]>
Commit | Line | Data |
---|---|---|
75411d23 SH |
1 | /* |
2 | * QEMU Enhanced Disk Format | |
3 | * | |
4 | * Copyright IBM, Corp. 2010 | |
5 | * | |
6 | * Authors: | |
7 | * Stefan Hajnoczi <[email protected]> | |
8 | * Anthony Liguori <[email protected]> | |
9 | * | |
10 | * This work is licensed under the terms of the GNU LGPL, version 2 or later. | |
11 | * See the COPYING.LIB file in the top-level directory. | |
12 | * | |
13 | */ | |
14 | ||
80c71a24 | 15 | #include "qemu/osdep.h" |
609f45ea | 16 | #include "block/qdict.h" |
da34e65c | 17 | #include "qapi/error.h" |
1de7afc9 | 18 | #include "qemu/timer.h" |
58369e22 | 19 | #include "qemu/bswap.h" |
db725815 | 20 | #include "qemu/main-loop.h" |
0b8fa32f | 21 | #include "qemu/module.h" |
922a01a0 | 22 | #include "qemu/option.h" |
eabba580 | 23 | #include "trace.h" |
75411d23 | 24 | #include "qed.h" |
8a56fdad | 25 | #include "sysemu/block-backend.h" |
959355a4 KW |
26 | #include "qapi/qmp/qdict.h" |
27 | #include "qapi/qobject-input-visitor.h" | |
28 | #include "qapi/qapi-visit-block-core.h" | |
29 | ||
30 | static QemuOptsList qed_create_opts; | |
75411d23 SH |
31 | |
32 | static int bdrv_qed_probe(const uint8_t *buf, int buf_size, | |
33 | const char *filename) | |
34 | { | |
35 | const QEDHeader *header = (const QEDHeader *)buf; | |
36 | ||
37 | if (buf_size < sizeof(*header)) { | |
38 | return 0; | |
39 | } | |
40 | if (le32_to_cpu(header->magic) != QED_MAGIC) { | |
41 | return 0; | |
42 | } | |
43 | return 100; | |
44 | } | |
45 | ||
46 | /** | |
47 | * Check whether an image format is raw | |
48 | * | |
49 | * @fmt: Backing file format, may be NULL | |
50 | */ | |
51 | static bool qed_fmt_is_raw(const char *fmt) | |
52 | { | |
53 | return fmt && strcmp(fmt, "raw") == 0; | |
54 | } | |
55 | ||
56 | static void qed_header_le_to_cpu(const QEDHeader *le, QEDHeader *cpu) | |
57 | { | |
58 | cpu->magic = le32_to_cpu(le->magic); | |
59 | cpu->cluster_size = le32_to_cpu(le->cluster_size); | |
60 | cpu->table_size = le32_to_cpu(le->table_size); | |
61 | cpu->header_size = le32_to_cpu(le->header_size); | |
62 | cpu->features = le64_to_cpu(le->features); | |
63 | cpu->compat_features = le64_to_cpu(le->compat_features); | |
64 | cpu->autoclear_features = le64_to_cpu(le->autoclear_features); | |
65 | cpu->l1_table_offset = le64_to_cpu(le->l1_table_offset); | |
66 | cpu->image_size = le64_to_cpu(le->image_size); | |
67 | cpu->backing_filename_offset = le32_to_cpu(le->backing_filename_offset); | |
68 | cpu->backing_filename_size = le32_to_cpu(le->backing_filename_size); | |
69 | } | |
70 | ||
71 | static void qed_header_cpu_to_le(const QEDHeader *cpu, QEDHeader *le) | |
72 | { | |
73 | le->magic = cpu_to_le32(cpu->magic); | |
74 | le->cluster_size = cpu_to_le32(cpu->cluster_size); | |
75 | le->table_size = cpu_to_le32(cpu->table_size); | |
76 | le->header_size = cpu_to_le32(cpu->header_size); | |
77 | le->features = cpu_to_le64(cpu->features); | |
78 | le->compat_features = cpu_to_le64(cpu->compat_features); | |
79 | le->autoclear_features = cpu_to_le64(cpu->autoclear_features); | |
80 | le->l1_table_offset = cpu_to_le64(cpu->l1_table_offset); | |
81 | le->image_size = cpu_to_le64(cpu->image_size); | |
82 | le->backing_filename_offset = cpu_to_le32(cpu->backing_filename_offset); | |
83 | le->backing_filename_size = cpu_to_le32(cpu->backing_filename_size); | |
84 | } | |
85 | ||
b10170ac | 86 | int qed_write_header_sync(BDRVQEDState *s) |
75411d23 SH |
87 | { |
88 | QEDHeader le; | |
89 | int ret; | |
90 | ||
91 | qed_header_cpu_to_le(&s->header, &le); | |
d9ca2ea2 | 92 | ret = bdrv_pwrite(s->bs->file, 0, &le, sizeof(le)); |
75411d23 SH |
93 | if (ret != sizeof(le)) { |
94 | return ret; | |
95 | } | |
96 | return 0; | |
97 | } | |
98 | ||
01979a98 SH |
99 | /** |
100 | * Update header in-place (does not rewrite backing filename or other strings) | |
101 | * | |
102 | * This function only updates known header fields in-place and does not affect | |
103 | * extra data after the QED header. | |
1f01e50b PB |
104 | * |
105 | * No new allocating reqs can start while this function runs. | |
01979a98 | 106 | */ |
87f0d882 | 107 | static int coroutine_fn qed_write_header(BDRVQEDState *s) |
01979a98 SH |
108 | { |
109 | /* We must write full sectors for O_DIRECT but cannot necessarily generate | |
110 | * the data following the header if an unrecognized compat feature is | |
111 | * active. Therefore, first read the sectors containing the header, update | |
112 | * them, and write back. | |
113 | */ | |
114 | ||
c41a73ff | 115 | int nsectors = DIV_ROUND_UP(sizeof(QEDHeader), BDRV_SECTOR_SIZE); |
01979a98 | 116 | size_t len = nsectors * BDRV_SECTOR_SIZE; |
7076309a | 117 | uint8_t *buf; |
7076309a KW |
118 | int ret; |
119 | ||
1f01e50b PB |
120 | assert(s->allocating_acb || s->allocating_write_reqs_plugged); |
121 | ||
7076309a | 122 | buf = qemu_blockalign(s->bs, len); |
7076309a | 123 | |
696e8cb2 | 124 | ret = bdrv_co_pread(s->bs->file, 0, len, buf, 0); |
7076309a KW |
125 | if (ret < 0) { |
126 | goto out; | |
127 | } | |
128 | ||
129 | /* Update header */ | |
130 | qed_header_cpu_to_le(&s->header, (QEDHeader *) buf); | |
131 | ||
696e8cb2 | 132 | ret = bdrv_co_pwrite(s->bs->file, 0, len, buf, 0); |
7076309a KW |
133 | if (ret < 0) { |
134 | goto out; | |
135 | } | |
136 | ||
137 | ret = 0; | |
138 | out: | |
139 | qemu_vfree(buf); | |
f13d712b | 140 | return ret; |
01979a98 SH |
141 | } |
142 | ||
75411d23 SH |
143 | static uint64_t qed_max_image_size(uint32_t cluster_size, uint32_t table_size) |
144 | { | |
145 | uint64_t table_entries; | |
146 | uint64_t l2_size; | |
147 | ||
148 | table_entries = (table_size * cluster_size) / sizeof(uint64_t); | |
149 | l2_size = table_entries * cluster_size; | |
150 | ||
151 | return l2_size * table_entries; | |
152 | } | |
153 | ||
154 | static bool qed_is_cluster_size_valid(uint32_t cluster_size) | |
155 | { | |
156 | if (cluster_size < QED_MIN_CLUSTER_SIZE || | |
157 | cluster_size > QED_MAX_CLUSTER_SIZE) { | |
158 | return false; | |
159 | } | |
160 | if (cluster_size & (cluster_size - 1)) { | |
161 | return false; /* not power of 2 */ | |
162 | } | |
163 | return true; | |
164 | } | |
165 | ||
166 | static bool qed_is_table_size_valid(uint32_t table_size) | |
167 | { | |
168 | if (table_size < QED_MIN_TABLE_SIZE || | |
169 | table_size > QED_MAX_TABLE_SIZE) { | |
170 | return false; | |
171 | } | |
172 | if (table_size & (table_size - 1)) { | |
173 | return false; /* not power of 2 */ | |
174 | } | |
175 | return true; | |
176 | } | |
177 | ||
178 | static bool qed_is_image_size_valid(uint64_t image_size, uint32_t cluster_size, | |
179 | uint32_t table_size) | |
180 | { | |
181 | if (image_size % BDRV_SECTOR_SIZE != 0) { | |
182 | return false; /* not multiple of sector size */ | |
183 | } | |
184 | if (image_size > qed_max_image_size(cluster_size, table_size)) { | |
185 | return false; /* image is too large */ | |
186 | } | |
187 | return true; | |
188 | } | |
189 | ||
190 | /** | |
191 | * Read a string of known length from the image file | |
192 | * | |
193 | * @file: Image file | |
194 | * @offset: File offset to start of string, in bytes | |
195 | * @n: String length in bytes | |
196 | * @buf: Destination buffer | |
197 | * @buflen: Destination buffer length in bytes | |
198 | * @ret: 0 on success, -errno on failure | |
199 | * | |
200 | * The string is NUL-terminated. | |
201 | */ | |
cf2ab8fc | 202 | static int qed_read_string(BdrvChild *file, uint64_t offset, size_t n, |
75411d23 SH |
203 | char *buf, size_t buflen) |
204 | { | |
205 | int ret; | |
206 | if (n >= buflen) { | |
207 | return -EINVAL; | |
208 | } | |
209 | ret = bdrv_pread(file, offset, buf, n); | |
210 | if (ret < 0) { | |
211 | return ret; | |
212 | } | |
213 | buf[n] = '\0'; | |
214 | return 0; | |
215 | } | |
216 | ||
eabba580 SH |
217 | /** |
218 | * Allocate new clusters | |
219 | * | |
220 | * @s: QED state | |
