![[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"
1.1.1.10! root 30: #include "trace.h"
1.1 root 31:
1.1.1.7 root 32: int qcow2_grow_l1_table(BlockDriverState *bs, int min_size, bool exact_size)
1.1 root 33: {
34: BDRVQcowState *s = bs->opaque;
35: int new_l1_size, new_l1_size2, ret, i;
36: uint64_t *new_l1_table;
1.1.1.3 root 37: int64_t new_l1_table_offset;
1.1 root 38: uint8_t data[12];
39:
1.1.1.7 root 40: if (min_size <= s->l1_size)
1.1 root 41: return 0;
1.1.1.7 root 42:
43: if (exact_size) {
44: new_l1_size = min_size;
45: } else {
46: /* Bump size up to reduce the number of times we have to grow */
47: new_l1_size = s->l1_size;
48: if (new_l1_size == 0) {
49: new_l1_size = 1;
50: }
51: while (min_size > new_l1_size) {
52: new_l1_size = (new_l1_size * 3 + 1) / 2;
53: }
1.1 root 54: }
1.1.1.7 root 55:
1.1 root 56: #ifdef DEBUG_ALLOC2
1.1.1.9 root 57: fprintf(stderr, "grow l1_table from %d to %d\n", s->l1_size, new_l1_size);
1.1 root 58: #endif
59:
60: new_l1_size2 = sizeof(uint64_t) * new_l1_size;
1.1.1.9 root 61: new_l1_table = g_malloc0(align_offset(new_l1_size2, 512));
1.1 root 62: memcpy(new_l1_table, s->l1_table, s->l1_size * sizeof(uint64_t));
63:
64: /* write new table (align to cluster) */
1.1.1.6 root 65: BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ALLOC_TABLE);
1.1 root 66: new_l1_table_offset = qcow2_alloc_clusters(bs, new_l1_size2);
1.1.1.3 root 67: if (new_l1_table_offset < 0) {
1.1.1.9 root 68: g_free(new_l1_table);
1.1.1.3 root 69: return new_l1_table_offset;
70: }
1.1 root 71:
1.1.1.7 root 72: ret = qcow2_cache_flush(bs, s->refcount_block_cache);
73: if (ret < 0) {
1.1.1.8 root 74: goto fail;
1.1.1.7 root 75: }
76:
1.1.1.6 root 77: BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_WRITE_TABLE);
1.1 root 78: for(i = 0; i < s->l1_size; i++)
79: new_l1_table[i] = cpu_to_be64(new_l1_table[i]);
1.1.1.6 root 80: ret = bdrv_pwrite_sync(bs->file, new_l1_table_offset, new_l1_table, new_l1_size2);
1.1.1.5 root 81: if (ret < 0)
1.1 root 82: goto fail;
83: for(i = 0; i < s->l1_size; i++)
84: new_l1_table[i] = be64_to_cpu(new_l1_table[i]);
85:
86: /* set new table */
1.1.1.6 root 87: BLKDBG_EVENT(bs->file, BLKDBG_L1_GROW_ACTIVATE_TABLE);
1.1 root 88: cpu_to_be32w((uint32_t*)data, new_l1_size);
1.1.1.7 root 89: cpu_to_be64wu((uint64_t*)(data + 4), new_l1_table_offset);
1.1.1.6 root 90: ret = bdrv_pwrite_sync(bs->file, offsetof(QCowHeader, l1_size), data,sizeof(data));
1.1.1.5 root 91: if (ret < 0) {
1.1 root 92: goto fail;
1.1.1.3 root 93: }
1.1.1.9 root 94: g_free(s->l1_table);
1.1 root 95: qcow2_free_clusters(bs, s->l1_table_offset, s->l1_size * sizeof(uint64_t));
96: s->l1_table_offset = new_l1_table_offset;
97: s->l1_table = new_l1_table;
98: s->l1_size = new_l1_size;
99: return 0;
100: fail:
1.1.1.9 root 101: g_free(new_l1_table);
1.1.1.3 root 102: qcow2_free_clusters(bs, new_l1_table_offset, new_l1_size2);
1.1.1.5 root 103: return ret;
1.1 root 104: }
105:
106: /*
107: * l2_load
108: *
109: * Loads a L2 table into memory. If the table is in the cache, the cache
110: * is used; otherwise the L2 table is loaded from the image file.
111: *
112: * Returns a pointer to the L2 table on success, or NULL if the read from
113: * the image file failed.
114: */
115:
1.1.1.6 root 116: static int l2_load(BlockDriverState *bs, uint64_t l2_offset,
117: uint64_t **l2_table)
1.1 root 118: {
119: BDRVQcowState *s = bs->opaque;
1.1.1.6 root 120: int ret;
1.1 root 121:
1.1.1.7 root 122: ret = qcow2_cache_get(bs, s->l2_table_cache, l2_offset, (void**) l2_table);
1.1 root 123:
1.1.1.7 root 124: return ret;
1.1 root 125: }
126:
127: /*
128: * Writes one sector of the L1 table to the disk (can't update single entries
129: * and we really don't want bdrv_pread to perform a read-modify-write)
130: */
131: #define L1_ENTRIES_PER_SECTOR (512 / 8)
1.1.1.6 root 132: static int write_l1_entry(BlockDriverState *bs, int l1_index)
1.1 root 133: {
1.1.1.6 root 134: BDRVQcowState *s = bs->opaque;
1.1 root 135: uint64_t buf[L1_ENTRIES_PER_SECTOR];
136: int l1_start_index;
1.1.1.5 root 137: int i, ret;
1.1 root 138:
139: l1_start_index = l1_index & ~(L1_ENTRIES_PER_SECTOR - 1);
140: for (i = 0; i < L1_ENTRIES_PER_SECTOR; i++) {
141: buf[i] = cpu_to_be64(s->l1_table[l1_start_index + i]);
142: }
143:
1.1.1.6 root 144: BLKDBG_EVENT(bs->file, BLKDBG_L1_UPDATE);
145: ret = bdrv_pwrite_sync(bs->file, s->l1_table_offset + 8 * l1_start_index,
1.1.1.5 root 146: buf, sizeof(buf));
147: if (ret < 0) {
148: return ret;
1.1 root 149: }
150:
151: return 0;
152: }
153:
154: /*
155: * l2_allocate
156: *
157: * Allocate a new l2 entry in the file. If l1_index points to an already
158: * used entry in the L2 table (i.e. we are doing a copy on write for the L2
159: * table) copy the contents of the old L2 table into the newly allocated one.
