• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #include "qemu-common.h"
25 #include "block_int.h"
26 #include <zlib.h>
27 #include "aes.h"
28 #include <assert.h>
29 
30 /*
31   Differences with QCOW:
32 
33   - Support for multiple incremental snapshots.
34   - Memory management by reference counts.
35   - Clusters which have a reference count of one have the bit
36     QCOW_OFLAG_COPIED to optimize write performance.
37   - Size of compressed clusters is stored in sectors to reduce bit usage
38     in the cluster offsets.
39   - Support for storing additional data (such as the VM state) in the
40     snapshots.
41   - If a backing store is used, the cluster size is not constrained
42     (could be backported to QCOW).
43   - L2 tables have always a size of one cluster.
44 */
45 
46 //#define DEBUG_ALLOC
47 //#define DEBUG_ALLOC2
48 
49 #define QCOW_MAGIC (('Q' << 24) | ('F' << 16) | ('I' << 8) | 0xfb)
50 #define QCOW_VERSION 2
51 
52 #define QCOW_CRYPT_NONE 0
53 #define QCOW_CRYPT_AES  1
54 
55 #define QCOW_MAX_CRYPT_CLUSTERS 32
56 
57 /* indicate that the refcount of the referenced cluster is exactly one. */
58 #define QCOW_OFLAG_COPIED     (1LL << 63)
59 /* indicate that the cluster is compressed (they never have the copied flag) */
60 #define QCOW_OFLAG_COMPRESSED (1LL << 62)
61 
62 #define REFCOUNT_SHIFT 1 /* refcount size is 2 bytes */
63 
64 typedef struct QCowHeader {
65     uint32_t magic;
66     uint32_t version;
67     uint64_t backing_file_offset;
68     uint32_t backing_file_size;
69     uint32_t cluster_bits;
70     uint64_t size; /* in bytes */
71     uint32_t crypt_method;
72     uint32_t l1_size; /* XXX: save number of clusters instead ? */
73     uint64_t l1_table_offset;
74     uint64_t refcount_table_offset;
75     uint32_t refcount_table_clusters;
76     uint32_t nb_snapshots;
77     uint64_t snapshots_offset;
78 } QCowHeader;
79 
80 typedef struct __attribute__((packed)) QCowSnapshotHeader {
81     /* header is 8 byte aligned */
82     uint64_t l1_table_offset;
83 
84     uint32_t l1_size;
85     uint16_t id_str_size;
86     uint16_t name_size;
87 
88     uint32_t date_sec;
89     uint32_t date_nsec;
90 
91     uint64_t vm_clock_nsec;
92 
93     uint32_t vm_state_size;
94     uint32_t extra_data_size; /* for extension */
95     /* extra data follows */
96     /* id_str follows */
97     /* name follows  */
98 } QCowSnapshotHeader;
99 
100 #define L2_CACHE_SIZE 16
101 
102 typedef struct QCowSnapshot {
103     uint64_t l1_table_offset;
104     uint32_t l1_size;
105     char *id_str;
106     char *name;
107     uint32_t vm_state_size;
108     uint32_t date_sec;
109     uint32_t date_nsec;
110     uint64_t vm_clock_nsec;
111 } QCowSnapshot;
112 
113 typedef struct BDRVQcowState {
114     BlockDriverState *hd;
115     int cluster_bits;
116     int cluster_size;
117     int cluster_sectors;
118     int l2_bits;
119     int l2_size;
120     int l1_size;
121     int l1_vm_state_index;
122     int csize_shift;
123     int csize_mask;
124     uint64_t cluster_offset_mask;
125     uint64_t l1_table_offset;
126     uint64_t *l1_table;
127     uint64_t *l2_cache;
128     uint64_t l2_cache_offsets[L2_CACHE_SIZE];
129     uint32_t l2_cache_counts[L2_CACHE_SIZE];
130     uint8_t *cluster_cache;
131     uint8_t *cluster_data;
132     uint64_t cluster_cache_offset;
133 
134     uint64_t *refcount_table;
135     uint64_t refcount_table_offset;
136     uint32_t refcount_table_size;
137     uint64_t refcount_block_cache_offset;
138     uint16_t *refcount_block_cache;
139     int64_t free_cluster_index;
140     int64_t free_byte_offset;
141 
142     uint32_t crypt_method; /* current crypt method, 0 if no key yet */
143     uint32_t crypt_method_header;
144     AES_KEY aes_encrypt_key;
145     AES_KEY aes_decrypt_key;
146     uint64_t snapshots_offset;
147     int snapshots_size;
148     int nb_snapshots;
149     QCowSnapshot *snapshots;
150 } BDRVQcowState;
151 
152 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset);
153 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
154                      uint8_t *buf, int nb_sectors);
155 static int qcow_read_snapshots(BlockDriverState *bs);
156 static void qcow_free_snapshots(BlockDriverState *bs);
157 static int refcount_init(BlockDriverState *bs);
158 static void refcount_close(BlockDriverState *bs);
159 static int get_refcount(BlockDriverState *bs, int64_t cluster_index);
160 static int update_cluster_refcount(BlockDriverState *bs,
161                                    int64_t cluster_index,
162                                    int addend);
163 static void update_refcount(BlockDriverState *bs,
164                             int64_t offset, int64_t length,
165                             int addend);
166 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size);
167 static int64_t alloc_bytes(BlockDriverState *bs, int size);
168 static void free_clusters(BlockDriverState *bs,
169                           int64_t offset, int64_t size);
170 #ifdef DEBUG_ALLOC
171 static void check_refcounts(BlockDriverState *bs);
172 #endif
173 
qcow_probe(const uint8_t * buf,int buf_size,const char * filename)174 static int qcow_probe(const uint8_t *buf, int buf_size, const char *filename)
175 {
176     const QCowHeader *cow_header = (const void *)buf;
177 
178     if (buf_size >= sizeof(QCowHeader) &&
179         be32_to_cpu(cow_header->magic) == QCOW_MAGIC &&
180         be32_to_cpu(cow_header->version) == QCOW_VERSION)
181         return 100;
182     else
183         return 0;
184 }
185 
qcow_open(BlockDriverState * bs,const char * filename,int flags)186 static int qcow_open(BlockDriverState *bs, const char *filename, int flags)
187 {
188     BDRVQcowState *s = bs->opaque;
189     int len, i, shift, ret;
190     QCowHeader header;
191 
192     ret = bdrv_file_open(&s->hd, filename, flags);
193     if (ret < 0)
194         return ret;
195     if (bdrv_pread(s->hd, 0, &header, sizeof(header)) != sizeof(header))
196         goto fail;
197     be32_to_cpus(&header.magic);
198     be32_to_cpus(&header.version);
199     be64_to_cpus(&header.backing_file_offset);
200     be32_to_cpus(&header.backing_file_size);
201     be64_to_cpus(&header.size);
202     be32_to_cpus(&header.cluster_bits);
203     be32_to_cpus(&header.crypt_method);
204     be64_to_cpus(&header.l1_table_offset);
205     be32_to_cpus(&header.l1_size);
206     be64_to_cpus(&header.refcount_table_offset);
207     be32_to_cpus(&header.refcount_table_clusters);
208     be64_to_cpus(&header.snapshots_offset);
209     be32_to_cpus(&header.nb_snapshots);
210 
211     if (header.magic != QCOW_MAGIC || header.version != QCOW_VERSION)
212         goto fail;
213     if (header.size <= 1 ||
214         header.cluster_bits < 9 ||
215         header.cluster_bits > 16)
216         goto fail;
217     if (header.crypt_method > QCOW_CRYPT_AES)
218         goto fail;
219     s->crypt_method_header = header.crypt_method;
220     if (s->crypt_method_header)
221         bs->encrypted = 1;
222     s->cluster_bits = header.cluster_bits;
223     s->cluster_size = 1 << s->cluster_bits;
224     s->cluster_sectors = 1 << (s->cluster_bits - 9);
225     s->l2_bits = s->cluster_bits - 3; /* L2 is always one cluster */
226     s->l2_size = 1 << s->l2_bits;
227     bs->total_sectors = header.size / 512;
228     s->csize_shift = (62 - (s->cluster_bits - 8));
229     s->csize_mask = (1 << (s->cluster_bits - 8)) - 1;
230     s->cluster_offset_mask = (1LL << s->csize_shift) - 1;
231     s->refcount_table_offset = header.refcount_table_offset;
232     s->refcount_table_size =
233         header.refcount_table_clusters << (s->cluster_bits - 3);
234 
235     s->snapshots_offset = header.snapshots_offset;
236     s->nb_snapshots = header.nb_snapshots;
237 
238     /* read the level 1 table */
239     s->l1_size = header.l1_size;
240     shift = s->cluster_bits + s->l2_bits;
241     s->l1_vm_state_index = (header.size + (1LL << shift) - 1) >> shift;
242     /* the L1 table must contain at least enough entries to put
243        header.size bytes */
244     if (s->l1_size < s->l1_vm_state_index)
245         goto fail;
246     s->l1_table_offset = header.l1_table_offset;
247     s->l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
248     if (!s->l1_table)
249         goto fail;
250     if (bdrv_pread(s->hd, s->l1_table_offset, s->l1_table, s->l1_size * sizeof(uint64_t)) !=
251         s->l1_size * sizeof(uint64_t))
252         goto fail;
253     for(i = 0;i < s->l1_size; i++) {
254         be64_to_cpus(&s->l1_table[i]);
255     }
256     /* alloc L2 cache */
257     s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
258     if (!s->l2_cache)
259         goto fail;
260     s->cluster_cache = qemu_malloc(s->cluster_size);
261     if (!s->cluster_cache)
262         goto fail;
263     /* one more sector for decompressed data alignment */
264     s->cluster_data = qemu_malloc(QCOW_MAX_CRYPT_CLUSTERS * s->cluster_size
265                                   + 512);
266     if (!s->cluster_data)
267         goto fail;
268     s->cluster_cache_offset = -1;
269 
270     if (refcount_init(bs) < 0)
271         goto fail;
272 
273     /* read the backing file name */
274     if (header.backing_file_offset != 0) {
275         len = header.backing_file_size;
276         if (len > 1023)
277             len = 1023;
278         if (bdrv_pread(s->hd, header.backing_file_offset, bs->backing_file, len) != len)
279             goto fail;
280         bs->backing_file[len] = '\0';
281     }
282     if (qcow_read_snapshots(bs) < 0)
283         goto fail;
284 
285 #ifdef DEBUG_ALLOC
286     check_refcounts(bs);
287 #endif
288     return 0;
289 
290  fail:
291     qcow_free_snapshots(bs);
292     refcount_close(bs);
293     qemu_free(s->l1_table);
294     qemu_free(s->l2_cache);
295     qemu_free(s->cluster_cache);
296     qemu_free(s->cluster_data);
297     bdrv_delete(s->hd);
298     return -1;
299 }
300 
qcow_set_key(BlockDriverState * bs,const char * key)301 static int qcow_set_key(BlockDriverState *bs, const char *key)
302 {
303     BDRVQcowState *s = bs->opaque;
304     uint8_t keybuf[16];
305     int len, i;
306 
307     memset(keybuf, 0, 16);
308     len = strlen(key);
309     if (len > 16)
310         len = 16;
311     /* XXX: we could compress the chars to 7 bits to increase
312        entropy */
313     for(i = 0;i < len;i++) {
314         keybuf[i] = key[i];
315     }
316     s->crypt_method = s->crypt_method_header;
317 
318     if (AES_set_encrypt_key(keybuf, 128, &s->aes_encrypt_key) != 0)
319         return -1;
320     if (AES_set_decrypt_key(keybuf, 128, &s->aes_decrypt_key) != 0)
321         return -1;
322 #if 0
323     /* test */
324     {
325         uint8_t in[16];
326         uint8_t out[16];
327         uint8_t tmp[16];
328         for(i=0;i<16;i++)
329             in[i] = i;
330         AES_encrypt(in, tmp, &s->aes_encrypt_key);
331         AES_decrypt(tmp, out, &s->aes_decrypt_key);
332         for(i = 0; i < 16; i++)
333             printf(" %02x", tmp[i]);
334         printf("\n");
335         for(i = 0; i < 16; i++)
336             printf(" %02x", out[i]);
337         printf("\n");
338     }
339 #endif
340     return 0;
341 }
342 
343 /* The crypt function is compatible with the linux cryptoloop
344    algorithm for < 4 GB images. NOTE: out_buf == in_buf is
345    supported */
encrypt_sectors(BDRVQcowState * s,int64_t sector_num,uint8_t * out_buf,const uint8_t * in_buf,int nb_sectors,int enc,const AES_KEY * key)346 static void encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
347                             uint8_t *out_buf, const uint8_t *in_buf,
348                             int nb_sectors, int enc,
349                             const AES_KEY *key)
350 {
351     union {
352         uint64_t ll[2];
353         uint8_t b[16];
354     } ivec;
355     int i;
356 
357     for(i = 0; i < nb_sectors; i++) {
358         ivec.ll[0] = cpu_to_le64(sector_num);
359         ivec.ll[1] = 0;
360         AES_cbc_encrypt(in_buf, out_buf, 512, key,
361                         ivec.b, enc);
362         sector_num++;
363         in_buf += 512;
364         out_buf += 512;
365     }
366 }
367 
copy_sectors(BlockDriverState * bs,uint64_t start_sect,uint64_t cluster_offset,int n_start,int n_end)368 static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
