]>
Commit | Line | Data |
---|---|---|
75411d23 SH |
1 | /* |
2 | * QEMU Enhanced Disk Format | |
3 | * | |
4 | * Copyright IBM, Corp. 2010 | |
5 | * | |
6 | * Authors: | |
7 | * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com> | |
8 | * Anthony Liguori <aliguori@us.ibm.com> | |
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" |
922a01a0 | 20 | #include "qemu/option.h" |
eabba580 | 21 | #include "trace.h" |
75411d23 | 22 | #include "qed.h" |
8a56fdad | 23 | #include "sysemu/block-backend.h" |
959355a4 KW |
24 | #include "qapi/qmp/qdict.h" |
25 | #include "qapi/qobject-input-visitor.h" | |
26 | #include "qapi/qapi-visit-block-core.h" | |
27 | ||
28 | static QemuOptsList qed_create_opts; | |
75411d23 SH |
29 | |
30 | static int bdrv_qed_probe(const uint8_t *buf, int buf_size, | |
31 | const char *filename) | |
32 | { | |
33 | const QEDHeader *header = (const QEDHeader *)buf; | |
34 | ||
35 | if (buf_size < sizeof(*header)) { | |
36 | return 0; | |
37 | } | |
38 | if (le32_to_cpu(header->magic) != QED_MAGIC) { | |
39 | return 0; | |
40 | } | |
41 | return 100; | |
42 | } | |
43 | ||
44 | /** | |
45 | * Check whether an image format is raw | |
46 | * | |
47 | * @fmt: Backing file format, may be NULL | |
48 | */ | |
49 | static bool qed_fmt_is_raw(const char *fmt) | |
50 | { | |
51 | return fmt && strcmp(fmt, "raw") == 0; | |
52 | } | |
53 | ||
54 | static void qed_header_le_to_cpu(const QEDHeader *le, QEDHeader *cpu) | |
55 | { | |
56 | cpu->magic = le32_to_cpu(le->magic); | |
57 | cpu->cluster_size = le32_to_cpu(le->cluster_size); | |
58 | cpu->table_size = le32_to_cpu(le->table_size); | |
59 | cpu->header_size = le32_to_cpu(le->header_size); | |
60 | cpu->features = le64_to_cpu(le->features); | |
61 | cpu->compat_features = le64_to_cpu(le->compat_features); | |
62 | cpu->autoclear_features = le64_to_cpu(le->autoclear_features); | |
63 | cpu->l1_table_offset = le64_to_cpu(le->l1_table_offset); | |
64 | cpu->image_size = le64_to_cpu(le->image_size); | |
65 | cpu->backing_filename_offset = le32_to_cpu(le->backing_filename_offset); | |
66 | cpu->backing_filename_size = le32_to_cpu(le->backing_filename_size); | |
67 | } | |
68 | ||
69 | static void qed_header_cpu_to_le(const QEDHeader *cpu, QEDHeader *le) | |
70 | { | |
71 | le->magic = cpu_to_le32(cpu->magic); | |
72 | le->cluster_size = cpu_to_le32(cpu->cluster_size); | |
73 | le->table_size = cpu_to_le32(cpu->table_size); | |
74 | le->header_size = cpu_to_le32(cpu->header_size); | |
75 | le->features = cpu_to_le64(cpu->features); | |
76 | le->compat_features = cpu_to_le64(cpu->compat_features); | |
77 | le->autoclear_features = cpu_to_le64(cpu->autoclear_features); | |
78 | le->l1_table_offset = cpu_to_le64(cpu->l1_table_offset); | |
79 | le->image_size = cpu_to_le64(cpu->image_size); | |
80 | le->backing_filename_offset = cpu_to_le32(cpu->backing_filename_offset); | |
81 | le->backing_filename_size = cpu_to_le32(cpu->backing_filename_size); | |
82 | } | |
83 | ||
b10170ac | 84 | int qed_write_header_sync(BDRVQEDState *s) |
75411d23 SH |
85 | { |
86 | QEDHeader le; | |
87 | int ret; | |
88 | ||
89 | qed_header_cpu_to_le(&s->header, &le); | |
d9ca2ea2 | 90 | ret = bdrv_pwrite(s->bs->file, 0, &le, sizeof(le)); |
75411d23 SH |
91 | if (ret != sizeof(le)) { |
92 | return ret; | |
93 | } | |
94 | return 0; | |
95 | } | |
96 | ||
01979a98 SH |
97 | /** |
98 | * Update header in-place (does not rewrite backing filename or other strings) | |
99 | * | |
100 | * This function only updates known header fields in-place and does not affect | |
101 | * extra data after the QED header. | |
1f01e50b PB |
102 | * |
103 | * No new allocating reqs can start while this function runs. | |
01979a98 | 104 | */ |
87f0d882 | 105 | static int coroutine_fn qed_write_header(BDRVQEDState *s) |
01979a98 SH |
106 | { |
107 | /* We must write full sectors for O_DIRECT but cannot necessarily generate | |
108 | * the data following the header if an unrecognized compat feature is | |
109 | * active. Therefore, first read the sectors containing the header, update | |
110 | * them, and write back. | |
111 | */ | |
112 | ||
c41a73ff | 113 | int nsectors = DIV_ROUND_UP(sizeof(QEDHeader), BDRV_SECTOR_SIZE); |
01979a98 | 114 | size_t len = nsectors * BDRV_SECTOR_SIZE; |
7076309a KW |
115 | uint8_t *buf; |
116 | struct iovec iov; | |
117 | QEMUIOVector qiov; | |
118 | int ret; | |
119 | ||
1f01e50b PB |
120 | assert(s->allocating_acb || s->allocating_write_reqs_plugged); |
121 | ||
7076309a KW |
122 | buf = qemu_blockalign(s->bs, len); |
123 | iov = (struct iovec) { | |
124 | .iov_base = buf, | |
125 | .iov_len = len, | |
126 | }; | |
127 | qemu_iovec_init_external(&qiov, &iov, 1); | |
128 | ||
0f714ec7 | 129 | ret = bdrv_co_preadv(s->bs->file, 0, qiov.size, &qiov, 0); |
7076309a KW |
130 | if (ret < 0) { |
131 | goto out; | |
132 | } | |
133 | ||
134 | /* Update header */ | |
135 | qed_header_cpu_to_le(&s->header, (QEDHeader *) buf); | |
136 | ||
0f714ec7 | 137 | ret = bdrv_co_pwritev(s->bs->file, 0, qiov.size, &qiov, 0); |
7076309a KW |
138 | if (ret < 0) { |
139 | goto out; | |
140 | } | |
141 | ||
142 | ret = 0; | |
143 | out: | |
144 | qemu_vfree(buf); | |
f13d712b | 145 | return ret; |
01979a98 SH |
146 | } |
147 | ||
75411d23 SH |
148 | static uint64_t qed_max_image_size(uint32_t cluster_size, uint32_t table_size) |
149 | { | |
150 | uint64_t table_entries; | |
151 | uint64_t l2_size; | |
152 | ||