160: * Otherwise the new table is initialized with zeros.
161: *
162: */
163:
1.1.1.6 root 164: static int l2_allocate(BlockDriverState *bs, int l1_index, uint64_t **table)
1.1 root 165: {
166: BDRVQcowState *s = bs->opaque;
167: uint64_t old_l2_offset;
1.1.1.3 root 168: uint64_t *l2_table;
169: int64_t l2_offset;
1.1.1.5 root 170: int ret;
1.1 root 171:
172: old_l2_offset = s->l1_table[l1_index];
173:
1.1.1.10! root 174: trace_qcow2_l2_allocate(bs, l1_index);
! 175:
1.1 root 176: /* allocate a new l2 entry */
177:
178: l2_offset = qcow2_alloc_clusters(bs, s->l2_size * sizeof(uint64_t));
1.1.1.3 root 179: if (l2_offset < 0) {
1.1.1.6 root 180: return l2_offset;
1.1.1.3 root 181: }
1.1 root 182:
1.1.1.7 root 183: ret = qcow2_cache_flush(bs, s->refcount_block_cache);
184: if (ret < 0) {
185: goto fail;
186: }
187:
1.1 root 188: /* allocate a new entry in the l2 cache */
189:
1.1.1.10! root 190: trace_qcow2_l2_allocate_get_empty(bs, l1_index);
1.1.1.7 root 191: ret = qcow2_cache_get_empty(bs, s->l2_table_cache, l2_offset, (void**) table);
192: if (ret < 0) {
193: return ret;
194: }
195:
196: l2_table = *table;
1.1 root 197:
1.1.1.10! root 198: if ((old_l2_offset & L1E_OFFSET_MASK) == 0) {
1.1 root 199: /* if there was no old l2 table, clear the new table */
200: memset(l2_table, 0, s->l2_size * sizeof(uint64_t));
201: } else {
1.1.1.7 root 202: uint64_t* old_table;
203:
1.1 root 204: /* if there was an old l2 table, read it from the disk */
1.1.1.6 root 205: BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_COW_READ);
1.1.1.10! root 206: ret = qcow2_cache_get(bs, s->l2_table_cache,
! 207: old_l2_offset & L1E_OFFSET_MASK,
1.1.1.7 root 208: (void**) &old_table);
209: if (ret < 0) {
210: goto fail;
211: }
212:
213: memcpy(l2_table, old_table, s->cluster_size);
214:
215: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &old_table);
1.1.1.6 root 216: if (ret < 0) {
1.1.1.5 root 217: goto fail;
1.1.1.6 root 218: }
1.1 root 219: }
1.1.1.7 root 220:
1.1 root 221: /* write the l2 table to the file */
1.1.1.6 root 222: BLKDBG_EVENT(bs->file, BLKDBG_L2_ALLOC_WRITE);
1.1.1.7 root 223:
1.1.1.10! root 224: trace_qcow2_l2_allocate_write_l2(bs, l1_index);
1.1.1.7 root 225: qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
226: ret = qcow2_cache_flush(bs, s->l2_table_cache);
1.1.1.5 root 227: if (ret < 0) {
228: goto fail;
229: }
230:
231: /* update the L1 entry */
1.1.1.10! root 232: trace_qcow2_l2_allocate_write_l1(bs, l1_index);
1.1.1.5 root 233: s->l1_table[l1_index] = l2_offset | QCOW_OFLAG_COPIED;
1.1.1.6 root 234: ret = write_l1_entry(bs, l1_index);
235: if (ret < 0) {
1.1.1.5 root 236: goto fail;
237: }
1.1 root 238:
1.1.1.6 root 239: *table = l2_table;
1.1.1.10! root 240: trace_qcow2_l2_allocate_done(bs, l1_index, 0);
1.1.1.6 root 241: return 0;
1.1.1.5 root 242:
243: fail:
1.1.1.10! root 244: trace_qcow2_l2_allocate_done(bs, l1_index, ret);
1.1.1.7 root 245: qcow2_cache_put(bs, s->l2_table_cache, (void**) table);
1.1.1.5 root 246: s->l1_table[l1_index] = old_l2_offset;
1.1.1.6 root 247: return ret;
1.1 root 248: }
249:
1.1.1.10! root 250: /*
! 251: * Checks how many clusters in a given L2 table are contiguous in the image
! 252: * file. As soon as one of the flags in the bitmask stop_flags changes compared
! 253: * to the first cluster, the search is stopped and the cluster is not counted
! 254: * as contiguous. (This allows it, for example, to stop at the first compressed
! 255: * cluster which may require a different handling)
! 256: */
1.1 root 257: static int count_contiguous_clusters(uint64_t nb_clusters, int cluster_size,
1.1.1.10! root 258: uint64_t *l2_table, uint64_t start, uint64_t stop_flags)
1.1 root 259: {
260: int i;
1.1.1.10! root 261: uint64_t mask = stop_flags | L2E_OFFSET_MASK;
! 262: uint64_t offset = be64_to_cpu(l2_table[0]) & mask;
1.1 root 263:
264: if (!offset)
265: return 0;
266:
1.1.1.10! root 267: for (i = start; i < start + nb_clusters; i++) {
! 268: uint64_t l2_entry = be64_to_cpu(l2_table[i]) & mask;
! 269: if (offset + (uint64_t) i * cluster_size != l2_entry) {
1.1 root 270: break;
1.1.1.10! root 271: }
! 272: }
1.1 root 273:
274: return (i - start);
275: }
276:
277: static int count_contiguous_free_clusters(uint64_t nb_clusters, uint64_t *l2_table)
278: {
1.1.1.10! root 279: int i;
1.1 root 280:
1.1.1.10! root 281: for (i = 0; i < nb_clusters; i++) {
! 282: int type = qcow2_get_cluster_type(be64_to_cpu(l2_table[i]));
! 283:
! 284: if (type != QCOW2_CLUSTER_UNALLOCATED) {
! 285: break;
! 286: }
! 287: }
1.1 root 288:
289: return i;
290: }
291:
292: /* The crypt function is compatible with the linux cryptoloop
293: algorithm for < 4 GB images. NOTE: out_buf == in_buf is
294: supported */
295: void qcow2_encrypt_sectors(BDRVQcowState *s, int64_t sector_num,
296: uint8_t *out_buf, const uint8_t *in_buf,
297: int nb_sectors, int enc,
298: const AES_KEY *key)
299: {
300: union {
301: uint64_t ll[2];
302: uint8_t b[16];
303: } ivec;
304: int i;
305:
306: for(i = 0; i < nb_sectors; i++) {
307: ivec.ll[0] = cpu_to_le64(sector_num);
308: ivec.ll[1] = 0;
309: AES_cbc_encrypt(in_buf, out_buf, 512, key,
310: ivec.b, enc);
311: sector_num++;
312: in_buf += 512;
313: out_buf += 512;
314: }
315: }
316:
1.1.1.10! root 317: static int coroutine_fn copy_sectors(BlockDriverState *bs,
! 318: uint64_t start_sect,
! 319: uint64_t cluster_offset,
! 320: int n_start, int n_end)
1.1 root 321: {
322: BDRVQcowState *s = bs->opaque;
1.1.1.7 root 323: QEMUIOVector qiov;
1.1.1.10! root 324: struct iovec iov;
! 325: int n, ret;
1.1 root 326:
1.1.1.10! root 327: /*
! 328: * If this is the last cluster and it is only partially used, we must only
! 329: * copy until the end of the image, or bdrv_check_request will fail for the
! 330: * bdrv_read/write calls below.
