![[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;
1.1.1.3 root 36: int64_t new_l1_table_offset;
1.1 root 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);
1.1.1.3 root 58: if (new_l1_table_offset < 0) {
59: qemu_free(new_l1_table);
60: return new_l1_table_offset;
61: }
1.1 root 62:
63: for(i = 0; i < s->l1_size; i++)
64: new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
1.1.1.5 ! root 65: ret = bdrv_pwrite_sync(s->hd, new_l1_table_offset, new_l1_table, new_l1_size2);
! 66: if (ret < 0)
1.1 root 67: goto fail;
68: for(i = 0; i < s->l1_size; i++)
69: new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
70:
71: /* set new table */
72: cpu_to_be32w((uint32_t*)data, new_l1_size);
73: cpu_to_be64w((uint64_t*)(data + 4), new_l1_table_offset);
1.1.1.5 ! root 74: ret = bdrv_pwrite_sync(s->hd, offsetof(QCowHeader, l1_size), data,sizeof(data));
! 75: if (ret < 0) {
1.1 root 76: goto fail;
1.1.1.3 root 77: }
1.1 root 78: qemu_free(s->l1_table);
79: qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
80: s->l1_table_offset = new_l1_table_offset;
81: s->l1_table = new_l1_table;
82: s->l1_size = new_l1_size;
83: return 0;
84: fail:
1.1.1.3 root 85: qemu_free(new_l1_table);
86: qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2);
1.1.1.5 ! root 87: return ret;
1.1 root 88: }
89:
90: void qcow2_l2_cache_reset(BlockDriverState *bs)
91: {
92: BDRVQcowState *s = bs->opaque;
93:
94: memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
95: memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
96: memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));
97: }
98:
99: static inline int l2_cache_new_entry(BlockDriverState *bs)
100: {
101: BDRVQcowState *s = bs->opaque;
102: uint32_t min_count;
103: int min_index, i;
104:
105: /* find a new entry in the least used one */
106: min_index = 0;
107: min_count = 0xffffffff;
108: for(i = 0; i < L2_CACHE_SIZE; i++) {
109: if (s->l2_cache_counts[i] < min_count) {
110: min_count = s->l2_cache_counts[i];
111: min_index = i;
112: }
113: }
114: return min_index;
115: }
116:
117: /*
118: * seek_l2_table
119: *
120: * seek l2_offset in the l2_cache table
121: * if not found, return NULL,
122: * if found,
123: * increments the l2 cache hit count of the entry,
124: * if counter overflow, divide by two all counters
125: * return the pointer to the l2 cache entry
126: *
127: */
128:
129: static uint64_t *seek_l2_table(BDRVQcowState *s, uint64_t l2_offset)
130: {
131: int i, j;
132:
133: for(i = 0; i < L2_CACHE_SIZE; i++) {
134: if (l2_offset == s->l2_cache_offsets[i]) {
135: /* increment the hit count */
136: if (++s->l2_cache_counts[i] == 0xffffffff) {
137: for(j = 0; j < L2_CACHE_SIZE; j++) {
138: s->l2_cache_counts[j] >>= 1;
139: }
140: }
141: return s->l2_cache + (i << s->l2_bits);
142: }
143: }
144: return NULL;
145: }
146:
147: /*
148: * l2_load
149: *
150: * Loads a L2 table into memory. If the table is in the cache, the cache
151: * is used; otherwise the L2 table is loaded from the image file.
152: *
153: * Returns a pointer to the L2 table on success, or NULL if the read from
154: * the image file failed.
