![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to qcow2-cluster.c CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [Qemu by Fabrice Bellard] / qemu / block|
Return to qcow2-cluster.c CVS log | Up to [Qemu by Fabrice Bellard] / qemu / block |
1.1 root 1: /*
2: * Block driver for the QCOW version 2 format
3: *
4: * Copyright (c) 2004-2006 Fabrice Bellard
5: *
6: * Permission is hereby granted, free of charge, to any person obtaining a copy
7: * of this software and associated documentation files (the "Software"), to deal
8: * in the Software without restriction, including without limitation the rights
9: * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10: * copies of the Software, and to permit persons to whom the Software is
11: * furnished to do so, subject to the following conditions:
12: *
13: * The above copyright notice and this permission notice shall be included in
14: * all copies or substantial portions of the Software.
15: *
16: * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17: * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18: * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19: * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20: * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21: * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22: * THE SOFTWARE.
23: */
24:
25: #include <zlib.h>
26:
27: #include "qemu-common.h"
28: #include "block_int.h"
29: #include "block/qcow2.h"
30:
31: int qcow2_grow_l1_table(BlockDriverState *bs, int min_size)
32: {
33: BDRVQcowState *s = bs->opaque;
34: int new_l1_size, new_l1_size2, ret, i;
35: uint64_t *new_l1_table;
36: uint64_t new_l1_table_offset;
37: uint8_t data[12];
38:
39: new_l1_size = s->l1_size;
40: if (min_size <= new_l1_size)
41: return 0;
1.1.1.2 ! root 42: if (new_l1_size == 0) {
! 43: new_l1_size = 1;
! 44: }
1.1 root 45: while (min_size > new_l1_size) {
46: new_l1_size = (new_l1_size * 3 + 1) / 2;
47: }
48: #ifdef DEBUG_ALLOC2
49: printf("grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
50: #endif
51:
52: new_l1_size2 = sizeof(uint64_t) * new_l1_size;
53: new_l1_table = qemu_mallocz(align_offset(new_l1_size2, 512));
54: memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
55:
56: /* write new table (align to cluster) */
57: new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
58:
59: for(i = 0; i < s->l1_size; i++)
60: new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
61: ret = bdrv_pwrite(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
62: if (ret != new_l1_size2)
63: goto fail;
64: for(i = 0; i < s->l1_size; i++)
65: new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
66:
67: /* set new table */
68: cpu_to_be32w((uint32_t*)data, new_l1_size);
69: cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
70: if (bdrv_pwrite(s->hd, offsetof(QCowHeader, l1_size), data,
71: sizeof(data)) != sizeof(data))
72: goto fail;
73: qemu_free(s->l1_table);
74: qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
75: s->l1_table_offset = new_l1_table_offset;
76: s->l1_table = new_l1_table;
77: s->l1_size = new_l1_size;
78: return 0;
79: fail:
80: qemu_free(s->l1_table);
81: return -EIO;
82: }
83:
84: void qcow2_l2_cache_reset(BlockDriverState *bs)
85: {
86: BDRVQcowState *s = bs->opaque;
87:
88: memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
89: memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
90: memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
91: }
92:
93: static inline int l2_cache_new_entry(BlockDriverState *bs)
94: {
95: BDRVQcowState *s = bs->opaque;
96: uint32_t min_count;
97: int min_index, i;
98:
99: /* find a new entry in the least used one */
100: min_index = 0;
101: min_count = 0xffffffff;
102: for(i = 0; i < L2_CACHE_SIZE; i++) {
103: if (s->l2_cache_counts[i] < min_count) {
104: min_count = s->l2_cache_counts[i];
105: min_index = i;
106: }
107: }
108: return min_index;
109: }
110:
111: /*
112: * seek_l2_table
113: *
114: * seek l2_offset in the l2_cache table
115: * if not found, return NULL,
116: * if found,
117: * increments the l2 cache hit count of the entry,
118: * if counter overflow, divide by two all counters
119: * return the pointer to the l2 cache entry
120: *
121: */
122:
123: static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
124: {
125: int i, j;
126:
127: for(i = 0; i < L2_CACHE_SIZE; i++) {
128: if (l2_offset == s->l2_cache_offsets[i]) {
129: /* increment the hit count */
130: if (++s->l2_cache_counts[i] == 0xffffffff) {
131: for(j = 0; j < L2_CACHE_SIZE; j++) {
132: s->l2_cache_counts[j] >>= 1;
133: }
134: }
135: return s->l2_cache + (i << s->l2_bits);
136: }
137: }
138: return NULL;
139: }
140:
141: /*
142: * l2_load
143: *
144: * Loads a L2 table into memory. If the table is in the cache, the cache
145: * is used; otherwise the L2 table is loaded from the image file.