! 331: */
! 332: if (start_sect + n_end > bs->total_sectors) {
! 333: n_end = bs->total_sectors - start_sect;
1.1 root 334: }
335:
336: n = n_end - n_start;
1.1.1.10! root 337: if (n <= 0) {
1.1 root 338: return 0;
1.1.1.10! root 339: }
! 340:
! 341: iov.iov_len = n * BDRV_SECTOR_SIZE;
! 342: iov.iov_base = qemu_blockalign(bs, iov.iov_len);
! 343:
! 344: qemu_iovec_init_external(&qiov, &iov, 1);
! 345:
1.1.1.6 root 346: BLKDBG_EVENT(bs->file, BLKDBG_COW_READ);
1.1.1.10! root 347:
! 348: /* Call .bdrv_co_readv() directly instead of using the public block-layer
! 349: * interface. This avoids double I/O throttling and request tracking,
! 350: * which can lead to deadlock when block layer copy-on-read is enabled.
! 351: */
! 352: ret = bs->drv->bdrv_co_readv(bs, start_sect + n_start, n, &qiov);
! 353: if (ret < 0) {
! 354: goto out;
! 355: }
! 356:
1.1 root 357: if (s->crypt_method) {
358: qcow2_encrypt_sectors(s, start_sect + n_start,
1.1.1.10! root 359: iov.iov_base, iov.iov_base, n, 1,
1.1 root 360: &s->aes_encrypt_key);
361: }
1.1.1.10! root 362:
1.1.1.6 root 363: BLKDBG_EVENT(bs->file, BLKDBG_COW_WRITE);
1.1.1.10! root 364: ret = bdrv_co_writev(bs->file, (cluster_offset >> 9) + n_start, n, &qiov);
! 365: if (ret < 0) {
! 366: goto out;
! 367: }
! 368:
! 369: ret = 0;
! 370: out:
! 371: qemu_vfree(iov.iov_base);
! 372: return ret;
1.1 root 373: }
374:
375:
376: /*
377: * get_cluster_offset
378: *
1.1.1.6 root 379: * For a given offset of the disk image, find the cluster offset in
380: * qcow2 file. The offset is stored in *cluster_offset.
1.1 root 381: *
1.1.1.9 root 382: * on entry, *num is the number of contiguous sectors we'd like to
1.1 root 383: * access following offset.
384: *
1.1.1.9 root 385: * on exit, *num is the number of contiguous sectors we can read.
1.1 root 386: *
1.1.1.10! root 387: * Returns the cluster type (QCOW2_CLUSTER_*) on success, -errno in error
! 388: * cases.
1.1 root 389: */
1.1.1.6 root 390: int qcow2_get_cluster_offset(BlockDriverState *bs, uint64_t offset,
391: int *num, uint64_t *cluster_offset)
1.1 root 392: {
393: BDRVQcowState *s = bs->opaque;
1.1.1.2 root 394: unsigned int l1_index, l2_index;
1.1.1.6 root 395: uint64_t l2_offset, *l2_table;
1.1 root 396: int l1_bits, c;
1.1.1.2 root 397: unsigned int index_in_cluster, nb_clusters;
398: uint64_t nb_available, nb_needed;
1.1.1.6 root 399: int ret;
1.1 root 400:
401: index_in_cluster = (offset >> 9) & (s->cluster_sectors - 1);
402: nb_needed = *num + index_in_cluster;
403:
404: l1_bits = s->l2_bits + s->cluster_bits;
405:
406: /* compute how many bytes there are between the offset and
407: * the end of the l1 entry
408: */
409:
1.1.1.2 root 410: nb_available = (1ULL << l1_bits) - (offset & ((1ULL << l1_bits) - 1));
1.1 root 411:
412: /* compute the number of available sectors */
413:
414: nb_available = (nb_available >> 9) + index_in_cluster;
415:
416: if (nb_needed > nb_available) {
417: nb_needed = nb_available;
418: }
419:
1.1.1.6 root 420: *cluster_offset = 0;
1.1 root 421:
422: /* seek the the l2 offset in the l1 table */
423:
424: l1_index = offset >> l1_bits;
1.1.1.10! root 425: if (l1_index >= s->l1_size) {
! 426: ret = QCOW2_CLUSTER_UNALLOCATED;
1.1 root 427: goto out;
1.1.1.10! root 428: }
1.1 root 429:
1.1.1.10! root 430: l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
! 431: if (!l2_offset) {
! 432: ret = QCOW2_CLUSTER_UNALLOCATED;
1.1 root 433: goto out;
1.1.1.10! root 434: }
1.1 root 435:
436: /* load the l2 table in memory */
437:
1.1.1.6 root 438: ret = l2_load(bs, l2_offset, &l2_table);
439: if (ret < 0) {
440: return ret;
441: }
1.1 root 442:
443: /* find the cluster offset for the given disk offset */
444:
445: l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
1.1.1.6 root 446: *cluster_offset = be64_to_cpu(l2_table[l2_index]);