369                         uint64_t cluster_offset, int n_start, int n_end)
370 {
371     BDRVQcowState *s = bs->opaque;
372     int n, ret;
373 
374     n = n_end - n_start;
375     if (n <= 0)
376         return 0;
377     ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
378     if (ret < 0)
379         return ret;
380     if (s->crypt_method) {
381         encrypt_sectors(s, start_sect + n_start,
382                         s->cluster_data,
383                         s->cluster_data, n, 1,
384                         &s->aes_encrypt_key);
385     }
386     ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
387                      s->cluster_data, n);
388     if (ret < 0)
389         return ret;
390     return 0;
391 }
392 
l2_cache_reset(BlockDriverState * bs)393 static void l2_cache_reset(BlockDriverState *bs)
394 {
395     BDRVQcowState *s = bs->opaque;
396 
397     memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
398     memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
399     memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
400 }
401 
l2_cache_new_entry(BlockDriverState * bs)402 static inline int l2_cache_new_entry(BlockDriverState *bs)
403 {
404     BDRVQcowState *s = bs->opaque;
405     uint32_t min_count;
406     int min_index, i;
407 
408     /* find a new entry in the least used one */
409     min_index = 0;
410     min_count = 0xffffffff;
411     for(i = 0; i < L2_CACHE_SIZE; i++) {
412         if (s->l2_cache_counts[i] < min_count) {
413             min_count = s->l2_cache_counts[i];
414             min_index = i;
415         }
416     }
417     return min_index;
418 }
419 
align_offset(int64_t offset,int n)420 static int64_t align_offset(int64_t offset, int n)
421 {
422     offset = (offset + n - 1) & ~(n - 1);
423     return offset;
424 }
425 
grow_l1_table(BlockDriverState * bs,int min_size)426 static int grow_l1_table(BlockDriverState *bs, int min_size)
427 {
428     BDRVQcowState *s = bs->opaque;
429     int new_l1_size, new_l1_size2, ret, i;
430     uint64_t *new_l1_table;
431     uint64_t new_l1_table_offset;
432     uint64_t data64;
433     uint32_t data32;
434 
435     new_l1_size = s->l1_size;
436     if (min_size <= new_l1_size)
437         return 0;
438     while (min_size > new_l1_size) {
439         new_l1_size = (new_l1_size * 3 + 1) / 2;
440     }
441 #ifdef DEBUG_ALLOC2
442     printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
443 #endif
444 
445     new_l1_size2 = sizeof(uint64_t) * new_l1_size;
446     new_l1_table = qemu_mallocz(new_l1_size2);
447     if (!new_l1_table)
448         return -ENOMEM;
449     memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
450 
451     /* write new table (align to cluster) */
452     new_l1_table_offset = alloc_clusters(bs, new_l1_size2);
453 
454     for(i = 0; i < s->l1_size; i++)
455         new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
456     ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
457     if (ret != new_l1_size2)
458         goto fail;
459     for(i = 0; i < s->l1_size; i++)
460         new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
461 
462     /* set new table */
463     data64 = cpu_to_be64(new_l1_table_offset);
464     if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_table_offset),
465                     &data64, sizeof(data64)) != sizeof(data64))
466         goto fail;
467     data32 = cpu_to_be32(new_l1_size);
468     if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size),
469                     &data32, sizeof(data32)) != sizeof(data32))
470         goto fail;
471     qemu_free(s->l1_table);
472     free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
473     s->l1_table_offset = new_l1_table_offset;
474     s->l1_table = new_l1_table;
475     s->l1_size = new_l1_size;
476     return 0;
477  fail:
478     qemu_free(s->l1_table);
479     return -EIO;
480 }
481 
482 /*
483  * seek_l2_table
484  *
485  * seek l2_offset in the l2_cache table
486  * if not found, return NULL,
487  * if found,
488  *   increments the l2 cache hit count of the entry,
489  *   if counter overflow, divide by two all counters
490  *   return the pointer to the l2 cache entry
491  *
492  */
493 
seek_l2_table(BDRVQcowState * s,uint64_t l2_offset)494 static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
495 {
496     int i, j;
497 
498     for(i = 0; i < L2_CACHE_SIZE; i++) {
499         if (l2_offset == s->l2_cache_offsets[i]) {
500             /* increment the hit count */
501             if (++s->l2_cache_counts[i] == 0xffffffff) {
502                 for(j = 0; j < L2_CACHE_SIZE; j++) {
503                     s->l2_cache_counts[j] >>= 1;
504                 }
505             }
506             return s->l2_cache + (i << s->l2_bits);
507         }
508     }
509     return NULL;
510 }
511 
512 /*
513  * l2_load
514  *
515  * Loads a L2 table into memory. If the table is in the cache, the cache
516  * is used; otherwise the L2 table is loaded from the image file.
517  *
518  * Returns a pointer to the L2 table on success, or NULL if the read from
519  * the image file failed.
520  */
521 
l2_load(BlockDriverState * bs,uint64_t l2_offset)522 static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
523 {
524     BDRVQcowState *s = bs->opaque;
525     int min_index;
526     uint64_t *l2_table;
527 
528     /* seek if the table for the given offset is in the cache */
529 
530     l2_table = seek_l2_table(s, l2_offset);
531     if (l2_table != NULL)
532         return l2_table;
533 
534     /* not found: load a new entry in the least used one */
535 
536     min_index = l2_cache_new_entry(bs);
537     l2_table = s->l2_cache + (min_index << s->l2_bits);
538     if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
539         s->l2_size * sizeof(uint64_t))
540         return NULL;
541     s->l2_cache_offsets[min_index] = l2_offset;
542     s->l2_cache_counts[min_index] = 1;
543 
544     return l2_table;
545 }
546 
547 /*
548  * l2_allocate
549  *
550  * Allocate a new l2 entry in the file. If l1_index points to an already
551  * used entry in the L2 table (i.e. we are doing a copy on write for the L2
552  * table) copy the contents of the old L2 table into the newly allocated one.
553  * Otherwise the new table is initialized with zeros.
554  *
555  */
556 
l2_allocate(BlockDriverState * bs,int l1_index)557 static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
558 {
559     BDRVQcowState *s = bs->opaque;
560     int min_index;
561     uint64_t old_l2_offset, tmp;
562     uint64_t *l2_table, l2_offset;
563 
564     old_l2_offset = s->l1_table[l1_index];
565 
566     /* allocate a new l2 entry */
567 
568     l2_offset = alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
569 
570     /* update the L1 entry */
571 
572     s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
573 
574     tmp = cpu_to_be64(l2_offset | QCOW_OFLAG_COPIED);
575     if (bdrv_pwrite(s->hd, s->l1_table_offset + l1_index * sizeof(tmp),
576                     &tmp, sizeof(tmp)) != sizeof(tmp))
577         return NULL;
578 
579     /* allocate a new entry in the l2 cache */
580 
581     min_index = l2_cache_new_entry(bs);
582     l2_table = s->l2_cache + (min_index << s->l2_bits);
583 
584     if (old_l2_offset == 0) {
585         /* if there was no old l2 table, clear the new table */
586         memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
587     } else {
588         /* if there was an old l2 table, read it from the disk */
589         if (bdrv_pread(s->hd, old_l2_offset,
590                        l2_table, s->l2_size * sizeof(uint64_t)) !=
591             s->l2_size * sizeof(uint64_t))
592             return NULL;
593     }
594     /* write the l2 table to the file */
595     if (bdrv_pwrite(s->hd, l2_offset,
596                     l2_table, s->l2_size * sizeof(uint64_t)) !=
597         s->l2_size * sizeof(uint64_t))
598         return NULL;
599 
600     /* update the l2 cache entry */
601 
602     s->l2_cache_offsets[min_index] = l2_offset;
603     s->l2_cache_counts[min_index] = 1;
604 
605     return l2_table;
606 }
607 
608 /*
609  * get_cluster_offset
610  *
611  * For a given offset of the disk image, return cluster offset in
612  * qcow2 file.
613  *
614  * on entry, *num is the number of contiguous clusters we'd like to
615  * access following offset.
616  *
617  * on exit, *num is the number of contiguous clusters we can read.
618  *
619  * Return 1, if the offset is found
620  * Return 0, otherwise.
621  *
622  */
623 
get_cluster_offset(BlockDriverState * bs,uint64_t offset,int * num)624 static uint64_t get_cluster_offset(BlockDriverState *bs,
625                                    uint64_t offset, int *num)
626 {
627     BDRVQcowState *s = bs->opaque;
628     int l1_index, l2_index;
629     uint64_t l2_offset, *l2_table, cluster_offset, next;
630     int l1_bits;
631     int index_in_cluster, nb_available, nb_needed;
632 
633     index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
634     nb_needed = *num + index_in_cluster;
635 
636     l1_bits = s->l2_bits + s->cluster_bits;
637 
638     /* compute how many bytes there are between the offset and
639      * and the end of the l1 entry
640      */
641 
642     nb_available = (1 << l1_bits) - (offset & ((1 << l1_bits) - 1));
643 
644     /* compute the number of available sectors */
645 
646     nb_available = (nb_available >> 9) + index_in_cluster;
647 
648     cluster_offset = 0;
649 
650     /* seek the the l2 offset in the l1 table */
651 
652     l1_index = offset >> l1_bits;
653     if (l1_index >= s->l1_size)
654         goto out;
655 
656     l2_offset = s->l1_table[l1_index];
657 
658     /* seek the l2 table of the given l2 offset */
659 
660     if (!l2_offset)
661         goto out;
662 
663     /* load the l2 table in memory */
664 
665     l2_offset &= ~QCOW_OFLAG_COPIED;
666     l2_table = l2_load(bs, l2_offset);
667     if (l2_table == NULL)
668         return 0;
669 
670     /* find the cluster offset for the given disk offset */
671 
672     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
673     cluster_offset = be64_to_cpu(l2_table[l2_index]);
674     nb_available = s->cluster_sectors;
675     l2_index++;
676 
677     if (!cluster_offset) {
678 
679        /* how many empty clusters ? */
680 
681        while (nb_available < nb_needed && !l2_table[l2_index]) {
682            l2_index++;
683            nb_available += s->cluster_sectors;
684        }
685     } else {
686 
687        /* how many allocated clusters ? */
688 
689        cluster_offset &= ~QCOW_OFLAG_COPIED;
690        while (nb_available < nb_needed) {
691            next = be64_to_cpu(l2_table[l2_index]) & ~QCOW_OFLAG_COPIED;
692            if (next != cluster_offset + (nb_available << 9))
693                break;
694            l2_index++;
695            nb_available += s->cluster_sectors;
696        }
697    }
698 
699 out:
700     if (nb_available > nb_needed)
701         nb_available = nb_needed;
702 
703     *num = nb_available - index_in_cluster;
704 
705     return cluster_offset;
706 }
707 
708 /*
709  * free_any_clusters
710  *
711  * free clusters according to its type: compressed or not
712  *
713  */
714 
free_any_clusters(BlockDriverState * bs,uint64_t cluster_offset,int nb_clusters)715 static void free_any_clusters(BlockDriverState *bs,
716                               uint64_t cluster_offset, int nb_clusters)
717 {
718     BDRVQcowState *s = bs->opaque;
719 
720     /* free the cluster */
721 
722     if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
723         int nb_csectors;
724         nb_csectors = ((cluster_offset >> s->csize_shift) &
725                        s->csize_mask) + 1;
726         free_clusters(bs, (cluster_offset & s->cluster_offset_mask) & ~511,
727                       nb_csectors * 512);
728         return;
729     }
730 
731     free_clusters(bs, cluster_offset, nb_clusters << s->cluster_bits);
732 
733     return;
734 }
735 
736 /*
737  * get_cluster_table
738  *
739  * for a given disk offset, load (and allocate if needed)
740  * the l2 table.
741  *
742  * the l2 table offset in the qcow2 file and the cluster index
743  * in the l2 table are given to the caller.
