debuggers.hg

view tools/ioemu/block-vmdk.c @ 10969:3ea0a6847198

[qemu] Make sure disk writes really made it to disk before we report I/O
completion to the guest domain. The DMA_MULTI_THREAD functionality
from the qemu-dm IDE emulation should make the performance overhead
of synchronous writes bearable, or at least comparable to native
hardware.

Signed-off-by: Rik van Riel <riel@redhat.com>
author chris@kneesaa.uk.xensource.com
date Fri Aug 04 10:28:51 2006 +0100 (2006-08-04)
parents b450f21472a0
children 08a11694b109
line source
1 /*
2 * Block driver for the VMDK format
3 *
4 * Copyright (c) 2004 Fabrice Bellard
5 * Copyright (c) 2005 Filip Navara
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
24 */
25 #include "vl.h"
26 #include "block_int.h"
28 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
29 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
31 typedef struct {
32 uint32_t version;
33 uint32_t flags;
34 uint32_t disk_sectors;
35 uint32_t granularity;
36 uint32_t l1dir_offset;
37 uint32_t l1dir_size;
38 uint32_t file_sectors;
39 uint32_t cylinders;
40 uint32_t heads;
41 uint32_t sectors_per_track;
42 } VMDK3Header;
44 typedef struct {
45 uint32_t version;
46 uint32_t flags;
47 int64_t capacity;
48 int64_t granularity;
49 int64_t desc_offset;
50 int64_t desc_size;
51 int32_t num_gtes_per_gte;
52 int64_t rgd_offset;
53 int64_t gd_offset;
54 int64_t grain_offset;
55 char filler[1];
56 char check_bytes[4];
57 } __attribute__((packed)) VMDK4Header;
59 #define L2_CACHE_SIZE 16
61 typedef struct BDRVVmdkState {
62 int fd;
63 int64_t l1_table_offset;
64 int64_t l1_backup_table_offset;
65 uint32_t *l1_table;
66 uint32_t *l1_backup_table;
67 unsigned int l1_size;
68 uint32_t l1_entry_sectors;
70 unsigned int l2_size;
71 uint32_t *l2_cache;
72 uint32_t l2_cache_offsets[L2_CACHE_SIZE];
73 uint32_t l2_cache_counts[L2_CACHE_SIZE];
75 unsigned int cluster_sectors;
76 } BDRVVmdkState;
78 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
79 {
80 uint32_t magic;
82 if (buf_size < 4)
83 return 0;
84 magic = be32_to_cpu(*(uint32_t *)buf);
85 if (magic == VMDK3_MAGIC ||
86 magic == VMDK4_MAGIC)
87 return 100;
88 else
89 return 0;
90 }
92 static int vmdk_open(BlockDriverState *bs, const char *filename)
93 {
94 BDRVVmdkState *s = bs->opaque;
95 int fd, i;
96 uint32_t magic;
97 int l1_size;
99 fd = open(filename, O_RDWR | O_BINARY | O_LARGEFILE | O_SYNC);
100 if (fd < 0) {
101 fd = open(filename, O_RDONLY | O_BINARY | O_LARGEFILE);
102 if (fd < 0)
103 return -1;
104 bs->read_only = 1;
105 }
106 if (read(fd, &magic, sizeof(magic)) != sizeof(magic))
107 goto fail;
108 magic = be32_to_cpu(magic);
109 if (magic == VMDK3_MAGIC) {
110 VMDK3Header header;
111 if (read(fd, &header, sizeof(header)) !=
112 sizeof(header))
113 goto fail;
114 s->cluster_sectors = le32_to_cpu(header.granularity);
115 s->l2_size = 1 << 9;
116 s->l1_size = 1 << 6;
117 bs->total_sectors = le32_to_cpu(header.disk_sectors);
118 s->l1_table_offset = le32_to_cpu(header.l1dir_offset) << 9;
119 s->l1_backup_table_offset = 0;
120 s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
121 } else if (magic == VMDK4_MAGIC) {
122 VMDK4Header header;
124 if (read(fd, &header, sizeof(header)) != sizeof(header))