1.1 root 447: nb_clusters = size_to_clusters(s, nb_needed << 9);
448:
1.1.1.10! root 449: ret = qcow2_get_cluster_type(*cluster_offset);
! 450: switch (ret) {
! 451: case QCOW2_CLUSTER_COMPRESSED:
! 452: /* Compressed clusters can only be processed one by one */
! 453: c = 1;
! 454: *cluster_offset &= L2E_COMPRESSED_OFFSET_SIZE_MASK;
! 455: break;
! 456: case QCOW2_CLUSTER_ZERO:
! 457: c = count_contiguous_clusters(nb_clusters, s->cluster_size,
! 458: &l2_table[l2_index], 0,
! 459: QCOW_OFLAG_COMPRESSED | QCOW_OFLAG_ZERO);
! 460: *cluster_offset = 0;
! 461: break;
! 462: case QCOW2_CLUSTER_UNALLOCATED:
1.1 root 463: /* how many empty clusters ? */
464: c = count_contiguous_free_clusters(nb_clusters, &l2_table[l2_index]);
1.1.1.10! root 465: *cluster_offset = 0;
! 466: break;
! 467: case QCOW2_CLUSTER_NORMAL:
1.1 root 468: /* how many allocated clusters ? */
469: c = count_contiguous_clusters(nb_clusters, s->cluster_size,
1.1.1.10! root 470: &l2_table[l2_index], 0,
! 471: QCOW_OFLAG_COMPRESSED | QCOW_OFLAG_ZERO);
! 472: *cluster_offset &= L2E_OFFSET_MASK;
! 473: break;
! 474: default:
! 475: abort();
1.1 root 476: }
477:
1.1.1.7 root 478: qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
479:
1.1.1.10! root 480: nb_available = (c * s->cluster_sectors);
! 481:
1.1 root 482: out:
483: if (nb_available > nb_needed)
484: nb_available = nb_needed;
485:
486: *num = nb_available - index_in_cluster;
487:
1.1.1.10! root 488: return ret;
1.1 root 489: }
490:
491: /*
492: * get_cluster_table
493: *
494: * for a given disk offset, load (and allocate if needed)
495: * the l2 table.
496: *
497: * the l2 table offset in the qcow2 file and the cluster index
498: * in the l2 table are given to the caller.
499: *
1.1.1.3 root 500: * Returns 0 on success, -errno in failure case
1.1 root 501: */
502: static int get_cluster_table(BlockDriverState *bs, uint64_t offset,
503: uint64_t **new_l2_table,
504: int *new_l2_index)
505: {
506: BDRVQcowState *s = bs->opaque;
1.1.1.2 root 507: unsigned int l1_index, l2_index;
1.1.1.6 root 508: uint64_t l2_offset;
509: uint64_t *l2_table = NULL;
1.1.1.2 root 510: int ret;
1.1 root 511:
512: /* seek the the l2 offset in the l1 table */
513:
514: l1_index = offset >> (s->l2_bits + s->cluster_bits);
515: if (l1_index >= s->l1_size) {
1.1.1.7 root 516: ret = qcow2_grow_l1_table(bs, l1_index + 1, false);
1.1.1.3 root 517: if (ret < 0) {
518: return ret;
519: }
1.1 root 520: }
1.1.1.10! root 521:
! 522: l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
1.1 root 523:
524: /* seek the l2 table of the given l2 offset */
525:
1.1.1.10! root 526: if (s->l1_table[l1_index] & QCOW_OFLAG_COPIED) {
1.1 root 527: /* load the l2 table in memory */
1.1.1.6 root 528: ret = l2_load(bs, l2_offset, &l2_table);
529: if (ret < 0) {
530: return ret;
1.1.1.3 root 531: }
1.1 root 532: } else {
1.1.1.7 root 533: /* First allocate a new L2 table (and do COW if needed) */
1.1.1.6 root 534: ret = l2_allocate(bs, l1_index, &l2_table);
535: if (ret < 0) {
536: return ret;
1.1.1.3 root 537: }
1.1.1.7 root 538:
539: /* Then decrease the refcount of the old table */
540: if (l2_offset) {
541: qcow2_free_clusters(bs, l2_offset, s->l2_size * sizeof(uint64_t));
542: }
1.1.1.10! root 543: l2_offset = s->l1_table[l1_index] & L1E_OFFSET_MASK;
1.1 root 544: }
545:
546: /* find the cluster offset for the given disk offset */
547:
548: l2_index = (offset >> s->cluster_bits) & (s->l2_size - 1);
549:
550: *new_l2_table = l2_table;
551: *new_l2_index = l2_index;
552:
1.1.1.3 root 553: return 0;
1.1 root 554: }
555:
556: /*
557: * alloc_compressed_cluster_offset
558: *
559: * For a given offset of the disk image, return cluster offset in
560: * qcow2 file.
561: *
562: * If the offset is not found, allocate a new compressed cluster.
563: *
564: * Return the cluster offset if successful,
565: * Return 0, otherwise.