146: *
147: * Returns a pointer to the L2 table on success, or NULL if the read from
148: * the image file failed.
149: */
150:
151: static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
152: {
153: BDRVQcowState *s = bs->opaque;
154: int min_index;
155: uint64_t *l2_table;
156:
157: /* seek if the table for the given offset is in the cache */
158:
159: l2_table = seek_l2_table(s, l2_offset);
160: if (l2_table != NULL)
161: return l2_table;
162:
163: /* not found: load a new entry in the least used one */
164:
165: min_index = l2_cache_new_entry(bs);
166: l2_table = s->l2_cache + (min_index << s->l2_bits);
167: if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
168: s->l2_size * sizeof(uint64_t))
169: return NULL;
170: s->l2_cache_offsets[min_index] = l2_offset;
171: s->l2_cache_counts[min_index] = 1;
172:
173: return l2_table;
174: }
175:
176: /*
177: * Writes one sector of the L1 table to the disk (can't update single entries
178: * and we really don't want bdrv_pread to perform a read-modify-write)
179: */
180: #define L1_ENTRIES_PER_SECTOR (512 / 8)
181: static int write_l1_entry(BDRVQcowState *s, int l1_index)
182: {
183: uint64_t buf[L1_ENTRIES_PER_SECTOR];
184: int l1_start_index;
185: int i;
186:
187: l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
188: for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
189: buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
190: }
191:
192: if (bdrv_pwrite(s->hd, s->l1_table_offset + 8 * l1_start_index,
193: buf, sizeof(buf)) != sizeof(buf))
194: {
195: return -1;
196: }
197:
198: return 0;
199: }
200:
201: /*
202: * l2_allocate
203: *
204: * Allocate a new l2 entry in the file. If l1_index points to an already
205: * used entry in the L2 table (i.e. we are doing a copy on write for the L2
206: * table) copy the contents of the old L2 table into the newly allocated one.
207: * Otherwise the new table is initialized with zeros.
208: *
209: */
210:
211: static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
212: {
213: BDRVQcowState *s = bs->opaque;
214: int min_index;
215: uint64_t old_l2_offset;
216: uint64_t *l2_table, l2_offset;
217:
218: old_l2_offset = s->l1_table[l1_index];
219:
220: /* allocate a new l2 entry */
221:
222: l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
223:
224: /* update the L1 entry */
225:
226: s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
227: if (write_l1_entry(s, l1_index) < 0) {
228: return NULL;
229: }
230:
231: /* allocate a new entry in the l2 cache */
232:
233: min_index = l2_cache_new_entry(bs);
234: l2_table = s->l2_cache + (min_index << s->l2_bits);
235:
236: if (old_l2_offset == 0) {
237: /* if there was no old l2 table, clear the new table */
238: memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
239: } else {
240: /* if there was an old l2 table, read it from the disk */
241: if (bdrv_pread(s->hd, old_l2_offset,
242: l2_table, s->l2_size * sizeof(uint64_t)) !=
243: s->l2_size * sizeof(uint64_t))
244: return NULL;
245: }
246: /* write the l2 table to the file */
247: if (bdrv_pwrite(s->hd, l2_offset,
248: l2_table, s->l2_size * sizeof(uint64_t)) !=
249: s->l2_size * sizeof(uint64_t))
250: return NULL;
251:
252: /* update the l2 cache entry */
253:
254: s->l2_cache_offsets[min_index] = l2_offset;
255: s->l2_cache_counts[min_index] = 1;
256:
257: return l2_table;
258: }
259:
260: static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
261: uint64_t *l2_table, uint64_t start, uint64_t mask)
