *   this leaks clusters, but is safe.
66  *   (The on-disk information must be updated before
67  *   qcow2_check_refcounts(), because that function relies on
68  *   s->nb_snapshots to reflect the on-disk value.)
69  *
70  * - If there were snapshots with too much extra metadata, increment
71  *   *extra_data_dropped for each.
72  *   This requires the caller to eventually rewrite the whole snapshot
73  *   table, which requires cluster allocation.  Therefore, this should
74  *   be done only after qcow2_check_refcounts() made sure the refcount
75  *   structures are valid.
76  *   (In the meantime, the image is still valid because
77  *   qcow2_check_refcounts() does not do anything with snapshots'
78  *   extra data.)
79  */
80 static int qcow2_do_read_snapshots(BlockDriverState *bs, bool repair,
81                                    int *nb_clusters_reduced,
82                                    int *extra_data_dropped,
83                                    Error **errp)
84 {
85     BDRVQcow2State *s = bs->opaque;
86     QCowSnapshotHeader h;
87     QCowSnapshotExtraData extra;
88     QCowSnapshot *sn;
89     int i, id_str_size, name_size;
90     int64_t offset, pre_sn_offset;
91     uint64_t table_length = 0;
92     int ret;
93 
94     if (!s->nb_snapshots) {
95         s->snapshots = NULL;
96         s->snapshots_size = 0;
97         return 0;
98     }
99 
100     offset = s->snapshots_offset;
101     s->snapshots = g_new0(QCowSnapshot, s->nb_snapshots);
102 
103     for(i = 0; i < s->nb_snapshots; i++) {
104         bool truncate_unknown_extra_data = false;
105 
106         pre_sn_offset = offset;
107         table_length = ROUND_UP(table_length, 8);
108 
109         /* Read statically sized part of the snapshot header */
110         offset = ROUND_UP(offset, 8);
111         ret = bdrv_pread(bs->file, offset, &h, sizeof(h));
112         if (ret < 0) {
113             error_setg_errno(errp, -ret, "Failed to read snapshot table");
114             goto fail;
115         }
116 
117         offset += sizeof(h);
118         sn = s->snapshots + i;
119         sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
120         sn->l1_size = be32_to_cpu(h.l1_size);
121         sn->vm_state_size = be32_to_cpu(h.vm_state_size);
122         sn->date_sec = be32_to_cpu(h.date_sec);
123         sn->date_nsec = be32_to_cpu(h.date_nsec);
124         sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
125         sn->extra_data_size = be32_to_cpu(h.extra_data_size);
126 
127         id_str_size = be16_to_cpu(h.id_str_size);
128         name_size = be16_to_cpu(h.name_size);
129 
130         if (sn->extra_data_size > QCOW_MAX_SNAPSHOT_EXTRA_DATA) {
131             if (!repair) {
132                 ret = -EFBIG;
133                 error_setg(errp, "Too much extra metadata in snapshot table "
134                            "entry %i", i);
135                 error_append_hint(errp, "You can force-remove this extra "
136                                   "metadata with qemu-img check -r all\n");
137                 goto fail;
138             }
139 
140             fprintf(stderr, "Discarding too much extra metadata in snapshot "
141                     "table entry %i (%" PRIu32 " > %u)\n",
142                     i, sn->extra_data_size, QCOW_MAX_SNAPSHOT_EXTRA_DATA);
143 
144             (*extra_data_dropped)++;
145             truncate_unknown_extra_data = true;
146         }
147 
148         /* Read known extra data */
149         ret = bdrv_pread(bs->file, offset, &extra,
150                          MIN(sizeof(extra), sn->extra_data_size));
151         if (ret < 0) {
152             error_setg_errno(errp, -ret, "Failed to read snapshot table");
153             goto fail;
154         }
155         offset += MIN(sizeof(extra), sn->extra_data_size);
156 
157         if (sn->extra_data_size >= endof(QCowSnapshotExtraData,
158                                          vm_state_size_large)) {
159             sn->vm_state_size = be64_to_cpu(extra.vm_state_size_large);
160         }
161 
162         if (sn->extra_data_size >= endof(QCowSnapshotExtraData, disk_size)) {
163             sn->disk_size = be64_to_cpu(extra.disk_size);
164         } else {
165             sn->disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