566: *
567: */
568:
569: uint64_t qcow2_alloc_compressed_cluster_offset(BlockDriverState *bs,
570: uint64_t offset,
571: int compressed_size)
572: {
573: BDRVQcowState *s = bs->opaque;
574: int l2_index, ret;
1.1.1.10! root 575: uint64_t *l2_table;
1.1.1.3 root 576: int64_t cluster_offset;
1.1 root 577: int nb_csectors;
578:
1.1.1.10! root 579: ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1.1.1.3 root 580: if (ret < 0) {
1.1 root 581: return 0;
1.1.1.3 root 582: }
1.1 root 583:
1.1.1.10! root 584: /* Compression can't overwrite anything. Fail if the cluster was already
! 585: * allocated. */
1.1 root 586: cluster_offset = be64_to_cpu(l2_table[l2_index]);
1.1.1.10! root 587: if (cluster_offset & L2E_OFFSET_MASK) {
1.1.1.9 root 588: qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
589: return 0;
590: }
1.1 root 591:
592: cluster_offset = qcow2_alloc_bytes(bs, compressed_size);
1.1.1.3 root 593: if (cluster_offset < 0) {
1.1.1.7 root 594: qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1.1.1.3 root 595: return 0;
596: }
597:
1.1 root 598: nb_csectors = ((cluster_offset + compressed_size - 1) >> 9) -
599: (cluster_offset >> 9);
600:
601: cluster_offset |= QCOW_OFLAG_COMPRESSED |
602: ((uint64_t)nb_csectors << s->csize_shift);
603:
604: /* update L2 table */
605:
606: /* compressed clusters never have the copied flag */
607:
1.1.1.6 root 608: BLKDBG_EVENT(bs->file, BLKDBG_L2_UPDATE_COMPRESSED);
1.1.1.7 root 609: qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
1.1 root 610: l2_table[l2_index] = cpu_to_be64(cluster_offset);
1.1.1.7 root 611: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1.1.1.5 root 612: if (ret < 0) {
1.1.1.7 root 613: return 0;
1.1 root 614: }
615:
1.1.1.7 root 616: return cluster_offset;
1.1 root 617: }
618:
1.1.1.3 root 619: int qcow2_alloc_cluster_link_l2(BlockDriverState *bs, QCowL2Meta *m)
1.1 root 620: {
621: BDRVQcowState *s = bs->opaque;
622: int i, j = 0, l2_index, ret;
1.1.1.10! root 623: uint64_t *old_cluster, start_sect, *l2_table;
! 624: uint64_t cluster_offset = m->alloc_offset;
1.1.1.7 root 625: bool cow = false;
1.1 root 626:
1.1.1.10! root 627: trace_qcow2_cluster_link_l2(qemu_coroutine_self(), m->nb_clusters);
! 628:
1.1 root 629: if (m->nb_clusters == 0)
630: return 0;
631:
1.1.1.9 root 632: old_cluster = g_malloc(m->nb_clusters * sizeof(uint64_t));
1.1 root 633:
634: /* copy content of unmodified sectors */
635: start_sect = (m->offset & ~(s->cluster_size - 1)) >> 9;
636: if (m->n_start) {
1.1.1.7 root 637: cow = true;
1.1.1.10! root 638: qemu_co_mutex_unlock(&s->lock);
1.1 root 639: ret = copy_sectors(bs, start_sect, cluster_offset, 0, m->n_start);
1.1.1.10! root 640: qemu_co_mutex_lock(&s->lock);
1.1 root 641: if (ret < 0)
642: goto err;
643: }
644:
645: if (m->nb_available & (s->cluster_sectors - 1)) {
646: uint64_t end = m->nb_available & ~(uint64_t)(s->cluster_sectors - 1);
1.1.1.7 root 647: cow = true;
1.1.1.10! root 648: qemu_co_mutex_unlock(&s->lock);
1.1 root 649: ret = copy_sectors(bs, start_sect + end, cluster_offset + (end << 9),
650: m->nb_available - end, s->cluster_sectors);
1.1.1.10! root 651: qemu_co_mutex_lock(&s->lock);
1.1 root 652: if (ret < 0)
653: goto err;
654: }
655:
1.1.1.7 root 656: /*
657: * Update L2 table.
658: *
659: * Before we update the L2 table to actually point to the new cluster, we
660: * need to be sure that the refcounts have been increased and COW was
661: * handled.
662: */
663: if (cow) {
664: qcow2_cache_depends_on_flush(s->l2_table_cache);
665: }
666:
667: qcow2_cache_set_dependency(bs, s->l2_table_cache, s->refcount_block_cache);
1.1.1.10! root 668: ret = get_cluster_table(bs, m->offset, &l2_table, &l2_index);
1.1.1.3 root 669: if (ret < 0) {
1.1 root 670: goto err;
1.1.1.3 root 671: }
1.1.1.7 root 672: qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
1.1 root 673:
674: for (i = 0; i < m->nb_clusters; i++) {
675: /* if two concurrent writes happen to the same unallocated cluster
676: * each write allocates separate cluster and writes data concurrently.
677: * The first one to complete updates l2 table with pointer to its
678: * cluster the second one has to do RMW (which is done above by
679: * copy_sectors()), update l2 table with its cluster pointer and free
680: * old cluster. This is what this loop does */
681: if(l2_table[l2_index + i] != 0)
682: old_cluster[j++] = l2_table[l2_index + i];
683:
684: l2_table[l2_index + i] = cpu_to_be64((cluster_offset +
685: (i << s->cluster_bits)) | QCOW_OFLAG_COPIED);
686: }
687:
1.1.1.7 root 688:
689: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1.1.1.5 root 690: if (ret < 0) {
1.1 root 691: goto err;
692: }
693:
1.1.1.6 root 694: /*
695: * If this was a COW, we need to decrease the refcount of the old cluster.
696: * Also flush bs->file to get the right order for L2 and refcount update.
697: */
698: if (j != 0) {
699: for (i = 0; i < j; i++) {
1.1.1.10! root 700: qcow2_free_any_clusters(bs, be64_to_cpu(old_cluster[i]), 1);
1.1.1.6 root 701: }
702: }
1.1 root 703:
704: ret = 0;
705: err:
1.1.1.9 root 706: g_free(old_cluster);
1.1 root 707: return ret;
708: }
709:
710: /*
1.1.1.10! root 711: * Returns the number of contiguous clusters that can be used for an allocating
! 712: * write, but require COW to be performed (this includes yet unallocated space,
! 713: * which must copy from the backing file)
1.1 root 714: */
1.1.1.10! root 715: static int count_cow_clusters(BDRVQcowState *s, int nb_clusters,
! 716: uint64_t *l2_table, int l2_index)
1.1 root 717: {
1.1.1.10! root 718: int i;
1.1 root 719:
1.1.1.10! root 720: for (i = 0; i < nb_clusters; i++) {
! 721: uint64_t l2_entry = be64_to_cpu(l2_table[l2_index + i]);
! 722: int cluster_type = qcow2_get_cluster_type(l2_entry);
1.1 root 723:
1.1.1.10! root 724: switch(cluster_type) {
! 725: case QCOW2_CLUSTER_NORMAL:
! 726: if (l2_entry & QCOW_OFLAG_COPIED) {
! 727: goto out;
! 728: }
1.1 root 729: break;
1.1.1.10! root 730: case QCOW2_CLUSTER_UNALLOCATED:
! 731: case QCOW2_CLUSTER_COMPRESSED:
! 732: case QCOW2_CLUSTER_ZERO:
1.1.1.3 root 733: break;
1.1.1.10! root 734: default:
! 735: abort();
1.1.1.3 root 736: }
1.1 root 737: }
1.1.1.10! root 738:
! 739: out:
1.1.1.3 root 740: assert(i <= nb_clusters);
1.1.1.10! root 741: return i;
! 742: }
! 743:
! 744: /*
! 745: * Allocates new clusters for the given guest_offset.
