![[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:
1.1.1.7 root 31: int qcow2_grow_l1_table(BlockDriverState *bs, int min_size, bool exact_size)
1.1 root 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:
1.1.1.7 root 39: if (min_size <= s->l1_size)
1.1 root 40: return 0;
1.1.1.7 root 41:
42: if (exact_size) {
43: new_l1_size = min_size;
44: } else {
45: /* Bump size up to reduce the number of times we have to grow */
46: new_l1_size = s->l1_size;
47: if (new_l1_size == 0) {
48: new_l1_size = 1;
49: }
50: while (min_size > new_l1_size) {
51: new_l1_size = (new_l1_size * 3 + 1) / 2;
52: }
1.1 root 53: }
1.1.1.7 root 54:
1.1 root 55: #ifdef DEBUG_ALLOC2
1.1.1.9 ! root 56: fprintf(stderr, "grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
1.1 root 57: #endif
58:
59: new_l1_size2 = sizeof(uint64_t) * new_l1_size;
1.1.1.9 ! root 60: new_l1_table = g_malloc0(align_offset(new_l1_size2, 512));
1.1 root 61: memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
62:
63: /* write new table (align to cluster) */
1.1.1.6 root 64: BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
1.1 root 65: new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
1.1.1.3 root 66: if (new_l1_table_offset < 0) {
1.1.1.9 ! root 67: g_free(new_l1_table);
1.1.1.3 root 68: return new_l1_table_offset;
69: }
1.1 root 70:
1.1.1.7 root 71: ret = qcow2_cache_flush(bs, s->refcount_block_cache);
72: if (ret < 0) {
1.1.1.8 root 73: goto fail;
1.1.1.7 root 74: }
75:
1.1.1.6 root 76: BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
1.1 root 77: for(i = 0; i < s->l1_size; i++)
78: new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
1.1.1.6 root 79: ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2);
1.1.1.5 root 80: if (ret < 0)
1.1 root 81: goto fail;
82: for(i = 0; i < s->l1_size; i++)
83: new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
84:
85: /* set new table */
1.1.1.6 root 86: BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
1.1 root 87: cpu_to_be32w((uint32_t*)data, new_l1_size);
1.1.1.7 root 88: cpu_to_be64wu((uint64_t*)(data + 4), new_l1_table_offset);
1.1.1.6 root 89: ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data));
1.1.1.5 root 90: if (ret < 0) {
1.1 root 91: goto fail;
1.1.1.3 root 92: }
1.1.1.9 ! root 93: g_free(s->l1_table);
1.1 root 94: qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
95: s->l1_table_offset = new_l1_table_offset;
96: s->l1_table = new_l1_table;
97: s->l1_size = new_l1_size;
98: return 0;
99: fail:
1.1.1.9 ! root 100: g_free(new_l1_table);
1.1.1.3 root 101: qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2);
1.1.1.5 root 102: return ret;
1.1 root 103: }
104:
105: /*
106: * l2_load
107: *
108: * Loads a L2 table into memory. If the table is in the cache, the cache
109: * is used; otherwise the L2 table is loaded from the image file.
110: *
111: * Returns a pointer to the L2 table on success, or NULL if the read from
112: * the image file failed.
113: */
114:
1.1.1.6 root 115: static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
116: uint64_t **l2_table)
1.1 root 117: {
118: BDRVQcowState *s = bs->opaque;
1.1.1.6 root 119: int ret;
1.1 root 120:
1.1.1.7 root 121: ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table);
1.1 root 122:
1.1.1.7 root 123: return ret;
1.1 root 124: }
125:
126: /*
127: * Writes one sector of the L1 table to the disk (can't update single entries
128: * and we really don't want bdrv_pread to perform a read-modify-write)
129: */
130: #define L1_ENTRIES_PER_SECTOR (512 / 8)
1.1.1.6 root 131: static int write_l1_entry(BlockDriverState *bs, int l1_index)
1.1 root 132: {
1.1.1.6 root 133: BDRVQcowState *s = bs->opaque;
1.1 root 134: uint64_t buf[L1_ENTRIES_PER_SECTOR];
135: int l1_start_index;
1.1.1.5 root 136: int i, ret;
1.1 root 137:
138: l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
139: for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
140: buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
141: }
142:
1.1.1.6 root 143: BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
144: ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index,
1.1.1.5 root 145: buf, sizeof(buf));
146: if (ret < 0) {
147: return ret;
1.1 root 148: }
149:
150: return 0;
151: }
152:
153: /*
154: * l2_allocate
155: *
156: * Allocate a new l2 entry in the file. If l1_index points to an already
157: * used entry in the L2 table (i.e. we are doing a copy on write for the L2
158: * table) copy the contents of the old L2 table into the newly allocated one.