125 goto fail;
126 bs->total_sectors = le64_to_cpu(header.capacity);
127 s->cluster_sectors = le64_to_cpu(header.granularity);
128 s->l2_size = le32_to_cpu(header.num_gtes_per_gte);
129 s->l1_entry_sectors = s->l2_size * s->cluster_sectors;
130 if (s->l1_entry_sectors <= 0)
131 goto fail;
132 s->l1_size = (bs->total_sectors + s->l1_entry_sectors - 1)
133 / s->l1_entry_sectors;
134 s->l1_table_offset = le64_to_cpu(header.rgd_offset) << 9;
135 s->l1_backup_table_offset = le64_to_cpu(header.gd_offset) << 9;
136 } else {
137 goto fail;
138 }
139 /* read the L1 table */
140 l1_size = s->l1_size * sizeof(uint32_t);
141 s->l1_table = qemu_malloc(l1_size);
142 if (!s->l1_table)
143 goto fail;
144 if (lseek(fd, s->l1_table_offset, SEEK_SET) == -1)
145 goto fail;
146 if (read(fd, s->l1_table, l1_size) != l1_size)
147 goto fail;
148 for(i = 0; i < s->l1_size; i++) {
149 le32_to_cpus(&s->l1_table[i]);
150 }
152 if (s->l1_backup_table_offset) {
153 s->l1_backup_table = qemu_malloc(l1_size);
154 if (!s->l1_backup_table)
155 goto fail;
156 if (lseek(fd, s->l1_backup_table_offset, SEEK_SET) == -1)
157 goto fail;
158 if (read(fd, s->l1_backup_table, l1_size) != l1_size)
159 goto fail;
160 for(i = 0; i < s->l1_size; i++) {
161 le32_to_cpus(&s->l1_backup_table[i]);
162 }
163 }
165 s->l2_cache = qemu_malloc(s->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
166 if (!s->l2_cache)
167 goto fail;
168 s->fd = fd;
169 return 0;
170 fail:
171 qemu_free(s->l1_backup_table);
172 qemu_free(s->l1_table);
173 qemu_free(s->l2_cache);
174 close(fd);
175 return -1;
176 }
178 static uint64_t get_cluster_offset(BlockDriverState *bs,
179 uint64_t offset, int allocate)
180 {
181 BDRVVmdkState *s = bs->opaque;
182 unsigned int l1_index, l2_offset, l2_index;
183 int min_index, i, j;
184 uint32_t min_count, *l2_table, tmp;
185 uint64_t cluster_offset;
187 l1_index = (offset >> 9) / s->l1_entry_sectors;
188 if (l1_index >= s->l1_size)
189 return 0;
190 l2_offset = s->l1_table[l1_index];
191 if (!l2_offset)
192 return 0;
193 for(i = 0; i < L2_CACHE_SIZE; i++) {
194 if (l2_offset == s->l2_cache_offsets[i]) {
195 /* increment the hit count */
196 if (++s->l2_cache_counts[i] == 0xffffffff) {
197 for(j = 0; j < L2_CACHE_SIZE; j++) {
198 s->l2_cache_counts[j] >>= 1;
199 }
200 }
201 l2_table = s->l2_cache + (i * s->l2_size);
202 goto found;
203 }
204 }
205 /* not found: load a new entry in the least used one */
206 min_index = 0;
207 min_count = 0xffffffff;
208 for(i = 0; i < L2_CACHE_SIZE; i++) {
209 if (s->l2_cache_counts[i] < min_count) {
210 min_count = s->l2_cache_counts[i];
211 min_index = i;
212 }
213 }
214 l2_table = s->l2_cache + (min_index * s->l2_size);
215 lseek(s->fd, (int64_t)l2_offset * 512, SEEK_SET);
216 if (read(s->fd, l2_table, s->l2_size * sizeof(uint32_t)) !=
217 s->l2_size * sizeof(uint32_t))
218 return 0;
219 s->l2_cache_offsets[min_index] = l2_offset;
220 s->l2_cache_counts[min_index] = 1;
221 found:
222 l2_index = ((offset >> 9) / s->cluster_sectors) % s->l2_size;
223 cluster_offset = le32_to_cpu(l2_table[l2_index]);
224 if (!cluster_offset) {
225 if (!allocate)
226 return 0;
227 cluster_offset = lseek(s->fd, 0, SEEK_END);