155: */
156:
157: static uint64_t *l2_load(BlockDriverState *bs, uint64_t l2_offset)
158: {
159: BDRVQcowState *s = bs->opaque;
160: int min_index;
161: uint64_t *l2_table;
162:
163: /* seek if the table for the given offset is in the cache */
164:
165: l2_table = seek_l2_table(s, l2_offset);
166: if (l2_table != NULL)
167: return l2_table;
168:
169: /* not found: load a new entry in the least used one */
170:
171: min_index = l2_cache_new_entry(bs);
172: l2_table = s->l2_cache + (min_index << s->l2_bits);
173: if (bdrv_pread(s->hd, l2_offset, l2_table, s->l2_size * sizeof(uint64_t)) !=
174: s->l2_size * sizeof(uint64_t))
175: return NULL;
176: s->l2_cache_offsets[min_index] = l2_offset;
177: s->l2_cache_counts[min_index] = 1;
178:
179: return l2_table;
180: }
181:
182: /*
183: * Writes one sector of the L1 table to the disk (can't update single entries
184: * and we really don't want bdrv_pread to perform a read-modify-write)
185: */
186: #define L1_ENTRIES_PER_SECTOR (512 / 8)
187: static int write_l1_entry(BDRVQcowState *s, int l1_index)
188: {
189: uint64_t buf[L1_ENTRIES_PER_SECTOR];
190: int l1_start_index;
1.1.1.5 ! root 191: int i, ret;
1.1 root 192:
193: l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
194: for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
195: buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
196: }
197:
1.1.1.5 ! root 198: ret = bdrv_pwrite_sync(s->hd, s->l1_table_offset + 8 * l1_start_index,
! 199: buf, sizeof(buf));
! 200: if (ret < 0) {
! 201: return ret;
1.1 root 202: }
203:
204: return 0;
205: }
206:
207: /*
208: * l2_allocate
209: *
210: * Allocate a new l2 entry in the file. If l1_index points to an already
211: * used entry in the L2 table (i.e. we are doing a copy on write for the L2
212: * table) copy the contents of the old L2 table into the newly allocated one.
213: * Otherwise the new table is initialized with zeros.
214: *
215: */
216:
217: static uint64_t *l2_allocate(BlockDriverState *bs, int l1_index)
218: {
219: BDRVQcowState *s = bs->opaque;
220: int min_index;
221: uint64_t old_l2_offset;
1.1.1.3 root 222: uint64_t *l2_table;
223: int64_t l2_offset;
1.1.1.5 ! root 224: int ret;
1.1 root 225:
226: old_l2_offset = s->l1_table[l1_index];
227:
228: /* allocate a new l2 entry */
229:
230: l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
1.1.1.3 root 231: if (l2_offset < 0) {
232: return NULL;
233: }
1.1 root 234:
235: /* allocate a new entry in the l2 cache */
236:
237: min_index = l2_cache_new_entry(bs);
238: l2_table = s->l2_cache + (min_index << s->l2_bits);
239:
240: if (old_l2_offset == 0) {
241: /* if there was no old l2 table, clear the new table */
242: memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
243: } else {
244: /* if there was an old l2 table, read it from the disk */
245: if (bdrv_pread(s->hd, old_l2_offset,
246: l2_table, s->l2_size * sizeof(uint64_t)) !=
247: s->l2_size * sizeof(uint64_t))
1.1.1.5 ! root 248: goto fail;
1.1 root 249: }
250: /* write the l2 table to the file */
1.1.1.5 ! root 251: ret = bdrv_pwrite_sync(s->hd, l2_offset, l2_table,
! 252: s->l2_size * sizeof(uint64_t));
! 253: if (ret < 0) {
! 254: goto fail;
! 255: }
! 256:
! 257: /* update the L1 entry */
! 258: s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
! 259: if (write_l1_entry(s, l1_index) < 0) {
! 260: goto fail;
! 261: }
1.1 root 262:
263: /* update the l2 cache entry */
264:
265: s->l2_cache_offsets[min_index] = l2_offset;
266: s->l2_cache_counts[min_index] = 1;
267:
268: return l2_table;
1.1.1.5 ! root 269:
! 270: fail:
! 271: s->l1_table[l1_index] = old_l2_offset;
! 272: qcow2_l2_cache_reset(bs);
! 273: return NULL;