159: * Otherwise the new table is initialized with zeros.
160: *
161: */
162:
1.1.1.6 root 163: static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
1.1 root 164: {
165: BDRVQcowState *s = bs->opaque;
166: uint64_t old_l2_offset;
1.1.1.3 root 167: uint64_t *l2_table;
168: int64_t l2_offset;
1.1.1.5 root 169: int ret;
1.1 root 170:
171: old_l2_offset = s->l1_table[l1_index];
172:
173: /* allocate a new l2 entry */
174:
175: l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
1.1.1.3 root 176: if (l2_offset < 0) {
1.1.1.6 root 177: return l2_offset;
1.1.1.3 root 178: }
1.1 root 179:
1.1.1.7 root 180: ret = qcow2_cache_flush(bs, s->refcount_block_cache);
181: if (ret < 0) {
182: goto fail;
183: }
184:
1.1 root 185: /* allocate a new entry in the l2 cache */
186:
1.1.1.7 root 187: ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
188: if (ret < 0) {
189: return ret;
190: }
191:
192: l2_table = *table;
1.1 root 193:
194: if (old_l2_offset == 0) {
195: /* if there was no old l2 table, clear the new table */
196: memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
197: } else {
1.1.1.7 root 198: uint64_t* old_table;
199:
1.1 root 200: /* if there was an old l2 table, read it from the disk */
1.1.1.6 root 201: BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
1.1.1.7 root 202: ret = qcow2_cache_get(bs, s->l2_table_cache, old_l2_offset,
203: (void**) &old_table);
204: if (ret < 0) {
205: goto fail;
206: }
207:
208: memcpy(l2_table, old_table, s->cluster_size);
209:
210: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &old_table);
1.1.1.6 root 211: if (ret < 0) {
1.1.1.5 root 212: goto fail;
1.1.1.6 root 213: }
1.1 root 214: }
1.1.1.7 root 215:
1.1 root 216: /* write the l2 table to the file */
1.1.1.6 root 217: BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
1.1.1.7 root 218:
219: qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
220: ret = qcow2_cache_flush(bs, s->l2_table_cache);
1.1.1.5 root 221: if (ret < 0) {
222: goto fail;
223: }
224:
225: /* update the L1 entry */
226: s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
1.1.1.6 root 227: ret = write_l1_entry(bs, l1_index);
228: if (ret < 0) {
1.1.1.5 root 229: goto fail;
230: }
1.1 root 231:
1.1.1.6 root 232: *table = l2_table;
233: return 0;
1.1.1.5 root 234:
235: fail:
1.1.1.7 root 236: qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
1.1.1.5 root 237: s->l1_table[l1_index] = old_l2_offset;
1.1.1.6 root 238: return ret;
1.1 root 239: }
240:
241: static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
242: uint64_t *l2_table, uint64_t start, uint64_t mask)
243: {
244: int i;
245: uint64_t offset = be64_to_cpu(l2_table[0]) & ~mask;
246:
247: if (!offset)
248: return 0;
249:
250: for (i = start; i < start + nb_clusters; i++)
1.1.1.2 root 251: if (offset + (uint64_t) i * cluster_size != (be64_to_cpu(l2_table[i]) & ~mask))
1.1 root 252: break;
253:
254: return (i - start);
255: }
256:
257: static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
258: {
259: int i = 0;
260:
261: while(nb_clusters-- && l2_table[i] == 0)
262: i++;
263:
264: return i;
265: }
266:
267: /* The crypt function is compatible with the linux cryptoloop
268: algorithm for < 4 GB images. NOTE: out_buf == in_buf is
269: supported */
270: void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