744  *
745  */
746 
get_cluster_table(BlockDriverState * bs,uint64_t offset,uint64_t ** new_l2_table,uint64_t * new_l2_offset,int * new_l2_index)747 static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
748                              uint64_t **new_l2_table,
749                              uint64_t *new_l2_offset,
750                              int *new_l2_index)
751 {
752     BDRVQcowState *s = bs->opaque;
753     int l1_index, l2_index, ret;
754     uint64_t l2_offset, *l2_table;
755 
756     /* seek the the l2 offset in the l1 table */
757 
758     l1_index = offset >> (s->l2_bits + s->cluster_bits);
759     if (l1_index >= s->l1_size) {
760         ret = grow_l1_table(bs, l1_index + 1);
761         if (ret < 0)
762             return 0;
763     }
764     l2_offset = s->l1_table[l1_index];
765 
766     /* seek the l2 table of the given l2 offset */
767 
768     if (l2_offset & QCOW_OFLAG_COPIED) {
769         /* load the l2 table in memory */
770         l2_offset &= ~QCOW_OFLAG_COPIED;
771         l2_table = l2_load(bs, l2_offset);
772         if (l2_table == NULL)
773             return 0;
774     } else {
775         if (l2_offset)
776             free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
777         l2_table = l2_allocate(bs, l1_index);
778         if (l2_table == NULL)
779             return 0;
780         l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
781     }
782 
783     /* find the cluster offset for the given disk offset */
784 
785     l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
786 
787     *new_l2_table = l2_table;
788     *new_l2_offset = l2_offset;
789     *new_l2_index = l2_index;
790 
791     return 1;
792 }
793 
794 /*
795  * alloc_compressed_cluster_offset
796  *
797  * For a given offset of the disk image, return cluster offset in
798  * qcow2 file.
799  *
800  * If the offset is not found, allocate a new compressed cluster.
801  *
802  * Return the cluster offset if successful,
803  * Return 0, otherwise.
804  *
805  */
806 
alloc_compressed_cluster_offset(BlockDriverState * bs,uint64_t offset,int compressed_size)807 static uint64_t alloc_compressed_cluster_offset(BlockDriverState *bs,
808                                                 uint64_t offset,
809                                                 int compressed_size)
810 {
811     BDRVQcowState *s = bs->opaque;
812     int l2_index, ret;
813     uint64_t l2_offset, *l2_table, cluster_offset;
814     int nb_csectors;
815 
816     ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
817     if (ret == 0)
818         return 0;
819 
820     cluster_offset = be64_to_cpu(l2_table[l2_index]);
821     if (cluster_offset & QCOW_OFLAG_COPIED)
822         return cluster_offset & ~QCOW_OFLAG_COPIED;
823 
824     if (cluster_offset)
825         free_any_clusters(bs, cluster_offset, 1);
826 
827     cluster_offset = alloc_bytes(bs, compressed_size);
828     nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
829                   (cluster_offset >> 9);
830 
831     cluster_offset |= QCOW_OFLAG_COMPRESSED |
832                       ((uint64_t)nb_csectors << s->csize_shift);
833 
834     /* update L2 table */
835 
836     /* compressed clusters never have the copied flag */
837 
838     l2_table[l2_index] = cpu_to_be64(cluster_offset);
839     if (bdrv_pwrite(s->hd,
840                     l2_offset + l2_index * sizeof(uint64_t),
841                     l2_table + l2_index,
842                     sizeof(uint64_t)) != sizeof(uint64_t))
843         return 0;
844 
845     return cluster_offset;
846 }
847 
848 /*
849  * alloc_cluster_offset
850  *
851  * For a given offset of the disk image, return cluster offset in
852  * qcow2 file.
853  *
854  * If the offset is not found, allocate a new cluster.
855  *
856  * Return the cluster offset if successful,
857  * Return 0, otherwise.
858  *
859  */
860 
alloc_cluster_offset(BlockDriverState * bs,uint64_t offset,int n_start,int n_end,int * num)861 static uint64_t alloc_cluster_offset(BlockDriverState *bs,
862                                      uint64_t offset,
863                                      int n_start, int n_end,
864                                      int *num)
865 {
866     BDRVQcowState *s = bs->opaque;
867     int l2_index, ret;
868     uint64_t l2_offset, *l2_table, cluster_offset;
869     int nb_available, nb_clusters, i, j;
870     uint64_t start_sect, current;
871 
872     ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
873     if (ret == 0)
874         return 0;
875 
876     nb_clusters = ((n_end << 9) + s->cluster_size - 1) >>
877                   s->cluster_bits;
878     if (nb_clusters > s->l2_size - l2_index)
879             nb_clusters = s->l2_size - l2_index;
880 
881     cluster_offset = be64_to_cpu(l2_table[l2_index]);
882 
883     /* We keep all QCOW_OFLAG_COPIED clusters */
884 
885     if (cluster_offset & QCOW_OFLAG_COPIED) {
886 
887         for (i = 1; i < nb_clusters; i++) {
888             current = be64_to_cpu(l2_table[l2_index + i]);
889             if (cluster_offset + (i << s->cluster_bits) != current)
890                 break;
891         }
892         nb_clusters = i;
893 
894         nb_available = nb_clusters << (s->cluster_bits - 9);
895         if (nb_available > n_end)
896             nb_available = n_end;
897 
898         cluster_offset &= ~QCOW_OFLAG_COPIED;
899 
900         goto out;
901     }
902 
903     /* for the moment, multiple compressed clusters are not managed */
904 
905     if (cluster_offset & QCOW_OFLAG_COMPRESSED)
906         nb_clusters = 1;
907 
908     /* how many available clusters ? */
909 
910     i = 0;
911     while (i < nb_clusters) {
912 
913         i++;
914 
915         if (!cluster_offset) {
916 
917             /* how many free clusters ? */
918 
919             while (i < nb_clusters) {
920                 cluster_offset = l2_table[l2_index + i];
921                 if (cluster_offset != 0)
922                     break;
923                 i++;
924             }
925 
926             if ((cluster_offset & QCOW_OFLAG_COPIED) ||
927                 (cluster_offset & QCOW_OFLAG_COMPRESSED))
928                 break;
929 
930         } else {
931 
932             /* how many contiguous clusters ? */
933 
934             j = 1;
935             current = 0;
936             while (i < nb_clusters) {
937                 current = be64_to_cpu(l2_table[l2_index + i]);
938                 if (cluster_offset + (j << s->cluster_bits) != current)
939                     break;
940 
941                 i++;
942                 j++;
943             }
944 
945             free_any_clusters(bs, cluster_offset, j);
946             if (current)
947                 break;
948             cluster_offset = current;
949         }
950     }
951     nb_clusters = i;
952 
953     /* allocate a new cluster */
954 
955     cluster_offset = alloc_clusters(bs, nb_clusters * s->cluster_size);
956 
957     /* we must initialize the cluster content which won't be
958        written */
959 
960     nb_available = nb_clusters << (s->cluster_bits - 9);
961     if (nb_available > n_end)
962         nb_available = n_end;
963 
964     /* copy content of unmodified sectors */
965 
966     start_sect = (offset & ~(s->cluster_size - 1)) >> 9;
967     if (n_start) {
968         ret = copy_sectors(bs, start_sect, cluster_offset, 0, n_start);
969         if (ret < 0)
970             return 0;
971     }
972 
973     if (nb_available & (s->cluster_sectors - 1)) {
974         uint64_t end = nb_available & ~(uint64_t)(s->cluster_sectors - 1);
975         ret = copy_sectors(bs, start_sect + end,
976                            cluster_offset + (end << 9),
977                            nb_available - end,
978                            s->cluster_sectors);
979         if (ret < 0)
980             return 0;
981     }
982 
983     /* update L2 table */
984 
985     for (i = 0; i < nb_clusters; i++)
986         l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
987                                              (i << s->cluster_bits)) |
988                                              QCOW_OFLAG_COPIED);
989 
990     if (bdrv_pwrite(s->hd,
991                     l2_offset + l2_index * sizeof(uint64_t),
992                     l2_table + l2_index,
993                     nb_clusters * sizeof(uint64_t)) !=
994                     nb_clusters * sizeof(uint64_t))
995         return 0;
996 
997 out:
998     *num = nb_available - n_start;
999 
1000     return cluster_offset;
1001 }
1002 
qcow_is_allocated(BlockDriverState * bs,int64_t sector_num,int nb_sectors,int * pnum)1003 static int qcow_is_allocated(BlockDriverState *bs, int64_t sector_num,
1004                              int nb_sectors, int *pnum)
1005 {
1006     uint64_t cluster_offset;
1007 
1008     *pnum = nb_sectors;
1009     cluster_offset = get_cluster_offset(bs, sector_num << 9, pnum);
1010 
1011     return (cluster_offset != 0);
1012 }
1013 
decompress_buffer(uint8_t * out_buf,int out_buf_size,const uint8_t * buf,int buf_size)1014 static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
1015                              const uint8_t *buf, int buf_size)
1016 {
1017     z_stream strm1, *strm = &strm1;
1018     int ret, out_len;
1019 
1020     memset(strm, 0, sizeof(*strm));
1021 
1022     strm->next_in = (uint8_t *)buf;
1023     strm->avail_in = buf_size;
1024     strm->next_out = out_buf;
1025     strm->avail_out = out_buf_size;
1026 
1027     ret = inflateInit2(strm, -12);
1028     if (ret != Z_OK)
1029         return -1;
1030     ret = inflate(strm, Z_FINISH);
1031     out_len = strm->next_out - out_buf;
1032     if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1033         out_len != out_buf_size) {
1034         inflateEnd(strm);
1035         return -1;
1036     }
1037     inflateEnd(strm);
1038     return 0;
1039 }
1040 
decompress_cluster(BDRVQcowState * s,uint64_t cluster_offset)1041 static int decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
1042 {
1043     int ret, csize, nb_csectors, sector_offset;
1044     uint64_t coffset;
1045 
1046     coffset = cluster_offset & s->cluster_offset_mask;
1047     if (s->cluster_cache_offset != coffset) {
1048         nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1049         sector_offset = coffset & 511;
1050         csize = nb_csectors * 512 - sector_offset;
1051         ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
1052         if (ret < 0) {
1053             return -1;
1054         }
1055         if (decompress_buffer(s->cluster_cache, s->cluster_size,
1056                               s->cluster_data + sector_offset, csize) < 0) {
1057             return -1;
1058         }
1059         s->cluster_cache_offset = coffset;
1060     }
1061     return 0;
1062 }
1063 
1064 /* handle reading after the end of the backing file */
backing_read1(BlockDriverState * bs,int64_t sector_num,uint8_t * buf,int nb_sectors)1065 static int backing_read1(BlockDriverState *bs,
1066                          int64_t sector_num, uint8_t *buf, int nb_sectors)
1067 {
1068     int n1;
1069     if ((sector_num + nb_sectors) <= bs->total_sectors)
1070         return nb_sectors;
1071     if (sector_num >= bs->total_sectors)
1072         n1 = 0;
1073     else
1074         n1 = bs->total_sectors - sector_num;
1075     memset(buf + n1 * 512, 0, 512 * (nb_sectors - n1));
1076     return n1;
1077 }
1078 
qcow_read(BlockDriverState * bs,int64_t sector_num,uint8_t * buf,int nb_sectors)1079 static int qcow_read(BlockDriverState *bs, int64_t sector_num,
1080                      uint8_t *buf, int nb_sectors)
1081 {
1082     BDRVQcowState *s = bs->opaque;
1083     int ret, index_in_cluster, n, n1;
1084     uint64_t cluster_offset;
1085 
1086     while (nb_sectors > 0) {
1087         n = nb_sectors;
1088         cluster_offset = get_cluster_offset(bs, sector_num << 9, &n);
1089         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1090         if (!cluster_offset) {