153 | table_entries = (table_size * cluster_size) / sizeof(uint64_t); | |
154 | l2_size = table_entries * cluster_size; | |
155 | ||
156 | return l2_size * table_entries; | |
157 | } | |
158 | ||
159 | static bool qed_is_cluster_size_valid(uint32_t cluster_size) | |
160 | { | |
161 | if (cluster_size < QED_MIN_CLUSTER_SIZE || | |
162 | cluster_size > QED_MAX_CLUSTER_SIZE) { | |
163 | return false; | |
164 | } | |
165 | if (cluster_size & (cluster_size - 1)) { | |
166 | return false; /* not power of 2 */ | |
167 | } | |
168 | return true; | |
169 | } | |
170 | ||
171 | static bool qed_is_table_size_valid(uint32_t table_size) | |
172 | { | |
173 | if (table_size < QED_MIN_TABLE_SIZE || | |
174 | table_size > QED_MAX_TABLE_SIZE) { | |
175 | return false; | |
176 | } | |
177 | if (table_size & (table_size - 1)) { | |
178 | return false; /* not power of 2 */ | |
179 | } | |
180 | return true; | |
181 | } | |
182 | ||
183 | static bool qed_is_image_size_valid(uint64_t image_size, uint32_t cluster_size, | |
184 | uint32_t table_size) | |
185 | { | |
186 | if (image_size % BDRV_SECTOR_SIZE != 0) { | |
187 | return false; /* not multiple of sector size */ | |
188 | } | |
189 | if (image_size > qed_max_image_size(cluster_size, table_size)) { | |
190 | return false; /* image is too large */ | |
191 | } | |
192 | return true; | |
193 | } | |
194 | ||
195 | /** | |
196 | * Read a string of known length from the image file | |
197 | * | |
198 | * @file: Image file | |
199 | * @offset: File offset to start of string, in bytes | |
200 | * @n: String length in bytes | |
201 | * @buf: Destination buffer | |
202 | * @buflen: Destination buffer length in bytes | |
203 | * @ret: 0 on success, -errno on failure | |
204 | * | |
205 | * The string is NUL-terminated. | |
206 | */ | |
cf2ab8fc | 207 | static int qed_read_string(BdrvChild *file, uint64_t offset, size_t n, |
75411d23 SH |
208 | char *buf, size_t buflen) |
209 | { | |
210 | int ret; | |
211 | if (n >= buflen) { | |
212 | return -EINVAL; | |
213 | } | |
214 | ret = bdrv_pread(file, offset, buf, n); | |
215 | if (ret < 0) { | |
216 | return ret; | |
217 | } | |
218 | buf[n] = '\0'; | |
219 | return 0; | |
220 | } | |
221 | ||
eabba580 SH |
222 | /** |
223 | * Allocate new clusters | |
224 | * | |
225 | * @s: QED state | |
226 | * @n: Number of contiguous clusters to allocate | |
227 | * @ret: Offset of first allocated cluster | |
228 | * | |
229 | * This function only produces the offset where the new clusters should be | |
230 | * written. It updates BDRVQEDState but does not make any changes to the image | |
231 | * file. | |
1f01e50b PB |
232 | * |
233 | * Called with table_lock held. | |
eabba580 SH |
234 | */ |
235 | static uint64_t qed_alloc_clusters(BDRVQEDState *s, unsigned int n) | |
236 | { | |
237 | uint64_t offset = s->file_size; | |
238 | s->file_size += n * s->header.cluster_size; | |
239 | return offset; | |
240 | } | |
241 | ||
298800ca SH |
242 | QEDTable *qed_alloc_table(BDRVQEDState *s) |
243 | { | |
244 | /* Honor O_DIRECT memory alignment requirements */ | |
245 | return qemu_blockalign(s->bs, | |
246 | s->header.cluster_size * s->header.table_size); | |
247 | } | |
248 | ||
eabba580 SH |
249 | /** |
250 | * Allocate a new zeroed L2 table | |
1f01e50b PB |
251 | * |
252 | * Called with table_lock held. | |
eabba580 SH |
253 | */ |
254 | static CachedL2Table *qed_new_l2_table(BDRVQEDState *s) | |
255 | { | |
256 | CachedL2Table *l2_table = qed_alloc_l2_cache_entry(&s->l2_cache); | |
257 | ||
258 | l2_table->table = qed_alloc_table(s); | |
259 | l2_table->offset = qed_alloc_clusters(s, s->header.table_size); | |
260 | ||
261 | memset(l2_table->table->offsets, 0, | |
262 | s->header.cluster_size * s->header.table_size); | |
263 | return l2_table; | |
264 | } | |
265 | ||
1f01e50b | 266 | static bool qed_plug_allocating_write_reqs(BDRVQEDState *s) |
6f321e93 | 267 | { |
1f01e50b PB |
268 | qemu_co_mutex_lock(&s->table_lock); |
269 | ||
270 | /* No reentrancy is allowed. */ | |
6f321e93 | 271 | assert(!s->allocating_write_reqs_plugged); |
1f01e50b PB |
272 | if (s->allocating_acb != NULL) { |
273 | /* Another allocating write came concurrently. This cannot happen | |
f8ea8dac | 274 | * from bdrv_qed_co_drain_begin, but it can happen when the timer runs. |
1f01e50b PB |
275 | */ |
276 | qemu_co_mutex_unlock(&s->table_lock); | |
277 | return false; | |
278 | } | |
6f321e93 SH |
279 | |
280 | s->allocating_write_reqs_plugged = true; | |
1f01e50b PB |
281 | qemu_co_mutex_unlock(&s->table_lock); |
282 | return true; | |
6f321e93 SH |
283 | } |
284 | ||
285 | static void qed_unplug_allocating_write_reqs(BDRVQEDState *s) | |
286 | { | |
1f01e50b | 287 | qemu_co_mutex_lock(&s->table_lock); |
6f321e93 | 288 | assert(s->allocating_write_reqs_plugged); |
6f321e93 | 289 | s->allocating_write_reqs_plugged = false; |
1f01e50b PB |
290 | qemu_co_queue_next(&s->allocating_write_reqs); |
291 | qemu_co_mutex_unlock(&s->table_lock); | |
6f321e93 SH |
292 | } |
293 | ||
87f0d882 | 294 | static void coroutine_fn qed_need_check_timer_entry(void *opaque) |
6f321e93 SH |
295 | { |
296 | BDRVQEDState *s = opaque; | |
c0e8f989 KW |
297 | int ret; |
298 | ||
c0e8f989 | 299 | trace_qed_need_check_timer_cb(s); |
6f321e93 | 300 | |
1f01e50b PB |
301 | if (!qed_plug_allocating_write_reqs(s)) { |
302 | return; | |
303 | } | |
c0e8f989 KW |
304 | |
305 | /* Ensure writes are on disk before clearing flag */ | |
306 | ret = bdrv_co_flush(s->bs->file->bs); | |