228 ftruncate(s->fd, cluster_offset + (s->cluster_sectors << 9));
229 cluster_offset >>= 9;
230 /* update L2 table */
231 tmp = cpu_to_le32(cluster_offset);
232 l2_table[l2_index] = tmp;
233 lseek(s->fd, ((int64_t)l2_offset * 512) + (l2_index * sizeof(tmp)), SEEK_SET);
234 if (write(s->fd, &tmp, sizeof(tmp)) != sizeof(tmp))
235 return 0;
236 /* update backup L2 table */
237 if (s->l1_backup_table_offset != 0) {
238 l2_offset = s->l1_backup_table[l1_index];
239 lseek(s->fd, ((int64_t)l2_offset * 512) + (l2_index * sizeof(tmp)), SEEK_SET);
240 if (write(s->fd, &tmp, sizeof(tmp)) != sizeof(tmp))
241 return 0;
242 }
243 }
244 cluster_offset <<= 9;
245 return cluster_offset;
246 }
248 static int vmdk_is_allocated(BlockDriverState *bs, int64_t sector_num,
249 int nb_sectors, int *pnum)
250 {
251 BDRVVmdkState *s = bs->opaque;
252 int index_in_cluster, n;
253 uint64_t cluster_offset;
255 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0);
256 index_in_cluster = sector_num % s->cluster_sectors;
257 n = s->cluster_sectors - index_in_cluster;
258 if (n > nb_sectors)
259 n = nb_sectors;
260 *pnum = n;
261 return (cluster_offset != 0);
262 }
264 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
265 uint8_t *buf, int nb_sectors)
266 {
267 BDRVVmdkState *s = bs->opaque;
268 int ret, index_in_cluster, n;
269 uint64_t cluster_offset;
271 while (nb_sectors > 0) {
272 cluster_offset = get_cluster_offset(bs, sector_num << 9, 0);
273 index_in_cluster = sector_num % s->cluster_sectors;
274 n = s->cluster_sectors - index_in_cluster;
275 if (n > nb_sectors)
276 n = nb_sectors;
277 if (!cluster_offset) {
278 memset(buf, 0, 512 * n);
279 } else {
280 lseek(s->fd, cluster_offset + index_in_cluster * 512, SEEK_SET);
281 ret = read(s->fd, buf, n * 512);
282 if (ret != n * 512)
283 return -1;
284 }
285 nb_sectors -= n;
286 sector_num += n;
287 buf += n * 512;
288 }
289 return 0;
290 }
292 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
293 const uint8_t *buf, int nb_sectors)
294 {
295 BDRVVmdkState *s = bs->opaque;
296 int ret, index_in_cluster, n;
297 uint64_t cluster_offset;
299 while (nb_sectors > 0) {
300 index_in_cluster = sector_num & (s->cluster_sectors - 1);
301 n = s->cluster_sectors - index_in_cluster;
302 if (n > nb_sectors)
303 n = nb_sectors;
304 cluster_offset = get_cluster_offset(bs, sector_num << 9, 1);
305 if (!cluster_offset)
306 return -1;
307 lseek(s->fd, cluster_offset + index_in_cluster * 512, SEEK_SET);
308 ret = write(s->fd, buf, n * 512);
309 if (ret != n * 512)
310 return -1;
311 nb_sectors -= n;
312 sector_num += n;
313 buf += n * 512;
314 }
315 return 0;
316 }
318 static int vmdk_create(const char *filename, int64_t total_size,
319 const char *backing_file, int flags)
320 {
321 int fd, i;
322 VMDK4Header header;
323 uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
324 char *desc_template =
325 "# Disk DescriptorFile\n"
326 "version=1\n"
327 "CID=%x\n"
328 "parentCID=ffffffff\n"
329 "createType=\"monolithicSparse\"\n"
330 "\n"
331 "# Extent description\n"
332 "RW %lu SPARSE \"%s\"\n"
333 "\n"
334 "# The Disk Data Base \n"
335 "#DDB\n"
336 "\n"
337 "ddb.virtualHWVersion = \"3\"\n"
338 "ddb.geometry.cylinders = \"%lu\"\n"
339 "ddb.geometry.heads = \"16\"\n"