1.1 root 274: }
275:
276: static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
277: uint64_t *l2_table, uint64_t start, uint64_t mask)
278: {
279: int i;
280: uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
281:
282: if (!offset)
283: return 0;
284:
285: for (i = start; i < start + nb_clusters; i++)
1.1.1.2 root 286: if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
1.1 root 287: break;
288:
289: return (i - start);
290: }
291:
292: static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
293: {
294: int i = 0;
295:
296: while(nb_clusters-- && l2_table[i] == 0)
297: i++;
298:
299: return i;
300: }
301:
302: /* The crypt function is compatible with the linux cryptoloop
303: algorithm for < 4 GB images. NOTE: out_buf == in_buf is
304: supported */
305: void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
306: uint8_t *out_buf, const uint8_t *in_buf,
307: int nb_sectors, int enc,
308: const AES_KEY *key)
309: {
310: union {
311: uint64_t ll[2];
312: uint8_t b[16];
313: } ivec;
314: int i;
315:
316: for(i = 0; i < nb_sectors; i++) {
317: ivec.ll[0] = cpu_to_le64(sector_num);
318: ivec.ll[1] = 0;
319: AES_cbc_encrypt(in_buf, out_buf, 512, key,
320: ivec.b, enc);
321: sector_num++;
322: in_buf += 512;
323: out_buf += 512;
324: }
325: }
326:
327:
328: static int qcow_read(BlockDriverState *bs, int64_t sector_num,
329: uint8_t *buf, int nb_sectors)
330: {
331: BDRVQcowState *s = bs->opaque;
332: int ret, index_in_cluster, n, n1;
333: uint64_t cluster_offset;
334:
335: while (nb_sectors > 0) {
336: n = nb_sectors;
337: cluster_offset = qcow2_get_cluster_offset(bs, sector_num << 9, &n);
338: index_in_cluster = sector_num & (s->cluster_sectors - 1);
339: if (!cluster_offset) {
340: if (bs->backing_hd) {
341: /* read from the base image */
342: n1 = qcow2_backing_read1(bs->backing_hd, sector_num, buf, n);
343: if (n1 > 0) {
344: ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
345: if (ret < 0)
346: return -1;
347: }
348: } else {
349: memset(buf, 0, 512 * n);
350: }
351: } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
352: if (qcow2_decompress_cluster(s, cluster_offset) < 0)
353: return -1;
354: memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
355: } else {
356: ret = bdrv_pread(s->hd, cluster_offset + index_in_cluster * 512, buf, n * 512);
357: if (ret != n * 512)
358: return -1;
359: if (s->crypt_method) {
360: qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0,
361: &s->aes_decrypt_key);
362: }
363: }
364: nb_sectors -= n;
365: sector_num += n;
366: buf += n * 512;
367: }
368: return 0;
369: }
370:
371: static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
372: uint64_t cluster_offset, int n_start, int n_end)
373: {
374: BDRVQcowState *s = bs->opaque;
375: int n, ret;
376:
377: n = n_end - n_start;
378: if (n <= 0)
379: return 0;
380: ret = qcow_read(bs, start_sect + n_start, s->cluster_data, n);
381: if (ret < 0)
382: return ret;
383: if (s->crypt_method) {
384: qcow2_encrypt_sectors(s, start_sect + n_start,
385: s->cluster_data,
386: s->cluster_data, n, 1,
387: &s->aes_encrypt_key);
388: }
1.1.1.5 ! root 389: ret = bdrv_write_sync(s->hd, (cluster_offset >> 9) + n_start,
! 390: s->cluster_data, n);
1.1 root 391: if (ret < 0)
392: return ret;
393: return 0;
394: }
395:
396:
397: /*
398: * get_cluster_offset
399: *
400: * For a given offset of the disk image, return cluster offset in
401: * qcow2 file.
402: *
403: * on entry, *num is the number of contiguous clusters we'd like to
404: * access following offset.
405: *
406: * on exit, *num is the number of contiguous clusters we can read.
407: *
408: * Return 1, if the offset is found
409: * Return 0, otherwise.