221 | * @n: Number of contiguous clusters to allocate | |
222 | * @ret: Offset of first allocated cluster | |
223 | * | |
224 | * This function only produces the offset where the new clusters should be | |
225 | * written. It updates BDRVQEDState but does not make any changes to the image | |
226 | * file. | |
1f01e50b PB |
227 | * |
228 | * Called with table_lock held. | |
eabba580 SH |
229 | */ |
230 | static uint64_t qed_alloc_clusters(BDRVQEDState *s, unsigned int n) | |
231 | { | |
232 | uint64_t offset = s->file_size; | |
233 | s->file_size += n * s->header.cluster_size; | |
234 | return offset; | |
235 | } | |
236 | ||
298800ca SH |
237 | QEDTable *qed_alloc_table(BDRVQEDState *s) |
238 | { | |
239 | /* Honor O_DIRECT memory alignment requirements */ | |
240 | return qemu_blockalign(s->bs, | |
241 | s->header.cluster_size * s->header.table_size); | |
242 | } | |
243 | ||
eabba580 SH |
244 | /** |
245 | * Allocate a new zeroed L2 table | |
1f01e50b PB |
246 | * |
247 | * Called with table_lock held. | |
eabba580 SH |
248 | */ |
249 | static CachedL2Table *qed_new_l2_table(BDRVQEDState *s) | |
250 | { | |
251 | CachedL2Table *l2_table = qed_alloc_l2_cache_entry(&s->l2_cache); | |
252 | ||
253 | l2_table->table = qed_alloc_table(s); | |
254 | l2_table->offset = qed_alloc_clusters(s, s->header.table_size); | |
255 | ||
256 | memset(l2_table->table->offsets, 0, | |
257 | s->header.cluster_size * s->header.table_size); | |
258 | return l2_table; | |
259 | } | |
260 | ||
1f01e50b | 261 | static bool qed_plug_allocating_write_reqs(BDRVQEDState *s) |
6f321e93 | 262 | { |
1f01e50b PB |
263 | qemu_co_mutex_lock(&s->table_lock); |
264 | ||
265 | /* No reentrancy is allowed. */ | |
6f321e93 | 266 | assert(!s->allocating_write_reqs_plugged); |
1f01e50b PB |
267 | if (s->allocating_acb != NULL) { |
268 | /* Another allocating write came concurrently. This cannot happen | |
f8ea8dac | 269 | * from bdrv_qed_co_drain_begin, but it can happen when the timer runs. |
1f01e50b PB |
270 | */ |
271 | qemu_co_mutex_unlock(&s->table_lock); | |
272 | return false; | |
273 | } | |
6f321e93 SH |
274 | |
275 | s->allocating_write_reqs_plugged = true; | |
1f01e50b PB |
276 | qemu_co_mutex_unlock(&s->table_lock); |
277 | return true; | |
6f321e93 SH |
278 | } |
279 | ||
280 | static void qed_unplug_allocating_write_reqs(BDRVQEDState *s) | |
281 | { | |
1f01e50b | 282 | qemu_co_mutex_lock(&s->table_lock); |
6f321e93 | 283 | assert(s->allocating_write_reqs_plugged); |
6f321e93 | 284 | s->allocating_write_reqs_plugged = false; |
1f01e50b PB |
285 | qemu_co_queue_next(&s->allocating_write_reqs); |
286 | qemu_co_mutex_unlock(&s->table_lock); | |
6f321e93 SH |
287 | } |
288 | ||
87f0d882 | 289 | static void coroutine_fn qed_need_check_timer_entry(void *opaque) |
6f321e93 SH |
290 | { |
291 | BDRVQEDState *s = opaque; | |
c0e8f989 KW |
292 | int ret; |
293 | ||
c0e8f989 | 294 | trace_qed_need_check_timer_cb(s); |
6f321e93 | 295 | |
1f01e50b PB |
296 | if (!qed_plug_allocating_write_reqs(s)) { |
297 | return; | |
298 | } | |
c0e8f989 KW |
299 | |
300 | /* Ensure writes are on disk before clearing flag */ | |
301 | ret = bdrv_co_flush(s->bs->file->bs); | |
c0e8f989 | 302 | if (ret < 0) { |
6f321e93 SH |
303 | qed_unplug_allocating_write_reqs(s); |
304 | return; | |
305 | } | |
306 | ||
307 | s->header.features &= ~QED_F_NEED_CHECK; | |
f13d712b KW |
308 | ret = qed_write_header(s); |
309 | (void) ret; | |
310 | ||
311 | qed_unplug_allocating_write_reqs(s); | |
312 | ||
c0e8f989 | 313 | ret = bdrv_co_flush(s->bs); |
f13d712b | 314 | (void) ret; |
6f321e93 SH |
315 | } |
316 | ||
317 | static void qed_need_check_timer_cb(void *opaque) | |
318 | { | |
c0e8f989 KW |
319 | Coroutine *co = qemu_coroutine_create(qed_need_check_timer_entry, opaque); |
320 | qemu_coroutine_enter(co); | |
2f47da5f PB |
321 | } |
322 | ||
6f321e93 SH |
323 | static void qed_start_need_check_timer(BDRVQEDState *s) |
324 | { | |
325 | trace_qed_start_need_check_timer(s); | |
326 | ||
bc72ad67 | 327 | /* Use QEMU_CLOCK_VIRTUAL so we don't alter the image file while suspended for |
6f321e93 SH |
328 | * migration. |
329 | */ | |
bc72ad67 | 330 | timer_mod(s->need_check_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + |
73bcb24d | 331 | NANOSECONDS_PER_SECOND * QED_NEED_CHECK_TIMEOUT); |
6f321e93 SH |
332 | } |
333 | ||
334 | /* It's okay to call this multiple times or when no timer is started */ | |
335 | static void qed_cancel_need_check_timer(BDRVQEDState *s) | |
336 | { | |
337 | trace_qed_cancel_need_check_timer(s); | |
bc72ad67 | 338 | timer_del(s->need_check_timer); |
6f321e93 SH |
339 | } |
340 | ||
a8c868c3 SH |
341 | static void bdrv_qed_detach_aio_context(BlockDriverState *bs) |
342 | { | |
343 | BDRVQEDState *s = bs->opaque; | |
344 | ||
345 | qed_cancel_need_check_timer(s); | |
346 | timer_free(s->need_check_timer); | |
347 | } | |
348 | ||
349 | static void bdrv_qed_attach_aio_context(BlockDriverState *bs, | |
350 | AioContext *new_context) | |
351 | { | |
352 | BDRVQEDState *s = bs->opaque; | |
353 | ||
354 | s->need_check_timer = aio_timer_new(new_context, | |
355 | QEMU_CLOCK_VIRTUAL, SCALE_NS, | |
356 | qed_need_check_timer_cb, s); | |
357 | if (s->header.features & QED_F_NEED_CHECK) { | |
358 | qed_start_need_check_timer(s); | |
359 | } | |
360 | } | |
361 | ||
f8ea8dac | 362 | static void coroutine_fn bdrv_qed_co_drain_begin(BlockDriverState *bs) |
6653a73d FZ |
363 | { |
364 | BDRVQEDState *s = bs->opaque; | |
365 | ||
366 | /* Fire the timer immediately in order to start doing I/O as soon as the | |
367 | * header is flushed. | |
368 | */ | |
369 | if (s->need_check_timer && timer_pending(s->need_check_timer)) { | |
370 | qed_cancel_need_check_timer(s); | |
61124f03 | 371 | qed_need_check_timer_entry(s); |
6653a73d FZ |
372 | } |
373 | } | |
374 | ||
61c7887e PB |
375 | static void bdrv_qed_init_state(BlockDriverState *bs) |
376 | { | |
377 | BDRVQEDState *s = bs->opaque; | |
378 | ||
379 | memset(s, 0, sizeof(BDRVQEDState)); | |
380 | s->bs = bs; | |
1f01e50b | 381 | qemu_co_mutex_init(&s->table_lock); |
61c7887e PB |
382 | qemu_co_queue_init(&s->allocating_write_reqs); |
383 | } | |
384 | ||
9fb4dfc5 PB |
385 | /* Called with table_lock held. */ |
386 | static int coroutine_fn bdrv_qed_do_open(BlockDriverState *bs, QDict *options, | |
387 | int flags, Error **errp) | |
75411d23 SH |
388 | { |
389 | BDRVQEDState *s = bs->opaque; | |
390 | QEDHeader le_header; | |
391 | int64_t file_size; | |
392 | int ret; | |
393 | ||
cf2ab8fc | 394 | ret = bdrv_pread(bs->file, 0, &le_header, sizeof(le_header)); |
75411d23 | 395 | if (ret < 0) { |
15ce94a6 | 396 | error_setg(errp, "Failed to read QED header"); |
75411d23 SH |
397 | return ret; |
398 | } | |
75411d23 SH |
399 | qed_header_le_to_cpu(&le_header, &s->header); |
400 | ||
401 | if (s->header.magic != QED_MAGIC) { | |
76abe407 PB |
402 | error_setg(errp, "Image not in QED format"); |
403 | return -EINVAL; | |
75411d23 SH |
404 | } |
405 | if (s->header.features & ~QED_FEATURE_MASK) { | |
10b758e8 | 406 | /* image uses unsupported feature bits */ |
a55448b3 HR |
407 | error_setg(errp, "Unsupported QED features: %" PRIx64, |
408 | s->header.features & ~QED_FEATURE_MASK); | |
10b758e8 | 409 | return -ENOTSUP; |
75411d23 SH |
410 | } |
411 | if (!qed_is_cluster_size_valid(s->header.cluster_size)) { | |
15ce94a6 | 412 | error_setg(errp, "QED cluster size is invalid"); |
75411d23 SH |
413 | return -EINVAL; |
414 | } | |
415 | ||
416 | /* Round down file size to the last cluster */ | |
9a4f4c31 | 417 | file_size = bdrv_getlength(bs->file->bs); |
75411d23 | 418 | if (file_size < 0) { |
15ce94a6 | 419 | error_setg(errp, "Failed to get file length"); |
75411d23 SH |
420 | return file_size; |
421 | } | |
422 | s->file_size = qed_start_of_cluster(s, file_size); | |
423 | ||
424 | if (!qed_is_table_size_valid(s->header.table_size)) { | |
15ce94a6 | 425 | error_setg(errp, "QED table size is invalid"); |
