]>
Commit | Line | Data |
---|---|---|
6ada7453 TS |
1 | /* |
2 | * Block driver for Parallels disk image format | |
3 | * | |
4 | * Copyright (c) 2007 Alex Beregszaszi | |
cc5690f2 | 5 | * Copyright (c) 2015 Denis V. Lunev <den@openvz.org> |
6ada7453 | 6 | * |
cc5690f2 DL |
7 | * This code was originally based on comparing different disk images created |
8 | * by Parallels. Currently it is based on opened OpenVZ sources | |
9 | * available at | |
10 | * http://git.openvz.org/?p=ploop;a=summary | |
6ada7453 TS |
11 | * |
12 | * Permission is hereby granted, free of charge, to any person obtaining a copy | |
13 | * of this software and associated documentation files (the "Software"), to deal | |
14 | * in the Software without restriction, including without limitation the rights | |
15 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell | |
16 | * copies of the Software, and to permit persons to whom the Software is | |
17 | * furnished to do so, subject to the following conditions: | |
18 | * | |
19 | * The above copyright notice and this permission notice shall be included in | |
20 | * all copies or substantial portions of the Software. | |
21 | * | |
22 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR | |
23 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, | |
24 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL | |
25 | * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER | |
26 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, | |
27 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | |
28 | * THE SOFTWARE. | |
29 | */ | |
922a01a0 | 30 | |
80c71a24 | 31 | #include "qemu/osdep.h" |
baefd977 | 32 | #include "qemu/error-report.h" |
da34e65c | 33 | #include "qapi/error.h" |
737e150e | 34 | #include "block/block_int.h" |
609f45ea | 35 | #include "block/qdict.h" |
8942764f | 36 | #include "sysemu/block-backend.h" |
1de7afc9 | 37 | #include "qemu/module.h" |
922a01a0 | 38 | #include "qemu/option.h" |
1511b490 KW |
39 | #include "qapi/qmp/qdict.h" |
40 | #include "qapi/qobject-input-visitor.h" | |
41 | #include "qapi/qapi-visit-block-core.h" | |
58369e22 | 42 | #include "qemu/bswap.h" |
0d31c7c2 | 43 | #include "qemu/bitmap.h" |
5df022cf | 44 | #include "qemu/memalign.h" |
1d0f37cf | 45 | #include "migration/blocker.h" |
90fe66f0 | 46 | #include "parallels.h" |
6ada7453 TS |
47 | |
48 | /**************************************************************/ | |
49 | ||
50 | #define HEADER_MAGIC "WithoutFreeSpace" | |
d25d5980 | 51 | #define HEADER_MAGIC2 "WithouFreSpacExt" |
6ada7453 | 52 | #define HEADER_VERSION 2 |
6dd6b9f1 | 53 | #define HEADER_INUSE_MAGIC (0x746F6E59) |
555a608c | 54 | #define MAX_PARALLELS_IMAGE_FACTOR (1ull << 32) |
6ada7453 | 55 | |
f7abe0ec MAL |
56 | static QEnumLookup prealloc_mode_lookup = { |
57 | .array = (const char *const[]) { | |
58 | "falloc", | |
59 | "truncate", | |
f7abe0ec MAL |
60 | }, |
61 | .size = PRL_PREALLOC_MODE__MAX | |
d6179011 DL |
62 | }; |
63 | ||
d6179011 DL |
64 | #define PARALLELS_OPT_PREALLOC_MODE "prealloc-mode" |
65 | #define PARALLELS_OPT_PREALLOC_SIZE "prealloc-size" | |
66 | ||
67 | static QemuOptsList parallels_runtime_opts = { | |
68 | .name = "parallels", | |
69 | .head = QTAILQ_HEAD_INITIALIZER(parallels_runtime_opts.head), | |
70 | .desc = { | |
71 | { | |
72 | .name = PARALLELS_OPT_PREALLOC_SIZE, | |
73 | .type = QEMU_OPT_SIZE, | |
74 | .help = "Preallocation size on image expansion", | |
ff5bbe56 | 75 | .def_value_str = "128M", |
d6179011 DL |
76 | }, |
77 | { | |
78 | .name = PARALLELS_OPT_PREALLOC_MODE, | |
79 | .type = QEMU_OPT_STRING, | |
80 | .help = "Preallocation mode on image expansion " | |
81 | "(allowed values: falloc, truncate)", | |
82 | .def_value_str = "falloc", | |
83 | }, | |
84 | { /* end of list */ }, | |
85 | }, | |
86 | }; | |
87 | ||
1511b490 KW |
88 | static QemuOptsList parallels_create_opts = { |
89 | .name = "parallels-create-opts", | |
90 | .head = QTAILQ_HEAD_INITIALIZER(parallels_create_opts.head), | |
91 | .desc = { | |
92 | { | |
93 | .name = BLOCK_OPT_SIZE, | |
94 | .type = QEMU_OPT_SIZE, | |
95 | .help = "Virtual disk size", | |
96 | }, | |
97 | { | |
98 | .name = BLOCK_OPT_CLUSTER_SIZE, | |
99 | .type = QEMU_OPT_SIZE, | |
100 | .help = "Parallels image cluster size", | |
101 | .def_value_str = stringify(DEFAULT_CLUSTER_SIZE), | |
102 | }, | |
103 | { /* end of list */ } | |
104 | } | |
105 | }; | |
106 | ||
d6179011 | 107 | |
555cc9d9 DL |
108 | static int64_t bat2sect(BDRVParallelsState *s, uint32_t idx) |
109 | { | |
dd97cdc0 | 110 | return (uint64_t)le32_to_cpu(s->bat_bitmap[idx]) * s->off_multiplier; |
555cc9d9 DL |
111 | } |
112 | ||
2d68e22e DL |
113 | static uint32_t bat_entry_off(uint32_t idx) |
114 | { | |
115 | return sizeof(ParallelsHeader) + sizeof(uint32_t) * idx; | |
116 | } | |
117 | ||
29442569 | 118 | static int64_t seek_to_sector(BDRVParallelsState *s, int64_t sector_num) |
6ada7453 | 119 | { |
c34d2451 | 120 | uint32_t index, offset; |
6ada7453 TS |
121 | |
122 | index = sector_num / s->tracks; | |
123 | offset = sector_num % s->tracks; | |
124 | ||
9d8b88f6 | 125 | /* not allocated */ |
369f7de9 | 126 | if ((index >= s->bat_size) || (s->bat_bitmap[index] == 0)) { |
f08e2f84 | 127 | return -1; |
369f7de9 | 128 | } |
555cc9d9 | 129 | return bat2sect(s, index) + offset; |
6ada7453 TS |
130 | } |
131 | ||
9de9da17 RK |
132 | static int cluster_remainder(BDRVParallelsState *s, int64_t sector_num, |
133 | int nb_sectors) | |
134 | { | |
135 | int ret = s->tracks - sector_num % s->tracks; | |
136 | return MIN(nb_sectors, ret); | |
137 | } | |
138 | ||
6bb8bc63 AI |
139 | static uint32_t host_cluster_index(BDRVParallelsState *s, int64_t off) |
140 | { | |
141 | off -= s->data_start << BDRV_SECTOR_BITS; | |
142 | return off / s->cluster_size; | |
143 | } | |
144 | ||
6953d920 DL |
145 | static int64_t block_status(BDRVParallelsState *s, int64_t sector_num, |
146 | int nb_sectors, int *pnum) | |
147 | { | |
148 | int64_t start_off = -2, prev_end_off = -2; | |
149 | ||
150 | *pnum = 0; | |
151 | while (nb_sectors > 0 || start_off == -2) { | |
152 | int64_t offset = seek_to_sector(s, sector_num); | |
153 | int to_end; | |
154 | ||
155 | if (start_off == -2) { | |
156 | start_off = offset; | |
157 | prev_end_off = offset; | |
158 | } else if (offset != prev_end_off) { | |
159 | break; | |
160 | } | |
161 | ||
162 | to_end = cluster_remainder(s, sector_num, nb_sectors); | |
163 | nb_sectors -= to_end; | |
164 | sector_num += to_end; | |
165 | *pnum += to_end; | |
166 | ||
167 | if (offset > 0) { | |
168 | prev_end_off += to_end; | |
169 | } | |
170 | } | |
171 | return start_off; | |
172 | } | |
173 | ||
b64b29b9 AI |
174 | static void parallels_set_bat_entry(BDRVParallelsState *s, |
175 | uint32_t index, uint32_t offset) | |
176 | { | |
177 | s->bat_bitmap[index] = cpu_to_le32(offset); | |
178 | bitmap_set(s->bat_dirty_bmap, bat_entry_off(index) / s->bat_dirty_block, 1); | |