c0e8f989 | 307 | if (ret < 0) { |
6f321e93 SH |
308 | qed_unplug_allocating_write_reqs(s); |
309 | return; | |
310 | } | |
311 | ||
312 | s->header.features &= ~QED_F_NEED_CHECK; | |
f13d712b KW |
313 | ret = qed_write_header(s); |
314 | (void) ret; | |
315 | ||
316 | qed_unplug_allocating_write_reqs(s); | |
317 | ||
c0e8f989 | 318 | ret = bdrv_co_flush(s->bs); |
f13d712b | 319 | (void) ret; |
6f321e93 SH |
320 | } |
321 | ||
322 | static void qed_need_check_timer_cb(void *opaque) | |
323 | { | |
c0e8f989 KW |
324 | Coroutine *co = qemu_coroutine_create(qed_need_check_timer_entry, opaque); |
325 | qemu_coroutine_enter(co); | |
2f47da5f PB |
326 | } |
327 | ||
6f321e93 SH |
328 | static void qed_start_need_check_timer(BDRVQEDState *s) |
329 | { | |
330 | trace_qed_start_need_check_timer(s); | |
331 | ||
bc72ad67 | 332 | /* Use QEMU_CLOCK_VIRTUAL so we don't alter the image file while suspended for |
6f321e93 SH |
333 | * migration. |
334 | */ | |
bc72ad67 | 335 | timer_mod(s->need_check_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + |
73bcb24d | 336 | NANOSECONDS_PER_SECOND * QED_NEED_CHECK_TIMEOUT); |
6f321e93 SH |
337 | } |
338 | ||
339 | /* It's okay to call this multiple times or when no timer is started */ | |
340 | static void qed_cancel_need_check_timer(BDRVQEDState *s) | |
341 | { | |
342 | trace_qed_cancel_need_check_timer(s); | |
bc72ad67 | 343 | timer_del(s->need_check_timer); |
6f321e93 SH |
344 | } |
345 | ||
a8c868c3 SH |
346 | static void bdrv_qed_detach_aio_context(BlockDriverState *bs) |
347 | { | |
348 | BDRVQEDState *s = bs->opaque; | |
349 | ||
350 | qed_cancel_need_check_timer(s); | |
351 | timer_free(s->need_check_timer); | |
352 | } | |
353 | ||
354 | static void bdrv_qed_attach_aio_context(BlockDriverState *bs, | |
355 | AioContext *new_context) | |
356 | { | |
357 | BDRVQEDState *s = bs->opaque; | |
358 | ||
359 | s->need_check_timer = aio_timer_new(new_context, | |
360 | QEMU_CLOCK_VIRTUAL, SCALE_NS, | |
361 | qed_need_check_timer_cb, s); | |
362 | if (s->header.features & QED_F_NEED_CHECK) { | |
363 | qed_start_need_check_timer(s); | |
364 | } | |
365 | } | |
366 | ||
f8ea8dac | 367 | static void coroutine_fn bdrv_qed_co_drain_begin(BlockDriverState *bs) |
6653a73d FZ |
368 | { |
369 | BDRVQEDState *s = bs->opaque; | |
370 | ||
371 | /* Fire the timer immediately in order to start doing I/O as soon as the | |
372 | * header is flushed. | |
373 | */ | |
374 | if (s->need_check_timer && timer_pending(s->need_check_timer)) { | |
375 | qed_cancel_need_check_timer(s); | |
61124f03 | 376 | qed_need_check_timer_entry(s); |
6653a73d FZ |
377 | } |
378 | } | |
379 | ||
61c7887e PB |
380 | static void bdrv_qed_init_state(BlockDriverState *bs) |
381 | { | |
382 | BDRVQEDState *s = bs->opaque; | |
383 | ||
384 | memset(s, 0, sizeof(BDRVQEDState)); | |
385 | s->bs = bs; | |
1f01e50b | 386 | qemu_co_mutex_init(&s->table_lock); |
61c7887e PB |
387 | qemu_co_queue_init(&s->allocating_write_reqs); |
388 | } | |
389 | ||
9fb4dfc5 PB |
390 | /* Called with table_lock held. */ |
391 | static int coroutine_fn bdrv_qed_do_open(BlockDriverState *bs, QDict *options, | |
392 | int flags, Error **errp) | |
75411d23 SH |
393 | { |
394 | BDRVQEDState *s = bs->opaque; | |
395 | QEDHeader le_header; | |
396 | int64_t file_size; | |
397 | int ret; | |
398 | ||
cf2ab8fc | 399 | ret = bdrv_pread(bs->file, 0, &le_header, sizeof(le_header)); |
75411d23 SH |
400 | if (ret < 0) { |
401 | return ret; | |
402 | } | |
75411d23 SH |
403 | qed_header_le_to_cpu(&le_header, &s->header); |
404 | ||
405 | if (s->header.magic != QED_MAGIC) { | |
76abe407 PB |
406 | error_setg(errp, "Image not in QED format"); |
407 | return -EINVAL; | |
75411d23 SH |
408 | } |
409 | if (s->header.features & ~QED_FEATURE_MASK) { | |
10b758e8 | 410 | /* image uses unsupported feature bits */ |
a55448b3 HR |
411 | error_setg(errp, "Unsupported QED features: %" PRIx64, |
412 | s->header.features & ~QED_FEATURE_MASK); | |
10b758e8 | 413 | return -ENOTSUP; |
75411d23 SH |
414 | } |
415 | if (!qed_is_cluster_size_valid(s->header.cluster_size)) { | |
416 | return -EINVAL; | |
417 | } | |
418 | ||
419 | /* Round down file size to the last cluster */ | |
9a4f4c31 | 420 | file_size = bdrv_getlength(bs->file->bs); |
75411d23 SH |
421 | if (file_size < 0) { |
422 | return file_size; | |
423 | } | |
424 | s->file_size = qed_start_of_cluster(s, file_size); | |
425 | ||
426 | if (!qed_is_table_size_valid(s->header.table_size)) { | |
427 | return -EINVAL; | |
428 | } | |
429 | if (!qed_is_image_size_valid(s->header.image_size, | |
430 | s->header.cluster_size, | |
431 | s->header.table_size)) { | |
432 | return -EINVAL; | |
433 | } | |
434 | if (!qed_check_table_offset(s, s->header.l1_table_offset)) { | |
435 | return -EINVAL; | |
436 | } | |
437 | ||
438 | s->table_nelems = (s->header.cluster_size * s->header.table_size) / | |
439 | sizeof(uint64_t); | |
786a4ea8 | 440 | s->l2_shift = ctz32(s->header.cluster_size); |
75411d23 | 441 | s->l2_mask = s->table_nelems - 1; |
786a4ea8 | 442 | s->l1_shift = s->l2_shift + ctz32(s->table_nelems); |
75411d23 | 443 | |
0adfa1ed SH |
444 | /* Header size calculation must not overflow uint32_t */ |
445 | if (s->header.header_size > UINT32_MAX / s->header.cluster_size) { | |
446 | return -EINVAL; | |
447 | } | |
448 | ||
75411d23 SH |
449 | if ((s->header.features & QED_F_BACKING_FILE)) { |
450 | if ((uint64_t)s->header.backing_filename_offset + | |