262: {
263: int i;
264: uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
265:
266: if (!offset)
267: return 0;
268:
269: for (i = start; i < start + nb_clusters; i++)
1.1.1.2 ! root 270: if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
1.1 root 271: break;
272:
273: return (i - start);
274: }
275:
276: static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
277: {
278: int i = 0;
279:
280: while(nb_clusters-- && l2_table[i] == 0)
281: i++;
282:
283: return i;
284: }
285:
286: /* The crypt function is compatible with the linux cryptoloop
287: algorithm for < 4 GB images. NOTE: out_buf == in_buf is
288: supported */
289: void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
290: uint8_t *out_buf, const uint8_t *in_buf,
291: int nb_sectors, int enc,
292: const AES_KEY *key)
293: {
294: union {
295: uint64_t ll[2];
296: uint8_t b[16];
297: } ivec;
298: int i;
299:
300: for(i = 0; i < nb_sectors; i++) {
301: ivec.ll[0] = cpu_to_le64(sector_num);
302: ivec.ll[1] = 0;
303: AES_cbc_encrypt(in_buf, out_buf, 512, key,
304: ivec.b, enc);
305: sector_num++;
306: in_buf += 512;
307: out_buf += 512;
308: }
309: }
310:
311:
312: static int qcow_read(BlockDriverState *bs, int64_t sector_num,
313: uint8_t *buf, int nb_sectors)
314: {
315: BDRVQcowState *s = bs->opaque;
316: int ret, index_in_cluster, n, n1;
317: uint64_t cluster_offset;
318:
319: while (nb_sectors > 0) {
320: n = nb_sectors;
321: cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, &n);
322: index_in_cluster = sector_num & (s->cluster_sectors - 1);
323: if (!cluster_offset) {
324: if (bs->backing_hd) {
325: /* read from the base image */
326: n1 = qcow2_backing_read1(bs->backing_hd, sector_num, buf, n);
327: if (n1 > 0) {
328: ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
329: if (ret < 0)
330: return -1;
331: }
332: } else {
333: memset(buf, 0, 512 * n);
334: }
335: } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
336: if (qcow2_decompress_cluster(s, cluster_offset) < 0)
337: return -1;
338: memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
339: } else {
340: ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
341: if (ret != n * 512)
342: return -1;
343: if (s->crypt_method) {
344: qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0,
345: &s->aes_decrypt_key);
346: }
347: }
348: nb_sectors -= n;
349: sector_num += n;
350: buf += n * 512;
351: }
352: return 0;
353: }
354:
355: static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
356: uint64_t cluster_offset, int n_start, int n_end)
357: {
358: BDRVQcowState *s = bs->opaque;
359: int n, ret;
360:
361: n = n_end - n_start;
362: if (n <= 0)
363: return 0;
364: ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
365: if (ret < 0)
366: return ret;
367: if (s->crypt_method) {
368: qcow2_encrypt_sectors(s, start_sect + n_start,
369: s->cluster_data,
370: s->cluster_data, n, 1,
371: &s->aes_encrypt_key);
372: }
373: ret = bdrv_write(s->hd, (cluster_offset >> 9) + n_start,
374: s->cluster_data, n);
375: if (ret < 0)
376: return ret;
377: return 0;
378: }
379:
380:
381: /*
382: * get_cluster_offset
383: *
384: * For a given offset of the disk image, return cluster offset in
385: * qcow2 file.
386: *
387: * on entry, *num is the number of contiguous clusters we'd like to
388: * access following offset.
389: *
390: * on exit, *num is the number of contiguous clusters we can read.
391: *
392: * Return 1, if the offset is found
393: * Return 0, otherwise.