271: uint8_t *out_buf, const uint8_t *in_buf,
272: int nb_sectors, int enc,
273: const AES_KEY *key)
274: {
275: union {
276: uint64_t ll[2];
277: uint8_t b[16];
278: } ivec;
279: int i;
280:
281: for(i = 0; i < nb_sectors; i++) {
282: ivec.ll[0] = cpu_to_le64(sector_num);
283: ivec.ll[1] = 0;
284: AES_cbc_encrypt(in_buf, out_buf, 512, key,
285: ivec.b, enc);
286: sector_num++;
287: in_buf += 512;
288: out_buf += 512;
289: }
290: }
291:
292:
1.1.1.7 root 293: static int qcow2_read(BlockDriverState *bs, int64_t sector_num,
294: uint8_t *buf, int nb_sectors)
1.1 root 295: {
296: BDRVQcowState *s = bs->opaque;
297: int ret, index_in_cluster, n, n1;
298: uint64_t cluster_offset;
1.1.1.7 root 299: struct iovec iov;
300: QEMUIOVector qiov;
1.1 root 301:
302: while (nb_sectors > 0) {
303: n = nb_sectors;
1.1.1.6 root 304:
305: ret = qcow2_get_cluster_offset(bs, sector_num << 9, &n,
306: &cluster_offset);
307: if (ret < 0) {
308: return ret;
309: }
310:
1.1 root 311: index_in_cluster = sector_num & (s->cluster_sectors - 1);
312: if (!cluster_offset) {
313: if (bs->backing_hd) {
314: /* read from the base image */
1.1.1.7 root 315: iov.iov_base = buf;
316: iov.iov_len = n * 512;
317: qemu_iovec_init_external(&qiov, &iov, 1);
318:
319: n1 = qcow2_backing_read1(bs->backing_hd, &qiov, sector_num, n);
1.1 root 320: if (n1 > 0) {
1.1.1.6 root 321: BLKDBG_EVENT(bs->file, BLKDBG_READ_BACKING);
1.1 root 322: ret = bdrv_read(bs->backing_hd, sector_num, buf, n1);
323: if (ret < 0)
324: return -1;
325: }
326: } else {
327: memset(buf, 0, 512 * n);
328: }
329: } else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
1.1.1.6 root 330: if (qcow2_decompress_cluster(bs, cluster_offset) < 0)
1.1 root 331: return -1;
332: memcpy(buf, s->cluster_cache + index_in_cluster * 512, 512 * n);
333: } else {
1.1.1.6 root 334: BLKDBG_EVENT(bs->file, BLKDBG_READ);
335: ret = bdrv_pread(bs->file, cluster_offset + index_in_cluster * 512, buf, n * 512);
1.1 root 336: if (ret != n * 512)
337: return -1;
338: if (s->crypt_method) {
339: qcow2_encrypt_sectors(s, sector_num, buf, buf, n, 0,
340: &s->aes_decrypt_key);
341: }
342: }
343: nb_sectors -= n;
344: sector_num += n;
345: buf += n * 512;
346: }
347: return 0;
348: }
349:
350: static int copy_sectors(BlockDriverState *bs, uint64_t start_sect,
351: uint64_t cluster_offset, int n_start, int n_end)
352: {
353: BDRVQcowState *s = bs->opaque;
354: int n, ret;
355:
356: n = n_end - n_start;
357: if (n <= 0)
358: return 0;
1.1.1.6 root 359: BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
1.1.1.7 root 360: ret = qcow2_read(bs, start_sect + n_start, s->cluster_data, n);
1.1 root 361: if (ret < 0)
362: return ret;
363: if (s->crypt_method) {
364: qcow2_encrypt_sectors(s, start_sect + n_start,
365: s->cluster_data,
366: s->cluster_data, n, 1,
367: &s->aes_encrypt_key);
368: }
1.1.1.6 root 369: BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
1.1.1.7 root 370: ret = bdrv_write(bs->file, (cluster_offset >> 9) + n_start,
1.1.1.5 root 371: s->cluster_data, n);
1.1 root 372: if (ret < 0)
373: return ret;
374: return 0;
375: }
376:
377:
378: /*
379: * get_cluster_offset
380: *
1.1.1.6 root 381: * For a given offset of the disk image, find the cluster offset in
382: * qcow2 file. The offset is stored in *cluster_offset.