166         }
167 
168         if (sn->extra_data_size >= endof(QCowSnapshotExtraData, icount)) {
169             sn->icount = be64_to_cpu(extra.icount);
170         } else {
171             sn->icount = -1ULL;
172         }
173 
174         if (sn->extra_data_size > sizeof(extra)) {
175             uint64_t extra_data_end;
176             size_t unknown_extra_data_size;
177 
178             extra_data_end = offset + sn->extra_data_size - sizeof(extra);
179 
180             if (truncate_unknown_extra_data) {
181                 sn->extra_data_size = QCOW_MAX_SNAPSHOT_EXTRA_DATA;
182             }
183 
184             /* Store unknown extra data */
185             unknown_extra_data_size = sn->extra_data_size - sizeof(extra);
186             sn->unknown_extra_data = g_malloc(unknown_extra_data_size);
187             ret = bdrv_pread(bs->file, offset, sn->unknown_extra_data,
188                              unknown_extra_data_size);
189             if (ret < 0) {
190                 error_setg_errno(errp, -ret,
191                                  "Failed to read snapshot table");
192                 goto fail;
193             }
194             offset = extra_data_end;
195         }
196 
197         /* Read snapshot ID */
198         sn->id_str = g_malloc(id_str_size + 1);
199         ret = bdrv_pread(bs->file, offset, sn->id_str, id_str_size);
200         if (ret < 0) {
201             error_setg_errno(errp, -ret, "Failed to read snapshot table");
202             goto fail;
203         }
204         offset += id_str_size;
205         sn->id_str[id_str_size] = '\0';
206 
207         /* Read snapshot name */
208         sn->name = g_malloc(name_size + 1);
209         ret = bdrv_pread(bs->file, offset, sn->name, name_size);
210         if (ret < 0) {
211             error_setg_errno(errp, -ret, "Failed to read snapshot table");
212             goto fail;
213         }
214         offset += name_size;
215         sn->name[name_size] = '\0';
216 
217         /* Note that the extra data may have been truncated */
218         table_length += sizeof(h) + sn->extra_data_size + id_str_size +
219                         name_size;
220         if (!repair) {
221             assert(table_length == offset - s->snapshots_offset);
222         }
223 
224         if (table_length > QCOW_MAX_SNAPSHOTS_SIZE ||
225             offset - s->snapshots_offset > INT_MAX)
226         {
227             if (!repair) {
228                 ret = -EFBIG;
229                 error_setg(errp, "Snapshot table is too big");
230                 error_append_hint(errp, "You can force-remove all %u "
231                                   "overhanging snapshots with qemu-img check "
232                                   "-r all\n", s->nb_snapshots - i);
233                 goto fail;
234             }
235 
236             fprintf(stderr, "Discarding %u overhanging snapshots (snapshot "
237                     "table is too big)\n", s->nb_snapshots - i);
238 
239             *nb_clusters_reduced += (s->nb_snapshots - i);
240 
241             /* Discard current snapshot also */
242             qcow2_free_single_snapshot(bs, i);
243 
244             /*
245              * This leaks all the rest of the snapshot table and the
246              * snapshots' clusters, but we run in check -r all mode,
247              * so qcow2_check_refcounts() will take care of it.
248              */
249             s->nb_snapshots = i;
250             offset = pre_sn_offset;
251             break;
252         }
253     }
254 
255     assert(offset - s->snapshots_offset <= INT_MAX);
256     s->snapshots_size = offset - s->snapshots_offset;
257     return 0;
258 
259 fail:
260     qcow2_free_snapshots(bs);
261     return ret;
262 }
263 
264 int qcow2_read_snapshots(BlockDriverState *bs, Error **errp)
265 {
266     return qcow2_do_read_snapshots(bs, false, NULL, NULL, errp);
267 }
268 
269 /* add at the end of the file a new list of snapshots */
270 int qcow2_write_snapshots(BlockDriverState *bs)
271 {
272     BDRVQcow2State *s = bs->opaque;
273     QCowSnapshot *sn;
274     QCowSnapshotHeader h;
275     QCowSnapshotExtraData extra;
276     int i, name_size, id_str_size, snapshots_size;
277     struct {
278         uint32_t nb_snapshots;
279         uint64_t snapshots_offset;