451 | s->header.backing_filename_size > | |
452 | s->header.cluster_size * s->header.header_size) { | |
453 | return -EINVAL; | |
454 | } | |
455 | ||
cf2ab8fc | 456 | ret = qed_read_string(bs->file, s->header.backing_filename_offset, |
75411d23 SH |
457 | s->header.backing_filename_size, bs->backing_file, |
458 | sizeof(bs->backing_file)); | |
459 | if (ret < 0) { | |
460 | return ret; | |
461 | } | |
462 | ||
463 | if (s->header.features & QED_F_BACKING_FORMAT_NO_PROBE) { | |
464 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), "raw"); | |
465 | } | |
466 | } | |
467 | ||
468 | /* Reset unknown autoclear feature bits. This is a backwards | |
469 | * compatibility mechanism that allows images to be opened by older | |
470 | * programs, which "knock out" unknown feature bits. When an image is | |
471 | * opened by a newer program again it can detect that the autoclear | |
472 | * feature is no longer valid. | |
473 | */ | |
474 | if ((s->header.autoclear_features & ~QED_AUTOCLEAR_FEATURE_MASK) != 0 && | |
04c01a5c | 475 | !bdrv_is_read_only(bs->file->bs) && !(flags & BDRV_O_INACTIVE)) { |
75411d23 SH |
476 | s->header.autoclear_features &= QED_AUTOCLEAR_FEATURE_MASK; |
477 | ||
478 | ret = qed_write_header_sync(s); | |
479 | if (ret) { | |
480 | return ret; | |
481 | } | |
482 | ||
483 | /* From here on only known autoclear feature bits are valid */ | |
9a4f4c31 | 484 | bdrv_flush(bs->file->bs); |
75411d23 SH |
485 | } |
486 | ||
298800ca SH |
487 | s->l1_table = qed_alloc_table(s); |
488 | qed_init_l2_cache(&s->l2_cache); | |
489 | ||
490 | ret = qed_read_l1_table_sync(s); | |
01979a98 SH |
491 | if (ret) { |
492 | goto out; | |
493 | } | |
494 | ||
495 | /* If image was not closed cleanly, check consistency */ | |
058f8f16 | 496 | if (!(flags & BDRV_O_CHECK) && (s->header.features & QED_F_NEED_CHECK)) { |
01979a98 SH |
497 | /* Read-only images cannot be fixed. There is no risk of corruption |
498 | * since write operations are not possible. Therefore, allow | |
499 | * potentially inconsistent images to be opened read-only. This can | |
500 | * aid data recovery from an otherwise inconsistent image. | |
501 | */ | |
9a4f4c31 | 502 | if (!bdrv_is_read_only(bs->file->bs) && |
04c01a5c | 503 | !(flags & BDRV_O_INACTIVE)) { |
01979a98 SH |
504 | BdrvCheckResult result = {0}; |
505 | ||
506 | ret = qed_check(s, &result, true); | |
6f321e93 SH |
507 | if (ret) { |
508 | goto out; | |
509 | } | |
01979a98 SH |
510 | } |
511 | } | |
512 | ||
a8c868c3 | 513 | bdrv_qed_attach_aio_context(bs, bdrv_get_aio_context(bs)); |
6f321e93 | 514 | |
01979a98 | 515 | out: |
298800ca SH |
516 | if (ret) { |
517 | qed_free_l2_cache(&s->l2_cache); | |
518 | qemu_vfree(s->l1_table); | |
519 | } | |
75411d23 SH |
520 | return ret; |
521 | } | |
522 | ||
9fb4dfc5 PB |
523 | typedef struct QEDOpenCo { |
524 | BlockDriverState *bs; | |
525 | QDict *options; | |
526 | int flags; | |
527 | Error **errp; | |
528 | int ret; | |
529 | } QEDOpenCo; | |
530 | ||
531 | static void coroutine_fn bdrv_qed_open_entry(void *opaque) | |
532 | { | |
533 | QEDOpenCo *qoc = opaque; | |
534 | BDRVQEDState *s = qoc->bs->opaque; | |
535 | ||
536 | qemu_co_mutex_lock(&s->table_lock); | |
537 | qoc->ret = bdrv_qed_do_open(qoc->bs, qoc->options, qoc->flags, qoc->errp); | |
538 | qemu_co_mutex_unlock(&s->table_lock); | |
539 | } | |
540 | ||
4e4bf5c4 KW |
541 | static int bdrv_qed_open(BlockDriverState *bs, QDict *options, int flags, |
542 | Error **errp) | |
543 | { | |
9fb4dfc5 PB |
544 | QEDOpenCo qoc = { |
545 | .bs = bs, | |
546 | .options = options, | |
547 | .flags = flags, | |
548 | .errp = errp, | |
549 | .ret = -EINPROGRESS | |
550 | }; | |
551 | ||
4e4bf5c4 KW |
552 | bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, |
553 | false, errp); | |
554 | if (!bs->file) { | |
555 | return -EINVAL; | |
556 | } | |
557 | ||
61c7887e | 558 | bdrv_qed_init_state(bs); |
9fb4dfc5 PB |
559 | if (qemu_in_coroutine()) { |
560 | bdrv_qed_open_entry(&qoc); | |
561 | } else { | |
4720cbee | 562 | assert(qemu_get_current_aio_context() == qemu_get_aio_context()); |
9fb4dfc5 PB |
563 | qemu_coroutine_enter(qemu_coroutine_create(bdrv_qed_open_entry, &qoc)); |
564 | BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); | |
565 | } | |
566 | BDRV_POLL_WHILE(bs, qoc.ret == -EINPROGRESS); | |
567 | return qoc.ret; | |
4e4bf5c4 KW |
568 | } |
569 | ||
3baca891 | 570 | static void bdrv_qed_refresh_limits(BlockDriverState *bs, Error **errp) |
d34682cd KW |
571 | { |
572 | BDRVQEDState *s = bs->opaque; | |
573 | ||
cf081fca | 574 | bs->bl.pwrite_zeroes_alignment = s->header.cluster_size; |
d34682cd KW |
575 | } |
576 | ||
f9cb20f1 JC |
577 | /* We have nothing to do for QED reopen, stubs just return |
578 | * success */ | |
579 | static int bdrv_qed_reopen_prepare(BDRVReopenState *state, | |
580 | BlockReopenQueue *queue, Error **errp) | |
581 | { | |
582 | return 0; | |
583 | } | |
584 | ||
75411d23 SH |
585 | static void bdrv_qed_close(BlockDriverState *bs) |
586 | { | |
298800ca SH |
587 | BDRVQEDState *s = bs->opaque; |
588 | ||
a8c868c3 | 589 | bdrv_qed_detach_aio_context(bs); |
6f321e93 | 590 | |
01979a98 | 591 | /* Ensure writes reach stable storage */ |
9a4f4c31 | 592 | bdrv_flush(bs->file->bs); |
01979a98 SH |
593 | |
594 | /* Clean shutdown, no check required on next open */ | |
595 | if (s->header.features & QED_F_NEED_CHECK) { | |
596 | s->header.features &= ~QED_F_NEED_CHECK; | |