179 | } | |
180 | ||
73194d3f DL |
181 | static int mark_used(BlockDriverState *bs, unsigned long *bitmap, |
182 | uint32_t bitmap_size, int64_t off, uint32_t count) | |
a398275e DL |
183 | { |
184 | BDRVParallelsState *s = bs->opaque; | |
185 | uint32_t cluster_index = host_cluster_index(s, off); | |
73194d3f DL |
186 | unsigned long next_used; |
187 | if (cluster_index + count > bitmap_size) { | |
a398275e DL |
188 | return -E2BIG; |
189 | } | |
73194d3f DL |
190 | next_used = find_next_bit(bitmap, bitmap_size, cluster_index); |
191 | if (next_used < cluster_index + count) { | |
a398275e DL |
192 | return -EBUSY; |
193 | } | |
73194d3f | 194 | bitmap_set(bitmap, cluster_index, count); |
a398275e DL |
195 | return 0; |
196 | } | |
197 | ||
e185100f DL |
198 | /* |
199 | * Collect used bitmap. The image can contain errors, we should fill the | |
200 | * bitmap anyway, as much as we can. This information will be used for | |
201 | * error resolution. | |
202 | */ | |
203 | static int parallels_fill_used_bitmap(BlockDriverState *bs) | |
204 | { | |
205 | BDRVParallelsState *s = bs->opaque; | |
206 | int64_t payload_bytes; | |
207 | uint32_t i; | |
208 | int err = 0; | |
209 | ||
210 | payload_bytes = bdrv_getlength(bs->file->bs); | |
211 | if (payload_bytes < 0) { | |
212 | return payload_bytes; | |
213 | } | |
214 | payload_bytes -= s->data_start * BDRV_SECTOR_SIZE; | |
215 | if (payload_bytes < 0) { | |
216 | return -EINVAL; | |
217 | } | |
218 | ||
219 | s->used_bmap_size = DIV_ROUND_UP(payload_bytes, s->cluster_size); | |
220 | if (s->used_bmap_size == 0) { | |
221 | return 0; | |
222 | } | |
223 | s->used_bmap = bitmap_try_new(s->used_bmap_size); | |
224 | if (s->used_bmap == NULL) { | |
225 | return -ENOMEM; | |
226 | } | |
227 | ||
228 | for (i = 0; i < s->bat_size; i++) { | |
229 | int err2; | |
230 | int64_t host_off = bat2sect(s, i) << BDRV_SECTOR_BITS; | |
231 | if (host_off == 0) { | |
232 | continue; | |
233 | } | |
234 | ||
73194d3f | 235 | err2 = mark_used(bs, s->used_bmap, s->used_bmap_size, host_off, 1); |
e185100f DL |
236 | if (err2 < 0 && err == 0) { |
237 | err = err2; | |
238 | } | |
239 | } | |
240 | return err; | |
241 | } | |
242 | ||
243 | static void parallels_free_used_bitmap(BlockDriverState *bs) | |
244 | { | |
245 | BDRVParallelsState *s = bs->opaque; | |
246 | s->used_bmap_size = 0; | |
247 | g_free(s->used_bmap); | |
248 | } | |
249 | ||
c2b8e315 KW |
250 | static int64_t coroutine_fn GRAPH_RDLOCK |
251 | allocate_clusters(BlockDriverState *bs, int64_t sector_num, | |
252 | int nb_sectors, int *pnum) | |
5a41e1fa | 253 | { |
bda4cdcb | 254 | int ret = 0; |
5a41e1fa | 255 | BDRVParallelsState *s = bs->opaque; |
3ac10d19 | 256 | int64_t i, pos, idx, to_allocate, first_free, host_off; |
5a41e1fa | 257 | |
ddd2ef2c DL |
258 | pos = block_status(s, sector_num, nb_sectors, pnum); |
259 | if (pos > 0) { | |
260 | return pos; | |
261 | } | |
5a41e1fa | 262 | |
ddd2ef2c | 263 | idx = sector_num / s->tracks; |
969401fe | 264 | to_allocate = DIV_ROUND_UP(sector_num + *pnum, s->tracks) - idx; |
86d1bd70 | 265 | |
a338dcbb AI |
266 | /* |
267 | * This function is called only by parallels_co_writev(), which will never | |
86d1bd70 HR |
268 | * pass a sector_num at or beyond the end of the image (because the block |
269 | * layer never passes such a sector_num to that function). Therefore, idx | |
270 | * is always below s->bat_size. | |
271 | * block_status() will limit *pnum so that sector_num + *pnum will not | |
272 | * exceed the image end. Therefore, idx + to_allocate cannot exceed | |
273 | * s->bat_size. | |
274 | * Note that s->bat_size is an unsigned int, therefore idx + to_allocate | |
a338dcbb AI |
275 | * will always fit into a uint32_t. |
276 | */ | |
86d1bd70 HR |
277 | assert(idx < s->bat_size && idx + to_allocate <= s->bat_size); |
278 | ||
3ac10d19 DL |
279 | first_free = find_first_zero_bit(s->used_bmap, s->used_bmap_size); |
280 | if (first_free == s->used_bmap_size) { | |
eeb1e6dc | 281 | uint32_t new_usedsize; |
66d201dd DL |
282 | int64_t bytes = to_allocate * s->cluster_size; |
283 | bytes += s->prealloc_size * BDRV_SECTOR_SIZE; | |
3ac10d19 DL |
284 | |
285 | host_off = s->data_end * BDRV_SECTOR_SIZE; | |
eeb1e6dc | 286 | |
bda4cdcb EB |
287 | /* |
288 | * We require the expanded size to read back as zero. If the | |
289 | * user permitted truncation, we try that; but if it fails, we | |
290 | * force the safer-but-slower fallocate. | |
291 | */ | |
292 | if (s->prealloc_mode == PRL_PREALLOC_MODE_TRUNCATE) { | |
66d201dd | 293 | ret = bdrv_co_truncate(bs->file, host_off + bytes, |
50688be0 AF |
294 | false, PREALLOC_MODE_OFF, |
295 | BDRV_REQ_ZERO_WRITE, NULL); | |
bda4cdcb EB |
296 | if (ret == -ENOTSUP) { |
297 | s->prealloc_mode = PRL_PREALLOC_MODE_FALLOCATE; | |
298 | } | |
299 | } | |
19f5dc15 | 300 | if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE) { |
66d201dd | 301 | ret = bdrv_co_pwrite_zeroes(bs->file, host_off, bytes, 0); |
19f5dc15 DL |
302 | } |
303 | if (ret < 0) { | |
304 | return ret; | |
305 | } | |
eeb1e6dc | 306 | |
66d201dd | 307 | new_usedsize = s->used_bmap_size + bytes / s->cluster_size; |
eeb1e6dc DL |
308 | s->used_bmap = bitmap_zero_extend(s->used_bmap, s->used_bmap_size, |
309 | new_usedsize); | |
310 | s->used_bmap_size = new_usedsize; | |
3ac10d19 DL |
311 | } else { |
312 | int64_t next_used; | |
313 | next_used = find_next_bit(s->used_bmap, s->used_bmap_size, first_free); | |
314 | ||
315 | /* Not enough continuous clusters in the middle, adjust the size */ | |
316 | if (next_used - first_free < to_allocate) { | |
317 | to_allocate = next_used - first_free; | |
318 | *pnum = (idx + to_allocate) * s->tracks - sector_num; | |
319 | } | |
320 | ||
321 | host_off = s->data_start * BDRV_SECTOR_SIZE; | |
322 | host_off += first_free * s->cluster_size; | |
323 | ||
324 | /* | |
325 | * No need to preallocate if we are using tail area from the above | |
326 | * branch. In the other case we are likely re-using hole. Preallocate | |
327 | * the space if required by the prealloc_mode. | |
328 | */ | |
329 | if (s->prealloc_mode == PRL_PREALLOC_MODE_FALLOCATE && | |
330 | host_off < s->data_end * BDRV_SECTOR_SIZE) { | |
331 | ret = bdrv_co_pwrite_zeroes(bs->file, host_off, | |
332 | s->cluster_size * to_allocate, 0); | |
333 | if (ret < 0) { | |
334 | return ret; | |
335 | } | |
336 | } | |
5a41e1fa DL |
337 | } |
338 | ||
a338dcbb AI |
339 | /* |
340 | * Try to read from backing to fill empty clusters | |
bcbb3866 EK |
341 | * FIXME: 1. previous write_zeroes may be redundant |
342 | * 2. most of data we read from backing will be rewritten by | |
343 | * parallels_co_writev. On aligned-to-cluster write we do not need | |
344 | * this read at all. | |
345 | * 3. it would be good to combine write of data from backing and new | |
a338dcbb AI |