280     } QEMU_PACKED header_data;
281     int64_t offset, snapshots_offset = 0;
282     int ret;
283 
284     /* compute the size of the snapshots */
285     offset = 0;
286     for(i = 0; i < s->nb_snapshots; i++) {
287         sn = s->snapshots + i;
288         offset = ROUND_UP(offset, 8);
289         offset += sizeof(h);
290         offset += MAX(sizeof(extra), sn->extra_data_size);
291         offset += strlen(sn->id_str);
292         offset += strlen(sn->name);
293 
294         if (offset > QCOW_MAX_SNAPSHOTS_SIZE) {
295             ret = -EFBIG;
296             goto fail;
297         }
298     }
299 
300     assert(offset <= INT_MAX);
301     snapshots_size = offset;
302 
303     /* Allocate space for the new snapshot list */
304     snapshots_offset = qcow2_alloc_clusters(bs, snapshots_size);
305     offset = snapshots_offset;
306     if (offset < 0) {
307         ret = offset;
308         goto fail;
309     }
310     ret = bdrv_flush(bs);
311     if (ret < 0) {
312         goto fail;
313     }
314 
315     /* The snapshot list position has not yet been updated, so these clusters
316      * must indeed be completely free */
317     ret = qcow2_pre_write_overlap_check(bs, 0, offset, snapshots_size, false);
318     if (ret < 0) {
319         goto fail;
320     }
321 
322 
323     /* Write all snapshots to the new list */
324     for(i = 0; i < s->nb_snapshots; i++) {
325         sn = s->snapshots + i;
326         memset(&h, 0, sizeof(h));
327         h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
328         h.l1_size = cpu_to_be32(sn->l1_size);
329         /* If it doesn't fit in 32 bit, older implementations should treat it
330          * as a disk-only snapshot rather than truncate the VM state */
331         if (sn->vm_state_size <= 0xffffffff) {
332             h.vm_state_size = cpu_to_be32(sn->vm_state_size);
333         }
334         h.date_sec = cpu_to_be32(sn->date_sec);
335         h.date_nsec = cpu_to_be32(sn->date_nsec);
336         h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
337         h.extra_data_size = cpu_to_be32(MAX(sizeof(extra),
338                                             sn->extra_data_size));
339 
340         memset(&extra, 0, sizeof(extra));
341         extra.vm_state_size_large = cpu_to_be64(sn->vm_state_size);
342         extra.disk_size = cpu_to_be64(sn->disk_size);
343         extra.icount = cpu_to_be64(sn->icount);
344 
345         id_str_size = strlen(sn->id_str);
346         name_size = strlen(sn->name);
347         assert(id_str_size <= UINT16_MAX && name_size <= UINT16_MAX);
348         h.id_str_size = cpu_to_be16(id_str_size);
349         h.name_size = cpu_to_be16(name_size);
350         offset = ROUND_UP(offset, 8);
351 
352         ret = bdrv_pwrite(bs->file, offset, &h, sizeof(h));
353         if (ret < 0) {
354             goto fail;
355         }
356         offset += sizeof(h);
357 
358         ret = bdrv_pwrite(bs->file, offset, &extra, sizeof(extra));
359         if (ret < 0) {
360             goto fail;
361         }
362         offset += sizeof(extra);
363 
364         if (sn->extra_data_size > sizeof(extra)) {
365             size_t unknown_extra_data_size =
366                 sn->extra_data_size - sizeof(extra);
367 
368             /* qcow2_read_snapshots() ensures no unbounded allocation */
369             assert(unknown_extra_data_size <= BDRV_REQUEST_MAX_BYTES);
370             assert(sn->unknown_extra_data);
371 
372             ret = bdrv_pwrite(bs->file, offset, sn->unknown_extra_data,
373                               unknown_extra_data_size);
374             if (ret < 0) {
375                 goto fail;
376             }
377             offset += unknown_extra_data_size;
378         }
379 
380         ret = bdrv_pwrite(bs->file, offset, sn->id_str, id_str_size);
381         if (ret < 0) {
382             goto fail;
383         }
384         offset += id_str_size;
385 
386         ret = bdrv_pwrite(bs->file, offset, sn->name, name_size);
387         if (ret < 0) {
388             goto fail;
389         }
390         offset += name_size;
391     }
392 
393     /*
394      * Update the header to point to the new snapshot table. This requires the
395      * new table and its refcounts to be stable on disk.