1.1 root 383: *
1.1.1.9 ! root 384: * on entry, *num is the number of contiguous sectors we'd like to
1.1 root 385: * access following offset.
386: *
1.1.1.9 ! root 387: * on exit, *num is the number of contiguous sectors we can read.
1.1 root 388: *
1.1.1.6 root 389: * Return 0, if the offset is found
390: * Return -errno, otherwise.
1.1 root 391: *
392: */
393:
1.1.1.6 root 394: int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
395: int *num, uint64_t *cluster_offset)
1.1 root 396: {
397: BDRVQcowState *s = bs->opaque;
1.1.1.2 root 398: unsigned int l1_index, l2_index;
1.1.1.6 root 399: uint64_t l2_offset, *l2_table;
1.1 root 400: int l1_bits, c;
1.1.1.2 root 401: unsigned int index_in_cluster, nb_clusters;
402: uint64_t nb_available, nb_needed;
1.1.1.6 root 403: int ret;
1.1 root 404:
405: index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
406: nb_needed = *num + index_in_cluster;
407:
408: l1_bits = s->l2_bits + s->cluster_bits;
409:
410: /* compute how many bytes there are between the offset and
411: * the end of the l1 entry
412: */
413:
1.1.1.2 root 414: nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
1.1 root 415:
416: /* compute the number of available sectors */
417:
418: nb_available = (nb_available >> 9) + index_in_cluster;
419:
420: if (nb_needed > nb_available) {
421: nb_needed = nb_available;
422: }
423:
1.1.1.6 root 424: *cluster_offset = 0;
1.1 root 425:
426: /* seek the the l2 offset in the l1 table */
427:
428: l1_index = offset >> l1_bits;
429: if (l1_index >= s->l1_size)
430: goto out;
431:
432: l2_offset = s->l1_table[l1_index];
433:
434: /* seek the l2 table of the given l2 offset */
435:
436: if (!l2_offset)
437: goto out;
438:
439: /* load the l2 table in memory */
440:
441: l2_offset &= ~QCOW_OFLAG_COPIED;
1.1.1.6 root 442: ret = l2_load(bs, l2_offset, &l2_table);
443: if (ret < 0) {
444: return ret;
445: }
1.1 root 446:
447: /* find the cluster offset for the given disk offset */
448:
449: l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
1.1.1.6 root 450: *cluster_offset = be64_to_cpu(l2_table[l2_index]);
1.1 root 451: nb_clusters = size_to_clusters(s, nb_needed << 9);
452:
1.1.1.6 root 453: if (!*cluster_offset) {
1.1 root 454: /* how many empty clusters ? */
455: c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
456: } else {
457: /* how many allocated clusters ? */
458: c = count_contiguous_clusters(nb_clusters, s->cluster_size,
459: &l2_table[l2_index], 0, QCOW_OFLAG_COPIED);
460: }
461:
1.1.1.7 root 462: qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
463:
1.1 root 464: nb_available = (c * s->cluster_sectors);
465: out:
466: if (nb_available > nb_needed)
467: nb_available = nb_needed;
468:
469: *num = nb_available - index_in_cluster;
470:
1.1.1.6 root 471: *cluster_offset &=~QCOW_OFLAG_COPIED;
472: return 0;
1.1 root 473: }
474:
475: /*
476: * get_cluster_table
477: *
478: * for a given disk offset, load (and allocate if needed)
479: * the l2 table.
480: *
481: * the l2 table offset in the qcow2 file and the cluster index
482: * in the l2 table are given to the caller.