346 | * data into one write call. |
347 | */ | |
bcbb3866 EK |
348 | if (bs->backing) { |
349 | int64_t nb_cow_sectors = to_allocate * s->tracks; | |
350 | int64_t nb_cow_bytes = nb_cow_sectors << BDRV_SECTOR_BITS; | |
a4072543 | 351 | void *buf = qemu_blockalign(bs, nb_cow_bytes); |
bcbb3866 | 352 | |
a4072543 VSO |
353 | ret = bdrv_co_pread(bs->backing, idx * s->tracks * BDRV_SECTOR_SIZE, |
354 | nb_cow_bytes, buf, 0); | |
bcbb3866 | 355 | if (ret < 0) { |
a4072543 | 356 | qemu_vfree(buf); |
bcbb3866 EK |
357 | return ret; |
358 | } | |
359 | ||
eba088f9 HR |
360 | ret = bdrv_co_pwrite(bs->file, s->data_end * BDRV_SECTOR_SIZE, |
361 | nb_cow_bytes, buf, 0); | |
a4072543 | 362 | qemu_vfree(buf); |
bcbb3866 EK |
363 | if (ret < 0) { |
364 | return ret; | |
365 | } | |
366 | } | |
367 | ||
3ac10d19 | 368 | ret = mark_used(bs, s->used_bmap, s->used_bmap_size, host_off, to_allocate); |
eeb1e6dc DL |
369 | if (ret < 0) { |
370 | /* Image consistency is broken. Alarm! */ | |
371 | return ret; | |
372 | } | |
ddd2ef2c | 373 | for (i = 0; i < to_allocate; i++) { |
3ac10d19 DL |
374 | parallels_set_bat_entry(s, idx + i, |
375 | host_off / BDRV_SECTOR_SIZE / s->off_multiplier); | |
376 | host_off += s->cluster_size; | |
377 | } | |
378 | if (host_off > s->data_end * BDRV_SECTOR_SIZE) { | |
379 | s->data_end = host_off / BDRV_SECTOR_SIZE; | |
ddd2ef2c | 380 | } |
0d31c7c2 | 381 | |
ddd2ef2c | 382 | return bat2sect(s, idx) + sector_num % s->tracks; |
5a41e1fa DL |
383 | } |
384 | ||
0d31c7c2 | 385 | |
b9b10c35 KW |
386 | static int coroutine_fn GRAPH_RDLOCK |
387 | parallels_co_flush_to_os(BlockDriverState *bs) | |
0d31c7c2 DL |
388 | { |
389 | BDRVParallelsState *s = bs->opaque; | |
390 | unsigned long size = DIV_ROUND_UP(s->header_size, s->bat_dirty_block); | |
391 | unsigned long bit; | |
392 | ||
393 | qemu_co_mutex_lock(&s->lock); | |
394 | ||
395 | bit = find_first_bit(s->bat_dirty_bmap, size); | |
396 | while (bit < size) { | |
397 | uint32_t off = bit * s->bat_dirty_block; | |
398 | uint32_t to_write = s->bat_dirty_block; | |
399 | int ret; | |
400 | ||
401 | if (off + to_write > s->header_size) { | |
402 | to_write = s->header_size - off; | |
403 | } | |
50688be0 AF |
404 | ret = bdrv_co_pwrite(bs->file, off, to_write, |
405 | (uint8_t *)s->header + off, 0); | |
0d31c7c2 DL |
406 | if (ret < 0) { |
407 | qemu_co_mutex_unlock(&s->lock); | |
408 | return ret; | |
409 | } | |
410 | bit = find_next_bit(s->bat_dirty_bmap, size, bit + 1); | |
411 | } | |
412 | bitmap_zero(s->bat_dirty_bmap, size); | |
413 | ||
414 | qemu_co_mutex_unlock(&s->lock); | |
415 | return 0; | |
416 | } | |
417 | ||
418 | ||
8e0cf59d EB |
419 | static int coroutine_fn parallels_co_block_status(BlockDriverState *bs, |
420 | bool want_zero, | |
421 | int64_t offset, | |
422 | int64_t bytes, | |
423 | int64_t *pnum, | |
424 | int64_t *map, | |
425 | BlockDriverState **file) | |
dd3bed16 RK |
426 | { |
427 | BDRVParallelsState *s = bs->opaque; | |
8e0cf59d | 428 | int count; |
dd3bed16 | 429 | |
8e0cf59d | 430 | assert(QEMU_IS_ALIGNED(offset | bytes, BDRV_SECTOR_SIZE)); |
dd3bed16 | 431 | qemu_co_mutex_lock(&s->lock); |
8e0cf59d EB |
432 | offset = block_status(s, offset >> BDRV_SECTOR_BITS, |
433 | bytes >> BDRV_SECTOR_BITS, &count); | |
dd3bed16 RK |
434 | qemu_co_mutex_unlock(&s->lock); |
435 | ||
8e0cf59d | 436 | *pnum = count * BDRV_SECTOR_SIZE; |
dd3bed16 RK |
437 | if (offset < 0) { |
438 | return 0; | |
439 | } | |
440 | ||
8e0cf59d | 441 | *map = offset * BDRV_SECTOR_SIZE; |
ddf4987d | 442 | *file = bs->file->bs; |
8e0cf59d | 443 | return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; |
dd3bed16 RK |
444 | } |
445 | ||
7b1fb72e KW |
446 | static int coroutine_fn GRAPH_RDLOCK |
447 | parallels_co_writev(BlockDriverState *bs, int64_t sector_num, int nb_sectors, | |
448 | QEMUIOVector *qiov, int flags) | |
5a41e1fa DL |
449 | { |
450 | BDRVParallelsState *s = bs->opaque; | |
451 | uint64_t bytes_done = 0; | |
452 | QEMUIOVector hd_qiov; | |
453 | int ret = 0; | |
454 | ||
455 | qemu_iovec_init(&hd_qiov, qiov->niov); | |
456 | ||
457 | while (nb_sectors > 0) { | |
458 | int64_t position; | |
459 | int n, nbytes; | |
460 | ||
461 | qemu_co_mutex_lock(&s->lock); | |
ddd2ef2c | 462 | position = allocate_clusters(bs, sector_num, nb_sectors, &n); |
5a41e1fa DL |
463 | qemu_co_mutex_unlock(&s->lock); |
464 | if (position < 0) { | |
465 | ret = (int)position; | |
466 | break; | |
467 | } | |
468 | ||
5a41e1fa DL |
469 | nbytes = n << BDRV_SECTOR_BITS; |
470 | ||
471 | qemu_iovec_reset(&hd_qiov); | |
472 | qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes); | |
473 | ||
d08c2a24 EB |
474 | ret = bdrv_co_pwritev(bs->file, position * BDRV_SECTOR_SIZE, nbytes, |
475 | &hd_qiov, 0); | |
5a41e1fa DL |
476 | if (ret < 0) { |
477 | break; | |
478 | } | |
479 | ||
480 | nb_sectors -= n; | |
481 | sector_num += n; | |
482 | bytes_done += nbytes; | |
483 | } | |
484 | ||
485 | qemu_iovec_destroy(&hd_qiov); | |
486 | return ret; | |
487 | } | |
488 | ||
b9b10c35 KW |
489 | static int coroutine_fn GRAPH_RDLOCK |
490 | parallels_co_readv(BlockDriverState *bs, int64_t sector_num, int nb_sectors, | |
491 | QEMUIOVector *qiov) | |
6ada7453 | 492 | { |
29442569 | 493 | BDRVParallelsState *s = bs->opaque; |
481fb9cf DL |
494 | uint64_t bytes_done = 0; |
495 | QEMUIOVector hd_qiov; | |
496 | int ret = 0; | |
497 | ||
498 | qemu_iovec_init(&hd_qiov, qiov->niov); | |
29442569 | 499 | |
6ada7453 | 500 | while (nb_sectors > 0) { |
481fb9cf DL |
501 | int64_t position; |
502 | int n, nbytes; | |
503 | ||
504 | qemu_co_mutex_lock(&s->lock); | |
6953d920 | 505 | position = block_status(s, sector_num, nb_sectors, &n); |
481fb9cf DL |
506 | qemu_co_mutex_unlock(&s->lock); |
507 | ||
481fb9cf DL |
508 | nbytes = n << BDRV_SECTOR_BITS; |
509 | ||
bcbb3866 EK |
510 | qemu_iovec_reset(&hd_qiov); |
511 | qemu_iovec_concat(&hd_qiov, qiov, bytes_done, nbytes); | |
512 | ||
481fb9cf | 513 | if (position < 0) { |
bcbb3866 | 514 | if (bs->backing) { |
d08c2a24 EB |
515 | ret = bdrv_co_preadv(bs->backing, sector_num * BDRV_SECTOR_SIZE, |
516 | nbytes, &hd_qiov, 0); | |
bcbb3866 EK |
517 | if (ret < 0) { |
518 | break; | |
519 | } | |
520 | } else { | |
521 | qemu_iovec_memset(&hd_qiov, 0, 0, nbytes); | |
522 | } | |
481fb9cf | 523 | } else { |
d08c2a24 EB |
524 | ret = bdrv_co_preadv(bs->file, position * BDRV_SECTOR_SIZE, nbytes, |
525 | &hd_qiov, 0); | |
29442569 | 526 | if (ret < 0) { |
481fb9cf | 527 | break; |
29442569 | 528 | } |
9d8b88f6 | 529 | } |
481fb9cf | 530 | |
9de9da17 RK |
531 | nb_sectors -= n; |
532 | sector_num += n; | |
481fb9cf | 533 | bytes_done += nbytes; |
6ada7453 | 534 | } |
6ada7453 | 535 | |
481fb9cf | 536 | qemu_iovec_destroy(&hd_qiov); |
2914caa0 PB |
537 | return ret; |
538 | } | |
539 | ||
c54fb435 DL |
540 | |
541 | static int coroutine_fn GRAPH_RDLOCK | |
542 | parallels_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes) | |