597 | qed_write_header_sync(s); | |
598 | } | |
599 | ||
298800ca SH |
600 | qed_free_l2_cache(&s->l2_cache); |
601 | qemu_vfree(s->l1_table); | |
75411d23 SH |
602 | } |
603 | ||
959355a4 KW |
604 | static int coroutine_fn bdrv_qed_co_create(BlockdevCreateOptions *opts, |
605 | Error **errp) | |
75411d23 | 606 | { |
959355a4 KW |
607 | BlockdevCreateOptionsQed *qed_opts; |
608 | BlockBackend *blk = NULL; | |
609 | BlockDriverState *bs = NULL; | |
610 | ||
611 | QEDHeader header; | |
75411d23 SH |
612 | QEDHeader le_header; |
613 | uint8_t *l1_table = NULL; | |
959355a4 | 614 | size_t l1_size; |
75411d23 | 615 | int ret = 0; |
75411d23 | 616 | |
959355a4 KW |
617 | assert(opts->driver == BLOCKDEV_DRIVER_QED); |
618 | qed_opts = &opts->u.qed; | |
619 | ||
620 | /* Validate options and set default values */ | |
621 | if (!qed_opts->has_cluster_size) { | |
622 | qed_opts->cluster_size = QED_DEFAULT_CLUSTER_SIZE; | |
623 | } | |
624 | if (!qed_opts->has_table_size) { | |
625 | qed_opts->table_size = QED_DEFAULT_TABLE_SIZE; | |
75411d23 SH |
626 | } |
627 | ||
959355a4 KW |
628 | if (!qed_is_cluster_size_valid(qed_opts->cluster_size)) { |
629 | error_setg(errp, "QED cluster size must be within range [%u, %u] " | |
630 | "and power of 2", | |
631 | QED_MIN_CLUSTER_SIZE, QED_MAX_CLUSTER_SIZE); | |
632 | return -EINVAL; | |
633 | } | |
634 | if (!qed_is_table_size_valid(qed_opts->table_size)) { | |
635 | error_setg(errp, "QED table size must be within range [%u, %u] " | |
636 | "and power of 2", | |
637 | QED_MIN_TABLE_SIZE, QED_MAX_TABLE_SIZE); | |
638 | return -EINVAL; | |
639 | } | |
640 | if (!qed_is_image_size_valid(qed_opts->size, qed_opts->cluster_size, | |
641 | qed_opts->table_size)) | |
642 | { | |
643 | error_setg(errp, "QED image size must be a non-zero multiple of " | |
644 | "cluster size and less than %" PRIu64 " bytes", | |
645 | qed_max_image_size(qed_opts->cluster_size, | |
646 | qed_opts->table_size)); | |
647 | return -EINVAL; | |
648 | } | |
649 | ||
650 | /* Create BlockBackend to write to the image */ | |
651 | bs = bdrv_open_blockdev_ref(qed_opts->file, errp); | |
652 | if (bs == NULL) { | |
8a56fdad | 653 | return -EIO; |
75411d23 SH |
654 | } |
655 | ||
959355a4 KW |
656 | blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL); |
657 | ret = blk_insert_bs(blk, bs, errp); | |
658 | if (ret < 0) { | |
659 | goto out; | |
660 | } | |
8a56fdad KW |
661 | blk_set_allow_write_beyond_eof(blk, true); |
662 | ||
959355a4 KW |
663 | /* Prepare image format */ |
664 | header = (QEDHeader) { | |
665 | .magic = QED_MAGIC, | |
666 | .cluster_size = qed_opts->cluster_size, | |
667 | .table_size = qed_opts->table_size, | |
668 | .header_size = 1, | |
669 | .features = 0, | |
670 | .compat_features = 0, | |
671 | .l1_table_offset = qed_opts->cluster_size, | |
672 | .image_size = qed_opts->size, | |
673 | }; | |
674 | ||
675 | l1_size = header.cluster_size * header.table_size; | |
676 | ||
c743849b | 677 | /* File must start empty and grow, check truncate is supported */ |
3a691c50 | 678 | ret = blk_truncate(blk, 0, PREALLOC_MODE_OFF, errp); |
c743849b SH |
679 | if (ret < 0) { |
680 | goto out; | |
681 | } | |
682 | ||
959355a4 | 683 | if (qed_opts->has_backing_file) { |
75411d23 SH |
684 | header.features |= QED_F_BACKING_FILE; |
685 | header.backing_filename_offset = sizeof(le_header); | |
959355a4 | 686 | header.backing_filename_size = strlen(qed_opts->backing_file); |
75411d23 | 687 | |
959355a4 KW |
688 | if (qed_opts->has_backing_fmt) { |
689 | const char *backing_fmt = BlockdevDriver_str(qed_opts->backing_fmt); | |
690 | if (qed_fmt_is_raw(backing_fmt)) { | |
691 | header.features |= QED_F_BACKING_FORMAT_NO_PROBE; | |
692 | } | |
75411d23 SH |
693 | } |
694 | } | |
695 | ||
696 | qed_header_cpu_to_le(&header, &le_header); | |
8341f00d | 697 | ret = blk_pwrite(blk, 0, &le_header, sizeof(le_header), 0); |
75411d23 SH |
698 | if (ret < 0) { |
699 | goto out; | |
700 | } | |
959355a4 | 701 | ret = blk_pwrite(blk, sizeof(le_header), qed_opts->backing_file, |
8341f00d | 702 | header.backing_filename_size, 0); |
75411d23 SH |
703 | if (ret < 0) { |
704 | goto out; | |
705 | } | |
706 | ||
7267c094 | 707 | l1_table = g_malloc0(l1_size); |
8341f00d | 708 | ret = blk_pwrite(blk, header.l1_table_offset, l1_table, l1_size, 0); |
75411d23 SH |
709 | if (ret < 0) { |
710 | goto out; | |
711 | } | |
712 | ||
713 | ret = 0; /* success */ | |
714 | out: | |
7267c094 | 715 | g_free(l1_table); |
8a56fdad | 716 | blk_unref(blk); |
959355a4 | 717 | bdrv_unref(bs); |
75411d23 SH |
718 | return ret; |
719 | } | |
720 | ||
efc75e2a SH |
721 | static int coroutine_fn bdrv_qed_co_create_opts(const char *filename, |
722 | QemuOpts *opts, | |
723 | Error **errp) | |
75411d23 | 724 | { |
959355a4 | 725 | BlockdevCreateOptions *create_options = NULL; |
92adf9db | 726 | QDict *qdict; |
959355a4 KW |
727 | Visitor *v; |
728 | BlockDriverState *bs = NULL; | |
729 | Error *local_err = NULL; | |
7ab74849 CL |
730 | int ret; |
731 | ||
959355a4 KW |
732 | static const QDictRenames opt_renames[] = { |
733 | { BLOCK_OPT_BACKING_FILE, "backing-file" }, | |
734 | { BLOCK_OPT_BACKING_FMT, "backing-fmt" }, | |
735 | { BLOCK_OPT_CLUSTER_SIZE, "cluster-size" }, | |
736 | { BLOCK_OPT_TABLE_SIZE, "table-size" }, | |
737 | { NULL, NULL }, | |
738 | }; | |
739 | ||
740 | /* Parse options and convert legacy syntax */ | |
741 | qdict = qemu_opts_to_qdict_filtered(opts, NULL, &qed_create_opts, true); | |