396      */
397     ret = bdrv_flush(bs);
398     if (ret < 0) {
399         goto fail;
400     }
401 
402     QEMU_BUILD_BUG_ON(offsetof(QCowHeader, snapshots_offset) !=
403                       endof(QCowHeader, nb_snapshots));
404 
405     header_data.nb_snapshots        = cpu_to_be32(s->nb_snapshots);
406     header_data.snapshots_offset    = cpu_to_be64(snapshots_offset);
407 
408     ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, nb_snapshots),
409                            &header_data, sizeof(header_data));
410     if (ret < 0) {
411         goto fail;
412     }
413 
414     /* free the old snapshot table */
415     qcow2_free_clusters(bs, s->snapshots_offset, s->snapshots_size,
416                         QCOW2_DISCARD_SNAPSHOT);
417     s->snapshots_offset = snapshots_offset;
418     s->snapshots_size = snapshots_size;
419     return 0;
420 
421 fail:
422     if (snapshots_offset > 0) {
423         qcow2_free_clusters(bs, snapshots_offset, snapshots_size,
424                             QCOW2_DISCARD_ALWAYS);
425     }
426     return ret;
427 }
428 
429 int coroutine_fn qcow2_check_read_snapshot_table(BlockDriverState *bs,
430                                                  BdrvCheckResult *result,
431                                                  BdrvCheckMode fix)
432 {
433     BDRVQcow2State *s = bs->opaque;
434     Error *local_err = NULL;
435     int nb_clusters_reduced = 0;
436     int extra_data_dropped = 0;
437     int ret;
438     struct {
439         uint32_t nb_snapshots;
440         uint64_t snapshots_offset;
441     } QEMU_PACKED snapshot_table_pointer;
442 
443     /* qcow2_do_open() discards this information in check mode */
444     ret = bdrv_pread(bs->file, offsetof(QCowHeader, nb_snapshots),
445                      &snapshot_table_pointer, sizeof(snapshot_table_pointer));
446     if (ret < 0) {
447         result->check_errors++;
448         fprintf(stderr, "ERROR failed to read the snapshot table pointer from "
449                 "the image header: %s\n", strerror(-ret));
450         return ret;
451     }
452 
453     s->snapshots_offset = be64_to_cpu(snapshot_table_pointer.snapshots_offset);
454     s->nb_snapshots = be32_to_cpu(snapshot_table_pointer.nb_snapshots);
455 
456     if (s->nb_snapshots > QCOW_MAX_SNAPSHOTS && (fix & BDRV_FIX_ERRORS)) {
457         fprintf(stderr, "Discarding %u overhanging snapshots\n",
458                 s->nb_snapshots - QCOW_MAX_SNAPSHOTS);
459 
460         nb_clusters_reduced += s->nb_snapshots - QCOW_MAX_SNAPSHOTS;
461         s->nb_snapshots = QCOW_MAX_SNAPSHOTS;
462     }
463 
464     ret = qcow2_validate_table(bs, s->snapshots_offset, s->nb_snapshots,
465                                sizeof(QCowSnapshotHeader),
466                                sizeof(QCowSnapshotHeader) * QCOW_MAX_SNAPSHOTS,
467                                "snapshot table", &local_err);
468     if (ret < 0) {
469         result->check_errors++;
470         error_reportf_err(local_err, "ERROR ");
471 
472         if (s->nb_snapshots > QCOW_MAX_SNAPSHOTS) {
473             fprintf(stderr, "You can force-remove all %u overhanging snapshots "
474                     "with qemu-img check -r all\n",
475                     s->nb_snapshots - QCOW_MAX_SNAPSHOTS);
476         }
477 
478         /* We did not read the snapshot table, so invalidate this information */
479         s->snapshots_offset = 0;
480         s->nb_snapshots = 0;
481 
482         return ret;
483     }
484 
485     qemu_co_mutex_unlock(&s->lock);
486     ret = qcow2_do_read_snapshots(bs, fix & BDRV_FIX_ERRORS,
487                                   &nb_clusters_reduced, &extra_data_dropped,
488                                   &local_err);
489     qemu_co_mutex_lock(&s->lock);
490     if (ret < 0) {
491         result->check_errors++;
492         error_reportf_err(local_err,
493                           "ERROR failed to read the snapshot table: ");
494 
495         /* We did not read the snapshot table, so invalidate this information */
496         s->snapshots_offset = 0;
497         s->nb_snapshots = 0;
498 
499         return ret;
500     }
501     result->corruptions += nb_clusters_reduced + extra_data_dropped;
502 
503     if (nb_clusters_reduced) {
504         /*
505          * Update image header now, because:
506          * (1) qcow2_check_refcounts() relies on s->nb_snapshots to be
507          *     the same as what the image header says,
508          * (2) this leaks clusters, but qcow2_check_refcounts() will
509          *     fix that.
510          */
511         assert(fix & BDRV_FIX_ERRORS);
512 
513         snapshot_table_pointer.nb_snapshots = cpu_to_be32(s->nb_snapshots);
514         ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, nb_snapshots),
515                                &snapshot_table_pointer.nb_snapshots,
516                                sizeof(snapshot_table_pointer.nb_snapshots));
517         if (ret < 0) {
518             result->check_errors++;
519             fprintf(stderr, "ERROR failed to update the snapshot count in the "
520                     "image header: %s\n", strerror(-ret));
521             return ret;
522         }
523 
524         result->corruptions_fixed += nb_clusters_reduced;
525         result->corruptions -= nb_clusters_reduced;
526     }
527 
528     /*
529      * All of v3 images' snapshot table entries need to have at least
530      * 16 bytes of extra data.