1091             if (bs->backing_hd) {
1092                 /* read from the base image */
1093                 n1 = backing_read1(bs->backing_hd, sector_num, buf, n);
1094                 if (n1 > 0) {
1095                     ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
1096                     if (ret < 0)
1097                         return -1;
1098                 }
1099             } else {
1100                 memset(buf, 0, 512 * n);
1101             }
1102         } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1103             if (decompress_cluster(s, cluster_offset) < 0)
1104                 return -1;
1105             memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
1106         } else {
1107             ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1108             if (ret != n * 512)
1109                 return -1;
1110             if (s->crypt_method) {
1111                 encrypt_sectors(s, sector_num, buf, buf, n, 0,
1112                                 &s->aes_decrypt_key);
1113             }
1114         }
1115         nb_sectors -= n;
1116         sector_num += n;
1117         buf += n * 512;
1118     }
1119     return 0;
1120 }
1121 
qcow_write(BlockDriverState * bs,int64_t sector_num,const uint8_t * buf,int nb_sectors)1122 static int qcow_write(BlockDriverState *bs, int64_t sector_num,
1123                      const uint8_t *buf, int nb_sectors)
1124 {
1125     BDRVQcowState *s = bs->opaque;
1126     int ret, index_in_cluster, n;
1127     uint64_t cluster_offset;
1128     int n_end;
1129 
1130     while (nb_sectors > 0) {
1131         index_in_cluster = sector_num & (s->cluster_sectors - 1);
1132         n_end = index_in_cluster + nb_sectors;
1133         if (s->crypt_method &&
1134             n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1135             n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1136         cluster_offset = alloc_cluster_offset(bs, sector_num << 9,
1137                                               index_in_cluster,
1138                                               n_end, &n);
1139         if (!cluster_offset)
1140             return -1;
1141         if (s->crypt_method) {
1142             encrypt_sectors(s, sector_num, s->cluster_data, buf, n, 1,
1143                             &s->aes_encrypt_key);
1144             ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512,
1145                               s->cluster_data, n * 512);
1146         } else {
1147             ret = bdrv_pwrite(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
1148         }
1149         if (ret != n * 512)
1150             return -1;
1151         nb_sectors -= n;
1152         sector_num += n;
1153         buf += n * 512;
1154     }
1155     s->cluster_cache_offset = -1; /* disable compressed cache */
1156     return 0;
1157 }
1158 
1159 typedef struct QCowAIOCB {
1160     BlockDriverAIOCB common;
1161     int64_t sector_num;
1162     uint8_t *buf;
1163     int nb_sectors;
1164     int n;
1165     uint64_t cluster_offset;
1166     uint8_t *cluster_data;
1167     BlockDriverAIOCB *hd_aiocb;
1168 } QCowAIOCB;
1169 
qcow_aio_read_cb(void * opaque,int ret)1170 static void qcow_aio_read_cb(void *opaque, int ret)
1171 {
1172     QCowAIOCB *acb = opaque;
1173     BlockDriverState *bs = acb->common.bs;
1174     BDRVQcowState *s = bs->opaque;
1175     int index_in_cluster, n1;
1176 
1177     acb->hd_aiocb = NULL;
1178     if (ret < 0) {
1179     fail:
1180         acb->common.cb(acb->common.opaque, ret);
1181         qemu_aio_release(acb);
1182         return;
1183     }
1184 
1185  redo:
1186     /* post process the read buffer */
1187     if (!acb->cluster_offset) {
1188         /* nothing to do */
1189     } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1190         /* nothing to do */
1191     } else {
1192         if (s->crypt_method) {
1193             encrypt_sectors(s, acb->sector_num, acb->buf, acb->buf,
1194                             acb->n, 0,
1195                             &s->aes_decrypt_key);
1196         }
1197     }
1198 
1199     acb->nb_sectors -= acb->n;
1200     acb->sector_num += acb->n;
1201     acb->buf += acb->n * 512;
1202 
1203     if (acb->nb_sectors == 0) {
1204         /* request completed */
1205         acb->common.cb(acb->common.opaque, 0);
1206         qemu_aio_release(acb);
1207         return;
1208     }
1209 
1210     /* prepare next AIO request */
1211     acb->n = acb->nb_sectors;
1212     acb->cluster_offset = get_cluster_offset(bs, acb->sector_num << 9, &acb->n);
1213     index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1214 
1215     if (!acb->cluster_offset) {
1216         if (bs->backing_hd) {
1217             /* read from the base image */
1218             n1 = backing_read1(bs->backing_hd, acb->sector_num,
1219                                acb->buf, acb->n);
1220             if (n1 > 0) {
1221                 acb->hd_aiocb = bdrv_aio_read(bs->backing_hd, acb->sector_num,
1222                                     acb->buf, acb->n, qcow_aio_read_cb, acb);
1223                 if (acb->hd_aiocb == NULL)
1224                     goto fail;
1225             } else {
1226                 goto redo;
1227             }
1228         } else {
1229             /* Note: in this case, no need to wait */
1230             memset(acb->buf, 0, 512 * acb->n);
1231             goto redo;
1232         }
1233     } else if (acb->cluster_offset & QCOW_OFLAG_COMPRESSED) {
1234         /* add AIO support for compressed blocks ? */
1235         if (decompress_cluster(s, acb->cluster_offset) < 0)
1236             goto fail;
1237         memcpy(acb->buf,
1238                s->cluster_cache + index_in_cluster * 512, 512 * acb->n);
1239         goto redo;
1240     } else {
1241         if ((acb->cluster_offset & 511) != 0) {
1242             ret = -EIO;
1243             goto fail;
1244         }
1245         acb->hd_aiocb = bdrv_aio_read(s->hd,
1246                             (acb->cluster_offset >> 9) + index_in_cluster,
1247                             acb->buf, acb->n, qcow_aio_read_cb, acb);
1248         if (acb->hd_aiocb == NULL)
1249             goto fail;
1250     }
1251 }
1252 
qcow_aio_setup(BlockDriverState * bs,int64_t sector_num,uint8_t * buf,int nb_sectors,BlockDriverCompletionFunc * cb,void * opaque)1253 static QCowAIOCB *qcow_aio_setup(BlockDriverState *bs,
1254         int64_t sector_num, uint8_t *buf, int nb_sectors,
1255         BlockDriverCompletionFunc *cb, void *opaque)
1256 {
1257     QCowAIOCB *acb;
1258 
1259     acb = qemu_aio_get(bs, cb, opaque);
1260     if (!acb)
1261         return NULL;
1262     acb->hd_aiocb = NULL;
1263     acb->sector_num = sector_num;
1264     acb->buf = buf;
1265     acb->nb_sectors = nb_sectors;
1266     acb->n = 0;
1267     acb->cluster_offset = 0;
1268     return acb;
1269 }
1270 
qcow_aio_read(BlockDriverState * bs,int64_t sector_num,uint8_t * buf,int nb_sectors,BlockDriverCompletionFunc * cb,void * opaque)1271 static BlockDriverAIOCB *qcow_aio_read(BlockDriverState *bs,
1272         int64_t sector_num, uint8_t *buf, int nb_sectors,
1273         BlockDriverCompletionFunc *cb, void *opaque)
1274 {
1275     QCowAIOCB *acb;
1276 
1277     acb = qcow_aio_setup(bs, sector_num, buf, nb_sectors, cb, opaque);
1278     if (!acb)
1279         return NULL;
1280 
1281     qcow_aio_read_cb(acb, 0);
1282     return &acb->common;
1283 }
1284 
qcow_aio_write_cb(void * opaque,int ret)1285 static void qcow_aio_write_cb(void *opaque, int ret)
1286 {
1287     QCowAIOCB *acb = opaque;
1288     BlockDriverState *bs = acb->common.bs;
1289     BDRVQcowState *s = bs->opaque;
1290     int index_in_cluster;
1291     uint64_t cluster_offset;
1292     const uint8_t *src_buf;
1293     int n_end;
1294 
1295     acb->hd_aiocb = NULL;
1296 
1297     if (ret < 0) {
1298     fail:
1299         acb->common.cb(acb->common.opaque, ret);
1300         qemu_aio_release(acb);
1301         return;
1302     }
1303 
1304     acb->nb_sectors -= acb->n;
1305     acb->sector_num += acb->n;
1306     acb->buf += acb->n * 512;
1307 
1308     if (acb->nb_sectors == 0) {
1309         /* request completed */
1310         acb->common.cb(acb->common.opaque, 0);
1311         qemu_aio_release(acb);
1312         return;
1313     }
1314 
1315     index_in_cluster = acb->sector_num & (s->cluster_sectors - 1);
1316     n_end = index_in_cluster + acb->nb_sectors;
1317     if (s->crypt_method &&
1318         n_end > QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors)
1319         n_end = QCOW_MAX_CRYPT_CLUSTERS * s->cluster_sectors;
1320 
1321     cluster_offset = alloc_cluster_offset(bs, acb->sector_num << 9,
1322                                           index_in_cluster,
1323                                           n_end, &acb->n);
1324     if (!cluster_offset || (cluster_offset & 511) != 0) {
1325         ret = -EIO;
1326         goto fail;
1327     }
1328     if (s->crypt_method) {
1329         if (!acb->cluster_data) {
1330             acb->cluster_data = qemu_mallocz(QCOW_MAX_CRYPT_CLUSTERS *
1331                                              s->cluster_size);
1332             if (!acb->cluster_data) {
1333                 ret = -ENOMEM;
1334                 goto fail;
1335             }
1336         }
1337         encrypt_sectors(s, acb->sector_num, acb->cluster_data, acb->buf,
1338                         acb->n, 1, &s->aes_encrypt_key);
1339         src_buf = acb->cluster_data;
1340     } else {
1341         src_buf = acb->buf;
1342     }
1343     acb->hd_aiocb = bdrv_aio_write(s->hd,
1344                                    (cluster_offset >> 9) + index_in_cluster,
1345                                    src_buf, acb->n,
1346                                    qcow_aio_write_cb, acb);
1347     if (acb->hd_aiocb == NULL)
1348         goto fail;
1349 }
1350 
qcow_aio_write(BlockDriverState * bs,int64_t sector_num,const uint8_t * buf,int nb_sectors,BlockDriverCompletionFunc * cb,void * opaque)1351 static BlockDriverAIOCB *qcow_aio_write(BlockDriverState *bs,
1352         int64_t sector_num, const uint8_t *buf, int nb_sectors,
1353         BlockDriverCompletionFunc *cb, void *opaque)
1354 {
1355     BDRVQcowState *s = bs->opaque;
1356     QCowAIOCB *acb;
1357 
1358     s->cluster_cache_offset = -1; /* disable compressed cache */
1359 
1360     acb = qcow_aio_setup(bs, sector_num, (uint8_t*)buf, nb_sectors, cb, opaque);
1361     if (!acb)
1362         return NULL;
1363 
1364     qcow_aio_write_cb(acb, 0);
1365     return &acb->common;
1366 }
1367 
qcow_aio_cancel(BlockDriverAIOCB * blockacb)1368 static void qcow_aio_cancel(BlockDriverAIOCB *blockacb)
1369 {
1370     QCowAIOCB *acb = (QCowAIOCB *)blockacb;
1371     if (acb->hd_aiocb)
1372         bdrv_aio_cancel(acb->hd_aiocb);
1373     qemu_aio_release(acb);
1374 }
1375 
qcow_close(BlockDriverState * bs)1376 static void qcow_close(BlockDriverState *bs)
1377 {
1378     BDRVQcowState *s = bs->opaque;
1379     qemu_free(s->l1_table);
1380     qemu_free(s->l2_cache);
1381     qemu_free(s->cluster_cache);
1382     qemu_free(s->cluster_data);
1383     refcount_close(bs);
1384     bdrv_delete(s->hd);
1385 }
1386 
1387 /* XXX: use std qcow open function ? */
1388 typedef struct QCowCreateState {
1389     int cluster_size;
1390     int cluster_bits;
1391     uint16_t *refcount_block;
1392     uint64_t *refcount_table;
1393     int64_t l1_table_offset;
1394     int64_t refcount_table_offset;
1395     int64_t refcount_block_offset;
1396 } QCowCreateState;
1397 
create_refcount_update(QCowCreateState * s,int64_t offset,int64_t size)1398 static void create_refcount_update(QCowCreateState *s,
1399                                    int64_t offset, int64_t size)
1400 {
1401     int refcount;
1402     int64_t start, last, cluster_offset;
1403     uint16_t *p;
1404 
1405     start = offset & ~(s->cluster_size - 1);
1406     last = (offset + size - 1)  & ~(s->cluster_size - 1);