75411d23 SH |
426 | return -EINVAL; |
427 | } | |
428 | if (!qed_is_image_size_valid(s->header.image_size, | |
429 | s->header.cluster_size, | |
430 | s->header.table_size)) { | |
15ce94a6 | 431 | error_setg(errp, "QED image size is invalid"); |
75411d23 SH |
432 | return -EINVAL; |
433 | } | |
434 | if (!qed_check_table_offset(s, s->header.l1_table_offset)) { | |
15ce94a6 | 435 | error_setg(errp, "QED table offset is invalid"); |
75411d23 SH |
436 | return -EINVAL; |
437 | } | |
438 | ||
439 | s->table_nelems = (s->header.cluster_size * s->header.table_size) / | |
440 | sizeof(uint64_t); | |
786a4ea8 | 441 | s->l2_shift = ctz32(s->header.cluster_size); |
75411d23 | 442 | s->l2_mask = s->table_nelems - 1; |
786a4ea8 | 443 | s->l1_shift = s->l2_shift + ctz32(s->table_nelems); |
75411d23 | 444 | |
0adfa1ed SH |
445 | /* Header size calculation must not overflow uint32_t */ |
446 | if (s->header.header_size > UINT32_MAX / s->header.cluster_size) { | |
15ce94a6 | 447 | error_setg(errp, "QED header size is too large"); |
0adfa1ed SH |
448 | return -EINVAL; |
449 | } | |
450 | ||
75411d23 SH |
451 | if ((s->header.features & QED_F_BACKING_FILE)) { |
452 | if ((uint64_t)s->header.backing_filename_offset + | |
453 | s->header.backing_filename_size > | |
454 | s->header.cluster_size * s->header.header_size) { | |
15ce94a6 | 455 | error_setg(errp, "QED backing filename offset is invalid"); |
75411d23 SH |
456 | return -EINVAL; |
457 | } | |
458 | ||
cf2ab8fc | 459 | ret = qed_read_string(bs->file, s->header.backing_filename_offset, |
998c2019 HR |
460 | s->header.backing_filename_size, |
461 | bs->auto_backing_file, | |
462 | sizeof(bs->auto_backing_file)); | |
75411d23 | 463 | if (ret < 0) { |
15ce94a6 | 464 | error_setg(errp, "Failed to read backing filename"); |
75411d23 SH |
465 | return ret; |
466 | } | |
998c2019 HR |
467 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), |
468 | bs->auto_backing_file); | |
75411d23 SH |
469 | |
470 | if (s->header.features & QED_F_BACKING_FORMAT_NO_PROBE) { | |
471 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), "raw"); | |
472 | } | |
473 | } | |
474 | ||
475 | /* Reset unknown autoclear feature bits. This is a backwards | |
476 | * compatibility mechanism that allows images to be opened by older | |
477 | * programs, which "knock out" unknown feature bits. When an image is | |
478 | * opened by a newer program again it can detect that the autoclear | |
479 | * feature is no longer valid. | |
480 | */ | |
481 | if ((s->header.autoclear_features & ~QED_AUTOCLEAR_FEATURE_MASK) != 0 && | |
04c01a5c | 482 | !bdrv_is_read_only(bs->file->bs) && !(flags & BDRV_O_INACTIVE)) { |
75411d23 SH |
483 | s->header.autoclear_features &= QED_AUTOCLEAR_FEATURE_MASK; |
484 | ||
485 | ret = qed_write_header_sync(s); | |
486 | if (ret) { | |
15ce94a6 | 487 | error_setg(errp, "Failed to update header"); |
75411d23 SH |
488 | return ret; |
489 | } | |
490 | ||
491 | /* From here on only known autoclear feature bits are valid */ | |
9a4f4c31 | 492 | bdrv_flush(bs->file->bs); |
75411d23 SH |
493 | } |
494 | ||
298800ca SH |
495 | s->l1_table = qed_alloc_table(s); |
496 | qed_init_l2_cache(&s->l2_cache); | |
497 | ||
498 | ret = qed_read_l1_table_sync(s); | |
01979a98 | 499 | if (ret) { |
15ce94a6 | 500 | error_setg(errp, "Failed to read L1 table"); |
01979a98 SH |
501 | goto out; |
502 | } | |
503 | ||
504 | /* If image was not closed cleanly, check consistency */ | |
058f8f16 | 505 | if (!(flags & BDRV_O_CHECK) && (s->header.features & QED_F_NEED_CHECK)) { |
01979a98 SH |
506 | /* Read-only images cannot be fixed. There is no risk of corruption |
507 | * since write operations are not possible. Therefore, allow | |
508 | * potentially inconsistent images to be opened read-only. This can | |
509 | * aid data recovery from an otherwise inconsistent image. | |
510 | */ | |
9a4f4c31 | 511 | if (!bdrv_is_read_only(bs->file->bs) && |
04c01a5c | 512 | !(flags & BDRV_O_INACTIVE)) { |
01979a98 SH |
513 | BdrvCheckResult result = {0}; |
514 | ||
515 | ret = qed_check(s, &result, true); | |
6f321e93 | 516 | if (ret) { |
15ce94a6 | 517 | error_setg(errp, "Image corrupted"); |
6f321e93 SH |
518 | goto out; |
519 | } | |
01979a98 SH |
520 | } |
521 | } | |
522 | ||
a8c868c3 | 523 | bdrv_qed_attach_aio_context(bs, bdrv_get_aio_context(bs)); |
6f321e93 | 524 | |
01979a98 | 525 | out: |
298800ca SH |
526 | if (ret) { |
527 | qed_free_l2_cache(&s->l2_cache); | |
528 | qemu_vfree(s->l1_table); | |
529 | } | |
75411d23 SH |
530 | return ret; |
531 | } | |
532 | ||
9fb4dfc5 PB |
533 | typedef struct QEDOpenCo { |
534 | BlockDriverState *bs; | |
535 | QDict *options; | |
536 | int flags; | |
537 | Error **errp; | |
538 | int ret; | |
539 | } QEDOpenCo; | |
540 | ||
541 | static void coroutine_fn bdrv_qed_open_entry(void *opaque) | |
542 | { | |
543 | QEDOpenCo *qoc = opaque; | |
544 | BDRVQEDState *s = qoc->bs->opaque; | |
545 | ||
546 | qemu_co_mutex_lock(&s->table_lock); | |
547 | qoc->ret = bdrv_qed_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp); | |
548 | qemu_co_mutex_unlock(&s->table_lock); | |
549 | } | |
550 | ||
4e4bf5c4 KW |
551 | static int bdrv_qed_open(BlockDriverState *bs, QDict *options, int flags, |
552 | Error **errp) | |
553 | { | |
9fb4dfc5 PB |
554 | QEDOpenCo qoc = { |
555 | .bs = bs, | |
556 | .options = options, | |
557 | .flags = flags, | |
558 | .errp = errp, | |
559 | .ret = -EINPROGRESS | |
560 | }; | |
561 | ||
8b1869da HR |
562 | bs->file = bdrv_open_child(NULL, options, "file", bs, &child_of_bds, |
563 | BDRV_CHILD_IMAGE, false, errp); | |
4e4bf5c4 KW |
564 | if (!bs->file) { |
565 | return -EINVAL; | |
566 | } | |
567 | ||
61c7887e | 568 | bdrv_qed_init_state(bs); |
9fb4dfc5 PB |
569 | if (qemu_in_coroutine()) { |
570 | bdrv_qed_open_entry(&qoc); | |
571 | } else { | |
4720cbee | 572 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
9fb4dfc5 PB |
573 | qemu_coroutine_enter(qemu_coroutine_create(bdrv_qed_open_entry, &qoc)); |
574 | BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); | |
575 | } | |
576 | BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); | |
577 | return qoc.ret; | |
4e4bf5c4 KW |
578 | } |
579 | ||
3baca891 | 580 | static void bdrv_qed_refresh_limits(BlockDriverState *bs, Error **errp) |
d34682cd KW |
581 | { |
582 | BDRVQEDState *s = bs->opaque; | |
583 | ||
cf081fca | 584 | bs->bl.pwrite_zeroes_alignment = s->header.cluster_size; |
d34682cd KW |
585 | } |
586 | ||
f9cb20f1 JC |
587 | /* We have nothing to do for QED reopen, stubs just return |
588 | * success */ | |
589 | static int bdrv_qed_reopen_prepare(BDRVReopenState *state, | |
590 | BlockReopenQueue *queue, Error **errp) | |
591 | { | |
592 | return 0; | |
593 | } | |
594 | ||
75411d23 SH |
595 | static void bdrv_qed_close(BlockDriverState *bs) |
596 | { | |
298800ca SH |
597 | BDRVQEDState *s = bs->opaque; |
598 | ||
a8c868c3 | 599 | bdrv_qed_detach_aio_context(bs); |
6f321e93 | 600 | |
01979a98 | 601 | /* Ensure writes reach stable storage */ |
9a4f4c31 | 602 | bdrv_flush(bs->file->bs); |
01979a98 SH |
603 | |
604 | /* Clean shutdown, no check required on next open */ | |
605 | if (s->header.features & QED_F_NEED_CHECK) { | |
606 | s->header.features &= ~QED_F_NEED_CHECK; | |
607 | qed_write_header_sync(s); | |
608 | } | |
609 | ||
298800ca SH |
610 | qed_free_l2_cache(&s->l2_cache); |
611 | qemu_vfree(s->l1_table); | |
75411d23 SH |
612 | } |
613 | ||
959355a4 KW |
614 | static int coroutine_fn bdrv_qed_co_create(BlockdevCreateOptions *opts, |
615 | Error **errp) | |
75411d23 | 616 | { |
959355a4 KW |
617 | BlockdevCreateOptionsQed *qed_opts; |
618 | BlockBackend *blk = NULL; | |
619 | BlockDriverState *bs = NULL; | |
620 | ||
621 | QEDHeader header; | |
75411d23 SH |
622 | QEDHeader le_header; |
623 | uint8_t *l1_table = NULL; | |
959355a4 | 624 | size_t l1_size; |
75411d23 | 625 | int ret = 0; |
75411d23 | 626 | |