742 | ||
743 | if (!qdict_rename_keys(qdict, opt_renames, errp)) { | |
7ab74849 | 744 | ret = -EINVAL; |
959355a4 | 745 | goto fail; |
75411d23 | 746 | } |
959355a4 KW |
747 | |
748 | /* Create and open the file (protocol layer) */ | |
749 | ret = bdrv_create_file(filename, opts, &local_err); | |
750 | if (ret < 0) { | |
751 | error_propagate(errp, local_err); | |
752 | goto fail; | |
753 | } | |
754 | ||
755 | bs = bdrv_open(filename, NULL, NULL, | |
756 | BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp); | |
757 | if (bs == NULL) { | |
758 | ret = -EIO; | |
759 | goto fail; | |
760 | } | |
761 | ||
762 | /* Now get the QAPI type BlockdevCreateOptions */ | |
763 | qdict_put_str(qdict, "driver", "qed"); | |
764 | qdict_put_str(qdict, "file", bs->node_name); | |
765 | ||
af91062e MA |
766 | v = qobject_input_visitor_new_flat_confused(qdict, errp); |
767 | if (!v) { | |
7ab74849 | 768 | ret = -EINVAL; |
959355a4 | 769 | goto fail; |
75411d23 | 770 | } |
959355a4 | 771 | |
959355a4 KW |
772 | visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err); |
773 | visit_free(v); | |
774 | ||
775 | if (local_err) { | |
776 | error_propagate(errp, local_err); | |
7ab74849 | 777 | ret = -EINVAL; |
959355a4 | 778 | goto fail; |
75411d23 SH |
779 | } |
780 | ||
959355a4 KW |
781 | /* Silently round up size */ |
782 | assert(create_options->driver == BLOCKDEV_DRIVER_QED); | |
783 | create_options->u.qed.size = | |
784 | ROUND_UP(create_options->u.qed.size, BDRV_SECTOR_SIZE); | |
785 | ||
786 | /* Create the qed image (format layer) */ | |
787 | ret = bdrv_qed_co_create(create_options, errp); | |
7ab74849 | 788 | |
959355a4 | 789 | fail: |
cb3e7f08 | 790 | qobject_unref(qdict); |
959355a4 KW |
791 | bdrv_unref(bs); |
792 | qapi_free_BlockdevCreateOptions(create_options); | |
7ab74849 | 793 | return ret; |
75411d23 SH |
794 | } |
795 | ||
b8d739fd EB |
796 | static int coroutine_fn bdrv_qed_co_block_status(BlockDriverState *bs, |
797 | bool want_zero, | |
798 | int64_t pos, int64_t bytes, | |
799 | int64_t *pnum, int64_t *map, | |
800 | BlockDriverState **file) | |
298800ca | 801 | { |
b8d739fd EB |
802 | BDRVQEDState *s = bs->opaque; |
803 | size_t len = MIN(bytes, SIZE_MAX); | |
804 | int status; | |
805 | QEDRequest request = { .l2_table = NULL }; | |
806 | uint64_t offset; | |
807 | int ret; | |
808 | ||
809 | qemu_co_mutex_lock(&s->table_lock); | |
810 | ret = qed_find_cluster(s, &request, pos, &len, &offset); | |
811 | ||
812 | *pnum = len; | |
4bc74be9 PB |
813 | switch (ret) { |
814 | case QED_CLUSTER_FOUND: | |
b8d739fd EB |
815 | *map = offset | qed_offset_into_cluster(s, pos); |
816 | status = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID; | |
817 | *file = bs->file->bs; | |
4bc74be9 PB |
818 | break; |
819 | case QED_CLUSTER_ZERO: | |
b8d739fd | 820 | status = BDRV_BLOCK_ZERO; |
4bc74be9 PB |
821 | break; |
822 | case QED_CLUSTER_L2: | |
823 | case QED_CLUSTER_L1: | |
b8d739fd | 824 | status = 0; |
4bc74be9 PB |
825 | break; |
826 | default: | |
827 | assert(ret < 0); | |
b8d739fd | 828 | status = ret; |
4bc74be9 PB |
829 | break; |
830 | } | |
831 | ||
298800ca | 832 | qed_unref_l2_cache_entry(request.l2_table); |
1f01e50b | 833 | qemu_co_mutex_unlock(&s->table_lock); |
298800ca | 834 | |
b8d739fd | 835 | return status; |
75411d23 SH |
836 | } |
837 | ||
eabba580 SH |
838 | static BDRVQEDState *acb_to_s(QEDAIOCB *acb) |
839 | { | |
48cc565e | 840 | return acb->bs->opaque; |
eabba580 SH |
841 | } |
842 | ||
843 | /** | |
844 | * Read from the backing file or zero-fill if no backing file | |
845 | * | |
f06ee3d4 KW |
846 | * @s: QED state |
847 | * @pos: Byte position in device | |
848 | * @qiov: Destination I/O vector | |
849 | * @backing_qiov: Possibly shortened copy of qiov, to be allocated here | |
850 | * @cb: Completion function | |
851 | * @opaque: User data for completion function | |
eabba580 SH |
852 | * |
853 | * This function reads qiov->size bytes starting at pos from the backing file. | |
854 | * If there is no backing file then zeroes are read. | |
855 | */ | |
87f0d882 KW |
856 | static int coroutine_fn qed_read_backing_file(BDRVQEDState *s, uint64_t pos, |
857 | QEMUIOVector *qiov, | |
858 | QEMUIOVector **backing_qiov) | |
eabba580 | 859 | { |
eabba580 SH |
860 | uint64_t backing_length = 0; |
861 | size_t size; | |
e85c5281 | 862 | int ret; |
eabba580 SH |
863 | |
864 | /* If there is a backing file, get its length. Treat the absence of a | |
865 | * backing file like a zero length backing file. | |
866 | */ | |
760e0063 KW |
867 | if (s->bs->backing) { |
868 | int64_t l = bdrv_getlength(s->bs->backing->bs); | |
eabba580 | 869 | if (l < 0) { |
e85c5281 | 870 | return l; |
eabba580 SH |
871 | } |
872 | backing_length = l; | |
873 | } | |
874 | ||
875 | /* Zero all sectors if reading beyond the end of the backing file */ | |
876 | if (pos >= backing_length || | |
877 | pos + qiov->size > backing_length) { | |
3d9b4925 | 878 | qemu_iovec_memset(qiov, 0, 0, qiov->size); |
eabba580 SH |
879 | } |
880 | ||
881 | /* Complete now if there are no backing file sectors to read */ | |
882 | if (pos >= backing_length) { | |
e85c5281 | 883 | return 0; |
eabba580 SH |
884 | } |
885 | ||
886 | /* If the read straddles the end of the backing file, shorten it */ | |
887 | size = MIN((uint64_t)backing_length - pos, qiov->size); | |
888 | ||
f06ee3d4 KW |
889 | assert(*backing_qiov == NULL); |
890 | *backing_qiov = g_new(QEMUIOVector, 1); | |