! 746: *
! 747: * At most *nb_clusters are allocated, and on return *nb_clusters is updated to
! 748: * contain the number of clusters that have been allocated and are contiguous
! 749: * in the image file.
! 750: *
! 751: * If *host_offset is non-zero, it specifies the offset in the image file at
! 752: * which the new clusters must start. *nb_clusters can be 0 on return in this
! 753: * case if the cluster at host_offset is already in use. If *host_offset is
! 754: * zero, the clusters can be allocated anywhere in the image file.
! 755: *
! 756: * *host_offset is updated to contain the offset into the image file at which
! 757: * the first allocated cluster starts.
! 758: *
! 759: * Return 0 on success and -errno in error cases. -EAGAIN means that the
! 760: * function has been waiting for another request and the allocation must be
! 761: * restarted, but the whole request should not be failed.
! 762: */
! 763: static int do_alloc_cluster_offset(BlockDriverState *bs, uint64_t guest_offset,
! 764: uint64_t *host_offset, unsigned int *nb_clusters)
! 765: {
! 766: BDRVQcowState *s = bs->opaque;
! 767: QCowL2Meta *old_alloc;
! 768:
! 769: trace_qcow2_do_alloc_clusters_offset(qemu_coroutine_self(), guest_offset,
! 770: *host_offset, *nb_clusters);
1.1 root 771:
772: /*
773: * Check if there already is an AIO write request in flight which allocates
774: * the same cluster. In this case we need to wait until the previous
775: * request has completed and updated the L2 table accordingly.
776: */
1.1.1.2 root 777: QLIST_FOREACH(old_alloc, &s->cluster_allocs, next_in_flight) {
1.1 root 778:
1.1.1.10! root 779: uint64_t start = guest_offset >> s->cluster_bits;
! 780: uint64_t end = start + *nb_clusters;
1.1.1.9 root 781: uint64_t old_start = old_alloc->offset >> s->cluster_bits;
782: uint64_t old_end = old_start + old_alloc->nb_clusters;
1.1 root 783:
1.1.1.9 root 784: if (end < old_start || start > old_end) {
1.1 root 785: /* No intersection */
786: } else {
1.1.1.9 root 787: if (start < old_start) {
1.1 root 788: /* Stop at the start of a running allocation */
1.1.1.10! root 789: *nb_clusters = old_start - start;
1.1 root 790: } else {
1.1.1.10! root 791: *nb_clusters = 0;
1.1 root 792: }
793:
1.1.1.10! root 794: if (*nb_clusters == 0) {
1.1.1.9 root 795: /* Wait for the dependency to complete. We need to recheck
796: * the free/allocated clusters when we continue. */
797: qemu_co_mutex_unlock(&s->lock);
798: qemu_co_queue_wait(&old_alloc->dependent_requests);
799: qemu_co_mutex_lock(&s->lock);
1.1.1.10! root 800: return -EAGAIN;
1.1 root 801: }
802: }
803: }
804:
1.1.1.10! root 805: if (!*nb_clusters) {
1.1 root 806: abort();
807: }
808:
1.1.1.10! root 809: /* Allocate new clusters */
! 810: trace_qcow2_cluster_alloc_phys(qemu_coroutine_self());
! 811: if (*host_offset == 0) {
! 812: int64_t cluster_offset =
! 813: qcow2_alloc_clusters(bs, *nb_clusters * s->cluster_size);
! 814: if (cluster_offset < 0) {
! 815: return cluster_offset;
! 816: }
! 817: *host_offset = cluster_offset;
! 818: return 0;
! 819: } else {
! 820: int ret = qcow2_alloc_clusters_at(bs, *host_offset, *nb_clusters);
! 821: if (ret < 0) {
! 822: return ret;
! 823: }
! 824: *nb_clusters = ret;
! 825: return 0;
! 826: }
! 827: }
! 828:
! 829: /*
! 830: * alloc_cluster_offset
! 831: *
! 832: * For a given offset on the virtual disk, find the cluster offset in qcow2
! 833: * file. If the offset is not found, allocate a new cluster.
! 834: *
! 835: * If the cluster was already allocated, m->nb_clusters is set to 0 and
! 836: * other fields in m are meaningless.
! 837: *
! 838: * If the cluster is newly allocated, m->nb_clusters is set to the number of
! 839: * contiguous clusters that have been allocated. In this case, the other
! 840: * fields of m are valid and contain information about the first allocated
! 841: * cluster.
! 842: *
! 843: * If the request conflicts with another write request in flight, the coroutine
! 844: * is queued and will be reentered when the dependency has completed.
! 845: *
! 846: * Return 0 on success and -errno in error cases
! 847: */
! 848: int qcow2_alloc_cluster_offset(BlockDriverState *bs, uint64_t offset,
! 849: int n_start, int n_end, int *num, QCowL2Meta *m)
! 850: {
! 851: BDRVQcowState *s = bs->opaque;
! 852: int l2_index, ret, sectors;
! 853: uint64_t *l2_table;
! 854: unsigned int nb_clusters, keep_clusters;
! 855: uint64_t cluster_offset;
! 856:
! 857: trace_qcow2_alloc_clusters_offset(qemu_coroutine_self(), offset,
! 858: n_start, n_end);
! 859:
! 860: /* Find L2 entry for the first involved cluster */
! 861: again:
! 862: ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
! 863: if (ret < 0) {
! 864: return ret;
! 865: }
! 866:
! 867: /*
! 868: * Calculate the number of clusters to look for. We stop at L2 table
! 869: * boundaries to keep things simple.
! 870: */
! 871: nb_clusters = MIN(size_to_clusters(s, n_end << BDRV_SECTOR_BITS),
! 872: s->l2_size - l2_index);
1.1.1.9 root 873:
1.1.1.10! root 874: cluster_offset = be64_to_cpu(l2_table[l2_index]);
1.1 root 875:
1.1.1.10! root 876: /*
! 877: * Check how many clusters are already allocated and don't need COW, and how
! 878: * many need a new allocation.