531      */
532     if (s->qcow_version >= 3) {
533         int i;
534         for (i = 0; i < s->nb_snapshots; i++) {
535             if (s->snapshots[i].extra_data_size <
536                 sizeof_field(QCowSnapshotExtraData, vm_state_size_large) +
537                 sizeof_field(QCowSnapshotExtraData, disk_size))
538             {
539                 result->corruptions++;
540                 fprintf(stderr, "%s snapshot table entry %i is incomplete\n",
541                         fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i);
542             }
543         }
544     }
545 
546     return 0;
547 }
548 
549 int coroutine_fn qcow2_check_fix_snapshot_table(BlockDriverState *bs,
550                                                 BdrvCheckResult *result,
551                                                 BdrvCheckMode fix)
552 {
553     BDRVQcow2State *s = bs->opaque;
554     int ret;
555 
556     if (result->corruptions && (fix & BDRV_FIX_ERRORS)) {
557         qemu_co_mutex_unlock(&s->lock);
558         ret = qcow2_write_snapshots(bs);
559         qemu_co_mutex_lock(&s->lock);
560         if (ret < 0) {
561             result->check_errors++;
562             fprintf(stderr, "ERROR failed to update snapshot table: %s\n",
563                     strerror(-ret));
564             return ret;
565         }
566 
567         result->corruptions_fixed += result->corruptions;
568         result->corruptions = 0;
569     }
570 
571     return 0;
572 }
573 
574 static void find_new_snapshot_id(BlockDriverState *bs,
575                                  char *id_str, int id_str_size)
576 {
577     BDRVQcow2State *s = bs->opaque;
578     QCowSnapshot *sn;
579     int i;
580     unsigned long id, id_max = 0;
581 
582     for(i = 0; i < s->nb_snapshots; i++) {
583         sn = s->snapshots + i;
584         id = strtoul(sn->id_str, NULL, 10);
585         if (id > id_max)
586             id_max = id;
587     }
588     snprintf(id_str, id_str_size, "%lu", id_max + 1);
589 }
590 
591 static int find_snapshot_by_id_and_name(BlockDriverState *bs,
592                                         const char *id,
593                                         const char *name)
594 {
595     BDRVQcow2State *s = bs->opaque;
596     int i;
597 
598     if (id && name) {
599         for (i = 0; i < s->nb_snapshots; i++) {
600             if (!strcmp(s->snapshots[i].id_str, id) &&
601                 !strcmp(s->snapshots[i].name, name)) {
602                 return i;
603             }
604         }
605     } else if (id) {
606         for (i = 0; i < s->nb_snapshots; i++) {
607             if (!strcmp(s->snapshots[i].id_str, id)) {
608                 return i;
609             }
610         }
611     } else if (name) {
612         for (i = 0; i < s->nb_snapshots; i++) {
613             if (!strcmp(s->snapshots[i].name, name)) {
614                 return i;
615             }
616         }
617     }
618 
619     return -1;
620 }
621 
622 static int find_snapshot_by_id_or_name(BlockDriverState *bs,
623                                        const char *id_or_name)
624 {
625     int ret;
626 
627     ret = find_snapshot_by_id_and_name(bs, id_or_name, NULL);
628     if (ret >= 0) {
629         return ret;
630     }
631     return find_snapshot_by_id_and_name(bs, NULL, id_or_name);
632 }
633 
634 /* if no id is provided, a new one is constructed */
635 int qcow2_snapshot_create(BlockDriverState *bs, QEMUSnapshotInfo *sn_info)
636 {
637     BDRVQcow2State *s = bs->opaque;
638     QCowSnapshot *new_snapshot_list = NULL;
639     QCowSnapshot *old_snapshot_list = NULL;
640     QCowSnapshot sn1, *sn = &sn1;
641     int i, ret;
642     uint64_t *l1_table = NULL;
643     int64_t l1_table_offset;
644 
645     if (s->nb_snapshots >= QCOW_MAX_SNAPSHOTS) {
646         return -EFBIG;
647     }
648 
649     if (has_data_file(bs)) {
650         return -ENOTSUP;
651     }
652 
653     memset(sn, 0, sizeof(*sn));
654 
655     /* Generate an ID */
656     find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
657 
658     /* Populate sn with passed data */
659     sn->id_str = g_strdup(sn_info->id_str);
660     sn->name = g_strdup(sn_info->name);
661 
662     sn->disk_size = bs->total_sectors * BDRV_SECTOR_SIZE;
663     sn->vm_state_size = sn_info->vm_state_size;
664     sn->date_sec = sn_info->date_sec;
665     sn->date_nsec = sn_info->date_nsec;
666     sn->vm_clock_nsec = sn_info->vm_clock_nsec;
667     sn->icount = sn_info->icount;
668     sn->extra_data_size = sizeof(QCowSnapshotExtraData);
669 
670     /* Allocate the L1 table of the snapshot and copy the current one there. */
671     l1_table_offset = qcow2_alloc_clusters(bs, s->l1_size * L1E_SIZE);
672     if (l1_table_offset < 0) {
673         ret = l1_table_offset;
674         goto fail;
675     }
676 
677     sn->l1_table_offset = l1_table_offset;
678     sn->l1_size = s->l1_size;
679 
680     l1_table = g_try_new(uint64_t, s->l1_size);
681     if (s->l1_size && l1_table == NULL) {
682         ret = -ENOMEM;
683         goto fail;
684     }
685 
686     for(i = 0; i < s->l1_size; i++) {
687         l1_table[i] = cpu_to_be64(s->l1_table[i]);
688     }
689 
690     ret = qcow2_pre_write_overlap_check(bs, 0, sn->l1_table_offset,
691                                         s->l1_size * L1E_SIZE, false);
692     if (ret < 0) {
693         goto fail;
694     }
695 
696     ret = bdrv_pwrite(bs->file, sn->l1_table_offset, l1_table,
697                       s->l1_size * L1E_SIZE);
698     if (ret < 0) {
699         goto fail;
700     }
701 
702     g_free(l1_table);
703     l1_table = NULL;
704 
705     /*
706      * Increase the refcounts of all clusters and make sure everything is
707      * stable on disk before updating the snapshot table to contain a pointer
708      * to the new L1 table.