410: *
411: */
412:
413: uint64_t qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
414: int *num)
415: {
416: BDRVQcowState *s = bs->opaque;
1.1.1.2 root 417: unsigned int l1_index, l2_index;
1.1 root 418: uint64_t l2_offset, *l2_table, cluster_offset;
419: int l1_bits, c;
1.1.1.2 root 420: unsigned int index_in_cluster, nb_clusters;
421: uint64_t nb_available, nb_needed;
1.1 root 422:
423: index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
424: nb_needed = *num + index_in_cluster;
425:
426: l1_bits = s->l2_bits + s->cluster_bits;
427:
428: /* compute how many bytes there are between the offset and
429: * the end of the l1 entry
430: */
431:
1.1.1.2 root 432: nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
1.1 root 433:
434: /* compute the number of available sectors */
435:
436: nb_available = (nb_available >> 9) + index_in_cluster;
437:
438: if (nb_needed > nb_available) {
439: nb_needed = nb_available;
440: }
441:
442: cluster_offset = 0;
443:
444: /* seek the the l2 offset in the l1 table */
445:
446: l1_index = offset >> l1_bits;
447: if (l1_index >= s->l1_size)
448: goto out;
449:
450: l2_offset = s->l1_table[l1_index];
451:
452: /* seek the l2 table of the given l2 offset */
453:
454: if (!l2_offset)
455: goto out;
456:
457: /* load the l2 table in memory */
458:
459: l2_offset &= ~QCOW_OFLAG_COPIED;
460: l2_table = l2_load(bs, l2_offset);
461: if (l2_table == NULL)
462: return 0;
463:
464: /* find the cluster offset for the given disk offset */
465:
466: l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
467: cluster_offset = be64_to_cpu(l2_table[l2_index]);
468: nb_clusters = size_to_clusters(s, nb_needed << 9);
469:
470: if (!cluster_offset) {
471: /* how many empty clusters ? */
472: c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
473: } else {
474: /* how many allocated clusters ? */
475: c = count_contiguous_clusters(nb_clusters, s->cluster_size,
476: &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
477: }
478:
479: nb_available = (c * s->cluster_sectors);
480: out:
481: if (nb_available > nb_needed)
482: nb_available = nb_needed;
483:
484: *num = nb_available - index_in_cluster;
485:
486: return cluster_offset & ~QCOW_OFLAG_COPIED;
487: }
488:
489: /*
490: * get_cluster_table
491: *
492: * for a given disk offset, load (and allocate if needed)
493: * the l2 table.
494: *
495: * the l2 table offset in the qcow2 file and the cluster index
496: * in the l2 table are given to the caller.
497: *
1.1.1.3 root 498: * Returns 0 on success, -errno in failure case
1.1 root 499: */
500: static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
501: uint64_t **new_l2_table,
502: uint64_t *new_l2_offset,
503: int *new_l2_index)
504: {
505: BDRVQcowState *s = bs->opaque;
1.1.1.2 root 506: unsigned int l1_index, l2_index;
1.1 root 507: uint64_t l2_offset, *l2_table;
1.1.1.2 root 508: int ret;
1.1 root 509:
510: /* seek the the l2 offset in the l1 table */
511:
512: l1_index = offset >> (s->l2_bits + s->cluster_bits);
513: if (l1_index >= s->l1_size) {
514: ret = qcow2_grow_l1_table(bs, l1_index + 1);
1.1.1.3 root 515: if (ret < 0) {
516: return ret;
517: }
1.1 root 518: }
519: l2_offset = s->l1_table[l1_index];
520:
521: /* seek the l2 table of the given l2 offset */
522:
523: if (l2_offset & QCOW_OFLAG_COPIED) {
524: /* load the l2 table in memory */
525: l2_offset &= ~QCOW_OFLAG_COPIED;
526: l2_table = l2_load(bs, l2_offset);
1.1.1.3 root 527: if (l2_table == NULL) {
528: return -EIO;
529: }
1.1 root 530: } else {
531: if (l2_offset)
532: qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
533: l2_table = l2_allocate(bs, l1_index);
1.1.1.3 root 534: if (l2_table == NULL) {
535: return -EIO;
536: }
1.1 root 537: l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
538: }
539:
540: /* find the cluster offset for the given disk offset */
541:
542: l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
543:
544: *new_l2_table = l2_table;
545: *new_l2_offset = l2_offset;
546: *new_l2_index = l2_index;
547:
1.1.1.3 root 548: return 0;
1.1 root 549: }
550:
551: /*
552: * alloc_compressed_cluster_offset
553: *
554: * For a given offset of the disk image, return cluster offset in
555: * qcow2 file.