1407     for(cluster_offset = start; cluster_offset <= last;
1408         cluster_offset += s->cluster_size) {
1409         p = &s->refcount_block[cluster_offset >> s->cluster_bits];
1410         refcount = be16_to_cpu(*p);
1411         refcount++;
1412         *p = cpu_to_be16(refcount);
1413     }
1414 }
1415 
qcow_create(const char * filename,int64_t total_size,const char * backing_file,int flags)1416 static int qcow_create(const char *filename, int64_t total_size,
1417                       const char *backing_file, int flags)
1418 {
1419     int fd, header_size, backing_filename_len, l1_size, i, shift, l2_bits;
1420     QCowHeader header;
1421     uint64_t tmp, offset;
1422     QCowCreateState s1, *s = &s1;
1423 
1424     memset(s, 0, sizeof(*s));
1425 
1426     fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY, 0644);
1427     if (fd < 0)
1428         return -1;
1429     memset(&header, 0, sizeof(header));
1430     header.magic = cpu_to_be32(QCOW_MAGIC);
1431     header.version = cpu_to_be32(QCOW_VERSION);
1432     header.size = cpu_to_be64(total_size * 512);
1433     header_size = sizeof(header);
1434     backing_filename_len = 0;
1435     if (backing_file) {
1436         header.backing_file_offset = cpu_to_be64(header_size);
1437         backing_filename_len = strlen(backing_file);
1438         header.backing_file_size = cpu_to_be32(backing_filename_len);
1439         header_size += backing_filename_len;
1440     }
1441     s->cluster_bits = 12;  /* 4 KB clusters */
1442     s->cluster_size = 1 << s->cluster_bits;
1443     header.cluster_bits = cpu_to_be32(s->cluster_bits);
1444     header_size = (header_size + 7) & ~7;
1445     if (flags & BLOCK_FLAG_ENCRYPT) {
1446         header.crypt_method = cpu_to_be32(QCOW_CRYPT_AES);
1447     } else {
1448         header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);
1449     }
1450     l2_bits = s->cluster_bits - 3;
1451     shift = s->cluster_bits + l2_bits;
1452     l1_size = (((total_size * 512) + (1LL << shift) - 1) >> shift);
1453     offset = align_offset(header_size, s->cluster_size);
1454     s->l1_table_offset = offset;
1455     header.l1_table_offset = cpu_to_be64(s->l1_table_offset);
1456     header.l1_size = cpu_to_be32(l1_size);
1457     offset += align_offset(l1_size * sizeof(uint64_t), s->cluster_size);
1458 
1459     s->refcount_table = qemu_mallocz(s->cluster_size);
1460     if (!s->refcount_table)
1461         goto fail;
1462     s->refcount_block = qemu_mallocz(s->cluster_size);
1463     if (!s->refcount_block)
1464         goto fail;
1465 
1466     s->refcount_table_offset = offset;
1467     header.refcount_table_offset = cpu_to_be64(offset);
1468     header.refcount_table_clusters = cpu_to_be32(1);
1469     offset += s->cluster_size;
1470 
1471     s->refcount_table[0] = cpu_to_be64(offset);
1472     s->refcount_block_offset = offset;
1473     offset += s->cluster_size;
1474 
1475     /* update refcounts */
1476     create_refcount_update(s, 0, header_size);
1477     create_refcount_update(s, s->l1_table_offset, l1_size * sizeof(uint64_t));
1478     create_refcount_update(s, s->refcount_table_offset, s->cluster_size);
1479     create_refcount_update(s, s->refcount_block_offset, s->cluster_size);
1480 
1481     /* write all the data */
1482     write(fd, &header, sizeof(header));
1483     if (backing_file) {
1484         write(fd, backing_file, backing_filename_len);
1485     }
1486     lseek(fd, s->l1_table_offset, SEEK_SET);
1487     tmp = 0;
1488     for(i = 0;i < l1_size; i++) {
1489         write(fd, &tmp, sizeof(tmp));
1490     }
1491     lseek(fd, s->refcount_table_offset, SEEK_SET);
1492     write(fd, s->refcount_table, s->cluster_size);
1493 
1494     lseek(fd, s->refcount_block_offset, SEEK_SET);
1495     write(fd, s->refcount_block, s->cluster_size);
1496 
1497     qemu_free(s->refcount_table);
1498     qemu_free(s->refcount_block);
1499     close(fd);
1500     return 0;
1501  fail:
1502     qemu_free(s->refcount_table);
1503     qemu_free(s->refcount_block);
1504     close(fd);
1505     return -ENOMEM;
1506 }
1507 
qcow_make_empty(BlockDriverState * bs)1508 static int qcow_make_empty(BlockDriverState *bs)
1509 {
1510 #if 0
1511     /* XXX: not correct */
1512     BDRVQcowState *s = bs->opaque;
1513     uint32_t l1_length = s->l1_size * sizeof(uint64_t);
1514     int ret;
1515 
1516     memset(s->l1_table, 0, l1_length);
1517     if (bdrv_pwrite(s->hd, s->l1_table_offset, s->l1_table, l1_length) < 0)
1518 	return -1;
1519     ret = bdrv_truncate(s->hd, s->l1_table_offset + l1_length);
1520     if (ret < 0)
1521         return ret;
1522 
1523     l2_cache_reset(bs);
1524 #endif
1525     return 0;
1526 }
1527 
1528 /* XXX: put compressed sectors first, then all the cluster aligned
1529    tables to avoid losing bytes in alignment */
qcow_write_compressed(BlockDriverState * bs,int64_t sector_num,const uint8_t * buf,int nb_sectors)1530 static int qcow_write_compressed(BlockDriverState *bs, int64_t sector_num,
1531                                  const uint8_t *buf, int nb_sectors)
1532 {
1533     BDRVQcowState *s = bs->opaque;
1534     z_stream strm;
1535     int ret, out_len;
1536     uint8_t *out_buf;
1537     uint64_t cluster_offset;
1538 
1539     if (nb_sectors == 0) {
1540         /* align end of file to a sector boundary to ease reading with
1541            sector based I/Os */
1542         cluster_offset = bdrv_getlength(s->hd);
1543         cluster_offset = (cluster_offset + 511) & ~511;
1544         bdrv_truncate(s->hd, cluster_offset);
1545         return 0;
1546     }
1547 
1548     if (nb_sectors != s->cluster_sectors)
1549         return -EINVAL;
1550 
1551     out_buf = qemu_malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
1552     if (!out_buf)
1553         return -ENOMEM;
1554 
1555     /* best compression, small window, no zlib header */
1556     memset(&strm, 0, sizeof(strm));
1557     ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
1558                        Z_DEFLATED, -12,
1559                        9, Z_DEFAULT_STRATEGY);
1560     if (ret != 0) {
1561         qemu_free(out_buf);
1562         return -1;
1563     }
1564 
1565     strm.avail_in = s->cluster_size;
1566     strm.next_in = (uint8_t *)buf;
1567     strm.avail_out = s->cluster_size;
1568     strm.next_out = out_buf;
1569 
1570     ret = deflate(&strm, Z_FINISH);
1571     if (ret != Z_STREAM_END && ret != Z_OK) {
1572         qemu_free(out_buf);
1573         deflateEnd(&strm);
1574         return -1;
1575     }
1576     out_len = strm.next_out - out_buf;
1577 
1578     deflateEnd(&strm);
1579 
1580     if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
1581         /* could not compress: write normal cluster */
1582         qcow_write(bs, sector_num, buf, s->cluster_sectors);
1583     } else {
1584         cluster_offset = alloc_compressed_cluster_offset(bs, sector_num << 9,
1585                                               out_len);
1586         if (!cluster_offset)
1587             return -1;
1588         cluster_offset &= s->cluster_offset_mask;
1589         if (bdrv_pwrite(s->hd, cluster_offset, out_buf, out_len) != out_len) {
1590             qemu_free(out_buf);
1591             return -1;
1592         }
1593     }
1594 
1595     qemu_free(out_buf);
1596     return 0;
1597 }
1598 
qcow_flush(BlockDriverState * bs)1599 static void qcow_flush(BlockDriverState *bs)
1600 {
1601     BDRVQcowState *s = bs->opaque;
1602     bdrv_flush(s->hd);
1603 }
1604 
qcow_get_info(BlockDriverState * bs,BlockDriverInfo * bdi)1605 static int qcow_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
1606 {
1607     BDRVQcowState *s = bs->opaque;
1608     bdi->cluster_size = s->cluster_size;
1609     bdi->vm_state_offset = (int64_t)s->l1_vm_state_index <<
1610         (s->cluster_bits + s->l2_bits);
1611     return 0;
1612 }
1613 
1614 /*********************************************************/
1615 /* snapshot support */
1616 
1617 /* update the refcounts of snapshots and the copied flag */
update_snapshot_refcount(BlockDriverState * bs,int64_t l1_table_offset,int l1_size,int addend)1618 static int update_snapshot_refcount(BlockDriverState *bs,
1619                                     int64_t l1_table_offset,
1620                                     int l1_size,
1621                                     int addend)
1622 {
1623     BDRVQcowState *s = bs->opaque;
1624     uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2, l1_allocated;
1625     int64_t old_offset, old_l2_offset;
1626     int l2_size, i, j, l1_modified, l2_modified, nb_csectors, refcount;
1627 
1628     l2_cache_reset(bs);
1629 
1630     l2_table = NULL;
1631     l1_table = NULL;
1632     l1_size2 = l1_size * sizeof(uint64_t);
1633     l1_allocated = 0;
1634     if (l1_table_offset != s->l1_table_offset) {
1635         l1_table = qemu_malloc(l1_size2);
1636         if (!l1_table)
1637             goto fail;
1638         l1_allocated = 1;
1639         if (bdrv_pread(s->hd, l1_table_offset,
1640                        l1_table, l1_size2) != l1_size2)
1641             goto fail;
1642         for(i = 0;i < l1_size; i++)
1643             be64_to_cpus(&l1_table[i]);
1644     } else {
1645         assert(l1_size == s->l1_size);
1646         l1_table = s->l1_table;
1647         l1_allocated = 0;
1648     }
1649 
1650     l2_size = s->l2_size * sizeof(uint64_t);
1651     l2_table = qemu_malloc(l2_size);
1652     if (!l2_table)
1653         goto fail;
1654     l1_modified = 0;
1655     for(i = 0; i < l1_size; i++) {
1656         l2_offset = l1_table[i];
1657         if (l2_offset) {
1658             old_l2_offset = l2_offset;
1659             l2_offset &= ~QCOW_OFLAG_COPIED;
1660             l2_modified = 0;
1661             if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
1662                 goto fail;
1663             for(j = 0; j < s->l2_size; j++) {
1664                 offset = be64_to_cpu(l2_table[j]);
1665                 if (offset != 0) {
1666                     old_offset = offset;
1667                     offset &= ~QCOW_OFLAG_COPIED;
1668                     if (offset & QCOW_OFLAG_COMPRESSED) {
1669                         nb_csectors = ((offset >> s->csize_shift) &
1670                                        s->csize_mask) + 1;
1671                         if (addend != 0)
1672                             update_refcount(bs, (offset & s->cluster_offset_mask) & ~511,
1673                                             nb_csectors * 512, addend);
1674                         /* compressed clusters are never modified */
1675                         refcount = 2;
1676                     } else {
1677                         if (addend != 0) {
1678                             refcount = update_cluster_refcount(bs, offset >> s->cluster_bits, addend);
1679                         } else {
1680                             refcount = get_refcount(bs, offset >> s->cluster_bits);
1681                         }
1682                     }
1683 
1684                     if (refcount == 1) {
1685                         offset |= QCOW_OFLAG_COPIED;
1686                     }
1687                     if (offset != old_offset) {
1688                         l2_table[j] = cpu_to_be64(offset);
1689                         l2_modified = 1;
1690                     }
1691                 }
1692             }
1693             if (l2_modified) {
1694                 if (bdrv_pwrite(s->hd,
1695                                 l2_offset, l2_table, l2_size) != l2_size)
1696                     goto fail;
1697             }
1698 
1699             if (addend != 0) {
1700                 refcount = update_cluster_refcount(bs, l2_offset >> s->cluster_bits, addend);
1701             } else {
1702                 refcount = get_refcount(bs, l2_offset >> s->cluster_bits);
1703             }
1704             if (refcount == 1) {
1705                 l2_offset |= QCOW_OFLAG_COPIED;