709      */
710     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
711     if (ret < 0) {
712         goto fail;
713     }
714 
715     /* Append the new snapshot to the snapshot list */
716     new_snapshot_list = g_new(QCowSnapshot, s->nb_snapshots + 1);
717     if (s->snapshots) {
718         memcpy(new_snapshot_list, s->snapshots,
719                s->nb_snapshots * sizeof(QCowSnapshot));
720         old_snapshot_list = s->snapshots;
721     }
722     s->snapshots = new_snapshot_list;
723     s->snapshots[s->nb_snapshots++] = *sn;
724 
725     ret = qcow2_write_snapshots(bs);
726     if (ret < 0) {
727         g_free(s->snapshots);
728         s->snapshots = old_snapshot_list;
729         s->nb_snapshots--;
730         goto fail;
731     }
732 
733     g_free(old_snapshot_list);
734 
735     /* The VM state isn't needed any more in the active L1 table; in fact, it
736      * hurts by causing expensive COW for the next snapshot. */
737     qcow2_cluster_discard(bs, qcow2_vm_state_offset(s),
738                           ROUND_UP(sn->vm_state_size, s->cluster_size),
739                           QCOW2_DISCARD_NEVER, false);
740 
741 #ifdef DEBUG_ALLOC
742     {
743       BdrvCheckResult result = {0};
744       qcow2_check_refcounts(bs, &result, 0);
745     }
746 #endif
747     return 0;
748 
749 fail:
750     g_free(sn->id_str);
751     g_free(sn->name);
752     g_free(l1_table);
753 
754     return ret;
755 }
756 
757 /* copy the snapshot 'snapshot_name' into the current disk image */
758 int qcow2_snapshot_goto(BlockDriverState *bs, const char *snapshot_id)
759 {
760     BDRVQcow2State *s = bs->opaque;
761     QCowSnapshot *sn;
762     Error *local_err = NULL;
763     int i, snapshot_index;
764     int cur_l1_bytes, sn_l1_bytes;
765     int ret;
766     uint64_t *sn_l1_table = NULL;
767 
768     if (has_data_file(bs)) {
769         return -ENOTSUP;
770     }
771 
772     /* Search the snapshot */
773     snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
774     if (snapshot_index < 0) {
775         return -ENOENT;
776     }
777     sn = &s->snapshots[snapshot_index];
778 
779     ret = qcow2_validate_table(bs, sn->l1_table_offset, sn->l1_size,
780                                L1E_SIZE, QCOW_MAX_L1_SIZE,
781                                "Snapshot L1 table", &local_err);
782     if (ret < 0) {
783         error_report_err(local_err);
784         goto fail;
785     }
786 
787     if (sn->disk_size != bs->total_sectors * BDRV_SECTOR_SIZE) {
788         BlockBackend *blk = blk_new_with_bs(bs, BLK_PERM_RESIZE, BLK_PERM_ALL,
789                                             &local_err);
790         if (!blk) {
791             error_report_err(local_err);
792             ret = -ENOTSUP;
793             goto fail;
794         }
795 
796         ret = blk_truncate(blk, sn->disk_size, true, PREALLOC_MODE_OFF, 0,
797                            &local_err);
798         blk_unref(blk);
799         if (ret < 0) {
800             error_report_err(local_err);
801             goto fail;
802         }
803     }
804 
805     /*
806      * Make sure that the current L1 table is big enough to contain the whole
807      * L1 table of the snapshot. If the snapshot L1 table is smaller, the
808      * current one must be padded with zeros.