483: *
1.1.1.3 root 484: * Returns 0 on success, -errno in failure case
1.1 root 485: */
486: static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
487: uint64_t **new_l2_table,
488: uint64_t *new_l2_offset,
489: int *new_l2_index)
490: {
491: BDRVQcowState *s = bs->opaque;
1.1.1.2 root 492: unsigned int l1_index, l2_index;
1.1.1.6 root 493: uint64_t l2_offset;
494: uint64_t *l2_table = NULL;
1.1.1.2 root 495: int ret;
1.1 root 496:
497: /* seek the the l2 offset in the l1 table */
498:
499: l1_index = offset >> (s->l2_bits + s->cluster_bits);
500: if (l1_index >= s->l1_size) {
1.1.1.7 root 501: ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
1.1.1.3 root 502: if (ret < 0) {
503: return ret;
504: }
1.1 root 505: }
506: l2_offset = s->l1_table[l1_index];
507:
508: /* seek the l2 table of the given l2 offset */
509:
510: if (l2_offset & QCOW_OFLAG_COPIED) {
511: /* load the l2 table in memory */
512: l2_offset &= ~QCOW_OFLAG_COPIED;
1.1.1.6 root 513: ret = l2_load(bs, l2_offset, &l2_table);
514: if (ret < 0) {
515: return ret;
1.1.1.3 root 516: }
1.1 root 517: } else {
1.1.1.7 root 518: /* First allocate a new L2 table (and do COW if needed) */
1.1.1.6 root 519: ret = l2_allocate(bs, l1_index, &l2_table);
520: if (ret < 0) {
521: return ret;
1.1.1.3 root 522: }
1.1.1.7 root 523:
524: /* Then decrease the refcount of the old table */
525: if (l2_offset) {
526: qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
527: }
1.1 root 528: l2_offset = s->l1_table[l1_index] & ~QCOW_OFLAG_COPIED;
529: }
530:
531: /* find the cluster offset for the given disk offset */
532:
533: l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
534:
535: *new_l2_table = l2_table;
536: *new_l2_offset = l2_offset;
537: *new_l2_index = l2_index;
538:
1.1.1.3 root 539: return 0;
1.1 root 540: }
541:
542: /*
543: * alloc_compressed_cluster_offset
544: *
545: * For a given offset of the disk image, return cluster offset in
546: * qcow2 file.
547: *
548: * If the offset is not found, allocate a new compressed cluster.
549: *
550: * Return the cluster offset if successful,
551: * Return 0, otherwise.
552: *
553: */
554:
555: uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
556: uint64_t offset,
557: int compressed_size)
558: {
559: BDRVQcowState *s = bs->opaque;
560: int l2_index, ret;
1.1.1.3 root 561: uint64_t l2_offset, *l2_table;
562: int64_t cluster_offset;
1.1 root 563: int nb_csectors;
564:
565: ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
1.1.1.3 root 566: if (ret < 0) {
1.1 root 567: return 0;
1.1.1.3 root 568: }
1.1 root 569:
570: cluster_offset = be64_to_cpu(l2_table[l2_index]);
1.1.1.9 ! root 571: if (cluster_offset & QCOW_OFLAG_COPIED) {
! 572: qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
! 573: return 0;
! 574: }
1.1 root 575:
576: if (cluster_offset)
577: qcow2_free_any_clusters(bs, cluster_offset, 1);
578:
579: cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
1.1.1.3 root 580: if (cluster_offset < 0) {
1.1.1.7 root 581: qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1.1.1.3 root 582: return 0;
583: }
584:
1.1 root 585: nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
586: (cluster_offset >> 9);
587:
588: cluster_offset |= QCOW_OFLAG_COMPRESSED |
589: ((uint64_t)nb_csectors << s->csize_shift);
590:
591: /* update L2 table */
592:
593: /* compressed clusters never have the copied flag */
594:
1.1.1.6 root 595: BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
1.1.1.7 root 596: qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
1.1 root 597: l2_table[l2_index] = cpu_to_be64(cluster_offset);
1.1.1.7 root 598: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1.1.1.5 root 599: if (ret < 0) {
1.1.1.7 root 600: return 0;
1.1 root 601: }
602:
1.1.1.7 root 603: return cluster_offset;
1.1 root 604: }
605:
1.1.1.3 root 606: int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
1.1 root 607: {
608: BDRVQcowState *s = bs->opaque;
609: int i, j = 0, l2_index, ret;
610: uint64_t *old_cluster, start_sect, l2_offset, *l2_table;
1.1.1.3 root 611: uint64_t cluster_offset = m->cluster_offset;
1.1.1.7 root 612: bool cow = false;
1.1 root 613:
614: if (m->nb_clusters == 0)
615: return 0;
616:
1.1.1.9 ! root 617: old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t));
1.1 root 618:
619: /* copy content of unmodified sectors */
620: start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
621: if (m->n_start) {
1.1.1.7 root 622: cow = true;
1.1 root 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);
1.1.1.7 root 630: cow = true;
1.1 root 631: ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
632: m->nb_available - end, s->cluster_sectors);
633: if (ret < 0)
634: goto err;
635: }
636:
1.1.1.7 root 637: /*
638: * Update L2 table.