! 879: */
! 880: if (qcow2_get_cluster_type(cluster_offset) == QCOW2_CLUSTER_NORMAL
! 881: && (cluster_offset & QCOW_OFLAG_COPIED))
! 882: {
! 883: /* We keep all QCOW_OFLAG_COPIED clusters */
! 884: keep_clusters =
! 885: count_contiguous_clusters(nb_clusters, s->cluster_size,
! 886: &l2_table[l2_index], 0,
! 887: QCOW_OFLAG_COPIED | QCOW_OFLAG_ZERO);
! 888: assert(keep_clusters <= nb_clusters);
! 889: nb_clusters -= keep_clusters;
! 890: } else {
! 891: keep_clusters = 0;
! 892: cluster_offset = 0;
! 893: }
1.1 root 894:
1.1.1.10! root 895: if (nb_clusters > 0) {
! 896: /* For the moment, overwrite compressed clusters one by one */
! 897: uint64_t entry = be64_to_cpu(l2_table[l2_index + keep_clusters]);
! 898: if (entry & QCOW_OFLAG_COMPRESSED) {
! 899: nb_clusters = 1;
! 900: } else {
! 901: nb_clusters = count_cow_clusters(s, nb_clusters, l2_table,
! 902: l2_index + keep_clusters);
! 903: }
1.1.1.3 root 904: }
1.1 root 905:
1.1.1.10! root 906: cluster_offset &= L2E_OFFSET_MASK;
! 907:
! 908: /*
! 909: * The L2 table isn't used any more after this. As long as the cache works
! 910: * synchronously, it's important to release it before calling
! 911: * do_alloc_cluster_offset, which may yield if we need to wait for another
! 912: * request to complete. If we still had the reference, we could use up the
! 913: * whole cache with sleeping requests.
! 914: */
1.1.1.7 root 915: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
916: if (ret < 0) {
1.1.1.10! root 917: return ret;
! 918: }
! 919:
! 920: /* If there is something left to allocate, do that now */
! 921: *m = (QCowL2Meta) {
! 922: .cluster_offset = cluster_offset,
! 923: .nb_clusters = 0,
! 924: };
! 925: qemu_co_queue_init(&m->dependent_requests);
! 926:
! 927: if (nb_clusters > 0) {
! 928: uint64_t alloc_offset;
! 929: uint64_t alloc_cluster_offset;
! 930: uint64_t keep_bytes = keep_clusters * s->cluster_size;
! 931:
! 932: /* Calculate start and size of allocation */
! 933: alloc_offset = offset + keep_bytes;
! 934:
! 935: if (keep_clusters == 0) {
! 936: alloc_cluster_offset = 0;
! 937: } else {
! 938: alloc_cluster_offset = cluster_offset + keep_bytes;
! 939: }
! 940:
! 941: /* Allocate, if necessary at a given offset in the image file */
! 942: ret = do_alloc_cluster_offset(bs, alloc_offset, &alloc_cluster_offset,
! 943: &nb_clusters);
! 944: if (ret == -EAGAIN) {
! 945: goto again;
! 946: } else if (ret < 0) {
! 947: goto fail;
! 948: }
! 949:
! 950: /* save info needed for meta data update */
! 951: if (nb_clusters > 0) {
! 952: int requested_sectors = n_end - keep_clusters * s->cluster_sectors;
! 953: int avail_sectors = (keep_clusters + nb_clusters)
! 954: << (s->cluster_bits - BDRV_SECTOR_BITS);
! 955:
! 956: *m = (QCowL2Meta) {
! 957: .cluster_offset = keep_clusters == 0 ?
! 958: alloc_cluster_offset : cluster_offset,
! 959: .alloc_offset = alloc_cluster_offset,
! 960: .offset = alloc_offset,
! 961: .n_start = keep_clusters == 0 ? n_start : 0,
! 962: .nb_clusters = nb_clusters,
! 963: .nb_available = MIN(requested_sectors, avail_sectors),
! 964: };
! 965: qemu_co_queue_init(&m->dependent_requests);
! 966: QLIST_INSERT_HEAD(&s->cluster_allocs, m, next_in_flight);
! 967: }
1.1.1.7 root 968: }
969:
1.1.1.10! root 970: /* Some cleanup work */
! 971: sectors = (keep_clusters + nb_clusters) << (s->cluster_bits - 9);
! 972: if (sectors > n_end) {
! 973: sectors = n_end;
! 974: }
1.1 root 975:
1.1.1.10! root 976: assert(sectors > n_start);
! 977: *num = sectors - n_start;
1.1 root 978:
1.1.1.3 root 979: return 0;
1.1.1.7 root 980:
981: fail:
1.1.1.10! root 982: if (m->nb_clusters > 0) {
! 983: QLIST_REMOVE(m, next_in_flight);
! 984: }
1.1.1.7 root 985: return ret;
1.1 root 986: }
987:
988: static int decompress_buffer(uint8_t *out_buf, int out_buf_size,
989: const uint8_t *buf, int buf_size)
990: {
991: z_stream strm1, *strm = &strm1;
992: int ret, out_len;
993:
994: memset(strm, 0, sizeof(*strm));
995:
996: strm->next_in = (uint8_t *)buf;
997: strm->avail_in = buf_size;
998: strm->next_out = out_buf;
999: strm->avail_out = out_buf_size;
1000:
1001: ret = inflateInit2(strm, -12);
1002: if (ret != Z_OK)
1003: return -1;
1004: ret = inflate(strm, Z_FINISH);
1005: out_len = strm->next_out - out_buf;
1006: if ((ret != Z_STREAM_END && ret != Z_BUF_ERROR) ||
1007: out_len != out_buf_size) {
1008: inflateEnd(strm);
1009: return -1;
1010: }
1011: inflateEnd(strm);
1012: return 0;
1013: }
1014:
1.1.1.6 root 1015: int qcow2_decompress_cluster(BlockDriverState *bs, uint64_t cluster_offset)
1.1 root 1016: {
1.1.1.6 root 1017: BDRVQcowState *s = bs->opaque;
1.1 root 1018: int ret, csize, nb_csectors, sector_offset;
1019: uint64_t coffset;
1020:
1021: coffset = cluster_offset & s->cluster_offset_mask;
1022: if (s->cluster_cache_offset != coffset) {
1023: nb_csectors = ((cluster_offset >> s->csize_shift) & s->csize_mask) + 1;
1024: sector_offset = coffset & 511;
1025: csize = nb_csectors * 512 - sector_offset;
1.1.1.6 root 1026: BLKDBG_EVENT(bs->file, BLKDBG_READ_COMPRESSED);
1027: ret = bdrv_read(bs->file, coffset >> 9, s->cluster_data, nb_csectors);
1.1 root 1028: if (ret < 0) {
1.1.1.7 root 1029: return ret;
1.1 root 1030: }
1031: if (decompress_buffer(s->cluster_cache, s->cluster_size,
1032: s->cluster_data + sector_offset, csize) < 0) {
1.1.1.7 root 1033: return -EIO;
1.1 root 1034: }
1035: s->cluster_cache_offset = coffset;
1036: }
1037: return 0;
1038: }
1.1.1.7 root 1039:
1040: /*
1041: * This discards as many clusters of nb_clusters as possible at once (i.e.