556: *
557: * If the offset is not found, allocate a new compressed cluster.
558: *
559: * Return the cluster offset if successful,
560: * Return 0, otherwise.
561: *
562: */
563:
564: uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
565: uint64_t offset,
566: int compressed_size)
567: {
568: BDRVQcowState *s = bs->opaque;
569: int l2_index, ret;
1.1.1.3 root 570: uint64_t l2_offset, *l2_table;
571: int64_t cluster_offset;
1.1 root 572: int nb_csectors;
573:
574: ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
1.1.1.3 root 575: if (ret < 0) {
1.1 root 576: return 0;
1.1.1.3 root 577: }
1.1 root 578:
579: cluster_offset = be64_to_cpu(l2_table[l2_index]);
580: if (cluster_offset & QCOW_OFLAG_COPIED)
581: return cluster_offset & ~QCOW_OFLAG_COPIED;
582:
583: if (cluster_offset)
584: qcow2_free_any_clusters(bs, cluster_offset, 1);
585:
586: cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
1.1.1.3 root 587: if (cluster_offset < 0) {
588: return 0;
589: }
590:
1.1 root 591: nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
592: (cluster_offset >> 9);
593:
594: cluster_offset |= QCOW_OFLAG_COMPRESSED |
595: ((uint64_t)nb_csectors << s->csize_shift);
596:
597: /* update L2 table */
598:
599: /* compressed clusters never have the copied flag */
600:
601: l2_table[l2_index] = cpu_to_be64(cluster_offset);
1.1.1.5 ! root 602: if (bdrv_pwrite_sync(s->hd,
1.1 root 603: l2_offset + l2_index * sizeof(uint64_t),
604: l2_table + l2_index,
1.1.1.5 ! root 605: sizeof(uint64_t)) < 0)
1.1 root 606: return 0;
607:
608: return cluster_offset;
609: }
610:
611: /*
612: * Write L2 table updates to disk, writing whole sectors to avoid a
613: * read-modify-write in bdrv_pwrite
614: */
615: #define L2_ENTRIES_PER_SECTOR (512 / 8)
616: static int write_l2_entries(BDRVQcowState *s, uint64_t *l2_table,
617: uint64_t l2_offset, int l2_index, int num)
618: {
619: int l2_start_index = l2_index & ~(L1_ENTRIES_PER_SECTOR - 1);
620: int start_offset = (8 * l2_index) & ~511;
621: int end_offset = (8 * (l2_index + num) + 511) & ~511;
622: size_t len = end_offset - start_offset;
1.1.1.5 ! root 623: int ret;
1.1 root 624:
1.1.1.5 ! root 625: ret = bdrv_pwrite_sync(s->hd, l2_offset + start_offset,
! 626: &l2_table[l2_start_index], len);
! 627: if (ret < 0) {
! 628: return ret;
1.1 root 629: }
630:
631: return 0;
632: }
633:
1.1.1.3 root 634: int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
1.1 root 635: {
636: BDRVQcowState *s = bs->opaque;
637: int i, j = 0, l2_index, ret;
638: uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
1.1.1.3 root 639: uint64_t cluster_offset = m->cluster_offset;
1.1 root 640:
641: if (m->nb_clusters == 0)
642: return 0;
643:
644: old_cluster = qemu_malloc(m->nb_clusters * sizeof(uint64_t));
645:
646: /* copy content of unmodified sectors */
647: start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
648: if (m->n_start) {
649: ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
650: if (ret < 0)
651: goto err;
652: }
653:
654: if (m->nb_available & (s->cluster_sectors - 1)) {
655: uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
656: ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
657: m->nb_available - end, s->cluster_sectors);
658: if (ret < 0)
659: goto err;
660: }
661:
662: /* update L2 table */
1.1.1.3 root 663: ret = get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index);
664: if (ret < 0) {
1.1 root 665: goto err;
1.1.1.3 root 666: }
1.1 root 667:
668: for (i = 0; i < m->nb_clusters; i++) {
669: /* if two concurrent writes happen to the same unallocated cluster
670: * each write allocates separate cluster and writes data concurrently.