639: *
640: * Before we update the L2 table to actually point to the new cluster, we
641: * need to be sure that the refcounts have been increased and COW was
642: * handled.
643: */
644: if (cow) {
645: qcow2_cache_depends_on_flush(s->l2_table_cache);
646: }
647:
648: qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
1.1.1.3 root 649: ret = get_cluster_table(bs, m->offset, &l2_table, &l2_offset, &l2_index);
650: if (ret < 0) {
1.1 root 651: goto err;
1.1.1.3 root 652: }
1.1.1.7 root 653: qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
1.1 root 654:
655: for (i = 0; i < m->nb_clusters; i++) {
656: /* if two concurrent writes happen to the same unallocated cluster
657: * each write allocates separate cluster and writes data concurrently.
658: * The first one to complete updates l2 table with pointer to its
659: * cluster the second one has to do RMW (which is done above by
660: * copy_sectors()), update l2 table with its cluster pointer and free
661: * old cluster. This is what this loop does */
662: if(l2_table[l2_index + i] != 0)
663: old_cluster[j++] = l2_table[l2_index + i];
664:
665: l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
666: (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
667: }
668:
1.1.1.7 root 669:
670: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1.1.1.5 root 671: if (ret < 0) {
1.1 root 672: goto err;
673: }
674:
1.1.1.6 root 675: /*
676: * If this was a COW, we need to decrease the refcount of the old cluster.
677: * Also flush bs->file to get the right order for L2 and refcount update.
678: */
679: if (j != 0) {
680: for (i = 0; i < j; i++) {
681: qcow2_free_any_clusters(bs,
682: be64_to_cpu(old_cluster[i]) & ~QCOW_OFLAG_COPIED, 1);
683: }
684: }
1.1 root 685:
686: ret = 0;
687: err:
1.1.1.9 ! root 688: g_free(old_cluster);
1.1 root 689: return ret;
690: }
691:
692: /*
693: * alloc_cluster_offset
694: *
1.1.1.3 root 695: * For a given offset of the disk image, return cluster offset in qcow2 file.
1.1 root 696: * If the offset is not found, allocate a new cluster.
697: *
1.1.1.3 root 698: * If the cluster was already allocated, m->nb_clusters is set to 0,
1.1.1.9 ! root 699: * other fields in m are meaningless.
1.1 root 700: *
1.1.1.3 root 701: * If the cluster is newly allocated, m->nb_clusters is set to the number of
1.1.1.9 ! root 702: * contiguous clusters that have been allocated. In this case, the other
! 703: * fields of m are valid and contain information about the first allocated
! 704: * cluster.
1.1.1.3 root 705: *
1.1.1.9 ! root 706: * If the request conflicts with another write request in flight, the coroutine
! 707: * is queued and will be reentered when the dependency has completed.