543 | { | |
544 | int ret = 0; | |
545 | uint32_t cluster, count; | |
546 | BDRVParallelsState *s = bs->opaque; | |
547 | ||
548 | /* | |
549 | * The image does not support ZERO mark inside the BAT, which means that | |
550 | * stale data could be exposed from the backing file. | |
551 | */ | |
552 | if (bs->backing) { | |
553 | return -ENOTSUP; | |
554 | } | |
555 | ||
556 | if (!QEMU_IS_ALIGNED(offset, s->cluster_size)) { | |
557 | return -ENOTSUP; | |
558 | } else if (!QEMU_IS_ALIGNED(bytes, s->cluster_size)) { | |
559 | return -ENOTSUP; | |
560 | } | |
561 | ||
562 | cluster = offset / s->cluster_size; | |
563 | count = bytes / s->cluster_size; | |
564 | ||
565 | qemu_co_mutex_lock(&s->lock); | |
566 | for (; count > 0; cluster++, count--) { | |
567 | int64_t host_off = bat2sect(s, cluster) << BDRV_SECTOR_BITS; | |
568 | if (host_off == 0) { | |
569 | continue; | |
570 | } | |
571 | ||
572 | ret = bdrv_co_pdiscard(bs->file, host_off, s->cluster_size); | |
573 | if (ret < 0) { | |
574 | goto done; | |
575 | } | |
576 | ||
577 | parallels_set_bat_entry(s, cluster, 0); | |
578 | bitmap_clear(s->used_bmap, host_cluster_index(s, host_off), 1); | |
579 | } | |
580 | done: | |
581 | qemu_co_mutex_unlock(&s->lock); | |
582 | return ret; | |
583 | } | |
584 | ||
a98b260a DL |
585 | static int coroutine_fn GRAPH_RDLOCK |
586 | parallels_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int64_t bytes, | |
587 | BdrvRequestFlags flags) | |
588 | { | |
589 | /* | |
590 | * The zero flag is missed in the Parallels format specification. We can | |
591 | * resort to discard if we have no backing file (this condition is checked | |
592 | * inside parallels_co_pdiscard(). | |
593 | */ | |
594 | return parallels_co_pdiscard(bs, offset, bytes); | |
595 | } | |
596 | ||
597 | ||
96de69c7 AI |
598 | static void parallels_check_unclean(BlockDriverState *bs, |
599 | BdrvCheckResult *res, | |
600 | BdrvCheckMode fix) | |
601 | { | |
602 | BDRVParallelsState *s = bs->opaque; | |
603 | ||
604 | if (!s->header_unclean) { | |
605 | return; | |
606 | } | |
607 | ||
608 | fprintf(stderr, "%s image was not closed correctly\n", | |
609 | fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR"); | |
610 | res->corruptions++; | |
611 | if (fix & BDRV_FIX_ERRORS) { | |
612 | /* parallels_close will do the job right */ | |
613 | res->corruptions_fixed++; | |
614 | s->header_unclean = false; | |
615 | } | |
616 | } | |
49ad6467 | 617 | |
8cd19203 AI |
618 | /* |
619 | * Returns true if data_off is correct, otherwise false. In both cases | |
620 | * correct_offset is set to the proper value. | |
621 | */ | |
622 | static bool parallels_test_data_off(BDRVParallelsState *s, | |
623 | int64_t file_nb_sectors, | |
624 | uint32_t *correct_offset) | |
625 | { | |
626 | uint32_t data_off, min_off; | |
627 | bool old_magic; | |
628 | ||
629 | /* | |
630 | * There are two slightly different image formats: with "WithoutFreeSpace" | |
631 | * or "WithouFreSpacExt" magic words. Call the first one as "old magic". | |
632 | * In such images data_off field can be zero. In this case the offset is | |
633 | * calculated as the end of BAT table plus some padding to ensure sector | |
634 | * size alignment. | |
635 | */ | |
636 | old_magic = !memcmp(s->header->magic, HEADER_MAGIC, 16); | |
637 | ||
638 | min_off = DIV_ROUND_UP(bat_entry_off(s->bat_size), BDRV_SECTOR_SIZE); | |
639 | if (!old_magic) { | |
640 | min_off = ROUND_UP(min_off, s->cluster_size / BDRV_SECTOR_SIZE); | |
641 | } | |
642 | ||
643 | if (correct_offset) { | |
644 | *correct_offset = min_off; | |
645 | } | |
646 | ||
647 | data_off = le32_to_cpu(s->header->data_off); | |
648 | if (data_off == 0 && old_magic) { | |
649 | return true; | |
650 | } | |
651 | ||
652 | if (data_off < min_off || data_off > file_nb_sectors) { | |
653 | return false; | |
654 | } | |
655 | ||
656 | if (correct_offset) { | |
657 | *correct_offset = data_off; | |
658 | } | |
659 | ||
660 | return true; | |
661 | } | |
662 | ||
663 | static int coroutine_fn GRAPH_RDLOCK | |
664 | parallels_check_data_off(BlockDriverState *bs, BdrvCheckResult *res, | |
665 | BdrvCheckMode fix) | |
666 | { | |
667 | BDRVParallelsState *s = bs->opaque; | |
668 | int64_t file_size; | |
669 | uint32_t data_off; | |
670 | ||
671 | file_size = bdrv_co_nb_sectors(bs->file->bs); | |
672 | if (file_size < 0) { | |
673 | res->check_errors++; | |
674 | return file_size; | |
675 | } | |
676 | ||
677 | if (parallels_test_data_off(s, file_size, &data_off)) { | |
678 | return 0; | |
679 | } | |
680 | ||
681 | res->corruptions++; | |
682 | if (fix & BDRV_FIX_ERRORS) { | |
e185100f | 683 | int err; |
8cd19203 | 684 | s->header->data_off = cpu_to_le32(data_off); |
6f2206b0 | 685 | s->data_start = data_off; |
e185100f DL |
686 | |
687 | parallels_free_used_bitmap(bs); | |
688 | err = parallels_fill_used_bitmap(bs); | |
689 | if (err == -ENOMEM) { | |
690 | res->check_errors++; | |
691 | return err; | |
692 | } | |
693 | ||
8cd19203 AI |
694 | res->corruptions_fixed++; |
695 | } | |
696 | ||
697 | fprintf(stderr, "%s data_off field has incorrect value\n", | |
698 | fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR"); | |
699 | ||
700 | return 0; | |
701 | } | |
702 | ||
6d416e56 AI |
703 | static int coroutine_fn GRAPH_RDLOCK |
704 | parallels_check_outside_image(BlockDriverState *bs, BdrvCheckResult *res, | |
705 | BdrvCheckMode fix) | |
706 | { | |
707 | BDRVParallelsState *s = bs->opaque; | |
708 | uint32_t i; | |
709 | int64_t off, high_off, size; | |
710 | ||
0af02bd1 | 711 | size = bdrv_co_getlength(bs->file->bs); |
6d416e56 AI |
712 | if (size < 0) { |
713 | res->check_errors++; | |
714 | return size; | |
715 | } | |
716 | ||
717 | high_off = 0; | |
718 | for (i = 0; i < s->bat_size; i++) { | |
719 | off = bat2sect(s, i) << BDRV_SECTOR_BITS; | |
029136f2 | 720 | if (off + s->cluster_size > size) { |
6d416e56 AI |
721 | fprintf(stderr, "%s cluster %u is outside image\n", |
722 | fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i); | |
723 | res->corruptions++; | |
724 | if (fix & BDRV_FIX_ERRORS) { | |
725 | parallels_set_bat_entry(s, i, 0); | |
726 | res->corruptions_fixed++; | |
727 | } | |
728 | continue; | |
729 | } | |
730 | if (high_off < off) { | |
731 | high_off = off; | |
732 | } | |
733 | } | |
734 | ||
735 | if (high_off == 0) { | |
736 | res->image_end_offset = s->data_end << BDRV_SECTOR_BITS; | |
737 | } else { | |
738 | res->image_end_offset = high_off + s->cluster_size; | |
739 | s->data_end = res->image_end_offset >> BDRV_SECTOR_BITS; | |
740 | } | |
741 | ||
742 | return 0; | |
743 | } | |
744 | ||
c2b8e315 | 745 | static int coroutine_fn GRAPH_RDLOCK |
09a21edf | 746 | parallels_check_leak(BlockDriverState *bs, BdrvCheckResult *res, |
728e1017 | 747 | BdrvCheckMode fix, bool explicit) |
49ad6467 DL |
748 | { |
749 | BDRVParallelsState *s = bs->opaque; | |