394: *
395: */
396:
397: uint64_t qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
398: int *num)
399: {
400: BDRVQcowState *s = bs->opaque;
1.1.1.2 ! root 401: unsigned int l1_index, l2_index;
1.1 root 402: uint64_t l2_offset, *l2_table, cluster_offset;
403: int l1_bits, c;
1.1.1.2 ! root 404: unsigned int index_in_cluster, nb_clusters;
! 405: uint64_t nb_available, nb_needed;
1.1 root 406:
407: index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
408: nb_needed = *num + index_in_cluster;
409:
410: l1_bits = s->l2_bits + s->cluster_bits;
411:
412: /* compute how many bytes there are between the offset and
413: * the end of the l1 entry
414: */
415:
1.1.1.2 ! root 416: nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
1.1 root 417:
418: /* compute the number of available sectors */
419:
420: nb_available = (nb_available >> 9) + index_in_cluster;
421:
422: if (nb_needed > nb_available) {
423: nb_needed = nb_available;
424: }
425:
426: cluster_offset = 0;
427:
428: /* seek the the l2 offset in the l1 table */
429:
430: l1_index = offset >> l1_bits;
431: if (l1_index >= s->l1_size)
432: goto out;
433:
434: l2_offset = s->l1_table[l1_index];
435:
436: /* seek the l2 table of the given l2 offset */
437:
438: if (!l2_offset)
439: goto out;
440:
441: /* load the l2 table in memory */
442:
443: l2_offset &= ~QCOW_OFLAG_COPIED;
444: l2_table = l2_load(bs, l2_offset);
445: if (l2_table == NULL)
446: return 0;
447:
448: /* find the cluster offset for the given disk offset */
449:
450: l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
451: cluster_offset = be64_to_cpu(l2_table[l2_index]);
452: nb_clusters = size_to_clusters(s, nb_needed << 9);
453:
454: if (!cluster_offset) {
455: /* how many empty clusters ? */
456: c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
457: } else {
458: /* how many allocated clusters ? */
459: c = count_contiguous_clusters(nb_clusters, s->cluster_size,
460: &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
461: }
462:
463: nb_available = (c * s->cluster_sectors);
464: out:
465: if (nb_available > nb_needed)
466: nb_available = nb_needed;
467:
468: *num = nb_available - index_in_cluster;
469:
470: return cluster_offset & ~QCOW_OFLAG_COPIED;
471: }
472:
473: /*
474: * get_cluster_table
475: *
476: * for a given disk offset, load (and allocate if needed)
477: * the l2 table.
478: *
479: * the l2 table offset in the qcow2 file and the cluster index
480: * in the l2 table are given to the caller.
481: *
482: */
483:
484: static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
485: uint64_t **new_l2_table,
486: uint64_t *new_l2_offset,
487: int *new_l2_index)
488: {
489: BDRVQcowState *s = bs->opaque;
1.1.1.2 ! root 490: unsigned int l1_index, l2_index;
1.1 root 491: uint64_t l2_offset, *l2_table;
1.1.1.2 ! root 492: int ret;
1.1 root 493:
494: /* seek the the l2 offset in the l1 table */
495:
496: l1_index = offset >> (s->l2_bits + s->cluster_bits);
497: if (l1_index >= s->l1_size) {
498: ret = qcow2_grow_l1_table(bs, l1_index + 1);
499: if (ret < 0)
500: return 0;
501: }
502: l2_offset = s->l1_table[l1_index];
503:
504: /* seek the l2 table of the given l2 offset */
505:
506: if (l2_offset & QCOW_OFLAG_COPIED) {
507: /* load the l2 table in memory */
508: l2_offset &= ~QCOW_OFLAG_COPIED;
509: l2_table = l2_load(bs, l2_offset);
510: if (l2_table == NULL)
511: return 0;
512: } else {
513: if (l2_offset)
514: qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
515: l2_table = l2_allocate(bs, l1_index);
516: if (l2_table == NULL)
517: return 0;
518: l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
519: }
520:
521: /* find the cluster offset for the given disk offset */
522:
523: l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
524:
525: *new_l2_table = l2_table;
526: *new_l2_offset = l2_offset;
527: *new_l2_index = l2_index;
528:
529: return 1;
530: }
531:
532: /*
533: * alloc_compressed_cluster_offset
534: *
535: * For a given offset of the disk image, return cluster offset in
536: * qcow2 file.
537: *
538: * If the offset is not found, allocate a new compressed cluster.
539: *
540: * Return the cluster offset if successful,
541: * Return 0, otherwise.