809      */
810     ret = qcow2_grow_l1_table(bs, sn->l1_size, true);
811     if (ret < 0) {
812         goto fail;
813     }
814 
815     cur_l1_bytes = s->l1_size * L1E_SIZE;
816     sn_l1_bytes = sn->l1_size * L1E_SIZE;
817 
818     /*
819      * Copy the snapshot L1 table to the current L1 table.
820      *
821      * Before overwriting the old current L1 table on disk, make sure to
822      * increase all refcounts for the clusters referenced by the new one.
823      * Decrease the refcount referenced by the old one only when the L1
824      * table is overwritten.
825      */
826     sn_l1_table = g_try_malloc0(cur_l1_bytes);
827     if (cur_l1_bytes && sn_l1_table == NULL) {
828         ret = -ENOMEM;
829         goto fail;
830     }
831 
832     ret = bdrv_pread(bs->file, sn->l1_table_offset,
833                      sn_l1_table, sn_l1_bytes);
834     if (ret < 0) {
835         goto fail;
836     }
837 
838     ret = qcow2_update_snapshot_refcount(bs, sn->l1_table_offset,
839                                          sn->l1_size, 1);
840     if (ret < 0) {
841         goto fail;
842     }
843 
844     ret = qcow2_pre_write_overlap_check(bs, QCOW2_OL_ACTIVE_L1,
845                                         s->l1_table_offset, cur_l1_bytes,
846                                         false);
847     if (ret < 0) {
848         goto fail;
849     }
850 
851     ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset, sn_l1_table,
852                            cur_l1_bytes);
853     if (ret < 0) {
854         goto fail;
855     }
856 
857     /*
858      * Decrease refcount of clusters of current L1 table.
859      *
860      * At this point, the in-memory s->l1_table points to the old L1 table,
861      * whereas on disk we already have the new one.
862      *
863      * qcow2_update_snapshot_refcount special cases the current L1 table to use
864      * the in-memory data instead of really using the offset to load a new one,
865      * which is why this works.
866      */
867     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset,
868                                          s->l1_size, -1);
869 
870     /*
871      * Now update the in-memory L1 table to be in sync with the on-disk one. We
872      * need to do this even if updating refcounts failed.
873      */
874     for(i = 0;i < s->l1_size; i++) {
875         s->l1_table[i] = be64_to_cpu(sn_l1_table[i]);
876     }
877 
878     if (ret < 0) {
879         goto fail;
880     }
881 
882     g_free(sn_l1_table);
883     sn_l1_table = NULL;
884 
885     /*
886      * Update QCOW_OFLAG_COPIED in the active L1 table (it may have changed
887      * when we decreased the refcount of the old snapshot.
888      */
889     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
890     if (ret < 0) {
891         goto fail;
892     }
893 
894 #ifdef DEBUG_ALLOC
895     {
896         BdrvCheckResult result = {0};
897         qcow2_check_refcounts(bs, &result, 0);
898     }
899 #endif
900     return 0;
901 
902 fail:
903     g_free(sn_l1_table);
904     return ret;
905 }
906 
907 int qcow2_snapshot_delete(BlockDriverState *bs,
908                           const char *snapshot_id,
909                           const char *name,
910                           Error **errp)
911 {
912     BDRVQcow2State *s = bs->opaque;
913     QCowSnapshot sn;
914     int snapshot_index, ret;
915 
916     if (has_data_file(bs)) {
917         return -ENOTSUP;
918     }
919 
920     /* Search the snapshot */
921     snapshot_index = find_snapshot_by_id_and_name(bs, snapshot_id, name);
922     if (snapshot_index < 0) {
923         error_setg(errp, "Can't find the snapshot");
924         return -ENOENT;
925     }
926     sn = s->snapshots[snapshot_index];
927 
928     ret = qcow2_validate_table(bs, sn.l1_table_offset, sn.l1_size,
929                                L1E_SIZE, QCOW_MAX_L1_SIZE,
930                                "Snapshot L1 table", errp);
931     if (ret < 0) {
932         return ret;
933     }
934 
935     /* Remove it from the snapshot list */
936     memmove(s->snapshots + snapshot_index,
937             s->snapshots + snapshot_index + 1,
938             (s->nb_snapshots - snapshot_index - 1) * sizeof(sn));
939     s->nb_snapshots--;
940     ret = qcow2_write_snapshots(bs);
941     if (ret < 0) {
942         error_setg_errno(errp, -ret,
943                          "Failed to remove snapshot from snapshot list");
944         return ret;
945     }
946 
947     /*
948      * The snapshot is now unused, clean up. If we fail after this point, we
949      * won't recover but just leak clusters.