959355a4 KW |
627 | assert(opts->driver == BLOCKDEV_DRIVER_QED); |
628 | qed_opts = &opts->u.qed; | |
629 | ||
630 | /* Validate options and set default values */ | |
631 | if (!qed_opts->has_cluster_size) { | |
632 | qed_opts->cluster_size = QED_DEFAULT_CLUSTER_SIZE; | |
633 | } | |
634 | if (!qed_opts->has_table_size) { | |
635 | qed_opts->table_size = QED_DEFAULT_TABLE_SIZE; | |
75411d23 SH |
636 | } |
637 | ||
959355a4 KW |
638 | if (!qed_is_cluster_size_valid(qed_opts->cluster_size)) { |
639 | error_setg(errp, "QED cluster size must be within range [%u, %u] " | |
640 | "and power of 2", | |
641 | QED_MIN_CLUSTER_SIZE, QED_MAX_CLUSTER_SIZE); | |
642 | return -EINVAL; | |
643 | } | |
644 | if (!qed_is_table_size_valid(qed_opts->table_size)) { | |
645 | error_setg(errp, "QED table size must be within range [%u, %u] " | |
646 | "and power of 2", | |
647 | QED_MIN_TABLE_SIZE, QED_MAX_TABLE_SIZE); | |
648 | return -EINVAL; | |
649 | } | |
650 | if (!qed_is_image_size_valid(qed_opts->size, qed_opts->cluster_size, | |
651 | qed_opts->table_size)) | |
652 | { | |
653 | error_setg(errp, "QED image size must be a non-zero multiple of " | |
654 | "cluster size and less than %" PRIu64 " bytes", | |
655 | qed_max_image_size(qed_opts->cluster_size, | |
656 | qed_opts->table_size)); | |
657 | return -EINVAL; | |
658 | } | |
659 | ||
660 | /* Create BlockBackend to write to the image */ | |
661 | bs = bdrv_open_blockdev_ref(qed_opts->file, errp); | |
662 | if (bs == NULL) { | |
8a56fdad | 663 | return -EIO; |
75411d23 SH |
664 | } |
665 | ||
a3aeeab5 EB |
666 | blk = blk_new_with_bs(bs, BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL, |
667 | errp); | |
668 | if (!blk) { | |
669 | ret = -EPERM; | |
959355a4 KW |
670 | goto out; |
671 | } | |
8a56fdad KW |
672 | blk_set_allow_write_beyond_eof(blk, true); |
673 | ||
959355a4 KW |
674 | /* Prepare image format */ |
675 | header = (QEDHeader) { | |
676 | .magic = QED_MAGIC, | |
677 | .cluster_size = qed_opts->cluster_size, | |
678 | .table_size = qed_opts->table_size, | |
679 | .header_size = 1, | |
680 | .features = 0, | |
681 | .compat_features = 0, | |
682 | .l1_table_offset = qed_opts->cluster_size, | |
683 | .image_size = qed_opts->size, | |
684 | }; | |
685 | ||
686 | l1_size = header.cluster_size * header.table_size; | |
687 | ||
e8d04f92 HR |
688 | /* |
689 | * The QED format associates file length with allocation status, | |
690 | * so a new file (which is empty) must have a length of 0. | |
691 | */ | |
8c6242b6 | 692 | ret = blk_truncate(blk, 0, true, PREALLOC_MODE_OFF, 0, errp); |
c743849b SH |
693 | if (ret < 0) { |
694 | goto out; | |
695 | } | |
696 | ||
959355a4 | 697 | if (qed_opts->has_backing_file) { |
75411d23 SH |
698 | header.features |= QED_F_BACKING_FILE; |
699 | header.backing_filename_offset = sizeof(le_header); | |
959355a4 | 700 | header.backing_filename_size = strlen(qed_opts->backing_file); |
75411d23 | 701 | |
959355a4 KW |
702 | if (qed_opts->has_backing_fmt) { |
703 | const char *backing_fmt = BlockdevDriver_str(qed_opts->backing_fmt); | |
704 | if (qed_fmt_is_raw(backing_fmt)) { | |
705 | header.features |= QED_F_BACKING_FORMAT_NO_PROBE; | |
706 | } | |
75411d23 SH |
707 | } |
708 | } | |
709 | ||
710 | qed_header_cpu_to_le(&header, &le_header); | |
8341f00d | 711 | ret = blk_pwrite(blk, 0, &le_header, sizeof(le_header), 0); |
75411d23 SH |
712 | if (ret < 0) { |
713 | goto out; | |
714 | } | |
959355a4 | 715 | ret = blk_pwrite(blk, sizeof(le_header), qed_opts->backing_file, |
8341f00d | 716 | header.backing_filename_size, 0); |
75411d23 SH |
717 | if (ret < 0) { |
718 | goto out; | |
719 | } | |
720 | ||
7267c094 | 721 | l1_table = g_malloc0(l1_size); |
8341f00d | 722 | ret = blk_pwrite(blk, header.l1_table_offset, l1_table, l1_size, 0); |
75411d23 SH |
723 | if (ret < 0) { |
724 | goto out; | |
725 | } | |
726 | ||
727 | ret = 0; /* success */ | |
728 | out: | |
7267c094 | 729 | g_free(l1_table); |
8a56fdad | 730 | blk_unref(blk); |
959355a4 | 731 | bdrv_unref(bs); |
75411d23 SH |
732 | return ret; |
733 | } | |
734 | ||
b92902df ML |
735 | static int coroutine_fn bdrv_qed_co_create_opts(BlockDriver *drv, |
736 | const char *filename, | |
efc75e2a SH |
737 | QemuOpts *opts, |
738 | Error **errp) | |
75411d23 | 739 | { |
959355a4 | 740 | BlockdevCreateOptions *create_options = NULL; |
92adf9db | 741 | QDict *qdict; |
959355a4 KW |
742 | Visitor *v; |
743 | BlockDriverState *bs = NULL; | |
7ab74849 CL |
744 | int ret; |
745 | ||
959355a4 KW |
746 | static const QDictRenames opt_renames[] = { |
747 | { BLOCK_OPT_BACKING_FILE, "backing-file" }, | |
748 | { BLOCK_OPT_BACKING_FMT, "backing-fmt" }, | |
749 | { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, | |
750 | { BLOCK_OPT_TABLE_SIZE, "table-size" }, | |
751 | { NULL, NULL }, | |
752 | }; | |
753 | ||
754 | /* Parse options and convert legacy syntax */ | |
755 | qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qed_create_opts, true); | |
756 | ||
757 | if (!qdict_rename_keys(qdict, opt_renames, errp)) { | |
7ab74849 | 758 | ret = -EINVAL; |
959355a4 | 759 | goto fail; |
75411d23 | 760 | } |
959355a4 KW |
761 | |
762 | /* Create and open the file (protocol layer) */ | |
668f62ec | 763 | ret = bdrv_create_file(filename, opts, errp); |
959355a4 | 764 | if (ret < 0) { |
959355a4 KW |
765 | goto fail; |
766 | } | |
767 | ||
768 | bs = bdrv_open(filename, NULL, NULL, | |
769 | BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); | |
770 | if (bs == NULL) { | |
771 | ret = -EIO; | |
772 | goto fail; | |
773 | } | |
774 | ||
775 | /* Now get the QAPI type BlockdevCreateOptions */ | |
776 | qdict_put_str(qdict, "driver", "qed"); | |
777 | qdict_put_str(qdict, "file", bs->node_name); | |
778 | ||
af91062e MA |
779 | v = qobject_input_visitor_new_flat_confused(qdict, errp); |
780 | if (!v) { | |
7ab74849 | 781 | ret = -EINVAL; |
959355a4 | 782 | goto fail; |
75411d23 | 783 | } |
959355a4 | 784 | |
b11a093c | 785 | visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp); |
959355a4 | 786 | visit_free(v); |
b11a093c | 787 | if (!create_options) { |
7ab74849 | 788 | ret = -EINVAL; |
959355a4 | 789 | goto fail; |
75411d23 SH |
790 | } |
791 | ||
959355a4 KW |
792 | /* Silently round up size */ |
793 | assert(create_options->driver == BLOCKDEV_DRIVER_QED); | |
794 | create_options->u.qed.size = | |
795 | ROUND_UP(create_options->u.qed.size, BDRV_SECTOR_SIZE); | |
796 | ||
797 | /* Create the qed image (format layer) */ | |
798 | ret = bdrv_qed_co_create(create_options, errp); | |
7ab74849 | 799 | |
959355a4 | 800 | fail: |
cb3e7f08 | 801 | qobject_unref(qdict); |
959355a4 KW |
802 | bdrv_unref(bs); |
803 | qapi_free_BlockdevCreateOptions(create_options); | |
7ab74849 | 804 | return ret; |
75411d23 SH |
805 | } |
806 | ||
b8d739fd EB |
807 | static int coroutine_fn bdrv_qed_co_block_status(BlockDriverState *bs, |
808 | bool want_zero, | |
809 | int64_t pos, int64_t bytes, | |
810 | int64_t *pnum, int64_t *map, | |
811 | BlockDriverState **file) | |
298800ca | 812 | { |
b8d739fd EB |
813 | BDRVQEDState *s = bs->opaque; |
814 | size_t len = MIN(bytes, SIZE_MAX); | |
815 | int status; | |
816 | QEDRequest request = { .l2_table = NULL }; | |
817 | uint64_t offset; | |
818 | int ret; | |
819 | ||
820 | qemu_co_mutex_lock(&s->table_lock); | |
821 | ret = qed_find_cluster(s, &request, pos, &len, &offset); | |
822 | ||
823 | *pnum = len; | |
4bc74be9 PB |
824 | switch (ret) { |
825 | case QED_CLUSTER_FOUND: | |
b8d739fd EB |
826 | *map = offset | qed_offset_into_cluster(s, pos); |
827 | status = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
828 | *file = bs->file->bs; | |
4bc74be9 PB |
829 | break; |
830 | case QED_CLUSTER_ZERO: | |
b8d739fd | 831 | status = BDRV_BLOCK_ZERO; |
4bc74be9 PB |
832 | break; |
833 | case QED_CLUSTER_L2: | |
834 | case QED_CLUSTER_L1: | |
b8d739fd | 835 | status = 0; |
4bc74be9 PB |
836 | break; |
837 | default: | |