891 | qemu_iovec_init(*backing_qiov, qiov->niov); | |
892 | qemu_iovec_concat(*backing_qiov, qiov, 0, size); | |
893 | ||
820100fd | 894 | BLKDBG_EVENT(s->bs->file, BLKDBG_READ_BACKING_AIO); |
0f714ec7 | 895 | ret = bdrv_co_preadv(s->bs->backing, pos, size, *backing_qiov, 0); |
e85c5281 KW |
896 | if (ret < 0) { |
897 | return ret; | |
898 | } | |
899 | return 0; | |
eabba580 SH |
900 | } |
901 | ||
eabba580 SH |
902 | /** |
903 | * Copy data from backing file into the image | |
904 | * | |
905 | * @s: QED state | |
906 | * @pos: Byte position in device | |
907 | * @len: Number of bytes | |
908 | * @offset: Byte offset in image file | |
eabba580 | 909 | */ |
87f0d882 KW |
910 | static int coroutine_fn qed_copy_from_backing_file(BDRVQEDState *s, |
911 | uint64_t pos, uint64_t len, | |
912 | uint64_t offset) | |
eabba580 | 913 | { |
0f7aa24d KW |
914 | QEMUIOVector qiov; |
915 | QEMUIOVector *backing_qiov = NULL; | |
916 | struct iovec iov; | |
e85c5281 | 917 | int ret; |
eabba580 SH |
918 | |
919 | /* Skip copy entirely if there is no work to do */ | |
920 | if (len == 0) { | |
b4ac32f3 | 921 | return 0; |
eabba580 SH |
922 | } |
923 | ||
0f7aa24d KW |
924 | iov = (struct iovec) { |
925 | .iov_base = qemu_blockalign(s->bs, len), | |
926 | .iov_len = len, | |
927 | }; | |
928 | qemu_iovec_init_external(&qiov, &iov, 1); | |
929 | ||
930 | ret = qed_read_backing_file(s, pos, &qiov, &backing_qiov); | |
931 | ||
932 | if (backing_qiov) { | |
933 | qemu_iovec_destroy(backing_qiov); | |
934 | g_free(backing_qiov); | |
935 | backing_qiov = NULL; | |
936 | } | |
937 | ||
938 | if (ret) { | |
939 | goto out; | |
940 | } | |
eabba580 | 941 | |
0f7aa24d | 942 | BLKDBG_EVENT(s->bs->file, BLKDBG_COW_WRITE); |
0f714ec7 | 943 | ret = bdrv_co_pwritev(s->bs->file, offset, qiov.size, &qiov, 0); |
0f7aa24d KW |
944 | if (ret < 0) { |
945 | goto out; | |
946 | } | |
947 | ret = 0; | |
948 | out: | |
949 | qemu_vfree(iov.iov_base); | |
b4ac32f3 | 950 | return ret; |
eabba580 SH |
951 | } |
952 | ||
953 | /** | |
954 | * Link one or more contiguous clusters into a table | |
955 | * | |
956 | * @s: QED state | |
957 | * @table: L2 table | |
958 | * @index: First cluster index | |
959 | * @n: Number of contiguous clusters | |
21df65b6 AL |
960 | * @cluster: First cluster offset |
961 | * | |
962 | * The cluster offset may be an allocated byte offset in the image file, the | |
963 | * zero cluster marker, or the unallocated cluster marker. | |
1f01e50b PB |
964 | * |
965 | * Called with table_lock held. | |
eabba580 | 966 | */ |
87f0d882 KW |
967 | static void coroutine_fn qed_update_l2_table(BDRVQEDState *s, QEDTable *table, |
968 | int index, unsigned int n, | |
969 | uint64_t cluster) | |
eabba580 SH |
970 | { |
971 | int i; | |
972 | for (i = index; i < index + n; i++) { | |
973 | table->offsets[i] = cluster; | |
21df65b6 AL |
974 | if (!qed_offset_is_unalloc_cluster(cluster) && |
975 | !qed_offset_is_zero_cluster(cluster)) { | |
976 | cluster += s->header.cluster_size; | |
977 | } | |
eabba580 SH |
978 | } |
979 | } | |
980 | ||
1f01e50b | 981 | /* Called with table_lock held. */ |
87f0d882 | 982 | static void coroutine_fn qed_aio_complete(QEDAIOCB *acb) |
eabba580 | 983 | { |
1919631e | 984 | BDRVQEDState *s = acb_to_s(acb); |
eabba580 SH |
985 | |
986 | /* Free resources */ | |
987 | qemu_iovec_destroy(&acb->cur_qiov); | |
988 | qed_unref_l2_cache_entry(acb->request.l2_table); | |
989 | ||
0e71be19 SH |
990 | /* Free the buffer we may have allocated for zero writes */ |
991 | if (acb->flags & QED_AIOCB_ZERO) { | |
992 | qemu_vfree(acb->qiov->iov[0].iov_base); | |
993 | acb->qiov->iov[0].iov_base = NULL; | |
994 | } | |
995 | ||
eabba580 SH |
996 | /* Start next allocating write request waiting behind this one. Note that |
997 | * requests enqueue themselves when they first hit an unallocated cluster | |
998 | * but they wait until the entire request is finished before waking up the | |
999 | * next request in the queue. This ensures that we don't cycle through | |
1000 | * requests multiple times but rather finish one at a time completely. | |
1001 | */ | |
0806c3b5 KW |
1002 | if (acb == s->allocating_acb) { |
1003 | s->allocating_acb = NULL; | |
1004 | if (!qemu_co_queue_empty(&s->allocating_write_reqs)) { | |
1f01e50b | 1005 | qemu_co_queue_next(&s->allocating_write_reqs); |
6f321e93 SH |
1006 | } else if (s->header.features & QED_F_NEED_CHECK) { |
1007 | qed_start_need_check_timer(s); | |
eabba580 SH |
1008 | } |
1009 | } | |
1010 | } | |
1011 | ||
1012 | /** | |
fae25ac7 | 1013 | * Update L1 table with new L2 table offset and write it out |
1f01e50b PB |
1014 | * |
1015 | * Called with table_lock held. | |
eabba580 | 1016 | */ |
87f0d882 | 1017 | static int coroutine_fn qed_aio_write_l1_update(QEDAIOCB *acb) |
eabba580 | 1018 | { |
eabba580 SH |
1019 | BDRVQEDState *s = acb_to_s(acb); |
1020 | CachedL2Table *l2_table = acb->request.l2_table; | |
e4fc8781 | 1021 | uint64_t l2_offset = l2_table->offset; |
fb18de21 | 1022 | int index, ret; |
eabba580 | 1023 | |
fae25ac7 KW |
1024 | index = qed_l1_index(s, acb->cur_pos); |
1025 | s->l1_table->offsets[index] = l2_table->offset; | |
1026 | ||
1027 | ret = qed_write_l1_table(s, index, 1); | |
1028 | ||
1029 | /* Commit the current L2 table to the cache */ | |
eabba580 SH |
1030 | qed_commit_l2_cache_entry(&s->l2_cache, l2_table); |
1031 | ||
1032 | /* This is guaranteed to succeed because we just committed the entry to the | |