542: *
543: */
544:
545: uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
546: uint64_t offset,
547: int compressed_size)
548: {
549: BDRVQcowState *s = bs->opaque;
550: int l2_index, ret;
551: uint64_t l2_offset, *l2_table, cluster_offset;
552: int nb_csectors;
553:
554: ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
555: if (ret == 0)
556: return 0;
557:
558: cluster_offset = be64_to_cpu(l2_table[l2_index]);
559: if (cluster_offset & QCOW_OFLAG_COPIED)
560: return cluster_offset & ~QCOW_OFLAG_COPIED;
561:
562: if (cluster_offset)
563: qcow2_free_any_clusters(bs, cluster_offset, 1);
564:
565: cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
566: nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
567: (cluster_offset >> 9);
568:
569: cluster_offset |= QCOW_OFLAG_COMPRESSED |
570: ((uint64_t)nb_csectors << s->csize_shift);
571:
572: /* update L2 table */
573:
574: /* compressed clusters never have the copied flag */
575:
576: l2_table[l2_index] = cpu_to_be64(cluster_offset);
577: if (bdrv_pwrite(s->hd,
578: l2_offset + l2_index * sizeof(uint64_t),
579: l2_table + l2_index,
580: sizeof(uint64_t)) != sizeof(uint64_t))
581: return 0;
582:
583: return cluster_offset;
584: }
585:
586: /*
587: * Write L2 table updates to disk, writing whole sectors to avoid a
588: * read-modify-write in bdrv_pwrite
589: */
590: #define L2_ENTRIES_PER_SECTOR (512 / 8)
591: static int write_l2_entries(BDRVQcowState *s, uint64_t *l2_table,
592: uint64_t l2_offset, int l2_index, int num)
593: {
594: int l2_start_index = l2_index & ~(L1_ENTRIES_PER_SECTOR - 1);
595: int start_offset = (8 * l2_index) & ~511;
596: int end_offset = (8 * (l2_index + num) + 511) & ~511;
597: size_t len = end_offset - start_offset;
598:
599: if (bdrv_pwrite(s->hd, l2_offset + start_offset, &l2_table[l2_start_index],
600: len) != len)
601: {
602: return -1;
603: }
604:
605: return 0;
606: }
607:
608: int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, uint64_t cluster_offset,
609: QCowL2Meta *m)
610: {
611: BDRVQcowState *s = bs->opaque;
612: int i, j = 0, l2_index, ret;
613: uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
614:
615: if (m->nb_clusters == 0)
616: return 0;
617:
618: old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
619:
620: /* copy content of unmodified sectors */
621: start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
622: if (m->n_start) {
623: ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
624: if (ret < 0)
625: goto err;
626: }
627:
628: if (m->nb_available & (s->cluster_sectors - 1)) {
629: uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
630: ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
631: m->nb_available - end, s->cluster_sectors);
632: if (ret < 0)
633: goto err;
634: }
635:
636: ret = -EIO;
637: /* update L2 table */
638: if (!get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index))
639: goto err;
640:
641: for (i = 0; i < m->nb_clusters; i++) {
642: /* if two concurrent writes happen to the same unallocated cluster
643: * each write allocates separate cluster and writes data concurrently.
644: * The first one to complete updates l2 table with pointer to its
645: * cluster the second one has to do RMW (which is done above by
646: * copy_sectors()), update l2 table with its cluster pointer and free
647: * old cluster. This is what this loop does */
648: if(l2_table[l2_index + i] != 0)
649: old_cluster[j++] = l2_table[l2_index + i];
650:
651: l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
652: (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
653: }
654:
655: if (write_l2_entries(s, l2_table, l2_offset, l2_index, m->nb_clusters) < 0) {
656: ret = -1;
657: goto err;
658: }
659:
660: for (i = 0; i < j; i++)
661: qcow2_free_any_clusters(bs,
662: be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1);
663:
664: ret = 0;
665: err:
666: qemu_free(old_cluster);
667: return ret;
668: }
669:
670: /*
671: * alloc_cluster_offset
672: *
673: * For a given offset of the disk image, return cluster offset in
674: * qcow2 file.
675: *
676: * If the offset is not found, allocate a new cluster.
677: *
678: * Return the cluster offset if successful,
679: * Return 0, otherwise.