838 | assert(ret < 0); | |
b8d739fd | 839 | status = ret; |
4bc74be9 PB |
840 | break; |
841 | } | |
842 | ||
298800ca | 843 | qed_unref_l2_cache_entry(request.l2_table); |
1f01e50b | 844 | qemu_co_mutex_unlock(&s->table_lock); |
298800ca | 845 | |
b8d739fd | 846 | return status; |
75411d23 SH |
847 | } |
848 | ||
eabba580 SH |
849 | static BDRVQEDState *acb_to_s(QEDAIOCB *acb) |
850 | { | |
48cc565e | 851 | return acb->bs->opaque; |
eabba580 SH |
852 | } |
853 | ||
854 | /** | |
855 | * Read from the backing file or zero-fill if no backing file | |
856 | * | |
f06ee3d4 KW |
857 | * @s: QED state |
858 | * @pos: Byte position in device | |
859 | * @qiov: Destination I/O vector | |
eabba580 SH |
860 | * |
861 | * This function reads qiov->size bytes starting at pos from the backing file. | |
862 | * If there is no backing file then zeroes are read. | |
863 | */ | |
87f0d882 | 864 | static int coroutine_fn qed_read_backing_file(BDRVQEDState *s, uint64_t pos, |
365fed51 | 865 | QEMUIOVector *qiov) |
eabba580 | 866 | { |
760e0063 | 867 | if (s->bs->backing) { |
365fed51 EB |
868 | BLKDBG_EVENT(s->bs->file, BLKDBG_READ_BACKING_AIO); |
869 | return bdrv_co_preadv(s->bs->backing, pos, qiov->size, qiov, 0); | |
e85c5281 | 870 | } |
365fed51 | 871 | qemu_iovec_memset(qiov, 0, 0, qiov->size); |
e85c5281 | 872 | return 0; |
eabba580 SH |
873 | } |
874 | ||
eabba580 SH |
875 | /** |
876 | * Copy data from backing file into the image | |
877 | * | |
878 | * @s: QED state | |
879 | * @pos: Byte position in device | |
880 | * @len: Number of bytes | |
881 | * @offset: Byte offset in image file | |
eabba580 | 882 | */ |
87f0d882 KW |
883 | static int coroutine_fn qed_copy_from_backing_file(BDRVQEDState *s, |
884 | uint64_t pos, uint64_t len, | |
885 | uint64_t offset) | |
eabba580 | 886 | { |
0f7aa24d | 887 | QEMUIOVector qiov; |
e85c5281 | 888 | int ret; |
eabba580 SH |
889 | |
890 | /* Skip copy entirely if there is no work to do */ | |
891 | if (len == 0) { | |
b4ac32f3 | 892 | return 0; |
eabba580 SH |
893 | } |
894 | ||
342544f9 | 895 | qemu_iovec_init_buf(&qiov, qemu_blockalign(s->bs, len), len); |
0f7aa24d | 896 | |
365fed51 | 897 | ret = qed_read_backing_file(s, pos, &qiov); |
0f7aa24d KW |
898 | |
899 | if (ret) { | |
900 | goto out; | |
901 | } | |
eabba580 | 902 | |
0f7aa24d | 903 | BLKDBG_EVENT(s->bs->file, BLKDBG_COW_WRITE); |
0f714ec7 | 904 | ret = bdrv_co_pwritev(s->bs->file, offset, qiov.size, &qiov, 0); |
0f7aa24d KW |
905 | if (ret < 0) { |
906 | goto out; | |
907 | } | |
908 | ret = 0; | |
909 | out: | |
342544f9 | 910 | qemu_vfree(qemu_iovec_buf(&qiov)); |
b4ac32f3 | 911 | return ret; |
eabba580 SH |
912 | } |
913 | ||
914 | /** | |
915 | * Link one or more contiguous clusters into a table | |
916 | * | |
917 | * @s: QED state | |
918 | * @table: L2 table | |
919 | * @index: First cluster index | |
920 | * @n: Number of contiguous clusters | |
21df65b6 AL |
921 | * @cluster: First cluster offset |
922 | * | |
923 | * The cluster offset may be an allocated byte offset in the image file, the | |
924 | * zero cluster marker, or the unallocated cluster marker. | |
1f01e50b PB |
925 | * |
926 | * Called with table_lock held. | |
eabba580 | 927 | */ |
87f0d882 KW |
928 | static void coroutine_fn qed_update_l2_table(BDRVQEDState *s, QEDTable *table, |
929 | int index, unsigned int n, | |
930 | uint64_t cluster) | |
eabba580 SH |
931 | { |
932 | int i; | |
933 | for (i = index; i < index + n; i++) { | |
934 | table->offsets[i] = cluster; | |
21df65b6 AL |
935 | if (!qed_offset_is_unalloc_cluster(cluster) && |
936 | !qed_offset_is_zero_cluster(cluster)) { | |
937 | cluster += s->header.cluster_size; | |
938 | } | |
eabba580 SH |
939 | } |
940 | } | |
941 | ||
1f01e50b | 942 | /* Called with table_lock held. */ |
87f0d882 | 943 | static void coroutine_fn qed_aio_complete(QEDAIOCB *acb) |
eabba580 | 944 | { |
1919631e | 945 | BDRVQEDState *s = acb_to_s(acb); |
eabba580 SH |
946 | |
947 | /* Free resources */ | |
948 | qemu_iovec_destroy(&acb->cur_qiov); | |
949 | qed_unref_l2_cache_entry(acb->request.l2_table); | |
950 | ||
0e71be19 SH |
951 | /* Free the buffer we may have allocated for zero writes */ |
952 | if (acb->flags & QED_AIOCB_ZERO) { | |
953 | qemu_vfree(acb->qiov->iov[0].iov_base); | |
954 | acb->qiov->iov[0].iov_base = NULL; | |
955 | } | |
956 | ||
eabba580 SH |
957 | /* Start next allocating write request waiting behind this one. Note that |
958 | * requests enqueue themselves when they first hit an unallocated cluster | |
959 | * but they wait until the entire request is finished before waking up the | |
960 | * next request in the queue. This ensures that we don't cycle through | |
961 | * requests multiple times but rather finish one at a time completely. | |
962 | */ | |
0806c3b5 KW |
963 | if (acb == s->allocating_acb) { |
964 | s->allocating_acb = NULL; | |
965 | if (!qemu_co_queue_empty(&s->allocating_write_reqs)) { | |
1f01e50b | 966 | qemu_co_queue_next(&s->allocating_write_reqs); |
6f321e93 SH |
967 | } else if (s->header.features & QED_F_NEED_CHECK) { |
968 | qed_start_need_check_timer(s); | |
eabba580 SH |
969 | } |
970 | } | |
971 | } | |
972 | ||
973 | /** | |
fae25ac7 | 974 | * Update L1 table with new L2 table offset and write it out |
1f01e50b PB |
975 | * |
976 | * Called with table_lock held. | |
eabba580 | 977 | */ |
87f0d882 | 978 | static int coroutine_fn qed_aio_write_l1_update(QEDAIOCB *acb) |
eabba580 | 979 | { |
eabba580 SH |
980 | BDRVQEDState *s = acb_to_s(acb); |
981 | CachedL2Table *l2_table = acb->request.l2_table; | |
e4fc8781 | 982 | uint64_t l2_offset = l2_table->offset; |
fb18de21 | 983 | int index, ret; |
eabba580 | 984 | |
fae25ac7 KW |
985 | index = qed_l1_index(s, acb->cur_pos); |
986 | s->l1_table->offsets[index] = l2_table->offset; | |
987 | ||
988 | ret = qed_write_l1_table(s, index, 1); | |
989 | ||
990 | /* Commit the current L2 table to the cache */ | |
eabba580 SH |
991 | qed_commit_l2_cache_entry(&s->l2_cache, l2_table); |
992 | ||
993 | /* This is guaranteed to succeed because we just committed the entry to the | |
994 | * cache. | |
995 | */ | |
e4fc8781 | 996 | acb->request.l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset); |
eabba580 SH |
997 | assert(acb->request.l2_table != NULL); |
998 | ||
fb18de21 | 999 | return ret; |
eabba580 SH |
1000 | } |
1001 | ||
eabba580 SH |
1002 | |
1003 | /** | |
1004 | * Update L2 table with new cluster offsets and write them out | |
1f01e50b PB |
1005 | * |
1006 | * Called with table_lock held. | |
eabba580 | 1007 | */ |
87f0d882 | 1008 | static int coroutine_fn qed_aio_write_l2_update(QEDAIOCB *acb, uint64_t offset) |
eabba580 | 1009 | { |
eabba580 SH |
1010 | BDRVQEDState *s = acb_to_s(acb); |
1011 | bool need_alloc = acb->find_cluster_ret == QED_CLUSTER_L1; | |
88d2dd72 | 1012 | int index, ret; |
eabba580 SH |
1013 | |
1014 | if (need_alloc) { | |
1015 | qed_unref_l2_cache_entry(acb->request.l2_table); | |
1016 | acb->request.l2_table = qed_new_l2_table(s); | |
1017 | } | |
1018 | ||
1019 | index = qed_l2_index(s, acb->cur_pos); | |
1020 | qed_update_l2_table(s, acb->request.l2_table->table, index, acb->cur_nclusters, | |
0e71be19 | 1021 | offset); |
eabba580 SH |
1022 | |
1023 | if (need_alloc) { | |
1024 | /* Write out the whole new L2 table */ | |
453e53e2 | 1025 | ret = qed_write_l2_table(s, &acb->request, 0, s->table_nelems, true); |
fb18de21 | 1026 | if (ret) { |
88d2dd72 | 1027 | return ret; |
fb18de21 | 1028 | } |
88d2dd72 | 1029 | return qed_aio_write_l1_update(acb); |
eabba580 SH |
1030 | } else { |
1031 | /* Write out only the updated part of the L2 table */ | |
453e53e2 KW |
1032 | ret = qed_write_l2_table(s, &acb->request, index, acb->cur_nclusters, |
1033 | false); | |
88d2dd72 KW |
1034 | if (ret) { |
1035 | return ret; | |
1036 | } | |
eabba580 | 1037 | } |
88d2dd72 | 1038 | return 0; |
eabba580 SH |
1039 | } |
1040 | ||
eabba580 SH |
1041 | /** |