1706             }
1707             if (l2_offset != old_l2_offset) {
1708                 l1_table[i] = l2_offset;
1709                 l1_modified = 1;
1710             }
1711         }
1712     }
1713     if (l1_modified) {
1714         for(i = 0; i < l1_size; i++)
1715             cpu_to_be64s(&l1_table[i]);
1716         if (bdrv_pwrite(s->hd, l1_table_offset, l1_table,
1717                         l1_size2) != l1_size2)
1718             goto fail;
1719         for(i = 0; i < l1_size; i++)
1720             be64_to_cpus(&l1_table[i]);
1721     }
1722     if (l1_allocated)
1723         qemu_free(l1_table);
1724     qemu_free(l2_table);
1725     return 0;
1726  fail:
1727     if (l1_allocated)
1728         qemu_free(l1_table);
1729     qemu_free(l2_table);
1730     return -EIO;
1731 }
1732 
qcow_free_snapshots(BlockDriverState * bs)1733 static void qcow_free_snapshots(BlockDriverState *bs)
1734 {
1735     BDRVQcowState *s = bs->opaque;
1736     int i;
1737 
1738     for(i = 0; i < s->nb_snapshots; i++) {
1739         qemu_free(s->snapshots[i].name);
1740         qemu_free(s->snapshots[i].id_str);
1741     }
1742     qemu_free(s->snapshots);
1743     s->snapshots = NULL;
1744     s->nb_snapshots = 0;
1745 }
1746 
qcow_read_snapshots(BlockDriverState * bs)1747 static int qcow_read_snapshots(BlockDriverState *bs)
1748 {
1749     BDRVQcowState *s = bs->opaque;
1750     QCowSnapshotHeader h;
1751     QCowSnapshot *sn;
1752     int i, id_str_size, name_size;
1753     int64_t offset;
1754     uint32_t extra_data_size;
1755 
1756     offset = s->snapshots_offset;
1757     s->snapshots = qemu_mallocz(s->nb_snapshots * sizeof(QCowSnapshot));
1758     if (!s->snapshots)
1759         goto fail;
1760     for(i = 0; i < s->nb_snapshots; i++) {
1761         offset = align_offset(offset, 8);
1762         if (bdrv_pread(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1763             goto fail;
1764         offset += sizeof(h);
1765         sn = s->snapshots + i;
1766         sn->l1_table_offset = be64_to_cpu(h.l1_table_offset);
1767         sn->l1_size = be32_to_cpu(h.l1_size);
1768         sn->vm_state_size = be32_to_cpu(h.vm_state_size);
1769         sn->date_sec = be32_to_cpu(h.date_sec);
1770         sn->date_nsec = be32_to_cpu(h.date_nsec);
1771         sn->vm_clock_nsec = be64_to_cpu(h.vm_clock_nsec);
1772         extra_data_size = be32_to_cpu(h.extra_data_size);
1773 
1774         id_str_size = be16_to_cpu(h.id_str_size);
1775         name_size = be16_to_cpu(h.name_size);
1776 
1777         offset += extra_data_size;
1778 
1779         sn->id_str = qemu_malloc(id_str_size + 1);
1780         if (!sn->id_str)
1781             goto fail;
1782         if (bdrv_pread(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1783             goto fail;
1784         offset += id_str_size;
1785         sn->id_str[id_str_size] = '\0';
1786 
1787         sn->name = qemu_malloc(name_size + 1);
1788         if (!sn->name)
1789             goto fail;
1790         if (bdrv_pread(s->hd, offset, sn->name, name_size) != name_size)
1791             goto fail;
1792         offset += name_size;
1793         sn->name[name_size] = '\0';
1794     }
1795     s->snapshots_size = offset - s->snapshots_offset;
1796     return 0;
1797  fail:
1798     qcow_free_snapshots(bs);
1799     return -1;
1800 }
1801 
1802 /* add at the end of the file a new list of snapshots */
qcow_write_snapshots(BlockDriverState * bs)1803 static int qcow_write_snapshots(BlockDriverState *bs)
1804 {
1805     BDRVQcowState *s = bs->opaque;
1806     QCowSnapshot *sn;
1807     QCowSnapshotHeader h;
1808     int i, name_size, id_str_size, snapshots_size;
1809     uint64_t data64;
1810     uint32_t data32;
1811     int64_t offset, snapshots_offset;
1812 
1813     /* compute the size of the snapshots */
1814     offset = 0;
1815     for(i = 0; i < s->nb_snapshots; i++) {
1816         sn = s->snapshots + i;
1817         offset = align_offset(offset, 8);
1818         offset += sizeof(h);
1819         offset += strlen(sn->id_str);
1820         offset += strlen(sn->name);
1821     }
1822     snapshots_size = offset;
1823 
1824     snapshots_offset = alloc_clusters(bs, snapshots_size);
1825     offset = snapshots_offset;
1826 
1827     for(i = 0; i < s->nb_snapshots; i++) {
1828         sn = s->snapshots + i;
1829         memset(&h, 0, sizeof(h));
1830         h.l1_table_offset = cpu_to_be64(sn->l1_table_offset);
1831         h.l1_size = cpu_to_be32(sn->l1_size);
1832         h.vm_state_size = cpu_to_be32(sn->vm_state_size);
1833         h.date_sec = cpu_to_be32(sn->date_sec);
1834         h.date_nsec = cpu_to_be32(sn->date_nsec);
1835         h.vm_clock_nsec = cpu_to_be64(sn->vm_clock_nsec);
1836 
1837         id_str_size = strlen(sn->id_str);
1838         name_size = strlen(sn->name);
1839         h.id_str_size = cpu_to_be16(id_str_size);
1840         h.name_size = cpu_to_be16(name_size);
1841         offset = align_offset(offset, 8);
1842         if (bdrv_pwrite(s->hd, offset, &h, sizeof(h)) != sizeof(h))
1843             goto fail;
1844         offset += sizeof(h);
1845         if (bdrv_pwrite(s->hd, offset, sn->id_str, id_str_size) != id_str_size)
1846             goto fail;
1847         offset += id_str_size;
1848         if (bdrv_pwrite(s->hd, offset, sn->name, name_size) != name_size)
1849             goto fail;
1850         offset += name_size;
1851     }
1852 
1853     /* update the various header fields */
1854     data64 = cpu_to_be64(snapshots_offset);
1855     if (bdrv_pwrite(s->hd, offsetof(QCowHeader, snapshots_offset),
1856                     &data64, sizeof(data64)) != sizeof(data64))
1857         goto fail;
1858     data32 = cpu_to_be32(s->nb_snapshots);
1859     if (bdrv_pwrite(s->hd, offsetof(QCowHeader, nb_snapshots),
1860                     &data32, sizeof(data32)) != sizeof(data32))
1861         goto fail;
1862 
1863     /* free the old snapshot table */
1864     free_clusters(bs, s->snapshots_offset, s->snapshots_size);
1865     s->snapshots_offset = snapshots_offset;
1866     s->snapshots_size = snapshots_size;
1867     return 0;
1868  fail:
1869     return -1;
1870 }
1871 
find_new_snapshot_id(BlockDriverState * bs,char * id_str,int id_str_size)1872 static void find_new_snapshot_id(BlockDriverState *bs,
1873                                  char *id_str, int id_str_size)
1874 {
1875     BDRVQcowState *s = bs->opaque;
1876     QCowSnapshot *sn;
1877     int i, id, id_max = 0;
1878 
1879     for(i = 0; i < s->nb_snapshots; i++) {
1880         sn = s->snapshots + i;
1881         id = strtoul(sn->id_str, NULL, 10);
1882         if (id > id_max)
1883             id_max = id;
1884     }
1885     snprintf(id_str, id_str_size, "%d", id_max + 1);
1886 }
1887 
find_snapshot_by_id(BlockDriverState * bs,const char * id_str)1888 static int find_snapshot_by_id(BlockDriverState *bs, const char *id_str)
1889 {
1890     BDRVQcowState *s = bs->opaque;
1891     int i;
1892 
1893     for(i = 0; i < s->nb_snapshots; i++) {
1894         if (!strcmp(s->snapshots[i].id_str, id_str))
1895             return i;
1896     }
1897     return -1;
1898 }
1899 
find_snapshot_by_id_or_name(BlockDriverState * bs,const char * name)1900 static int find_snapshot_by_id_or_name(BlockDriverState *bs, const char *name)
1901 {
1902     BDRVQcowState *s = bs->opaque;
1903     int i, ret;
1904 
1905     ret = find_snapshot_by_id(bs, name);
1906     if (ret >= 0)
1907         return ret;
1908     for(i = 0; i < s->nb_snapshots; i++) {
1909         if (!strcmp(s->snapshots[i].name, name))
1910             return i;
1911     }
1912     return -1;
1913 }
1914 
1915 /* if no id is provided, a new one is constructed */
qcow_snapshot_create(BlockDriverState * bs,QEMUSnapshotInfo * sn_info)1916 static int qcow_snapshot_create(BlockDriverState *bs,
1917                                 QEMUSnapshotInfo *sn_info)
1918 {
1919     BDRVQcowState *s = bs->opaque;
1920     QCowSnapshot *snapshots1, sn1, *sn = &sn1;
1921     int i, ret;
1922     uint64_t *l1_table = NULL;
1923 
1924     memset(sn, 0, sizeof(*sn));
1925 
1926     if (sn_info->id_str[0] == '\0') {
1927         /* compute a new id */
1928         find_new_snapshot_id(bs, sn_info->id_str, sizeof(sn_info->id_str));
1929     }
1930 
1931     /* check that the ID is unique */
1932     if (find_snapshot_by_id(bs, sn_info->id_str) >= 0)
1933         return -ENOENT;
1934 
1935     sn->id_str = qemu_strdup(sn_info->id_str);
1936     if (!sn->id_str)
1937         goto fail;
1938     sn->name = qemu_strdup(sn_info->name);
1939     if (!sn->name)
1940         goto fail;
1941     sn->vm_state_size = sn_info->vm_state_size;
1942     sn->date_sec = sn_info->date_sec;
1943     sn->date_nsec = sn_info->date_nsec;
1944     sn->vm_clock_nsec = sn_info->vm_clock_nsec;
1945 
1946     ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1);
1947     if (ret < 0)
1948         goto fail;
1949 
1950     /* create the L1 table of the snapshot */
1951     sn->l1_table_offset = alloc_clusters(bs, s->l1_size * sizeof(uint64_t));
1952     sn->l1_size = s->l1_size;
1953 
1954     l1_table = qemu_malloc(s->l1_size * sizeof(uint64_t));
1955     if (!l1_table)
1956         goto fail;
1957     for(i = 0; i < s->l1_size; i++) {
1958         l1_table[i] = cpu_to_be64(s->l1_table[i]);
1959     }
1960     if (bdrv_pwrite(s->hd, sn->l1_table_offset,
1961                     l1_table, s->l1_size * sizeof(uint64_t)) !=
1962         (s->l1_size * sizeof(uint64_t)))
1963         goto fail;
1964     qemu_free(l1_table);
1965     l1_table = NULL;
1966 
1967     snapshots1 = qemu_malloc((s->nb_snapshots + 1) * sizeof(QCowSnapshot));
1968     if (!snapshots1)
1969         goto fail;
1970     memcpy(snapshots1, s->snapshots, s->nb_snapshots * sizeof(QCowSnapshot));
1971     s->snapshots = snapshots1;
1972     s->snapshots[s->nb_snapshots++] = *sn;
1973 
1974     if (qcow_write_snapshots(bs) < 0)
1975         goto fail;
1976 #ifdef DEBUG_ALLOC
1977     check_refcounts(bs);
1978 #endif
1979     return 0;
1980  fail:
1981     qemu_free(sn->name);
1982     qemu_free(l1_table);
1983     return -1;
1984 }
1985 
1986 /* copy the snapshot 'snapshot_name' into the current disk image */
qcow_snapshot_goto(BlockDriverState * bs,const char * snapshot_id)1987 static int qcow_snapshot_goto(BlockDriverState *bs,
1988                               const char *snapshot_id)
1989 {
1990     BDRVQcowState *s = bs->opaque;
1991     QCowSnapshot *sn;
1992     int i, snapshot_index, l1_size2;
1993 
1994     snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
1995     if (snapshot_index < 0)
1996         return -ENOENT;
1997     sn = &s->snapshots[snapshot_index];
1998 
1999     if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, -1) < 0)
2000         goto fail;
2001 
2002     if (grow_l1_table(bs, sn->l1_size) < 0)
2003         goto fail;
2004 
2005     s->l1_size = sn->l1_size;
2006     l1_size2 = s->l1_size * sizeof(uint64_t);
2007     /* copy the snapshot l1 table to the current l1 table */
2008     if (bdrv_pread(s->hd, sn->l1_table_offset,
2009                    s->l1_table, l1_size2) != l1_size2)
2010         goto fail;
2011     if (bdrv_pwrite(s->hd, s->l1_table_offset,
2012                     s->l1_table, l1_size2) != l1_size2)
2013         goto fail;
2014     for(i = 0;i < s->l1_size; i++) {
2015         be64_to_cpus(&s->l1_table[i]);
2016     }
2017 
2018     if (update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 1) < 0)
2019         goto fail;
2020 
2021 #ifdef DEBUG_ALLOC
2022     check_refcounts(bs);
2023 #endif
2024     return 0;
2025  fail:
2026     return -EIO;
2027 }
2028 
qcow_snapshot_delete(BlockDriverState * bs,const char * snapshot_id)2029 static int qcow_snapshot_delete(BlockDriverState *bs, const char *snapshot_id)
2030 {
2031     BDRVQcowState *s = bs->opaque;
2032     QCowSnapshot *sn;
2033     int snapshot_index, ret;
2034 
2035     snapshot_index = find_snapshot_by_id_or_name(bs, snapshot_id);