09a21edf | 750 | int64_t size; |
6d416e56 | 751 | int ret; |
49ad6467 | 752 | |
7c5f86d5 | 753 | size = bdrv_co_getlength(bs->file->bs); |
49ad6467 DL |
754 | if (size < 0) { |
755 | res->check_errors++; | |
756 | return size; | |
757 | } | |
758 | ||
09a21edf AI |
759 | if (size > res->image_end_offset) { |
760 | int64_t count; | |
761 | count = DIV_ROUND_UP(size - res->image_end_offset, s->cluster_size); | |
728e1017 AI |
762 | if (explicit) { |
763 | fprintf(stderr, | |
764 | "%s space leaked at the end of the image %" PRId64 "\n", | |
765 | fix & BDRV_FIX_LEAKS ? "Repairing" : "ERROR", | |
766 | size - res->image_end_offset); | |
767 | res->leaks += count; | |
768 | } | |
09a21edf AI |
769 | if (fix & BDRV_FIX_LEAKS) { |
770 | Error *local_err = NULL; | |
771 | ||
772 | /* | |
773 | * In order to really repair the image, we must shrink it. | |
774 | * That means we have to pass exact=true. | |
775 | */ | |
776 | ret = bdrv_co_truncate(bs->file, res->image_end_offset, true, | |
777 | PREALLOC_MODE_OFF, 0, &local_err); | |
778 | if (ret < 0) { | |
779 | error_report_err(local_err); | |
780 | res->check_errors++; | |
781 | return ret; | |
782 | } | |
728e1017 AI |
783 | if (explicit) { |
784 | res->leaks_fixed += count; | |
785 | } | |
09a21edf AI |
786 | } |
787 | } | |
788 | ||
789 | return 0; | |
790 | } | |
791 | ||
6bb8bc63 AI |
792 | static int coroutine_fn GRAPH_RDLOCK |
793 | parallels_check_duplicate(BlockDriverState *bs, BdrvCheckResult *res, | |
794 | BdrvCheckMode fix) | |
795 | { | |
796 | BDRVParallelsState *s = bs->opaque; | |
797 | int64_t host_off, host_sector, guest_sector; | |
798 | unsigned long *bitmap; | |
a398275e | 799 | uint32_t i, bitmap_size, bat_entry; |
6bb8bc63 AI |
800 | int n, ret = 0; |
801 | uint64_t *buf = NULL; | |
802 | bool fixed = false; | |
803 | ||
804 | /* | |
805 | * Create a bitmap of used clusters. | |
806 | * If a bit is set, there is a BAT entry pointing to this cluster. | |
807 | * Loop through the BAT entries, check bits relevant to an entry offset. | |
808 | * If bit is set, this entry is duplicated. Otherwise set the bit. | |
809 | * | |
810 | * We shouldn't worry about newly allocated clusters outside the image | |
811 | * because they are created higher then any existing cluster pointed by | |
812 | * a BAT entry. | |
813 | */ | |
814 | bitmap_size = host_cluster_index(s, res->image_end_offset); | |
815 | if (bitmap_size == 0) { | |
816 | return 0; | |
817 | } | |
818 | if (res->image_end_offset % s->cluster_size) { | |
819 | /* A not aligned image end leads to a bitmap shorter by 1 */ | |
820 | bitmap_size++; | |
821 | } | |
822 | ||
823 | bitmap = bitmap_new(bitmap_size); | |
824 | ||
825 | buf = qemu_blockalign(bs, s->cluster_size); | |
826 | ||
827 | for (i = 0; i < s->bat_size; i++) { | |
828 | host_off = bat2sect(s, i) << BDRV_SECTOR_BITS; | |
829 | if (host_off == 0) { | |
830 | continue; | |
831 | } | |
832 | ||
73194d3f | 833 | ret = mark_used(bs, bitmap, bitmap_size, host_off, 1); |
a398275e DL |
834 | assert(ret != -E2BIG); |
835 | if (ret == 0) { | |
6bb8bc63 AI |
836 | continue; |
837 | } | |
838 | ||
839 | /* this cluster duplicates another one */ | |
840 | fprintf(stderr, "%s duplicate offset in BAT entry %u\n", | |
841 | fix & BDRV_FIX_ERRORS ? "Repairing" : "ERROR", i); | |
842 | ||
843 | res->corruptions++; | |
844 | ||
845 | if (!(fix & BDRV_FIX_ERRORS)) { | |
846 | continue; | |
847 | } | |
848 | ||
849 | /* | |
850 | * Reset the entry and allocate a new cluster | |
851 | * for the relevant guest offset. In this way we let | |
852 | * the lower layer to place the new cluster properly. | |
853 | * Copy the original cluster to the allocated one. | |
854 | * But before save the old offset value for repairing | |
855 | * if we have an error. | |
856 | */ | |
857 | bat_entry = s->bat_bitmap[i]; | |
858 | parallels_set_bat_entry(s, i, 0); | |
859 | ||
860 | ret = bdrv_co_pread(bs->file, host_off, s->cluster_size, buf, 0); | |
861 | if (ret < 0) { | |
862 | res->check_errors++; | |
863 | goto out_repair_bat; | |
864 | } | |
865 | ||
866 | guest_sector = (i * (int64_t)s->cluster_size) >> BDRV_SECTOR_BITS; | |
867 | host_sector = allocate_clusters(bs, guest_sector, s->tracks, &n); | |
868 | if (host_sector < 0) { | |
869 | res->check_errors++; | |
870 | goto out_repair_bat; | |
871 | } | |
872 | host_off = host_sector << BDRV_SECTOR_BITS; | |
873 | ||
874 | ret = bdrv_co_pwrite(bs->file, host_off, s->cluster_size, buf, 0); | |
875 | if (ret < 0) { | |
876 | res->check_errors++; | |
877 | goto out_repair_bat; | |
878 | } | |
879 | ||
880 | if (host_off + s->cluster_size > res->image_end_offset) { | |
881 | res->image_end_offset = host_off + s->cluster_size; | |
882 | } | |
883 | ||
884 | /* | |
885 | * In the future allocate_cluster() will reuse holed offsets | |
886 | * inside the image. Keep the used clusters bitmap content | |
887 | * consistent for the new allocated clusters too. | |
888 | * | |
889 | * Note, clusters allocated outside the current image are not | |
a398275e DL |
890 | * considered, and the bitmap size doesn't change. This specifically |
891 | * means that -E2BIG is OK. | |
6bb8bc63 | 892 | */ |
73194d3f | 893 | ret = mark_used(bs, bitmap, bitmap_size, host_off, 1); |
a398275e DL |
894 | if (ret == -EBUSY) { |
895 | res->check_errors++; | |
896 | goto out_repair_bat; | |
6bb8bc63 AI |
897 | } |
898 | ||
899 | fixed = true; | |
900 | res->corruptions_fixed++; | |
901 | ||
902 | } | |
903 | ||
904 | if (fixed) { | |
905 | /* | |
906 | * When new clusters are allocated, the file size increases by | |
907 | * 128 Mb. We need to truncate the file to the right size. Let | |
908 | * the leak fix code make its job without res changing. | |
909 | */ | |
910 | ret = parallels_check_leak(bs, res, fix, false); | |
911 | } | |
912 | ||
913 | out_free: | |
914 | g_free(buf); | |
915 | g_free(bitmap); | |
916 | return ret; | |
917 | /* | |
918 | * We can get here only from places where index and old_offset have | |
919 | * meaningful values. | |
920 | */ | |
921 | out_repair_bat: | |
922 | s->bat_bitmap[i] = bat_entry; | |
923 | goto out_free; | |
924 | } | |
925 | ||
7e259e25 AI |
926 | static void parallels_collect_statistics(BlockDriverState *bs, |
927 | BdrvCheckResult *res, | |
928 | BdrvCheckMode fix) | |
09a21edf AI |
929 | { |
930 | BDRVParallelsState *s = bs->opaque; | |
7e259e25 | 931 | int64_t off, prev_off; |
09a21edf AI |
932 | uint32_t i; |
933 | ||
49ad6467 DL |
934 | res->bfi.total_clusters = s->bat_size; |
935 | res->bfi.compressed_clusters = 0; /* compression is not supported */ | |
936 | ||
49ad6467 DL |
937 | prev_off = 0; |
938 | for (i = 0; i < s->bat_size; i++) { | |
7e259e25 | 939 | off = bat2sect(s, i) << BDRV_SECTOR_BITS; |
9616f7a6 AI |
940 | /* |
941 | * If BDRV_FIX_ERRORS is not set, out-of-image BAT entries were not | |
942 | * fixed. Skip not allocated and out-of-image BAT entries. | |