1042 | * Write data to the image file | |
1f01e50b PB |
1043 | * |
1044 | * Called with table_lock *not* held. | |
eabba580 | 1045 | */ |
87f0d882 | 1046 | static int coroutine_fn qed_aio_write_main(QEDAIOCB *acb) |
eabba580 | 1047 | { |
eabba580 SH |
1048 | BDRVQEDState *s = acb_to_s(acb); |
1049 | uint64_t offset = acb->cur_cluster + | |
1050 | qed_offset_into_cluster(s, acb->cur_pos); | |
eabba580 | 1051 | |
eaf0bc56 | 1052 | trace_qed_aio_write_main(s, acb, 0, offset, acb->cur_qiov.size); |
eabba580 | 1053 | |
a4d8f1ae | 1054 | BLKDBG_EVENT(s->bs->file, BLKDBG_WRITE_AIO); |
e7569c18 PB |
1055 | return bdrv_co_pwritev(s->bs->file, offset, acb->cur_qiov.size, |
1056 | &acb->cur_qiov, 0); | |
eabba580 SH |
1057 | } |
1058 | ||
1059 | /** | |
b4ac32f3 | 1060 | * Populate untouched regions of new data cluster |
1f01e50b PB |
1061 | * |
1062 | * Called with table_lock held. | |
eabba580 | 1063 | */ |
87f0d882 | 1064 | static int coroutine_fn qed_aio_write_cow(QEDAIOCB *acb) |
eabba580 | 1065 | { |
eabba580 | 1066 | BDRVQEDState *s = acb_to_s(acb); |
b4ac32f3 | 1067 | uint64_t start, len, offset; |
a101341a | 1068 | int ret; |
eabba580 | 1069 | |
1f01e50b PB |
1070 | qemu_co_mutex_unlock(&s->table_lock); |
1071 | ||
b4ac32f3 KW |
1072 | /* Populate front untouched region of new data cluster */ |
1073 | start = qed_start_of_cluster(s, acb->cur_pos); | |
1074 | len = qed_offset_into_cluster(s, acb->cur_pos); | |
1075 | ||
1076 | trace_qed_aio_write_prefill(s, acb, start, len, acb->cur_cluster); | |
1077 | ret = qed_copy_from_backing_file(s, start, len, acb->cur_cluster); | |
a101341a | 1078 | if (ret < 0) { |
1f01e50b | 1079 | goto out; |
eabba580 SH |
1080 | } |
1081 | ||
b4ac32f3 KW |
1082 | /* Populate back untouched region of new data cluster */ |
1083 | start = acb->cur_pos + acb->cur_qiov.size; | |
1084 | len = qed_start_of_cluster(s, start + s->header.cluster_size - 1) - start; | |
1085 | offset = acb->cur_cluster + | |
1086 | qed_offset_into_cluster(s, acb->cur_pos) + | |
1087 | acb->cur_qiov.size; | |
eabba580 | 1088 | |
b4ac32f3 KW |
1089 | trace_qed_aio_write_postfill(s, acb, start, len, offset); |
1090 | ret = qed_copy_from_backing_file(s, start, len, offset); | |
eaf0bc56 | 1091 | if (ret < 0) { |
1f01e50b | 1092 | goto out; |
eaf0bc56 | 1093 | } |
a101341a | 1094 | |
e7569c18 PB |
1095 | ret = qed_aio_write_main(acb); |
1096 | if (ret < 0) { | |
1f01e50b | 1097 | goto out; |
e7569c18 PB |
1098 | } |
1099 | ||
1100 | if (s->bs->backing) { | |
1101 | /* | |
1102 | * Flush new data clusters before updating the L2 table | |
1103 | * | |
1104 | * This flush is necessary when a backing file is in use. A crash | |
1105 | * during an allocating write could result in empty clusters in the | |
1106 | * image. If the write only touched a subregion of the cluster, | |
1107 | * then backing image sectors have been lost in the untouched | |
1108 | * region. The solution is to flush after writing a new data | |
1109 | * cluster and before updating the L2 table. | |
1110 | */ | |
1111 | ret = bdrv_co_flush(s->bs->file->bs); | |
e7569c18 PB |
1112 | } |
1113 | ||
1f01e50b PB |
1114 | out: |
1115 | qemu_co_mutex_lock(&s->table_lock); | |
1116 | return ret; | |
eabba580 SH |
1117 | } |
1118 | ||
0d09c797 SH |
1119 | /** |
1120 | * Check if the QED_F_NEED_CHECK bit should be set during allocating write | |
1121 | */ | |
1122 | static bool qed_should_set_need_check(BDRVQEDState *s) | |
1123 | { | |
1124 | /* The flush before L2 update path ensures consistency */ | |
760e0063 | 1125 | if (s->bs->backing) { |
0d09c797 SH |
1126 | return false; |
1127 | } | |
1128 | ||
1129 | return !(s->header.features & QED_F_NEED_CHECK); | |
1130 | } | |
1131 | ||
eabba580 SH |
1132 | /** |
1133 | * Write new data cluster | |
1134 | * | |
1135 | * @acb: Write request | |
1136 | * @len: Length in bytes | |
1137 | * | |
1138 | * This path is taken when writing to previously unallocated clusters. | |
1f01e50b PB |
1139 | * |
1140 | * Called with table_lock held. | |
eabba580 | 1141 | */ |
87f0d882 | 1142 | static int coroutine_fn qed_aio_write_alloc(QEDAIOCB *acb, size_t len) |
eabba580 SH |
1143 | { |
1144 | BDRVQEDState *s = acb_to_s(acb); | |
f13d712b | 1145 | int ret; |
eabba580 | 1146 | |
6f321e93 | 1147 | /* Cancel timer when the first allocating request comes in */ |
0806c3b5 | 1148 | if (s->allocating_acb == NULL) { |
6f321e93 SH |
1149 | qed_cancel_need_check_timer(s); |
1150 | } | |
1151 | ||
eabba580 | 1152 | /* Freeze this request if another allocating write is in progress */ |
0806c3b5 KW |
1153 | if (s->allocating_acb != acb || s->allocating_write_reqs_plugged) { |
1154 | if (s->allocating_acb != NULL) { | |
1f01e50b | 1155 | qemu_co_queue_wait(&s->allocating_write_reqs, &s->table_lock); |
0806c3b5 KW |
1156 | assert(s->allocating_acb == NULL); |
1157 | } | |
1158 | s->allocating_acb = acb; | |
1159 | return -EAGAIN; /* start over with looking up table entries */ | |
eabba580 SH |
1160 | } |
1161 | ||
1162 | acb->cur_nclusters = qed_bytes_to_clusters(s, | |
1163 | qed_offset_into_cluster(s, acb->cur_pos) + len); | |
1b093c48 | 1164 | qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
eabba580 | 1165 | |
0e71be19 SH |
1166 | if (acb->flags & QED_AIOCB_ZERO) { |
1167 | /* Skip ahead if the clusters are already zero */ | |
1168 | if (acb->find_cluster_ret == QED_CLUSTER_ZERO) { | |
d6daddcd | 1169 | return 0; |
0e71be19 | 1170 | } |
e7569c18 | 1171 | acb->cur_cluster = 1; |
0e71be19 | 1172 | } else { |
0e71be19 SH |
1173 | acb->cur_cluster = qed_alloc_clusters(s, acb->cur_nclusters); |
1174 | } | |
1175 | ||
0d09c797 SH |
1176 | if (qed_should_set_need_check(s)) { |
1177 | s->header.features |= QED_F_NEED_CHECK; | |
f13d712b | 1178 | ret = qed_write_header(s); |
a101341a | 1179 | if (ret < 0) { |
d6daddcd | 1180 | return ret; |
a101341a KW |
1181 | } |
1182 | } | |
1183 | ||
e7569c18 | 1184 | if (!(acb->flags & QED_AIOCB_ZERO)) { |
a101341a | 1185 | ret = qed_aio_write_cow(acb); |
e7569c18 PB |
1186 | if (ret < 0) { |
1187 | return ret; | |
1188 | } | |
01979a98 | 1189 | } |
e7569c18 PB |
1190 | |
1191 | return qed_aio_write_l2_update(acb, acb->cur_cluster); | |
eabba580 SH |
1192 | } |
1193 | ||
1194 | /** | |
1195 | * Write data cluster in place | |
1196 | * | |
1197 | * @acb: Write request | |
1198 | * @offset: Cluster offset in bytes | |
1199 | * @len: Length in bytes | |
1200 | * | |
1201 | * This path is taken when writing to already allocated clusters. | |
1f01e50b PB |
1202 | * |
1203 | * Called with table_lock held. | |
eabba580 | 1204 | */ |
87f0d882 KW |
1205 | static int coroutine_fn qed_aio_write_inplace(QEDAIOCB *acb, uint64_t offset, |
1206 | size_t len) | |
eabba580 | 1207 | { |
1f01e50b PB |
1208 | BDRVQEDState *s = acb_to_s(acb); |
1209 | int r; | |
1210 | ||
1211 | qemu_co_mutex_unlock(&s->table_lock); | |
1212 | ||
0e71be19 SH |
1213 | /* Allocate buffer for zero writes */ |
1214 | if (acb->flags & QED_AIOCB_ZERO) { | |
1215 | struct iovec *iov = acb->qiov->iov; | |
1216 | ||
1217 | if (!iov->iov_base) { | |
48cc565e | 1218 | iov->iov_base = qemu_try_blockalign(acb->bs, iov->iov_len); |
4f4896db | 1219 | if (iov->iov_base == NULL) { |
1f01e50b PB |
1220 | r = -ENOMEM; |
1221 | goto out; | |
4f4896db | 1222 | } |
0e71be19 SH |
1223 | memset(iov->iov_base, 0, iov->iov_len); |
1224 | } | |
1225 | } | |
1226 | ||
eabba580 SH |
1227 | /* Calculate the I/O vector */ |
1228 | acb->cur_cluster = offset; | |
1b093c48 | 1229 | qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
eabba580 | 1230 | |
1f01e50b PB |
1231 | /* Do the actual write. */ |
1232 | r = qed_aio_write_main(acb); | |
1233 | out: | |
1234 | qemu_co_mutex_lock(&s->table_lock); | |
1235 | return r; | |
eabba580 SH |
1236 | } |
1237 | ||
1238 | /** | |
1239 | * Write data cluster | |
1240 | * | |
1241 | * @opaque: Write request | |
0596be7e | 1242 | * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2 or QED_CLUSTER_L1 |
eabba580 SH |