340 "ddb.geometry.sectors = \"63\"\n"
341 "ddb.adapterType = \"ide\"\n";
342 char desc[1024];
343 const char *real_filename, *temp_str;
345 /* XXX: add support for backing file */
347 fd = open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
348 0644);
349 if (fd < 0)
350 return -1;
351 magic = cpu_to_be32(VMDK4_MAGIC);
352 memset(&header, 0, sizeof(header));
353 header.version = cpu_to_le32(1);
354 header.flags = cpu_to_le32(3); /* ?? */
355 header.capacity = cpu_to_le64(total_size);
356 header.granularity = cpu_to_le64(128);
357 header.num_gtes_per_gte = cpu_to_le32(512);
359 grains = (total_size + header.granularity - 1) / header.granularity;
360 gt_size = ((header.num_gtes_per_gte * sizeof(uint32_t)) + 511) >> 9;
361 gt_count = (grains + header.num_gtes_per_gte - 1) / header.num_gtes_per_gte;
362 gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
364 header.desc_offset = 1;
365 header.desc_size = 20;
366 header.rgd_offset = header.desc_offset + header.desc_size;
367 header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
368 header.grain_offset =
369 ((header.gd_offset + gd_size + (gt_size * gt_count) +
370 header.granularity - 1) / header.granularity) *
371 header.granularity;
373 header.desc_offset = cpu_to_le64(header.desc_offset);
374 header.desc_size = cpu_to_le64(header.desc_size);
375 header.rgd_offset = cpu_to_le64(header.rgd_offset);
376 header.gd_offset = cpu_to_le64(header.gd_offset);
377 header.grain_offset = cpu_to_le64(header.grain_offset);
379 header.check_bytes[0] = 0xa;
380 header.check_bytes[1] = 0x20;
381 header.check_bytes[2] = 0xd;
382 header.check_bytes[3] = 0xa;
384 /* write all the data */
385 write(fd, &magic, sizeof(magic));
386 write(fd, &header, sizeof(header));
388 ftruncate(fd, header.grain_offset << 9);
390 /* write grain directory */
391 lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
392 for (i = 0, tmp = header.rgd_offset + gd_size;
393 i < gt_count; i++, tmp += gt_size)
394 write(fd, &tmp, sizeof(tmp));
396 /* write backup grain directory */
397 lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
398 for (i = 0, tmp = header.gd_offset + gd_size;
399 i < gt_count; i++, tmp += gt_size)
400 write(fd, &tmp, sizeof(tmp));
402 /* compose the descriptor */
403 real_filename = filename;
404 if ((temp_str = strrchr(real_filename, '\\')) != NULL)
405 real_filename = temp_str + 1;
406 if ((temp_str = strrchr(real_filename, '/')) != NULL)
407 real_filename = temp_str + 1;
408 if ((temp_str = strrchr(real_filename, ':')) != NULL)
409 real_filename = temp_str + 1;
410 sprintf(desc, desc_template, time(NULL), (unsigned long)total_size,
411 real_filename, total_size / (63 * 16));
413 /* write the descriptor */
414 lseek(fd, le64_to_cpu(header.desc_offset) << 9, SEEK_SET);
415 write(fd, desc, strlen(desc));
417 close(fd);
418 return 0;
419 }
421 static void vmdk_close(BlockDriverState *bs)
422 {
423 BDRVVmdkState *s = bs->opaque;
424 qemu_free(s->l1_table);
425 qemu_free(s->l2_cache);
426 close(s->fd);
427 }
429 BlockDriver bdrv_vmdk = {
430 "vmdk",
431 sizeof(BDRVVmdkState),
432 vmdk_probe,
433 vmdk_open,
434 vmdk_read,
435 vmdk_write,
436 vmdk_close,
437 vmdk_create,
438 vmdk_is_allocated,
439 };