943 | */ | |
944 | if (off == 0 || off + s->cluster_size > res->image_end_offset) { | |
49ad6467 DL |
945 | prev_off = 0; |
946 | continue; | |
947 | } | |
948 | ||
e0b5207f | 949 | if (prev_off != 0 && (prev_off + s->cluster_size) != off) { |
49ad6467 DL |
950 | res->bfi.fragmented_clusters++; |
951 | } | |
952 | prev_off = off; | |
7e259e25 | 953 | res->bfi.allocated_clusters++; |
49ad6467 | 954 | } |
7e259e25 AI |
955 | } |
956 | ||
957 | static int coroutine_fn GRAPH_RDLOCK | |
958 | parallels_co_check(BlockDriverState *bs, BdrvCheckResult *res, | |
959 | BdrvCheckMode fix) | |
960 | { | |
961 | BDRVParallelsState *s = bs->opaque; | |
962 | int ret; | |
963 | ||
c0fc051d AI |
964 | WITH_QEMU_LOCK_GUARD(&s->lock) { |
965 | parallels_check_unclean(bs, res, fix); | |
7e259e25 | 966 | |
8cd19203 AI |
967 | ret = parallels_check_data_off(bs, res, fix); |
968 | if (ret < 0) { | |
969 | return ret; | |
970 | } | |
971 | ||
c0fc051d AI |
972 | ret = parallels_check_outside_image(bs, res, fix); |
973 | if (ret < 0) { | |
974 | return ret; | |
975 | } | |
7e259e25 | 976 | |
728e1017 | 977 | ret = parallels_check_leak(bs, res, fix, true); |
c0fc051d AI |
978 | if (ret < 0) { |
979 | return ret; | |
980 | } | |
7e259e25 | 981 | |
6bb8bc63 AI |
982 | ret = parallels_check_duplicate(bs, res, fix); |
983 | if (ret < 0) { | |
984 | return ret; | |
985 | } | |
986 | ||
c0fc051d | 987 | parallels_collect_statistics(bs, res, fix); |
7e259e25 AI |
988 | } |
989 | ||
c0fc051d AI |
990 | ret = bdrv_co_flush(bs); |
991 | if (ret < 0) { | |
992 | res->check_errors++; | |
3569cb7b AI |
993 | } |
994 | ||
2fd61638 | 995 | return ret; |
49ad6467 DL |
996 | } |
997 | ||
998 | ||
4db7ba3b KW |
999 | static int coroutine_fn GRAPH_UNLOCKED |
1000 | parallels_co_create(BlockdevCreateOptions* opts, Error **errp) | |
74cf6c50 | 1001 | { |
1511b490 KW |
1002 | BlockdevCreateOptionsParallels *parallels_opts; |
1003 | BlockDriverState *bs; | |
1004 | BlockBackend *blk; | |
74cf6c50 | 1005 | int64_t total_size, cl_size; |
369f7de9 | 1006 | uint32_t bat_entries, bat_sectors; |
74cf6c50 | 1007 | ParallelsHeader header; |
1511b490 | 1008 | uint8_t tmp[BDRV_SECTOR_SIZE]; |
74cf6c50 DL |
1009 | int ret; |
1010 | ||
1511b490 KW |
1011 | assert(opts->driver == BLOCKDEV_DRIVER_PARALLELS); |
1012 | parallels_opts = &opts->u.parallels; | |
1013 | ||
1014 | /* Sanity checks */ | |
1015 | total_size = parallels_opts->size; | |
1016 | ||
1017 | if (parallels_opts->has_cluster_size) { | |
1018 | cl_size = parallels_opts->cluster_size; | |
1019 | } else { | |
1020 | cl_size = DEFAULT_CLUSTER_SIZE; | |
1021 | } | |
1022 | ||
2332d825 KW |
1023 | /* XXX What is the real limit here? This is an insanely large maximum. */ |
1024 | if (cl_size >= INT64_MAX / MAX_PARALLELS_IMAGE_FACTOR) { | |
1025 | error_setg(errp, "Cluster size is too large"); | |
1026 | return -EINVAL; | |
1027 | } | |
555a608c | 1028 | if (total_size >= MAX_PARALLELS_IMAGE_FACTOR * cl_size) { |
1511b490 | 1029 | error_setg(errp, "Image size is too large for this cluster size"); |
555a608c KK |
1030 | return -E2BIG; |
1031 | } | |
74cf6c50 | 1032 | |
1511b490 KW |
1033 | if (!QEMU_IS_ALIGNED(total_size, BDRV_SECTOR_SIZE)) { |
1034 | error_setg(errp, "Image size must be a multiple of 512 bytes"); | |
1035 | return -EINVAL; | |
74cf6c50 DL |
1036 | } |
1037 | ||
1511b490 KW |
1038 | if (!QEMU_IS_ALIGNED(cl_size, BDRV_SECTOR_SIZE)) { |
1039 | error_setg(errp, "Cluster size must be a multiple of 512 bytes"); | |
1040 | return -EINVAL; | |
1041 | } | |
1042 | ||
1043 | /* Create BlockBackend to write to the image */ | |
48a4e92d | 1044 | bs = bdrv_co_open_blockdev_ref(parallels_opts->file, errp); |
1511b490 | 1045 | if (bs == NULL) { |
8942764f | 1046 | return -EIO; |
74cf6c50 | 1047 | } |
8942764f | 1048 | |
48a4e92d KW |
1049 | blk = blk_co_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL, |
1050 | errp); | |
a3aeeab5 EB |
1051 | if (!blk) { |
1052 | ret = -EPERM; | |
1511b490 KW |
1053 | goto out; |
1054 | } | |
1055 | blk_set_allow_write_beyond_eof(blk, true); | |
8942764f | 1056 | |
1511b490 | 1057 | /* Create image format */ |
369f7de9 | 1058 | bat_entries = DIV_ROUND_UP(total_size, cl_size); |
2d68e22e | 1059 | bat_sectors = DIV_ROUND_UP(bat_entry_off(bat_entries), cl_size); |
369f7de9 | 1060 | bat_sectors = (bat_sectors * cl_size) >> BDRV_SECTOR_BITS; |
74cf6c50 DL |
1061 | |
1062 | memset(&header, 0, sizeof(header)); | |
1063 | memcpy(header.magic, HEADER_MAGIC2, sizeof(header.magic)); | |
1064 | header.version = cpu_to_le32(HEADER_VERSION); | |
1065 | /* don't care much about geometry, it is not used on image level */ | |
908b1c84 KK |
1066 | header.heads = cpu_to_le32(HEADS_NUMBER); |
1067 | header.cylinders = cpu_to_le32(total_size / BDRV_SECTOR_SIZE | |
1068 | / HEADS_NUMBER / SEC_IN_CYL); | |
74cf6c50 | 1069 | header.tracks = cpu_to_le32(cl_size >> BDRV_SECTOR_BITS); |
369f7de9 | 1070 | header.bat_entries = cpu_to_le32(bat_entries); |
74cf6c50 | 1071 | header.nb_sectors = cpu_to_le64(DIV_ROUND_UP(total_size, BDRV_SECTOR_SIZE)); |
369f7de9 | 1072 | header.data_off = cpu_to_le32(bat_sectors); |
74cf6c50 DL |
1073 | |
1074 | /* write all the data */ | |
1075 | memset(tmp, 0, sizeof(tmp)); | |
1076 | memcpy(tmp, &header, sizeof(header)); | |
1077 | ||
50688be0 | 1078 | ret = blk_co_pwrite(blk, 0, BDRV_SECTOR_SIZE, tmp, 0); |
74cf6c50 DL |
1079 | if (ret < 0) { |
1080 | goto exit; | |
1081 | } | |
50688be0 AF |
1082 | ret = blk_co_pwrite_zeroes(blk, BDRV_SECTOR_SIZE, |
1083 | (bat_sectors - 1) << BDRV_SECTOR_BITS, 0); | |
74cf6c50 DL |
1084 | if (ret < 0) { |
1085 | goto exit; | |
1086 | } | |
74cf6c50 | 1087 | |
1511b490 KW |
1088 | ret = 0; |
1089 | out: | |
b2ab5f54 KW |
1090 | blk_co_unref(blk); |
1091 | bdrv_co_unref(bs); | |
74cf6c50 DL |
1092 | return ret; |
1093 | ||
1094 | exit: | |
1095 | error_setg_errno(errp, -ret, "Failed to create Parallels image"); | |
1511b490 KW |
1096 | goto out; |
1097 | } | |
1098 | ||
4db7ba3b | 1099 | static int coroutine_fn GRAPH_UNLOCKED |
4ec8df01 KW |
1100 | parallels_co_create_opts(BlockDriver *drv, const char *filename, |
1101 | QemuOpts *opts, Error **errp) | |
1511b490 KW |
1102 | { |
1103 | BlockdevCreateOptions *create_options = NULL; | |
1511b490 | 1104 | BlockDriverState *bs = NULL; |
92adf9db | 1105 | QDict *qdict; |
1511b490 KW |
1106 | Visitor *v; |
1107 | int ret; | |
1108 | ||
1109 | static const QDictRenames opt_renames[] = { | |
1110 | { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, | |
1111 | { NULL, NULL }, | |
1112 | }; | |
1113 | ||
1114 | /* Parse options and convert legacy syntax */ | |
1115 | qdict = qemu_opts_to_qdict_filtered(opts, NULL, ¶llels_create_opts, | |
1116 | true); | |
1117 | ||
1118 | if (!qdict_rename_keys(qdict, opt_renames, errp)) { | |
1119 | ret = -EINVAL; | |
1120 | goto done; | |
1121 | } | |
1122 | ||