1243 | * @offset: Cluster offset in bytes |
1244 | * @len: Length in bytes | |
1f01e50b PB |
1245 | * |
1246 | * Called with table_lock held. | |
eabba580 | 1247 | */ |
87f0d882 KW |
1248 | static int coroutine_fn qed_aio_write_data(void *opaque, int ret, |
1249 | uint64_t offset, size_t len) | |
eabba580 SH |
1250 | { |
1251 | QEDAIOCB *acb = opaque; | |
1252 | ||
1253 | trace_qed_aio_write_data(acb_to_s(acb), acb, ret, offset, len); | |
1254 | ||
1255 | acb->find_cluster_ret = ret; | |
1256 | ||
1257 | switch (ret) { | |
1258 | case QED_CLUSTER_FOUND: | |
0596be7e | 1259 | return qed_aio_write_inplace(acb, offset, len); |
eabba580 SH |
1260 | |
1261 | case QED_CLUSTER_L2: | |
1262 | case QED_CLUSTER_L1: | |
21df65b6 | 1263 | case QED_CLUSTER_ZERO: |
0596be7e | 1264 | return qed_aio_write_alloc(acb, len); |
eabba580 SH |
1265 | |
1266 | default: | |
0596be7e | 1267 | g_assert_not_reached(); |
d6daddcd | 1268 | } |
eabba580 SH |
1269 | } |
1270 | ||
1271 | /** | |
1272 | * Read data cluster | |
1273 | * | |
1274 | * @opaque: Read request | |
0596be7e | 1275 | * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2 or QED_CLUSTER_L1 |
eabba580 SH |
1276 | * @offset: Cluster offset in bytes |
1277 | * @len: Length in bytes | |
1f01e50b PB |
1278 | * |
1279 | * Called with table_lock held. | |
eabba580 | 1280 | */ |
87f0d882 KW |
1281 | static int coroutine_fn qed_aio_read_data(void *opaque, int ret, |
1282 | uint64_t offset, size_t len) | |
eabba580 SH |
1283 | { |
1284 | QEDAIOCB *acb = opaque; | |
1285 | BDRVQEDState *s = acb_to_s(acb); | |
48cc565e | 1286 | BlockDriverState *bs = acb->bs; |
1f01e50b PB |
1287 | int r; |
1288 | ||
1289 | qemu_co_mutex_unlock(&s->table_lock); | |
eabba580 SH |
1290 | |
1291 | /* Adjust offset into cluster */ | |
1292 | offset += qed_offset_into_cluster(s, acb->cur_pos); | |
1293 | ||
1294 | trace_qed_aio_read_data(s, acb, ret, offset, len); | |
1295 | ||
1b093c48 | 1296 | qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
eabba580 | 1297 | |
1f01e50b PB |
1298 | /* Handle zero cluster and backing file reads, otherwise read |
1299 | * data cluster directly. | |
1300 | */ | |
21df65b6 | 1301 | if (ret == QED_CLUSTER_ZERO) { |
3d9b4925 | 1302 | qemu_iovec_memset(&acb->cur_qiov, 0, 0, acb->cur_qiov.size); |
1f01e50b | 1303 | r = 0; |
21df65b6 | 1304 | } else if (ret != QED_CLUSTER_FOUND) { |
365fed51 | 1305 | r = qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov); |
1f01e50b PB |
1306 | } else { |
1307 | BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); | |
1308 | r = bdrv_co_preadv(bs->file, offset, acb->cur_qiov.size, | |
1309 | &acb->cur_qiov, 0); | |
eabba580 SH |
1310 | } |
1311 | ||
1f01e50b PB |
1312 | qemu_co_mutex_lock(&s->table_lock); |
1313 | return r; | |
eabba580 SH |
1314 | } |
1315 | ||
1316 | /** | |
1317 | * Begin next I/O or complete the request | |
1318 | */ | |
87f0d882 | 1319 | static int coroutine_fn qed_aio_next_io(QEDAIOCB *acb) |
eabba580 | 1320 | { |
eabba580 | 1321 | BDRVQEDState *s = acb_to_s(acb); |
0f21b7a1 KW |
1322 | uint64_t offset; |
1323 | size_t len; | |
dddf8db1 | 1324 | int ret; |
eabba580 | 1325 | |
1f01e50b | 1326 | qemu_co_mutex_lock(&s->table_lock); |
01859874 KW |
1327 | while (1) { |
1328 | trace_qed_aio_next_io(s, acb, 0, acb->cur_pos + acb->cur_qiov.size); | |
eabba580 | 1329 | |
01859874 KW |
1330 | acb->qiov_offset += acb->cur_qiov.size; |
1331 | acb->cur_pos += acb->cur_qiov.size; | |
1332 | qemu_iovec_reset(&acb->cur_qiov); | |
eabba580 | 1333 | |
01859874 KW |
1334 | /* Complete request */ |
1335 | if (acb->cur_pos >= acb->end_pos) { | |
48cc565e KW |
1336 | ret = 0; |
1337 | break; | |
01859874 | 1338 | } |
eabba580 | 1339 | |
01859874 KW |
1340 | /* Find next cluster and start I/O */ |
1341 | len = acb->end_pos - acb->cur_pos; | |
1342 | ret = qed_find_cluster(s, &acb->request, acb->cur_pos, &len, &offset); | |
1343 | if (ret < 0) { | |
48cc565e | 1344 | break; |
01859874 | 1345 | } |
0596be7e | 1346 | |
01859874 KW |
1347 | if (acb->flags & QED_AIOCB_WRITE) { |
1348 | ret = qed_aio_write_data(acb, ret, offset, len); | |
1349 | } else { | |
1350 | ret = qed_aio_read_data(acb, ret, offset, len); | |
1351 | } | |
0596be7e | 1352 | |
0806c3b5 | 1353 | if (ret < 0 && ret != -EAGAIN) { |
48cc565e | 1354 | break; |
0596be7e | 1355 | } |
0596be7e | 1356 | } |
eabba580 | 1357 | |
48cc565e KW |
1358 | trace_qed_aio_complete(s, acb, ret); |
1359 | qed_aio_complete(acb); | |
1f01e50b | 1360 | qemu_co_mutex_unlock(&s->table_lock); |
48cc565e | 1361 | return ret; |
89f89709 KW |
1362 | } |
1363 | ||
1364 | static int coroutine_fn qed_co_request(BlockDriverState *bs, int64_t sector_num, | |
1365 | QEMUIOVector *qiov, int nb_sectors, | |
1366 | int flags) | |
1367 | { | |
48cc565e KW |
1368 | QEDAIOCB acb = { |
1369 | .bs = bs, | |
1370 | .cur_pos = (uint64_t) sector_num * BDRV_SECTOR_SIZE, | |
1371 | .end_pos = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE, | |
1372 | .qiov = qiov, | |
1373 | .flags = flags, | |
89f89709 | 1374 | }; |
48cc565e | 1375 | qemu_iovec_init(&acb.cur_qiov, qiov->niov); |
eabba580 | 1376 | |
48cc565e | 1377 | trace_qed_aio_setup(bs->opaque, &acb, sector_num, nb_sectors, NULL, flags); |
eabba580 SH |
1378 | |
1379 | /* Start request */ | |
48cc565e | 1380 | return qed_aio_next_io(&acb); |
75411d23 SH |
1381 | } |
1382 | ||
89f89709 KW |
1383 | static int coroutine_fn bdrv_qed_co_readv(BlockDriverState *bs, |
1384 | int64_t sector_num, int nb_sectors, | |
1385 | QEMUIOVector *qiov) | |
75411d23 | 1386 | { |
89f89709 | 1387 | return qed_co_request(bs, sector_num, qiov, nb_sectors, 0); |
75411d23 SH |
1388 | } |
1389 | ||
89f89709 KW |
1390 | static int coroutine_fn bdrv_qed_co_writev(BlockDriverState *bs, |
1391 | int64_t sector_num, int nb_sectors, | |
e18a58b4 | 1392 | QEMUIOVector *qiov, int flags) |
0e71be19 | 1393 | { |
e18a58b4 | 1394 | assert(!flags); |
89f89709 | 1395 | return qed_co_request(bs, sector_num, qiov, nb_sectors, QED_AIOCB_WRITE); |
0e71be19 SH |
1396 | } |
1397 | ||
49a2e483 EB |
1398 | static int coroutine_fn bdrv_qed_co_pwrite_zeroes(BlockDriverState *bs, |
1399 | int64_t offset, | |
f5a5ca79 | 1400 | int bytes, |
49a2e483 | 1401 | BdrvRequestFlags flags) |
0e71be19 | 1402 | { |
ef72f76e | 1403 | BDRVQEDState *s = bs->opaque; |
342544f9 VSO |
1404 | |
1405 | /* | |
1406 | * Zero writes start without an I/O buffer. If a buffer becomes necessary | |
1407 | * then it will be allocated during request processing. | |
1408 | */ | |
1409 | QEMUIOVector qiov = QEMU_IOVEC_INIT_BUF(qiov, NULL, bytes); | |
0e71be19 | 1410 | |
49a2e483 EB |
1411 | /* Fall back if the request is not aligned */ |
1412 | if (qed_offset_into_cluster(s, offset) || | |
f5a5ca79 | 1413 | qed_offset_into_cluster(s, bytes)) { |
49a2e483 | 1414 | return -ENOTSUP; |
ef72f76e SH |
1415 | } |
1416 | ||
89f89709 | 1417 | return qed_co_request(bs, offset >> BDRV_SECTOR_BITS, &qiov, |
f5a5ca79 | 1418 | bytes >> BDRV_SECTOR_BITS, |
89f89709 | 1419 | QED_AIOCB_WRITE | QED_AIOCB_ZERO); |
0e71be19 SH |
1420 | } |
1421 | ||
061ca8a3 KW |
1422 | static int coroutine_fn bdrv_qed_co_truncate(BlockDriverState *bs, |
1423 | int64_t offset, | |
c80d8b06 | 1424 | bool exact, |
061ca8a3 | 1425 | PreallocMode prealloc, |
92b92799 | 1426 | BdrvRequestFlags flags, |
061ca8a3 | 1427 | Error **errp) |
75411d23 | 1428 | { |
77a5a000 SH |
1429 | BDRVQEDState *s = bs->opaque; |
1430 | uint64_t old_image_size; | |
1431 | int ret; | |
1432 | ||
8243ccb7 HR |
1433 | if (prealloc != PREALLOC_MODE_OFF) { |
1434 | error_setg(errp, "Unsupported preallocation mode '%s'", | |
977c736f | 1435 | PreallocMode_str(prealloc)); |
8243ccb7 HR |
1436 | return -ENOTSUP; |
1437 | } | |
1438 | ||
77a5a000 SH |
1439 | if (!qed_is_image_size_valid(offset, s->header.cluster_size, |
1440 | s->header.table_size)) { | |
f59adb32 | 1441 | error_setg(errp, "Invalid image size specified"); |
77a5a000 SH |
1442 | return -EINVAL; |