1.1.1.3 root 708: *
709: * Return 0 on success and -errno in error cases
1.1 root 710: */
1.1.1.3 root 711: int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
712: int n_start, int n_end, int *num, QCowL2Meta *m)
1.1 root 713: {
714: BDRVQcowState *s = bs->opaque;
715: int l2_index, ret;
1.1.1.3 root 716: uint64_t l2_offset, *l2_table;
717: int64_t cluster_offset;
1.1.1.2 root 718: unsigned int nb_clusters, i = 0;
1.1 root 719: QCowL2Meta *old_alloc;
720:
721: ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
1.1.1.3 root 722: if (ret < 0) {
723: return ret;
724: }
1.1 root 725:
1.1.1.9 ! root 726: again:
1.1 root 727: nb_clusters = size_to_clusters(s, n_end << 9);
728:
729: nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
730:
731: cluster_offset = be64_to_cpu(l2_table[l2_index]);
732:
733: /* We keep all QCOW_OFLAG_COPIED clusters */
734:
735: if (cluster_offset & QCOW_OFLAG_COPIED) {
736: nb_clusters = count_contiguous_clusters(nb_clusters, s->cluster_size,
737: &l2_table[l2_index], 0, 0);
738:
739: cluster_offset &= ~QCOW_OFLAG_COPIED;
740: m->nb_clusters = 0;
741:
742: goto out;
743: }
744:
745: /* for the moment, multiple compressed clusters are not managed */
746:
747: if (cluster_offset & QCOW_OFLAG_COMPRESSED)
748: nb_clusters = 1;
749:
750: /* how many available clusters ? */
751:
752: while (i < nb_clusters) {
753: i += count_contiguous_clusters(nb_clusters - i, s->cluster_size,
754: &l2_table[l2_index], i, 0);
1.1.1.3 root 755: if ((i >= nb_clusters) || be64_to_cpu(l2_table[l2_index + i])) {
1.1 root 756: break;
1.1.1.3 root 757: }
1.1 root 758:
759: i += count_contiguous_free_clusters(nb_clusters - i,
760: &l2_table[l2_index + i]);
1.1.1.3 root 761: if (i >= nb_clusters) {
762: break;
763: }
1.1 root 764:
765: cluster_offset = be64_to_cpu(l2_table[l2_index + i]);
766:
767: if ((cluster_offset & QCOW_OFLAG_COPIED) ||
768: (cluster_offset & QCOW_OFLAG_COMPRESSED))
769: break;
770: }
1.1.1.3 root 771: assert(i <= nb_clusters);
1.1 root 772: nb_clusters = i;
773:
774: /*
775: * Check if there already is an AIO write request in flight which allocates
776: * the same cluster. In this case we need to wait until the previous
777: * request has completed and updated the L2 table accordingly.
778: */
1.1.1.2 root 779: QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
1.1 root 780:
1.1.1.9 ! root 781: uint64_t start = offset >> s->cluster_bits;
! 782: uint64_t end = start + nb_clusters;
! 783: uint64_t old_start = old_alloc->offset >> s->cluster_bits;
! 784: uint64_t old_end = old_start + old_alloc->nb_clusters;
1.1 root 785:
1.1.1.9 ! root 786: if (end < old_start || start > old_end) {
1.1 root 787: /* No intersection */
788: } else {
1.1.1.9 ! root 789: if (start < old_start) {
1.1 root 790: /* Stop at the start of a running allocation */
1.1.1.9 ! root 791: nb_clusters = old_start - start;
1.1 root 792: } else {
793: nb_clusters = 0;
794: }
795:
796: if (nb_clusters == 0) {
1.1.1.9 ! root 797: /* Wait for the dependency to complete. We need to recheck
! 798: * the free/allocated clusters when we continue. */
! 799: qemu_co_mutex_unlock(&s->lock);
! 800: qemu_co_queue_wait(&old_alloc->dependent_requests);
! 801: qemu_co_mutex_lock(&s->lock);
! 802: goto again;
1.1 root 803: }
804: }
805: }
806:
807: if (!nb_clusters) {
808: abort();
809: }
810:
1.1.1.9 ! root 811: /* save info needed for meta data update */
! 812: m->offset = offset;
! 813: m->n_start = n_start;
! 814: m->nb_clusters = nb_clusters;
! 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) {
1.1.1.7 root 822: ret = cluster_offset;
823: goto fail;
1.1.1.3 root 824: }
1.1 root 825:
826: out:
1.1.1.7 root 827: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
828: if (ret < 0) {
1.1.1.8 root 829: goto fail_put;
1.1.1.7 root 830: }
831:
1.1 root 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;
1.1.1.7 root 838:
839: fail:
840: qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1.1.1.8 root 841: fail_put:
842: QLIST_REMOVE(m, next_in_flight);
1.1.1.7 root 843: return ret;
1.1 root 844: }
845:
846: static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
847: const uint8_t *buf, int buf_size)
848: {
849: z_stream strm1, *strm = &strm1;
850: int ret, out_len;
851:
852: memset(strm, 0, sizeof(*strm));
853:
854: strm->next_in = (uint8_t *)buf;
855: strm->avail_in = buf_size;
856: strm->next_out = out_buf;
857: strm->avail_out = out_buf_size;
858:
859: ret = inflateInit2(strm, -12);
860: if (ret != Z_OK)
861: return -1;
862: ret = inflate(strm, Z_FINISH);
863: out_len = strm->next_out - out_buf;
864: if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
865: out_len != out_buf_size) {
866: inflateEnd(strm);
867: return -1;
868: }
869: inflateEnd(strm);
870: return 0;
871: }
872:
1.1.1.6 root 873: int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
1.1 root 874: {
1.1.1.6 root 875: BDRVQcowState *s = bs->opaque;
1.1 root 876: int ret, csize, nb_csectors, sector_offset;
877: uint64_t coffset;
878:
879: coffset = cluster_offset & s->cluster_offset_mask;
880: if (s->cluster_cache_offset != coffset) {
881: nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
882: sector_offset = coffset & 511;
883: csize = nb_csectors * 512 - sector_offset;
1.1.1.6 root 884: BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
885: ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors);
1.1 root 886: if (ret < 0) {
1.1.1.7 root 887: return ret;
1.1 root 888: }
889: if (decompress_buffer(s->cluster_cache, s->cluster_size,
890: s->cluster_data + sector_offset, csize) < 0) {
1.1.1.7 root 891: return -EIO;
1.1 root 892: }
893: s->cluster_cache_offset = coffset;
894: }
895: return 0;
896: }
1.1.1.7 root 897:
898: /*
899: * This discards as many clusters of nb_clusters as possible at once (i.e.
900: * all clusters in the same L2 table) and returns the number of discarded
901: * clusters.
902: */
903: static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
904: unsigned int nb_clusters)
905: {
906: BDRVQcowState *s = bs->opaque;
907: uint64_t l2_offset, *l2_table;
908: int l2_index;
909: int ret;
910: int i;
911:
912: ret = get_cluster_table(bs, offset, &l2_table, &l2_offset, &l2_index);
913: if (ret < 0) {
914: return ret;
915: }
916:
917: /* Limit nb_clusters to one L2 table */
918: nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
919:
920: for (i = 0; i < nb_clusters; i++) {
921: uint64_t old_offset;
922:
923: old_offset = be64_to_cpu(l2_table[l2_index + i]);
924: old_offset &= ~QCOW_OFLAG_COPIED;
925:
926: if (old_offset == 0) {
927: continue;
928: }
929:
930: /* First remove L2 entries */
931: qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
932: l2_table[l2_index + i] = cpu_to_be64(0);
933:
934: /* Then decrease the refcount */
935: qcow2_free_any_clusters(bs, old_offset, 1);
936: }
937:
938: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
939: if (ret < 0) {
940: return ret;
941: }
942:
943: return nb_clusters;
944: }
945:
946: int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
947: int nb_sectors)
948: {
949: BDRVQcowState *s = bs->opaque;
950: uint64_t end_offset;
951: unsigned int nb_clusters;
952: int ret;
953:
954: end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS);
955:
956: /* Round start up and end down */
957: offset = align_offset(offset, s->cluster_size);
958: end_offset &= ~(s->cluster_size - 1);
959:
960: if (offset > end_offset) {
961: return 0;
962: }
963:
964: nb_clusters = size_to_clusters(s, end_offset - offset);
965:
966: /* Each L2 table is handled by its own loop iteration */
967: while (nb_clusters > 0) {
968: ret = discard_single_l2(bs, offset, nb_clusters);
969: if (ret < 0) {
970: return ret;
971: }
972:
973: nb_clusters -= ret;
974: offset += (ret * s->cluster_size);
975: }
976:
977: return 0;
978: }
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.