671: * The first one to complete updates l2 table with pointer to its
672: * cluster the second one has to do RMW (which is done above by
673: * copy_sectors()), update l2 table with its cluster pointer and free
674: * old cluster. This is what this loop does */
675: if(l2_table[l2_index + i] != 0)
676: old_cluster[j++] = l2_table[l2_index + i];
677:
678: l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
679: (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
680: }
681:
1.1.1.5 ! root 682: ret = write_l2_entries(s, l2_table, l2_offset, l2_index, m->nb_clusters);
! 683: if (ret < 0) {
! 684: qcow2_l2_cache_reset(bs);
1.1 root 685: goto err;
686: }
687:
688: for (i = 0; i < j; i++)
689: qcow2_free_any_clusters(bs,
690: be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1);
691:
692: ret = 0;
693: err:
694: qemu_free(old_cluster);
695: return ret;
696: }
697:
698: /*
699: * alloc_cluster_offset
700: *
1.1.1.3 root 701: * For a given offset of the disk image, return cluster offset in qcow2 file.
1.1 root 702: * If the offset is not found, allocate a new cluster.
703: *
1.1.1.3 root 704: * If the cluster was already allocated, m->nb_clusters is set to 0,
705: * m->depends_on is set to NULL and the other fields in m are meaningless.
1.1 root 706: *
1.1.1.3 root 707: * If the cluster is newly allocated, m->nb_clusters is set to the number of
708: * contiguous clusters that have been allocated. This may be 0 if the request
709: * conflict with another write request in flight; in this case, m->depends_on
710: * is set and the remaining fields of m are meaningless.
711: *
712: * If m->nb_clusters is non-zero, the other fields of m are valid and contain
713: * information about the first allocated cluster.
714: *
715: * Return 0 on success and -errno in error cases
1.1 root 716: */
1.1.1.3 root 717: int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
718: int n_start, int n_end, int *num, QCowL2Meta *m)
1.1 root 719: {
720: BDRVQcowState *s = bs->opaque;
721: int l2_index, ret;
1.1.1.3 root 722: uint64_t l2_offset, *l2_table;
723: int64_t cluster_offset;
1.1.1.2 root 724: unsigned int nb_clusters, i = 0;
1.1 root 725: QCowL2Meta *old_alloc;
726:
727: ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
1.1.1.3 root 728: if (ret < 0) {
729: return ret;
730: }
1.1 root 731:
732: nb_clusters = size_to_clusters(s, n_end << 9);
733:
734: nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
735:
736: cluster_offset = be64_to_cpu(l2_table[l2_index]);
737:
738: /* We keep all QCOW_OFLAG_COPIED clusters */
739:
740: if (cluster_offset & QCOW_OFLAG_COPIED) {
741: nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
742: &l2_table[l2_index], 0, 0);
743:
744: cluster_offset &= ~QCOW_OFLAG_COPIED;
745: m->nb_clusters = 0;
1.1.1.3 root 746: m->depends_on = NULL;
1.1 root 747:
748: goto out;
749: }
750:
751: /* for the moment, multiple compressed clusters are not managed */
752:
753: if (cluster_offset & QCOW_OFLAG_COMPRESSED)
754: nb_clusters = 1;
755:
756: /* how many available clusters ? */
757:
758: while (i < nb_clusters) {
759: i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
760: &l2_table[l2_index], i, 0);
1.1.1.3 root 761: if ((i >= nb_clusters) || be64_to_cpu(l2_table[l2_index + i])) {
1.1 root 762: break;
1.1.1.3 root 763: }
1.1 root 764:
765: i += count_contiguous_free_clusters(nb_clusters - i,
766: &l2_table[l2_index + i]);
1.1.1.3 root 767: if (i >= nb_clusters) {
768: break;
769: }
1.1 root 770:
771: cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
772:
773: if ((cluster_offset & QCOW_OFLAG_COPIED) ||
774: (cluster_offset & QCOW_OFLAG_COMPRESSED))
775: break;
776: }
1.1.1.3 root 777: assert(i <= nb_clusters);
1.1 root 778: nb_clusters = i;
779:
780: /*
781: * Check if there already is an AIO write request in flight which allocates
782: * the same cluster. In this case we need to wait until the previous
783: * request has completed and updated the L2 table accordingly.