1123 | /* Create and open the file (protocol layer) */ | |
2475a0d0 | 1124 | ret = bdrv_co_create_file(filename, opts, errp); |
1511b490 | 1125 | if (ret < 0) { |
1511b490 KW |
1126 | goto done; |
1127 | } | |
1128 | ||
48a4e92d KW |
1129 | bs = bdrv_co_open(filename, NULL, NULL, |
1130 | BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); | |
1511b490 KW |
1131 | if (bs == NULL) { |
1132 | ret = -EIO; | |
1133 | goto done; | |
1134 | } | |
1135 | ||
1136 | /* Now get the QAPI type BlockdevCreateOptions */ | |
1137 | qdict_put_str(qdict, "driver", "parallels"); | |
1138 | qdict_put_str(qdict, "file", bs->node_name); | |
1139 | ||
af91062e MA |
1140 | v = qobject_input_visitor_new_flat_confused(qdict, errp); |
1141 | if (!v) { | |
1511b490 KW |
1142 | ret = -EINVAL; |
1143 | goto done; | |
1144 | } | |
1145 | ||
b11a093c | 1146 | visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); |
1511b490 | 1147 | visit_free(v); |
b11a093c | 1148 | if (!create_options) { |
1511b490 KW |
1149 | ret = -EINVAL; |
1150 | goto done; | |
1151 | } | |
1152 | ||
1153 | /* Silently round up sizes */ | |
1154 | create_options->u.parallels.size = | |
1155 | ROUND_UP(create_options->u.parallels.size, BDRV_SECTOR_SIZE); | |
1156 | create_options->u.parallels.cluster_size = | |
1157 | ROUND_UP(create_options->u.parallels.cluster_size, BDRV_SECTOR_SIZE); | |
1158 | ||
1159 | /* Create the Parallels image (format layer) */ | |
1160 | ret = parallels_co_create(create_options, errp); | |
1161 | if (ret < 0) { | |
1162 | goto done; | |
1163 | } | |
1164 | ret = 0; | |
1165 | ||
1166 | done: | |
cb3e7f08 | 1167 | qobject_unref(qdict); |
b2ab5f54 | 1168 | bdrv_co_unref(bs); |
1511b490 KW |
1169 | qapi_free_BlockdevCreateOptions(create_options); |
1170 | return ret; | |
74cf6c50 DL |
1171 | } |
1172 | ||
23d6bd3b DL |
1173 | |
1174 | static int parallels_probe(const uint8_t *buf, int buf_size, | |
1175 | const char *filename) | |
1176 | { | |
1177 | const ParallelsHeader *ph = (const void *)buf; | |
1178 | ||
1179 | if (buf_size < sizeof(ParallelsHeader)) { | |
1180 | return 0; | |
1181 | } | |
1182 | ||
1183 | if ((!memcmp(ph->magic, HEADER_MAGIC, 16) || | |
1184 | !memcmp(ph->magic, HEADER_MAGIC2, 16)) && | |
1185 | (le32_to_cpu(ph->version) == HEADER_VERSION)) { | |
1186 | return 100; | |
1187 | } | |
1188 | ||
1189 | return 0; | |
1190 | } | |
1191 | ||
6dd6b9f1 DL |
1192 | static int parallels_update_header(BlockDriverState *bs) |
1193 | { | |
1194 | BDRVParallelsState *s = bs->opaque; | |
9a4f4c31 KW |
1195 | unsigned size = MAX(bdrv_opt_mem_align(bs->file->bs), |
1196 | sizeof(ParallelsHeader)); | |
6dd6b9f1 DL |
1197 | |
1198 | if (size > s->header_size) { | |
1199 | size = s->header_size; | |
1200 | } | |
32cc71de | 1201 | return bdrv_pwrite_sync(bs->file, 0, size, s->header, 0); |
6dd6b9f1 DL |
1202 | } |
1203 | ||
93361b7e DL |
1204 | |
1205 | static int parallels_opts_prealloc(BlockDriverState *bs, QDict *options, | |
1206 | Error **errp) | |
1207 | { | |
1208 | int err; | |
1209 | char *buf; | |
1210 | int64_t bytes; | |
1211 | BDRVParallelsState *s = bs->opaque; | |
1212 | Error *local_err = NULL; | |
1213 | QemuOpts *opts = qemu_opts_create(¶llels_runtime_opts, NULL, 0, errp); | |
1214 | if (!opts) { | |
1215 | return -ENOMEM; | |
1216 | } | |
1217 | ||
1218 | err = -EINVAL; | |
1219 | if (!qemu_opts_absorb_qdict(opts, options, errp)) { | |
1220 | goto done; | |
1221 | } | |
1222 | ||
1223 | bytes = qemu_opt_get_size_del(opts, PARALLELS_OPT_PREALLOC_SIZE, 0); | |
1224 | s->prealloc_size = bytes >> BDRV_SECTOR_BITS; | |
1225 | buf = qemu_opt_get_del(opts, PARALLELS_OPT_PREALLOC_MODE); | |
1226 | /* prealloc_mode can be downgraded later during allocate_clusters */ | |
1227 | s->prealloc_mode = qapi_enum_parse(&prealloc_mode_lookup, buf, | |
1228 | PRL_PREALLOC_MODE_FALLOCATE, | |
1229 | &local_err); | |
1230 | g_free(buf); | |
1231 | if (local_err != NULL) { | |
1232 | error_propagate(errp, local_err); | |
1233 | goto done; | |
1234 | } | |
1235 | err = 0; | |
1236 | ||
1237 | done: | |
1238 | qemu_opts_del(opts); | |
1239 | return err; | |
1240 | } | |
1241 | ||
23d6bd3b DL |
1242 | static int parallels_open(BlockDriverState *bs, QDict *options, int flags, |
1243 | Error **errp) | |
1244 | { | |
1245 | BDRVParallelsState *s = bs->opaque; | |
1246 | ParallelsHeader ph; | |
19f5dc15 | 1247 | int ret, size, i; |
cfce1091 | 1248 | int64_t file_nb_sectors, sector; |
c89d4362 | 1249 | uint32_t data_start; |
8f5f5326 | 1250 | bool need_check = false; |
23d6bd3b | 1251 | |
93361b7e DL |
1252 | ret = parallels_opts_prealloc(bs, options, errp); |
1253 | if (ret < 0) { | |
1254 | return ret; | |
1255 | } | |
1256 | ||
83930780 VSO |
1257 | ret = bdrv_open_file_child(NULL, options, "file", bs, errp); |
1258 | if (ret < 0) { | |
1259 | return ret; | |
4e4bf5c4 KW |
1260 | } |
1261 | ||
f5e715db AI |
1262 | file_nb_sectors = bdrv_nb_sectors(bs->file->bs); |
1263 | if (file_nb_sectors < 0) { | |
1264 | return -EINVAL; | |
1265 | } | |
1266 | ||
32cc71de | 1267 | ret = bdrv_pread(bs->file, 0, sizeof(ph), &ph, 0); |
23d6bd3b | 1268 | if (ret < 0) { |
e17b9d08 | 1269 | return ret; |
23d6bd3b DL |
1270 | } |
1271 | ||
1272 | bs->total_sectors = le64_to_cpu(ph.nb_sectors); | |
1273 | ||
1274 | if (le32_to_cpu(ph.version) != HEADER_VERSION) { | |
1275 | goto fail_format; | |
1276 | } | |
1277 | if (!memcmp(ph.magic, HEADER_MAGIC, 16)) { | |
1278 | s->off_multiplier = 1; | |
1279 | bs->total_sectors = 0xffffffff & bs->total_sectors; | |
1280 | } else if (!memcmp(ph.magic, HEADER_MAGIC2, 16)) { | |
1281 | s->off_multiplier = le32_to_cpu(ph.tracks); | |
1282 | } else { | |
1283 | goto fail_format; | |
1284 | } | |
1285 | ||
1286 | s->tracks = le32_to_cpu(ph.tracks); | |
1287 | if (s->tracks == 0) { | |
1288 | error_setg(errp, "Invalid image: Zero sectors per track"); | |
e17b9d08 | 1289 | return -EINVAL; |
23d6bd3b DL |
1290 | } |
1291 | if (s->tracks > INT32_MAX/513) { | |
1292 | error_setg(errp, "Invalid image: Too big cluster"); | |
e17b9d08 | 1293 | return -EFBIG; |
23d6bd3b | 1294 | } |
93361b7e | 1295 | s->prealloc_size = MAX(s->tracks, s->prealloc_size); |
e0b5207f | 1296 | s->cluster_size = s->tracks << BDRV_SECTOR_BITS; |
23d6bd3b DL |
1297 | |
1298 | s->bat_size = le32_to_cpu(ph.bat_entries); | |
1299 | if (s->bat_size > INT_MAX / sizeof(uint32_t)) { | |
1300 | error_setg(errp, "Catalog too large"); | |
e17b9d08 | 1301 | return -EFBIG; |
23d6bd3b DL |
1302 | } |
1303 | ||
2d68e22e | 1304 | size = bat_entry_off(s->bat_size); |
9a4f4c31 KW |
1305 | s->header_size = ROUND_UP(size, bdrv_opt_mem_align(bs->file->bs)); |
1306 | s->header = qemu_try_blockalign(bs->file->bs, s->header_size); | |
23d6bd3b | 1307 | if (s->header == NULL) { |
e17b9d08 | 1308 | return -ENOMEM; |
23d6bd3b | 1309 | } |
23d6bd3b | 1310 | |
32cc71de | 1311 | ret = bdrv_pread(bs->file, 0, s->header_size, s->header, 0); |
23d6bd3b DL |