2036     if (snapshot_index < 0)
2037         return -ENOENT;
2038     sn = &s->snapshots[snapshot_index];
2039 
2040     ret = update_snapshot_refcount(bs, sn->l1_table_offset, sn->l1_size, -1);
2041     if (ret < 0)
2042         return ret;
2043     /* must update the copied flag on the current cluster offsets */
2044     ret = update_snapshot_refcount(bs, s->l1_table_offset, s->l1_size, 0);
2045     if (ret < 0)
2046         return ret;
2047     free_clusters(bs, sn->l1_table_offset, sn->l1_size * sizeof(uint64_t));
2048 
2049     qemu_free(sn->id_str);
2050     qemu_free(sn->name);
2051     memmove(sn, sn + 1, (s->nb_snapshots - snapshot_index - 1) * sizeof(*sn));
2052     s->nb_snapshots--;
2053     ret = qcow_write_snapshots(bs);
2054     if (ret < 0) {
2055         /* XXX: restore snapshot if error ? */
2056         return ret;
2057     }
2058 #ifdef DEBUG_ALLOC
2059     check_refcounts(bs);
2060 #endif
2061     return 0;
2062 }
2063 
qcow_snapshot_list(BlockDriverState * bs,QEMUSnapshotInfo ** psn_tab)2064 static int qcow_snapshot_list(BlockDriverState *bs,
2065                               QEMUSnapshotInfo **psn_tab)
2066 {
2067     BDRVQcowState *s = bs->opaque;
2068     QEMUSnapshotInfo *sn_tab, *sn_info;
2069     QCowSnapshot *sn;
2070     int i;
2071 
2072     sn_tab = qemu_mallocz(s->nb_snapshots * sizeof(QEMUSnapshotInfo));
2073     if (!sn_tab)
2074         goto fail;
2075     for(i = 0; i < s->nb_snapshots; i++) {
2076         sn_info = sn_tab + i;
2077         sn = s->snapshots + i;
2078         pstrcpy(sn_info->id_str, sizeof(sn_info->id_str),
2079                 sn->id_str);
2080         pstrcpy(sn_info->name, sizeof(sn_info->name),
2081                 sn->name);
2082         sn_info->vm_state_size = sn->vm_state_size;
2083         sn_info->date_sec = sn->date_sec;
2084         sn_info->date_nsec = sn->date_nsec;
2085         sn_info->vm_clock_nsec = sn->vm_clock_nsec;
2086     }
2087     *psn_tab = sn_tab;
2088     return s->nb_snapshots;
2089  fail:
2090     qemu_free(sn_tab);
2091     *psn_tab = NULL;
2092     return -ENOMEM;
2093 }
2094 
2095 /*********************************************************/
2096 /* refcount handling */
2097 
refcount_init(BlockDriverState * bs)2098 static int refcount_init(BlockDriverState *bs)
2099 {
2100     BDRVQcowState *s = bs->opaque;
2101     int ret, refcount_table_size2, i;
2102 
2103     s->refcount_block_cache = qemu_malloc(s->cluster_size);
2104     if (!s->refcount_block_cache)
2105         goto fail;
2106     refcount_table_size2 = s->refcount_table_size * sizeof(uint64_t);
2107     s->refcount_table = qemu_malloc(refcount_table_size2);
2108     if (!s->refcount_table)
2109         goto fail;
2110     if (s->refcount_table_size > 0) {
2111         ret = bdrv_pread(s->hd, s->refcount_table_offset,
2112                          s->refcount_table, refcount_table_size2);
2113         if (ret != refcount_table_size2)
2114             goto fail;
2115         for(i = 0; i < s->refcount_table_size; i++)
2116             be64_to_cpus(&s->refcount_table[i]);
2117     }
2118     return 0;
2119  fail:
2120     return -ENOMEM;
2121 }
2122 
refcount_close(BlockDriverState * bs)2123 static void refcount_close(BlockDriverState *bs)
2124 {
2125     BDRVQcowState *s = bs->opaque;
2126     qemu_free(s->refcount_block_cache);
2127     qemu_free(s->refcount_table);
2128 }
2129 
2130 
load_refcount_block(BlockDriverState * bs,int64_t refcount_block_offset)2131 static int load_refcount_block(BlockDriverState *bs,
2132                                int64_t refcount_block_offset)
2133 {
2134     BDRVQcowState *s = bs->opaque;
2135     int ret;
2136     ret = bdrv_pread(s->hd, refcount_block_offset, s->refcount_block_cache,
2137                      s->cluster_size);
2138     if (ret != s->cluster_size)
2139         return -EIO;
2140     s->refcount_block_cache_offset = refcount_block_offset;
2141     return 0;
2142 }
2143 
get_refcount(BlockDriverState * bs,int64_t cluster_index)2144 static int get_refcount(BlockDriverState *bs, int64_t cluster_index)
2145 {
2146     BDRVQcowState *s = bs->opaque;
2147     int refcount_table_index, block_index;
2148     int64_t refcount_block_offset;
2149 
2150     refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2151     if (refcount_table_index >= s->refcount_table_size)
2152         return 0;
2153     refcount_block_offset = s->refcount_table[refcount_table_index];
2154     if (!refcount_block_offset)
2155         return 0;
2156     if (refcount_block_offset != s->refcount_block_cache_offset) {
2157         /* better than nothing: return allocated if read error */
2158         if (load_refcount_block(bs, refcount_block_offset) < 0)
2159             return 1;
2160     }
2161     block_index = cluster_index &
2162         ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2163     return be16_to_cpu(s->refcount_block_cache[block_index]);
2164 }
2165 
2166 /* return < 0 if error */
alloc_clusters_noref(BlockDriverState * bs,int64_t size)2167 static int64_t alloc_clusters_noref(BlockDriverState *bs, int64_t size)
2168 {
2169     BDRVQcowState *s = bs->opaque;
2170     int i, nb_clusters;
2171 
2172     nb_clusters = (size + s->cluster_size - 1) >> s->cluster_bits;
2173     for(;;) {
2174         if (get_refcount(bs, s->free_cluster_index) == 0) {
2175             s->free_cluster_index++;
2176             for(i = 1; i < nb_clusters; i++) {
2177                 if (get_refcount(bs, s->free_cluster_index) != 0)
2178                     goto not_found;
2179                 s->free_cluster_index++;
2180             }
2181 #ifdef DEBUG_ALLOC2
2182             printf("alloc_clusters: size=%lld -> %lld\n",
2183                    size,
2184                    (s->free_cluster_index - nb_clusters) << s->cluster_bits);
2185 #endif
2186             return (s->free_cluster_index - nb_clusters) << s->cluster_bits;
2187         } else {
2188         not_found:
2189             s->free_cluster_index++;
2190         }
2191     }
2192 }
2193 
alloc_clusters(BlockDriverState * bs,int64_t size)2194 static int64_t alloc_clusters(BlockDriverState *bs, int64_t size)
2195 {
2196     int64_t offset;
2197 
2198     offset = alloc_clusters_noref(bs, size);
2199     update_refcount(bs, offset, size, 1);
2200     return offset;
2201 }
2202 
2203 /* only used to allocate compressed sectors. We try to allocate
2204    contiguous sectors. size must be <= cluster_size */
alloc_bytes(BlockDriverState * bs,int size)2205 static int64_t alloc_bytes(BlockDriverState *bs, int size)
2206 {
2207     BDRVQcowState *s = bs->opaque;
2208     int64_t offset, cluster_offset;
2209     int free_in_cluster;
2210 
2211     assert(size > 0 && size <= s->cluster_size);
2212     if (s->free_byte_offset == 0) {
2213         s->free_byte_offset = alloc_clusters(bs, s->cluster_size);
2214     }
2215  redo:
2216     free_in_cluster = s->cluster_size -
2217         (s->free_byte_offset & (s->cluster_size - 1));
2218     if (size <= free_in_cluster) {
2219         /* enough space in current cluster */
2220         offset = s->free_byte_offset;
2221         s->free_byte_offset += size;
2222         free_in_cluster -= size;
2223         if (free_in_cluster == 0)
2224             s->free_byte_offset = 0;
2225         if ((offset & (s->cluster_size - 1)) != 0)
2226             update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2227     } else {
2228         offset = alloc_clusters(bs, s->cluster_size);
2229         cluster_offset = s->free_byte_offset & ~(s->cluster_size - 1);
2230         if ((cluster_offset + s->cluster_size) == offset) {
2231             /* we are lucky: contiguous data */
2232             offset = s->free_byte_offset;
2233             update_cluster_refcount(bs, offset >> s->cluster_bits, 1);
2234             s->free_byte_offset += size;
2235         } else {
2236             s->free_byte_offset = offset;
2237             goto redo;
2238         }
2239     }
2240     return offset;
2241 }
2242 
free_clusters(BlockDriverState * bs,int64_t offset,int64_t size)2243 static void free_clusters(BlockDriverState *bs,
2244                           int64_t offset, int64_t size)
2245 {
2246     update_refcount(bs, offset, size, -1);
2247 }
2248 
grow_refcount_table(BlockDriverState * bs,int min_size)2249 static int grow_refcount_table(BlockDriverState *bs, int min_size)
2250 {
2251     BDRVQcowState *s = bs->opaque;
2252     int new_table_size, new_table_size2, refcount_table_clusters, i, ret;
2253     uint64_t *new_table;
2254     int64_t table_offset;
2255     uint64_t data64;
2256     uint32_t data32;
2257     int old_table_size;
2258     int64_t old_table_offset;
2259 
2260     if (min_size <= s->refcount_table_size)
2261         return 0;
2262     /* compute new table size */
2263     refcount_table_clusters = s->refcount_table_size >> (s->cluster_bits - 3);
2264     for(;;) {
2265         if (refcount_table_clusters == 0) {
2266             refcount_table_clusters = 1;
2267         } else {
2268             refcount_table_clusters = (refcount_table_clusters * 3 + 1) / 2;
2269         }
2270         new_table_size = refcount_table_clusters << (s->cluster_bits - 3);
2271         if (min_size <= new_table_size)
2272             break;
2273     }
2274 #ifdef DEBUG_ALLOC2
2275     printf("grow_refcount_table from %d to %d\n",
2276            s->refcount_table_size,
2277            new_table_size);
2278 #endif
2279     new_table_size2 = new_table_size * sizeof(uint64_t);
2280     new_table = qemu_mallocz(new_table_size2);
2281     if (!new_table)
2282         return -ENOMEM;
2283     memcpy(new_table, s->refcount_table,
2284            s->refcount_table_size * sizeof(uint64_t));
2285     for(i = 0; i < s->refcount_table_size; i++)
2286         cpu_to_be64s(&new_table[i]);
2287     /* Note: we cannot update the refcount now to avoid recursion */
2288     table_offset = alloc_clusters_noref(bs, new_table_size2);
2289     ret = bdrv_pwrite(s->hd, table_offset, new_table, new_table_size2);
2290     if (ret != new_table_size2)
2291         goto fail;
2292     for(i = 0; i < s->refcount_table_size; i++)
2293         be64_to_cpus(&new_table[i]);
2294 
2295     data64 = cpu_to_be64(table_offset);
2296     if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_offset),
2297                     &data64, sizeof(data64)) != sizeof(data64))
2298         goto fail;
2299     data32 = cpu_to_be32(refcount_table_clusters);
2300     if (bdrv_pwrite(s->hd, offsetof(QCowHeader, refcount_table_clusters),
2301                     &data32, sizeof(data32)) != sizeof(data32))
2302         goto fail;
2303     qemu_free(s->refcount_table);
2304     old_table_offset = s->refcount_table_offset;
2305     old_table_size = s->refcount_table_size;
2306     s->refcount_table = new_table;
2307     s->refcount_table_size = new_table_size;
2308     s->refcount_table_offset = table_offset;
2309 
2310     update_refcount(bs, table_offset, new_table_size2, 1);
2311     free_clusters(bs, old_table_offset, old_table_size * sizeof(uint64_t));
2312     return 0;
2313  fail:
2314     free_clusters(bs, table_offset, new_table_size2);
2315     qemu_free(new_table);
2316     return -EIO;
2317 }
2318 
2319 /* addend must be 1 or -1 */
2320 /* XXX: cache several refcount block clusters ? */
update_cluster_refcount(BlockDriverState * bs,int64_t cluster_index,int addend)2321 static int update_cluster_refcount(BlockDriverState *bs,
2322                                    int64_t cluster_index,
2323                                    int addend)
2324 {
2325     BDRVQcowState *s = bs->opaque;
2326     int64_t offset, refcount_block_offset;
2327     int ret, refcount_table_index, block_index, refcount;
2328     uint64_t data64;
2329 
2330     refcount_table_index = cluster_index >> (s->cluster_bits - REFCOUNT_SHIFT);
2331     if (refcount_table_index >= s->refcount_table_size) {
2332         if (addend < 0)
2333             return -EINVAL;