784: */
1.1.1.2 root 785: QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
1.1 root 786:
787: uint64_t end_offset = offset + nb_clusters * s->cluster_size;
788: uint64_t old_offset = old_alloc->offset;
789: uint64_t old_end_offset = old_alloc->offset +
790: old_alloc->nb_clusters * s->cluster_size;
791:
792: if (end_offset < old_offset || offset > old_end_offset) {
793: /* No intersection */
794: } else {
795: if (offset < old_offset) {
796: /* Stop at the start of a running allocation */
797: nb_clusters = (old_offset - offset) >> s->cluster_bits;
798: } else {
799: nb_clusters = 0;
800: }
801:
802: if (nb_clusters == 0) {
803: /* Set dependency and wait for a callback */
804: m->depends_on = old_alloc;
805: m->nb_clusters = 0;
806: *num = 0;
807: return 0;
808: }
809: }
810: }
811:
812: if (!nb_clusters) {
813: abort();
814: }
815:
1.1.1.2 root 816: QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
1.1 root 817:
818: /* allocate a new cluster */
819:
820: cluster_offset = qcow2_alloc_clusters(bs, nb_clusters * s->cluster_size);
1.1.1.3 root 821: if (cluster_offset < 0) {
1.1.1.4 root 822: QLIST_REMOVE(m, next_in_flight);
1.1.1.3 root 823: return cluster_offset;
824: }
1.1 root 825:
826: /* save info needed for meta data update */
827: m->offset = offset;
828: m->n_start = n_start;
829: m->nb_clusters = nb_clusters;
830:
831: out:
832: m->nb_available = MIN(nb_clusters << (s->cluster_bits - 9), n_end);
1.1.1.3 root 833: m->cluster_offset = cluster_offset;
1.1 root 834:
835: *num = m->nb_available - n_start;
836:
1.1.1.3 root 837: return 0;
1.1 root 838: }
839:
840: static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
841: const uint8_t *buf, int buf_size)
842: {
843: z_stream strm1, *strm = &strm1;
844: int ret, out_len;
845:
846: memset(strm, 0, sizeof(*strm));
847:
848: strm->next_in = (uint8_t *)buf;
849: strm->avail_in = buf_size;
850: strm->next_out = out_buf;
851: strm->avail_out = out_buf_size;
852:
853: ret = inflateInit2(strm, -12);
854: if (ret != Z_OK)
855: return -1;
856: ret = inflate(strm, Z_FINISH);
857: out_len = strm->next_out - out_buf;
858: if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
859: out_len != out_buf_size) {
860: inflateEnd(strm);
861: return -1;
862: }
863: inflateEnd(strm);
864: return 0;
865: }
866:
867: int qcow2_decompress_cluster(BDRVQcowState *s, uint64_t cluster_offset)
868: {
869: int ret, csize, nb_csectors, sector_offset;
870: uint64_t coffset;
871:
872: coffset = cluster_offset & s->cluster_offset_mask;
873: if (s->cluster_cache_offset != coffset) {
874: nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
875: sector_offset = coffset & 511;
876: csize = nb_csectors * 512 - sector_offset;
877: ret = bdrv_read(s->hd, coffset >> 9, s->cluster_data, nb_csectors);
878: if (ret < 0) {
879: return -1;
880: }
881: if (decompress_buffer(s->cluster_cache, s->cluster_size,
882: s->cluster_data + sector_offset, csize) < 0) {
883: return -1;
884: }
885: s->cluster_cache_offset = coffset;
886: }
887: return 0;
888: }
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.