1312 | if (ret < 0) { |
1313 | goto fail; | |
1314 | } | |
1315 | s->bat_bitmap = (uint32_t *)(s->header + 1); | |
1316 | ||
6dd6b9f1 | 1317 | if (le32_to_cpu(ph.inuse) == HEADER_INUSE_MAGIC) { |
8f5f5326 DL |
1318 | need_check = s->header_unclean = true; |
1319 | } | |
1320 | ||
1321 | { | |
1322 | bool ok = parallels_test_data_off(s, file_nb_sectors, &data_start); | |
1323 | need_check = need_check || !ok; | |
6dd6b9f1 DL |
1324 | } |
1325 | ||
c89d4362 AI |
1326 | s->data_start = data_start; |
1327 | s->data_end = s->data_start; | |
1328 | if (s->data_end < (s->header_size >> BDRV_SECTOR_BITS)) { | |
1329 | /* | |
1330 | * There is not enough unused space to fit to block align between BAT | |
1331 | * and actual data. We can't avoid read-modify-write... | |
1332 | */ | |
1333 | s->header_size = size; | |
1334 | } | |
1335 | ||
baefd977 VSO |
1336 | if (ph.ext_off) { |
1337 | if (flags & BDRV_O_RDWR) { | |
1338 | /* | |
1339 | * It's unsafe to open image RW if there is an extension (as we | |
1340 | * don't support it). But parallels driver in QEMU historically | |
1341 | * ignores the extension, so print warning and don't care. | |
1342 | */ | |
1343 | warn_report("Format Extension ignored in RW mode"); | |
1344 | } else { | |
1345 | ret = parallels_read_format_extension( | |
1346 | bs, le64_to_cpu(ph.ext_off) << BDRV_SECTOR_BITS, errp); | |
1347 | if (ret < 0) { | |
1348 | goto fail; | |
1349 | } | |
1350 | } | |
1351 | } | |
1352 | ||
6c7d390b | 1353 | if ((flags & BDRV_O_RDWR) && !(flags & BDRV_O_INACTIVE)) { |
6dd6b9f1 DL |
1354 | s->header->inuse = cpu_to_le32(HEADER_INUSE_MAGIC); |
1355 | ret = parallels_update_header(bs); | |
1356 | if (ret < 0) { | |
1357 | goto fail; | |
1358 | } | |
1359 | } | |
1360 | ||
8e3b0cbb | 1361 | s->bat_dirty_block = 4 * qemu_real_host_page_size(); |
0d31c7c2 DL |
1362 | s->bat_dirty_bmap = |
1363 | bitmap_new(DIV_ROUND_UP(s->header_size, s->bat_dirty_block)); | |
1364 | ||
a94750d9 | 1365 | /* Disable migration until bdrv_activate method is added */ |
1d0f37cf JC |
1366 | error_setg(&s->migration_blocker, "The Parallels format used by node '%s' " |
1367 | "does not support live migration", | |
1368 | bdrv_get_device_or_node_name(bs)); | |
386f6c07 MA |
1369 | ret = migrate_add_blocker(s->migration_blocker, errp); |
1370 | if (ret < 0) { | |
cfce1091 | 1371 | error_setg(errp, "Migration blocker error"); |
1d0f37cf JC |
1372 | goto fail; |
1373 | } | |
23d6bd3b | 1374 | qemu_co_mutex_init(&s->lock); |
cfce1091 AI |
1375 | |
1376 | for (i = 0; i < s->bat_size; i++) { | |
1377 | sector = bat2sect(s, i); | |
1378 | if (sector + s->tracks > s->data_end) { | |
1379 | s->data_end = sector + s->tracks; | |
1380 | } | |
1381 | } | |
8f5f5326 | 1382 | need_check = need_check || s->data_end > file_nb_sectors; |
cfce1091 | 1383 | |
e185100f DL |
1384 | if (!need_check) { |
1385 | ret = parallels_fill_used_bitmap(bs); | |
1386 | if (ret == -ENOMEM) { | |
1387 | goto fail; | |
1388 | } | |
1389 | need_check = need_check || ret < 0; /* These are correctable errors */ | |
1390 | } | |
1391 | ||
cfce1091 AI |
1392 | /* |
1393 | * We don't repair the image here if it's opened for checks. Also we don't | |
1394 | * want to change inactive images and can't change readonly images. | |
1395 | */ | |
1396 | if ((flags & (BDRV_O_CHECK | BDRV_O_INACTIVE)) || !(flags & BDRV_O_RDWR)) { | |
1397 | return 0; | |
1398 | } | |
1399 | ||
8f5f5326 DL |
1400 | /* Repair the image if corruption was detected. */ |
1401 | if (need_check) { | |
cfce1091 AI |
1402 | BdrvCheckResult res; |
1403 | ret = bdrv_check(bs, &res, BDRV_FIX_ERRORS | BDRV_FIX_LEAKS); | |
1404 | if (ret < 0) { | |
1405 | error_setg_errno(errp, -ret, "Could not repair corrupted image"); | |
1406 | migrate_del_blocker(s->migration_blocker); | |
1407 | goto fail; | |
1408 | } | |
1409 | } | |
23d6bd3b DL |
1410 | return 0; |
1411 | ||
1412 | fail_format: | |
1413 | error_setg(errp, "Image not in Parallels format"); | |
9c398781 DL |
1414 | return -EINVAL; |
1415 | ||
23d6bd3b | 1416 | fail: |
cfce1091 AI |
1417 | /* |
1418 | * "s" object was allocated by g_malloc0 so we can safely | |
1419 | * try to free its fields even they were not allocated. | |
1420 | */ | |
e185100f DL |
1421 | parallels_free_used_bitmap(bs); |
1422 | ||
cfce1091 AI |
1423 | error_free(s->migration_blocker); |
1424 | g_free(s->bat_dirty_bmap); | |
23d6bd3b DL |
1425 | qemu_vfree(s->header); |
1426 | return ret; | |
1427 | } | |
1428 | ||
1429 | ||
6ada7453 TS |
1430 | static void parallels_close(BlockDriverState *bs) |
1431 | { | |
1432 | BDRVParallelsState *s = bs->opaque; | |
6dd6b9f1 | 1433 | |
6c7d390b | 1434 | if ((bs->open_flags & BDRV_O_RDWR) && !(bs->open_flags & BDRV_O_INACTIVE)) { |
6dd6b9f1 DL |
1435 | s->header->inuse = 0; |
1436 | parallels_update_header(bs); | |
e8d04f92 HR |
1437 | |
1438 | /* errors are ignored, so we might as well pass exact=true */ | |
1439 | bdrv_truncate(bs->file, s->data_end << BDRV_SECTOR_BITS, true, | |
7b8e4857 | 1440 | PREALLOC_MODE_OFF, 0, NULL); |
19f5dc15 DL |
1441 | } |
1442 | ||
e185100f DL |
1443 | parallels_free_used_bitmap(bs); |
1444 | ||
0d31c7c2 | 1445 | g_free(s->bat_dirty_bmap); |
9eae9cca | 1446 | qemu_vfree(s->header); |
1d0f37cf JC |
1447 | |
1448 | migrate_del_blocker(s->migration_blocker); | |
1449 | error_free(s->migration_blocker); | |
6ada7453 TS |
1450 | } |
1451 | ||
73f3e136 DL |
1452 | static bool parallels_is_support_dirty_bitmaps(BlockDriverState *bs) |
1453 | { | |
1454 | return 1; | |
1455 | } | |
1456 | ||
5efa9d5a | 1457 | static BlockDriver bdrv_parallels = { |
bb16991f DL |
1458 | .format_name = "parallels", |
1459 | .instance_size = sizeof(BDRVParallelsState), | |
1460 | .create_opts = ¶llels_create_opts, | |
1461 | .is_format = true, | |
1462 | .supports_backing = true, | |
1463 | ||
1464 | .bdrv_has_zero_init = bdrv_has_zero_init_1, | |
73f3e136 | 1465 | .bdrv_supports_persistent_dirty_bitmap = parallels_is_support_dirty_bitmaps, |
bb16991f DL |
1466 | |
1467 | .bdrv_probe = parallels_probe, | |
1468 | .bdrv_open = parallels_open, | |
1469 | .bdrv_close = parallels_close, | |
1470 | .bdrv_child_perm = bdrv_default_perms, | |
1471 | .bdrv_co_block_status = parallels_co_block_status, | |
1472 | .bdrv_co_flush_to_os = parallels_co_flush_to_os, | |
1473 | .bdrv_co_readv = parallels_co_readv, | |
1474 | .bdrv_co_writev = parallels_co_writev, | |
1475 | .bdrv_co_create = parallels_co_create, | |
1476 | .bdrv_co_create_opts = parallels_co_create_opts, | |
1477 | .bdrv_co_check = parallels_co_check, | |
c54fb435 | 1478 | .bdrv_co_pdiscard = parallels_co_pdiscard, |
a98b260a | 1479 | .bdrv_co_pwrite_zeroes = parallels_co_pwrite_zeroes, |
6ada7453 | 1480 | }; |
5efa9d5a AL |
1481 | |
1482 | static void bdrv_parallels_init(void) | |
1483 | { | |
1484 | bdrv_register(&bdrv_parallels); | |
1485 | } | |
1486 | ||
1487 | block_init(bdrv_parallels_init); |