950      */
951     g_free(sn.unknown_extra_data);
952     g_free(sn.id_str);
953     g_free(sn.name);
954 
955     /*
956      * Now decrease the refcounts of clusters referenced by the snapshot and
957      * free the L1 table.
958      */
959     ret = qcow2_update_snapshot_refcount(bs, sn.l1_table_offset,
960                                          sn.l1_size, -1);
961     if (ret < 0) {
962         error_setg_errno(errp, -ret, "Failed to free the cluster and L1 table");
963         return ret;
964     }
965     qcow2_free_clusters(bs, sn.l1_table_offset, sn.l1_size * L1E_SIZE,
966                         QCOW2_DISCARD_SNAPSHOT);
967 
968     /* must update the copied flag on the current cluster offsets */
969     ret = qcow2_update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
970     if (ret < 0) {
971         error_setg_errno(errp, -ret,
972                          "Failed to update snapshot status in disk");
973         return ret;
974     }
975 
976 #ifdef DEBUG_ALLOC
977     {
978         BdrvCheckResult result = {0};
979         qcow2_check_refcounts(bs, &result, 0);
980     }
981 #endif
982     return 0;
983 }
984 
985 int qcow2_snapshot_list(BlockDriverState *bs, QEMUSnapshotInfo **psn_tab)
986 {
987     BDRVQcow2State *s = bs->opaque;
988     QEMUSnapshotInfo *sn_tab, *sn_info;
989     QCowSnapshot *sn;
990     int i;
991 
992     if (has_data_file(bs)) {
993         return -ENOTSUP;
994     }
995     if (!s->nb_snapshots) {
996         *psn_tab = NULL;
997         return s->nb_snapshots;
998     }
999 
1000     sn_tab = g_new0(QEMUSnapshotInfo, s->nb_snapshots);
1001     for(i = 0; i < s->nb_snapshots; i++) {
1002         sn_info = sn_tab + i;
1003         sn = s->snapshots + i;
1004         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
1005                 sn->id_str);
1006         pstrcpy(sn_info->name, sizeof(sn_info->name),
1007                 sn->name);
1008         sn_info->vm_state_size = sn->vm_state_size;
1009         sn_info->date_sec = sn->date_sec;
1010         sn_info->date_nsec = sn->date_nsec;
1011         sn_info->vm_clock_nsec = sn->vm_clock_nsec;
1012         sn_info->icount = sn->icount;
1013     }
1014     *psn_tab = sn_tab;
1015     return s->nb_snapshots;
1016 }
1017 
1018 int qcow2_snapshot_load_tmp(BlockDriverState *bs,
1019                             const char *snapshot_id,
1020                             const char *name,
1021                             Error **errp)
1022 {
1023     int i, snapshot_index;
1024     BDRVQcow2State *s = bs->opaque;
1025     QCowSnapshot *sn;
1026     uint64_t *new_l1_table;
1027     int new_l1_bytes;
1028     int ret;
1029 
1030     assert(bdrv_is_read_only(bs));
1031 
1032     /* Search the snapshot */
1033     snapshot_index = find_snapshot_by_id_and_name(bs, snapshot_id, name);
1034     if (snapshot_index < 0) {
1035         error_setg(errp,
1036                    "Can't find snapshot");
1037         return -ENOENT;
1038     }
1039     sn = &s->snapshots[snapshot_index];
1040 
1041     /* Allocate and read in the snapshot's L1 table */
1042     ret = qcow2_validate_table(bs, sn->l1_table_offset, sn->l1_size,
1043                                L1E_SIZE, QCOW_MAX_L1_SIZE,
1044                                "Snapshot L1 table", errp);
1045     if (ret < 0) {
1046         return ret;
1047     }
1048     new_l1_bytes = sn->l1_size * L1E_SIZE;
1049     new_l1_table = qemu_try_blockalign(bs->file->bs, new_l1_bytes);
1050     if (new_l1_table == NULL) {
1051         return -ENOMEM;
1052     }
1053 
1054     ret = bdrv_pread(bs->file, sn->l1_table_offset,
1055                      new_l1_table, new_l1_bytes);
1056     if (ret < 0) {
1057         error_setg(errp, "Failed to read l1 table for snapshot");
1058         qemu_vfree(new_l1_table);
1059         return ret;
1060     }
1061 
1062     /* Switch the L1 table */
1063     qemu_vfree(s->l1_table);
1064 
1065     s->l1_size = sn->l1_size;
1066     s->l1_table_offset = sn->l1_table_offset;
1067     s->l1_table = new_l1_table;
1068 
1069     for(i = 0;i < s->l1_size; i++) {
1070         be64_to_cpus(&s->l1_table[i]);
1071     }
1072 
1073     return 0;
1074 }
1075