1443 | } | |
1444 | ||
77a5a000 | 1445 | if ((uint64_t)offset < s->header.image_size) { |
f59adb32 | 1446 | error_setg(errp, "Shrinking images is currently not supported"); |
77a5a000 SH |
1447 | return -ENOTSUP; |
1448 | } | |
1449 | ||
1450 | old_image_size = s->header.image_size; | |
1451 | s->header.image_size = offset; | |
1452 | ret = qed_write_header_sync(s); | |
1453 | if (ret < 0) { | |
1454 | s->header.image_size = old_image_size; | |
f59adb32 | 1455 | error_setg_errno(errp, -ret, "Failed to update the image size"); |
77a5a000 SH |
1456 | } |
1457 | return ret; | |
75411d23 SH |
1458 | } |
1459 | ||
1460 | static int64_t bdrv_qed_getlength(BlockDriverState *bs) | |
1461 | { | |
1462 | BDRVQEDState *s = bs->opaque; | |
1463 | return s->header.image_size; | |
1464 | } | |
1465 | ||
1466 | static int bdrv_qed_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
1467 | { | |
1468 | BDRVQEDState *s = bs->opaque; | |
1469 | ||
1470 | memset(bdi, 0, sizeof(*bdi)); | |
1471 | bdi->cluster_size = s->header.cluster_size; | |
d68dbee8 | 1472 | bdi->is_dirty = s->header.features & QED_F_NEED_CHECK; |
75411d23 SH |
1473 | return 0; |
1474 | } | |
1475 | ||
1476 | static int bdrv_qed_change_backing_file(BlockDriverState *bs, | |
1477 | const char *backing_file, | |
1478 | const char *backing_fmt) | |
1479 | { | |
1480 | BDRVQEDState *s = bs->opaque; | |
1481 | QEDHeader new_header, le_header; | |
1482 | void *buffer; | |
1483 | size_t buffer_len, backing_file_len; | |
1484 | int ret; | |
1485 | ||
1486 | /* Refuse to set backing filename if unknown compat feature bits are | |
1487 | * active. If the image uses an unknown compat feature then we may not | |
1488 | * know the layout of data following the header structure and cannot safely | |
1489 | * add a new string. | |
1490 | */ | |
1491 | if (backing_file && (s->header.compat_features & | |
1492 | ~QED_COMPAT_FEATURE_MASK)) { | |
1493 | return -ENOTSUP; | |
1494 | } | |
1495 | ||
1496 | memcpy(&new_header, &s->header, sizeof(new_header)); | |
1497 | ||
1498 | new_header.features &= ~(QED_F_BACKING_FILE | | |
1499 | QED_F_BACKING_FORMAT_NO_PROBE); | |
1500 | ||
1501 | /* Adjust feature flags */ | |
1502 | if (backing_file) { | |
1503 | new_header.features |= QED_F_BACKING_FILE; | |
1504 | ||
1505 | if (qed_fmt_is_raw(backing_fmt)) { | |
1506 | new_header.features |= QED_F_BACKING_FORMAT_NO_PROBE; | |
1507 | } | |
1508 | } | |
1509 | ||
1510 | /* Calculate new header size */ | |
1511 | backing_file_len = 0; | |
1512 | ||
1513 | if (backing_file) { | |
1514 | backing_file_len = strlen(backing_file); | |
1515 | } | |
1516 | ||
1517 | buffer_len = sizeof(new_header); | |
1518 | new_header.backing_filename_offset = buffer_len; | |
1519 | new_header.backing_filename_size = backing_file_len; | |
1520 | buffer_len += backing_file_len; | |
1521 | ||
1522 | /* Make sure we can rewrite header without failing */ | |
1523 | if (buffer_len > new_header.header_size * new_header.cluster_size) { | |
1524 | return -ENOSPC; | |
1525 | } | |
1526 | ||
1527 | /* Prepare new header */ | |
7267c094 | 1528 | buffer = g_malloc(buffer_len); |
75411d23 SH |
1529 | |
1530 | qed_header_cpu_to_le(&new_header, &le_header); | |
1531 | memcpy(buffer, &le_header, sizeof(le_header)); | |
1532 | buffer_len = sizeof(le_header); | |
1533 | ||
feba23b1 PB |
1534 | if (backing_file) { |
1535 | memcpy(buffer + buffer_len, backing_file, backing_file_len); | |
1536 | buffer_len += backing_file_len; | |
1537 | } | |
75411d23 SH |
1538 | |
1539 | /* Write new header */ | |
d9ca2ea2 | 1540 | ret = bdrv_pwrite_sync(bs->file, 0, buffer, buffer_len); |
7267c094 | 1541 | g_free(buffer); |
75411d23 SH |
1542 | if (ret == 0) { |
1543 | memcpy(&s->header, &new_header, sizeof(new_header)); | |
1544 | } | |
1545 | return ret; | |
1546 | } | |
1547 | ||
2b148f39 PB |
1548 | static void coroutine_fn bdrv_qed_co_invalidate_cache(BlockDriverState *bs, |
1549 | Error **errp) | |
c82954e5 | 1550 | { |
1f01e50b | 1551 | BDRVQEDState *s = bs->opaque; |
5a8a30db | 1552 | int ret; |
c82954e5 BC |
1553 | |
1554 | bdrv_qed_close(bs); | |
3456a8d1 | 1555 | |
61c7887e | 1556 | bdrv_qed_init_state(bs); |
2b148f39 | 1557 | qemu_co_mutex_lock(&s->table_lock); |
15ce94a6 | 1558 | ret = bdrv_qed_do_open(bs, NULL, bs->open_flags, errp); |
2b148f39 | 1559 | qemu_co_mutex_unlock(&s->table_lock); |
15ce94a6 VSO |
1560 | if (ret < 0) { |
1561 | error_prepend(errp, "Could not reopen qed layer: "); | |
5a8a30db | 1562 | } |
c82954e5 BC |
1563 | } |
1564 | ||
54277a2a VSO |
1565 | static int coroutine_fn bdrv_qed_co_check(BlockDriverState *bs, |
1566 | BdrvCheckResult *result, | |
1567 | BdrvCheckMode fix) | |
75411d23 | 1568 | { |
01979a98 | 1569 | BDRVQEDState *s = bs->opaque; |
2fd61638 | 1570 | int ret; |
01979a98 | 1571 | |
2fd61638 PB |
1572 | qemu_co_mutex_lock(&s->table_lock); |
1573 | ret = qed_check(s, result, !!fix); | |
1574 | qemu_co_mutex_unlock(&s->table_lock); | |
1575 | ||
1576 | return ret; | |
75411d23 SH |
1577 | } |
1578 | ||
7ab74849 CL |
1579 | static QemuOptsList qed_create_opts = { |
1580 | .name = "qed-create-opts", | |
1581 | .head = QTAILQ_HEAD_INITIALIZER(qed_create_opts.head), | |
1582 | .desc = { | |
1583 | { | |
1584 | .name = BLOCK_OPT_SIZE, | |
1585 | .type = QEMU_OPT_SIZE, | |
1586 | .help = "Virtual disk size" | |
1587 | }, | |
1588 | { | |
1589 | .name = BLOCK_OPT_BACKING_FILE, | |
1590 | .type = QEMU_OPT_STRING, | |
1591 | .help = "File name of a base image" | |
1592 | }, | |
1593 | { | |
1594 | .name = BLOCK_OPT_BACKING_FMT, | |
1595 | .type = QEMU_OPT_STRING, | |
1596 | .help = "Image format of the base image" | |
1597 | }, | |
1598 | { | |
1599 | .name = BLOCK_OPT_CLUSTER_SIZE, | |
1600 | .type = QEMU_OPT_SIZE, | |
1601 | .help = "Cluster size (in bytes)", | |
1602 | .def_value_str = stringify(QED_DEFAULT_CLUSTER_SIZE) | |
1603 | }, | |
1604 | { | |
1605 | .name = BLOCK_OPT_TABLE_SIZE, | |
1606 | .type = QEMU_OPT_SIZE, | |
1607 | .help = "L1/L2 table size (in clusters)" | |
1608 | }, | |
1609 | { /* end of list */ } | |
1610 | } | |
75411d23 SH |
1611 | }; |
1612 | ||
1613 | static BlockDriver bdrv_qed = { | |
1614 | .format_name = "qed", | |
1615 | .instance_size = sizeof(BDRVQEDState), | |
7ab74849 | 1616 | .create_opts = &qed_create_opts, |
d67066d8 | 1617 | .is_format = true, |
8ee79e70 | 1618 | .supports_backing = true, |
75411d23 SH |
1619 | |
1620 | .bdrv_probe = bdrv_qed_probe, | |
1621 | .bdrv_open = bdrv_qed_open, | |
1622 | .bdrv_close = bdrv_qed_close, | |
f9cb20f1 | 1623 | .bdrv_reopen_prepare = bdrv_qed_reopen_prepare, |
69dca43d | 1624 | .bdrv_child_perm = bdrv_default_perms, |
959355a4 | 1625 | .bdrv_co_create = bdrv_qed_co_create, |
efc75e2a | 1626 | .bdrv_co_create_opts = bdrv_qed_co_create_opts, |
3ac21627 | 1627 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
b8d739fd | 1628 | .bdrv_co_block_status = bdrv_qed_co_block_status, |
89f89709 KW |
1629 | .bdrv_co_readv = bdrv_qed_co_readv, |
1630 | .bdrv_co_writev = bdrv_qed_co_writev, | |
49a2e483 | 1631 | .bdrv_co_pwrite_zeroes = bdrv_qed_co_pwrite_zeroes, |
061ca8a3 | 1632 | .bdrv_co_truncate = bdrv_qed_co_truncate, |
75411d23 SH |
1633 | .bdrv_getlength = bdrv_qed_getlength, |
1634 | .bdrv_get_info = bdrv_qed_get_info, | |
d34682cd | 1635 | .bdrv_refresh_limits = bdrv_qed_refresh_limits, |
75411d23 | 1636 | .bdrv_change_backing_file = bdrv_qed_change_backing_file, |
2b148f39 | 1637 | .bdrv_co_invalidate_cache = bdrv_qed_co_invalidate_cache, |
2fd61638 | 1638 | .bdrv_co_check = bdrv_qed_co_check, |
a8c868c3 SH |
1639 | .bdrv_detach_aio_context = bdrv_qed_detach_aio_context, |
1640 | .bdrv_attach_aio_context = bdrv_qed_attach_aio_context, | |
f8ea8dac | 1641 | .bdrv_co_drain_begin = bdrv_qed_co_drain_begin, |
75411d23 SH |
1642 | }; |
1643 | ||
1644 | static void bdrv_qed_init(void) | |
1645 | { | |
1646 | bdrv_register(&bdrv_qed); | |
1647 | } | |
1648 | ||
1649 | block_init(bdrv_qed_init); |