680: *
681: */
682:
683: uint64_t qcow2_alloc_cluster_offset(BlockDriverState *bs,
684: uint64_t offset,
685: int n_start, int n_end,
686: int *num, QCowL2Meta *m)
687: {
688: BDRVQcowState *s = bs->opaque;
689: int l2_index, ret;
690: uint64_t l2_offset, *l2_table, cluster_offset;
1.1.1.2 ! root 691: unsigned int nb_clusters, i = 0;
1.1 root 692: QCowL2Meta *old_alloc;
693:
694: ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
695: if (ret == 0)
696: return 0;
697:
698: nb_clusters = size_to_clusters(s, n_end << 9);
699:
700: nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
701:
702: cluster_offset = be64_to_cpu(l2_table[l2_index]);
703:
704: /* We keep all QCOW_OFLAG_COPIED clusters */
705:
706: if (cluster_offset & QCOW_OFLAG_COPIED) {
707: nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
708: &l2_table[l2_index], 0, 0);
709:
710: cluster_offset &= ~QCOW_OFLAG_COPIED;
711: m->nb_clusters = 0;
712:
713: goto out;
714: }
715:
716: /* for the moment, multiple compressed clusters are not managed */
717:
718: if (cluster_offset & QCOW_OFLAG_COMPRESSED)
719: nb_clusters = 1;
720:
721: /* how many available clusters ? */
722:
723: while (i < nb_clusters) {
724: i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
725: &l2_table[l2_index], i, 0);
726:
727: if(be64_to_cpu(l2_table[l2_index + i]))
728: break;
729:
730: i += count_contiguous_free_clusters(nb_clusters - i,
731: &l2_table[l2_index + i]);
732:
733: cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
734:
735: if ((cluster_offset & QCOW_OFLAG_COPIED) ||
736: (cluster_offset & QCOW_OFLAG_COMPRESSED))
737: break;
738: }
739: nb_clusters = i;
740:
741: /*
742: * Check if there already is an AIO write request in flight which allocates
743: * the same cluster. In this case we need to wait until the previous
744: * request has completed and updated the L2 table accordingly.
745: */
1.1.1.2 ! root 746: QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
1.1 root 747:
748: uint64_t end_offset = offset + nb_clusters * s->cluster_size;
749: uint64_t old_offset = old_alloc->offset;
750: uint64_t old_end_offset = old_alloc->offset +
751: old_alloc->nb_clusters * s->cluster_size;
752:
753: if (end_offset < old_offset || offset > old_end_offset) {
754: /* No intersection */
755: } else {
756: if (offset < old_offset) {
757: /* Stop at the start of a running allocation */
758: nb_clusters = (old_offset - offset) >> s->cluster_bits;
759: } else {
760: nb_clusters = 0;
761: }
762:
763: if (nb_clusters == 0) {
764: /* Set dependency and wait for a callback */
765: m->depends_on = old_alloc;
766: m->nb_clusters = 0;
767: *num = 0;
768: return 0;
769: }
770: }
771: }
772:
773: if (!nb_clusters) {
774: abort();
775: }
776:
1.1.1.2 ! root 777: QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
1.1 root 778:
779: /* allocate a new cluster */
780:
781: cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size);
782:
783: /* save info needed for meta data update */
784: m->offset = offset;
785: m->n_start = n_start;
786: m->nb_clusters = nb_clusters;
787:
788: out:
789: m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
790:
791: *num = m->nb_available - n_start;
792:
793: return cluster_offset;
794: }
795:
796: static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
797: const uint8_t *buf, int buf_size)
798: {
799: z_stream strm1, *strm = &strm1;
800: int ret, out_len;
801:
802: memset(strm, 0, sizeof(*strm));
803:
804: strm->next_in = (uint8_t *)buf;
805: strm->avail_in = buf_size;
806: strm->next_out = out_buf;
807: strm->avail_out = out_buf_size;
808:
809: ret = inflateInit2(strm, -12);
810: if (ret != Z_OK)
811: return -1;
812: ret = inflate(strm, Z_FINISH);
813: out_len = strm->next_out - out_buf;
814: if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
815: out_len != out_buf_size) {
816: inflateEnd(strm);
817: return -1;
818: }
819: inflateEnd(strm);
820: return 0;
821: }
822:
823: int qcow2_decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
824: {
825: int ret, csize, nb_csectors, sector_offset;
826: uint64_t coffset;
827:
828: coffset = cluster_offset & s->cluster_offset_mask;
829: if (s->cluster_cache_offset != coffset) {
830: nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
831: sector_offset = coffset & 511;
832: csize = nb_csectors * 512 - sector_offset;
833: ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
834: if (ret < 0) {
835: return -1;
836: }
837: if (decompress_buffer(s->cluster_cache, s->cluster_size,
838: s->cluster_data + sector_offset, csize) < 0) {
839: return -1;
840: }
841: s->cluster_cache_offset = coffset;
842: }
843: return 0;
844: }
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.