2334         ret = grow_refcount_table(bs, refcount_table_index + 1);
2335         if (ret < 0)
2336             return ret;
2337     }
2338     refcount_block_offset = s->refcount_table[refcount_table_index];
2339     if (!refcount_block_offset) {
2340         if (addend < 0)
2341             return -EINVAL;
2342         /* create a new refcount block */
2343         /* Note: we cannot update the refcount now to avoid recursion */
2344         offset = alloc_clusters_noref(bs, s->cluster_size);
2345         memset(s->refcount_block_cache, 0, s->cluster_size);
2346         ret = bdrv_pwrite(s->hd, offset, s->refcount_block_cache, s->cluster_size);
2347         if (ret != s->cluster_size)
2348             return -EINVAL;
2349         s->refcount_table[refcount_table_index] = offset;
2350         data64 = cpu_to_be64(offset);
2351         ret = bdrv_pwrite(s->hd, s->refcount_table_offset +
2352                           refcount_table_index * sizeof(uint64_t),
2353                           &data64, sizeof(data64));
2354         if (ret != sizeof(data64))
2355             return -EINVAL;
2356 
2357         refcount_block_offset = offset;
2358         s->refcount_block_cache_offset = offset;
2359         update_refcount(bs, offset, s->cluster_size, 1);
2360     } else {
2361         if (refcount_block_offset != s->refcount_block_cache_offset) {
2362             if (load_refcount_block(bs, refcount_block_offset) < 0)
2363                 return -EIO;
2364         }
2365     }
2366     /* we can update the count and save it */
2367     block_index = cluster_index &
2368         ((1 << (s->cluster_bits - REFCOUNT_SHIFT)) - 1);
2369     refcount = be16_to_cpu(s->refcount_block_cache[block_index]);
2370     refcount += addend;
2371     if (refcount < 0 || refcount > 0xffff)
2372         return -EINVAL;
2373     if (refcount == 0 && cluster_index < s->free_cluster_index) {
2374         s->free_cluster_index = cluster_index;
2375     }
2376     s->refcount_block_cache[block_index] = cpu_to_be16(refcount);
2377     if (bdrv_pwrite(s->hd,
2378                     refcount_block_offset + (block_index << REFCOUNT_SHIFT),
2379                     &s->refcount_block_cache[block_index], 2) != 2)
2380         return -EIO;
2381     return refcount;
2382 }
2383 
update_refcount(BlockDriverState * bs,int64_t offset,int64_t length,int addend)2384 static void update_refcount(BlockDriverState *bs,
2385                             int64_t offset, int64_t length,
2386                             int addend)
2387 {
2388     BDRVQcowState *s = bs->opaque;
2389     int64_t start, last, cluster_offset;
2390 
2391 #ifdef DEBUG_ALLOC2
2392     printf("update_refcount: offset=%lld size=%lld addend=%d\n",
2393            offset, length, addend);
2394 #endif
2395     if (length <= 0)
2396         return;
2397     start = offset & ~(s->cluster_size - 1);
2398     last = (offset + length - 1) & ~(s->cluster_size - 1);
2399     for(cluster_offset = start; cluster_offset <= last;
2400         cluster_offset += s->cluster_size) {
2401         update_cluster_refcount(bs, cluster_offset >> s->cluster_bits, addend);
2402     }
2403 }
2404 
2405 #ifdef DEBUG_ALLOC
inc_refcounts(BlockDriverState * bs,uint16_t * refcount_table,int refcount_table_size,int64_t offset,int64_t size)2406 static void inc_refcounts(BlockDriverState *bs,
2407                           uint16_t *refcount_table,
2408                           int refcount_table_size,
2409                           int64_t offset, int64_t size)
2410 {
2411     BDRVQcowState *s = bs->opaque;
2412     int64_t start, last, cluster_offset;
2413     int k;
2414 
2415     if (size <= 0)
2416         return;
2417 
2418     start = offset & ~(s->cluster_size - 1);
2419     last = (offset + size - 1) & ~(s->cluster_size - 1);
2420     for(cluster_offset = start; cluster_offset <= last;
2421         cluster_offset += s->cluster_size) {
2422         k = cluster_offset >> s->cluster_bits;
2423         if (k < 0 || k >= refcount_table_size) {
2424             printf("ERROR: invalid cluster offset=0x%llx\n", cluster_offset);
2425         } else {
2426             if (++refcount_table[k] == 0) {
2427                 printf("ERROR: overflow cluster offset=0x%llx\n", cluster_offset);
2428             }
2429         }
2430     }
2431 }
2432 
check_refcounts_l1(BlockDriverState * bs,uint16_t * refcount_table,int refcount_table_size,int64_t l1_table_offset,int l1_size,int check_copied)2433 static int check_refcounts_l1(BlockDriverState *bs,
2434                               uint16_t *refcount_table,
2435                               int refcount_table_size,
2436                               int64_t l1_table_offset, int l1_size,
2437                               int check_copied)
2438 {
2439     BDRVQcowState *s = bs->opaque;
2440     uint64_t *l1_table, *l2_table, l2_offset, offset, l1_size2;
2441     int l2_size, i, j, nb_csectors, refcount;
2442 
2443     l2_table = NULL;
2444     l1_size2 = l1_size * sizeof(uint64_t);
2445 
2446     inc_refcounts(bs, refcount_table, refcount_table_size,
2447                   l1_table_offset, l1_size2);
2448 
2449     l1_table = qemu_malloc(l1_size2);
2450     if (!l1_table)
2451         goto fail;
2452     if (bdrv_pread(s->hd, l1_table_offset,
2453                    l1_table, l1_size2) != l1_size2)
2454         goto fail;
2455     for(i = 0;i < l1_size; i++)
2456         be64_to_cpus(&l1_table[i]);
2457 
2458     l2_size = s->l2_size * sizeof(uint64_t);
2459     l2_table = qemu_malloc(l2_size);
2460     if (!l2_table)
2461         goto fail;
2462     for(i = 0; i < l1_size; i++) {
2463         l2_offset = l1_table[i];
2464         if (l2_offset) {
2465             if (check_copied) {
2466                 refcount = get_refcount(bs, (l2_offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2467                 if ((refcount == 1) != ((l2_offset & QCOW_OFLAG_COPIED) != 0)) {
2468                     printf("ERROR OFLAG_COPIED: l2_offset=%llx refcount=%d\n",
2469                            l2_offset, refcount);
2470                 }
2471             }
2472             l2_offset &= ~QCOW_OFLAG_COPIED;
2473             if (bdrv_pread(s->hd, l2_offset, l2_table, l2_size) != l2_size)
2474                 goto fail;
2475             for(j = 0; j < s->l2_size; j++) {
2476                 offset = be64_to_cpu(l2_table[j]);
2477                 if (offset != 0) {
2478                     if (offset & QCOW_OFLAG_COMPRESSED) {
2479                         if (offset & QCOW_OFLAG_COPIED) {
2480                             printf("ERROR: cluster %lld: copied flag must never be set for compressed clusters\n",
2481                                    offset >> s->cluster_bits);
2482                             offset &= ~QCOW_OFLAG_COPIED;
2483                         }
2484                         nb_csectors = ((offset >> s->csize_shift) &
2485                                        s->csize_mask) + 1;
2486                         offset &= s->cluster_offset_mask;
2487                         inc_refcounts(bs, refcount_table,
2488                                       refcount_table_size,
2489                                       offset & ~511, nb_csectors * 512);
2490                     } else {
2491                         if (check_copied) {
2492                             refcount = get_refcount(bs, (offset & ~QCOW_OFLAG_COPIED) >> s->cluster_bits);
2493                             if ((refcount == 1) != ((offset & QCOW_OFLAG_COPIED) != 0)) {
2494                                 printf("ERROR OFLAG_COPIED: offset=%llx refcount=%d\n",
2495                                        offset, refcount);
2496                             }
2497                         }
2498                         offset &= ~QCOW_OFLAG_COPIED;
2499                         inc_refcounts(bs, refcount_table,
2500                                       refcount_table_size,
2501                                       offset, s->cluster_size);
2502                     }
2503                 }
2504             }
2505             inc_refcounts(bs, refcount_table,
2506                           refcount_table_size,
2507                           l2_offset,
2508                           s->cluster_size);
2509         }
2510     }
2511     qemu_free(l1_table);
2512     qemu_free(l2_table);
2513     return 0;
2514  fail:
2515     printf("ERROR: I/O error in check_refcounts_l1\n");
2516     qemu_free(l1_table);
2517     qemu_free(l2_table);
2518     return -EIO;
2519 }
2520 
check_refcounts(BlockDriverState * bs)2521 static void check_refcounts(BlockDriverState *bs)
2522 {
2523     BDRVQcowState *s = bs->opaque;
2524     int64_t size;
2525     int nb_clusters, refcount1, refcount2, i;
2526     QCowSnapshot *sn;
2527     uint16_t *refcount_table;
2528 
2529     size = bdrv_getlength(s->hd);
2530     nb_clusters = (size + s->cluster_size - 1) >> s->cluster_bits;
2531     refcount_table = qemu_mallocz(nb_clusters * sizeof(uint16_t));
2532 
2533     /* header */
2534     inc_refcounts(bs, refcount_table, nb_clusters,
2535                   0, s->cluster_size);
2536 
2537     check_refcounts_l1(bs, refcount_table, nb_clusters,
2538                        s->l1_table_offset, s->l1_size, 1);
2539 
2540     /* snapshots */
2541     for(i = 0; i < s->nb_snapshots; i++) {
2542         sn = s->snapshots + i;
2543         check_refcounts_l1(bs, refcount_table, nb_clusters,
2544                            sn->l1_table_offset, sn->l1_size, 0);
2545     }
2546     inc_refcounts(bs, refcount_table, nb_clusters,
2547                   s->snapshots_offset, s->snapshots_size);
2548 
2549     /* refcount data */
2550     inc_refcounts(bs, refcount_table, nb_clusters,
2551                   s->refcount_table_offset,
2552                   s->refcount_table_size * sizeof(uint64_t));
2553     for(i = 0; i < s->refcount_table_size; i++) {
2554         int64_t offset;
2555         offset = s->refcount_table[i];
2556         if (offset != 0) {
2557             inc_refcounts(bs, refcount_table, nb_clusters,
2558                           offset, s->cluster_size);
2559         }
2560     }
2561 
2562     /* compare ref counts */
2563     for(i = 0; i < nb_clusters; i++) {
2564         refcount1 = get_refcount(bs, i);
2565         refcount2 = refcount_table[i];
2566         if (refcount1 != refcount2)
2567             printf("ERROR cluster %d refcount=%d reference=%d\n",
2568                    i, refcount1, refcount2);
2569     }
2570 
2571     qemu_free(refcount_table);
2572 }
2573 
2574 #if 0
2575 static void dump_refcounts(BlockDriverState *bs)
2576 {
2577     BDRVQcowState *s = bs->opaque;
2578     int64_t nb_clusters, k, k1, size;
2579     int refcount;
2580 
2581     size = bdrv_getlength(s->hd);
2582     nb_clusters = (size + s->cluster_size - 1) >> s->cluster_bits;
2583     for(k = 0; k < nb_clusters;) {
2584         k1 = k;
2585         refcount = get_refcount(bs, k);
2586         k++;
2587         while (k < nb_clusters && get_refcount(bs, k) == refcount)
2588             k++;
2589         printf("%lld: refcount=%d nb=%lld\n", k, refcount, k - k1);
2590     }
2591 }
2592 #endif
2593 #endif
2594 
2595 BlockDriver bdrv_qcow2 = {
2596     "qcow2",
2597     sizeof(BDRVQcowState),
2598     qcow_probe,
2599     qcow_open,
2600     NULL,
2601     NULL,
2602     qcow_close,
2603     qcow_create,
2604     qcow_flush,
2605     qcow_is_allocated,
2606     qcow_set_key,
2607     qcow_make_empty,
2608 
2609     .bdrv_aio_read = qcow_aio_read,
2610     .bdrv_aio_write = qcow_aio_write,
2611     .bdrv_aio_cancel = qcow_aio_cancel,
2612     .aiocb_size = sizeof(QCowAIOCB),
2613     .bdrv_write_compressed = qcow_write_compressed,
2614 
2615     .bdrv_snapshot_create = qcow_snapshot_create,
2616     .bdrv_snapshot_goto = qcow_snapshot_goto,
2617     .bdrv_snapshot_delete = qcow_snapshot_delete,
2618     .bdrv_snapshot_list = qcow_snapshot_list,
2619     .bdrv_get_info = qcow_get_info,
2620 };
2621