1042: * all clusters in the same L2 table) and returns the number of discarded
1043: * clusters.
1044: */
1045: static int discard_single_l2(BlockDriverState *bs, uint64_t offset,
1046: unsigned int nb_clusters)
1047: {
1048: BDRVQcowState *s = bs->opaque;
1.1.1.10! root 1049: uint64_t *l2_table;
1.1.1.7 root 1050: int l2_index;
1051: int ret;
1052: int i;
1053:
1.1.1.10! root 1054: ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
1.1.1.7 root 1055: if (ret < 0) {
1056: return ret;
1057: }
1058:
1059: /* Limit nb_clusters to one L2 table */
1060: nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
1061:
1062: for (i = 0; i < nb_clusters; i++) {
1063: uint64_t old_offset;
1064:
1065: old_offset = be64_to_cpu(l2_table[l2_index + i]);
1.1.1.10! root 1066: if ((old_offset & L2E_OFFSET_MASK) == 0) {
1.1.1.7 root 1067: continue;
1068: }
1069:
1070: /* First remove L2 entries */
1071: qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
1072: l2_table[l2_index + i] = cpu_to_be64(0);
1073:
1074: /* Then decrease the refcount */
1075: qcow2_free_any_clusters(bs, old_offset, 1);
1076: }
1077:
1078: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
1079: if (ret < 0) {
1080: return ret;
1081: }
1082:
1083: return nb_clusters;
1084: }
1085:
1086: int qcow2_discard_clusters(BlockDriverState *bs, uint64_t offset,
1087: int nb_sectors)
1088: {
1089: BDRVQcowState *s = bs->opaque;
1090: uint64_t end_offset;
1091: unsigned int nb_clusters;
1092: int ret;
1093:
1094: end_offset = offset + (nb_sectors << BDRV_SECTOR_BITS);
1095:
1096: /* Round start up and end down */
1097: offset = align_offset(offset, s->cluster_size);
1098: end_offset &= ~(s->cluster_size - 1);
1099:
1100: if (offset > end_offset) {
1101: return 0;
1102: }
1103:
1104: nb_clusters = size_to_clusters(s, end_offset - offset);
1105:
1106: /* Each L2 table is handled by its own loop iteration */
1107: while (nb_clusters > 0) {
1108: ret = discard_single_l2(bs, offset, nb_clusters);
1109: if (ret < 0) {
1110: return ret;
1111: }
1112:
1113: nb_clusters -= ret;
1114: offset += (ret * s->cluster_size);
1115: }
1116:
1117: return 0;
1118: }
1.1.1.10! root 1119:
! 1120: /*
! 1121: * This zeroes as many clusters of nb_clusters as possible at once (i.e.
! 1122: * all clusters in the same L2 table) and returns the number of zeroed
! 1123: * clusters.
! 1124: */
! 1125: static int zero_single_l2(BlockDriverState *bs, uint64_t offset,
! 1126: unsigned int nb_clusters)
! 1127: {
! 1128: BDRVQcowState *s = bs->opaque;
! 1129: uint64_t *l2_table;
! 1130: int l2_index;
! 1131: int ret;
! 1132: int i;
! 1133:
! 1134: ret = get_cluster_table(bs, offset, &l2_table, &l2_index);
! 1135: if (ret < 0) {
! 1136: return ret;
! 1137: }
! 1138:
! 1139: /* Limit nb_clusters to one L2 table */
! 1140: nb_clusters = MIN(nb_clusters, s->l2_size - l2_index);
! 1141:
! 1142: for (i = 0; i < nb_clusters; i++) {
! 1143: uint64_t old_offset;
! 1144:
! 1145: old_offset = be64_to_cpu(l2_table[l2_index + i]);
! 1146:
! 1147: /* Update L2 entries */
! 1148: qcow2_cache_entry_mark_dirty(s->l2_table_cache, l2_table);
! 1149: if (old_offset & QCOW_OFLAG_COMPRESSED) {
! 1150: l2_table[l2_index + i] = cpu_to_be64(QCOW_OFLAG_ZERO);
! 1151: qcow2_free_any_clusters(bs, old_offset, 1);
! 1152: } else {
! 1153: l2_table[l2_index + i] |= cpu_to_be64(QCOW_OFLAG_ZERO);
! 1154: }
! 1155: }
! 1156:
! 1157: ret = qcow2_cache_put(bs, s->l2_table_cache, (void**) &l2_table);
! 1158: if (ret < 0) {
! 1159: return ret;
! 1160: }
! 1161:
! 1162: return nb_clusters;
! 1163: }
! 1164:
! 1165: int qcow2_zero_clusters(BlockDriverState *bs, uint64_t offset, int nb_sectors)
! 1166: {
! 1167: BDRVQcowState *s = bs->opaque;
! 1168: unsigned int nb_clusters;
! 1169: int ret;
! 1170:
! 1171: /* The zero flag is only supported by version 3 and newer */
! 1172: if (s->qcow_version < 3) {
! 1173: return -ENOTSUP;
! 1174: }
! 1175:
! 1176: /* Each L2 table is handled by its own loop iteration */
! 1177: nb_clusters = size_to_clusters(s, nb_sectors << BDRV_SECTOR_BITS);
! 1178:
! 1179: while (nb_clusters > 0) {
! 1180: ret = zero_single_l2(bs, offset, nb_clusters);
! 1181: if (ret < 0) {
! 1182: return ret;
! 1183: }
! 1184:
! 1185: nb_clusters -= ret;
! 1186: offset += (ret * s->cluster_size);
! 1187: }
! 1188:
! 1189: return 0;
! 1190: }
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.