1033 | * cache. | |
1034 | */ | |
e4fc8781 | 1035 | acb->request.l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset); |
eabba580 SH |
1036 | assert(acb->request.l2_table != NULL); |
1037 | ||
fb18de21 | 1038 | return ret; |
eabba580 SH |
1039 | } |
1040 | ||
eabba580 SH |
1041 | |
1042 | /** | |
1043 | * Update L2 table with new cluster offsets and write them out | |
1f01e50b PB |
1044 | * |
1045 | * Called with table_lock held. | |
eabba580 | 1046 | */ |
87f0d882 | 1047 | static int coroutine_fn qed_aio_write_l2_update(QEDAIOCB *acb, uint64_t offset) |
eabba580 | 1048 | { |
eabba580 SH |
1049 | BDRVQEDState *s = acb_to_s(acb); |
1050 | bool need_alloc = acb->find_cluster_ret == QED_CLUSTER_L1; | |
88d2dd72 | 1051 | int index, ret; |
eabba580 SH |
1052 | |
1053 | if (need_alloc) { | |
1054 | qed_unref_l2_cache_entry(acb->request.l2_table); | |
1055 | acb->request.l2_table = qed_new_l2_table(s); | |
1056 | } | |
1057 | ||
1058 | index = qed_l2_index(s, acb->cur_pos); | |
1059 | qed_update_l2_table(s, acb->request.l2_table->table, index, acb->cur_nclusters, | |
0e71be19 | 1060 | offset); |
eabba580 SH |
1061 | |
1062 | if (need_alloc) { | |
1063 | /* Write out the whole new L2 table */ | |
453e53e2 | 1064 | ret = qed_write_l2_table(s, &acb->request, 0, s->table_nelems, true); |
fb18de21 | 1065 | if (ret) { |
88d2dd72 | 1066 | return ret; |
fb18de21 | 1067 | } |
88d2dd72 | 1068 | return qed_aio_write_l1_update(acb); |
eabba580 SH |
1069 | } else { |
1070 | /* Write out only the updated part of the L2 table */ | |
453e53e2 KW |
1071 | ret = qed_write_l2_table(s, &acb->request, index, acb->cur_nclusters, |
1072 | false); | |
88d2dd72 KW |
1073 | if (ret) { |
1074 | return ret; | |
1075 | } | |
eabba580 | 1076 | } |
88d2dd72 | 1077 | return 0; |
eabba580 SH |
1078 | } |
1079 | ||
eabba580 SH |
1080 | /** |
1081 | * Write data to the image file | |
1f01e50b PB |
1082 | * |
1083 | * Called with table_lock *not* held. | |
eabba580 | 1084 | */ |
87f0d882 | 1085 | static int coroutine_fn qed_aio_write_main(QEDAIOCB *acb) |
eabba580 | 1086 | { |
eabba580 SH |
1087 | BDRVQEDState *s = acb_to_s(acb); |
1088 | uint64_t offset = acb->cur_cluster + | |
1089 | qed_offset_into_cluster(s, acb->cur_pos); | |
eabba580 | 1090 | |
eaf0bc56 | 1091 | trace_qed_aio_write_main(s, acb, 0, offset, acb->cur_qiov.size); |
eabba580 | 1092 | |
a4d8f1ae | 1093 | BLKDBG_EVENT(s->bs->file, BLKDBG_WRITE_AIO); |
e7569c18 PB |
1094 | return bdrv_co_pwritev(s->bs->file, offset, acb->cur_qiov.size, |
1095 | &acb->cur_qiov, 0); | |
eabba580 SH |
1096 | } |
1097 | ||
1098 | /** | |
b4ac32f3 | 1099 | * Populate untouched regions of new data cluster |
1f01e50b PB |
1100 | * |
1101 | * Called with table_lock held. | |
eabba580 | 1102 | */ |
87f0d882 | 1103 | static int coroutine_fn qed_aio_write_cow(QEDAIOCB *acb) |
eabba580 | 1104 | { |
eabba580 | 1105 | BDRVQEDState *s = acb_to_s(acb); |
b4ac32f3 | 1106 | uint64_t start, len, offset; |
a101341a | 1107 | int ret; |
eabba580 | 1108 | |
1f01e50b PB |
1109 | qemu_co_mutex_unlock(&s->table_lock); |
1110 | ||
b4ac32f3 KW |
1111 | /* Populate front untouched region of new data cluster */ |
1112 | start = qed_start_of_cluster(s, acb->cur_pos); | |
1113 | len = qed_offset_into_cluster(s, acb->cur_pos); | |
1114 | ||
1115 | trace_qed_aio_write_prefill(s, acb, start, len, acb->cur_cluster); | |
1116 | ret = qed_copy_from_backing_file(s, start, len, acb->cur_cluster); | |
a101341a | 1117 | if (ret < 0) { |
1f01e50b | 1118 | goto out; |
eabba580 SH |
1119 | } |
1120 | ||
b4ac32f3 KW |
1121 | /* Populate back untouched region of new data cluster */ |
1122 | start = acb->cur_pos + acb->cur_qiov.size; | |
1123 | len = qed_start_of_cluster(s, start + s->header.cluster_size - 1) - start; | |
1124 | offset = acb->cur_cluster + | |
1125 | qed_offset_into_cluster(s, acb->cur_pos) + | |
1126 | acb->cur_qiov.size; | |
eabba580 | 1127 | |
b4ac32f3 KW |
1128 | trace_qed_aio_write_postfill(s, acb, start, len, offset); |
1129 | ret = qed_copy_from_backing_file(s, start, len, offset); | |
eaf0bc56 | 1130 | if (ret < 0) { |
1f01e50b | 1131 | goto out; |
eaf0bc56 | 1132 | } |
a101341a | 1133 | |
e7569c18 PB |
1134 | ret = qed_aio_write_main(acb); |
1135 | if (ret < 0) { | |
1f01e50b | 1136 | goto out; |
e7569c18 PB |
1137 | } |
1138 | ||
1139 | if (s->bs->backing) { | |
1140 | /* | |
1141 | * Flush new data clusters before updating the L2 table | |
1142 | * | |
1143 | * This flush is necessary when a backing file is in use. A crash | |
1144 | * during an allocating write could result in empty clusters in the | |
1145 | * image. If the write only touched a subregion of the cluster, | |
1146 | * then backing image sectors have been lost in the untouched | |
1147 | * region. The solution is to flush after writing a new data | |
1148 | * cluster and before updating the L2 table. | |
1149 | */ | |
1150 | ret = bdrv_co_flush(s->bs->file->bs); | |
e7569c18 PB |
1151 | } |
1152 | ||
1f01e50b PB |
1153 | out: |
1154 | qemu_co_mutex_lock(&s->table_lock); | |
1155 | return ret; | |
eabba580 SH |
1156 | } |
1157 | ||
0d09c797 SH |
1158 | /** |
1159 | * Check if the QED_F_NEED_CHECK bit should be set during allocating write | |
1160 | */ | |
1161 | static bool qed_should_set_need_check(BDRVQEDState *s) | |
1162 | { | |
1163 | /* The flush before L2 update path ensures consistency */ | |
760e0063 | 1164 | if (s->bs->backing) { |
0d09c797 SH |
1165 | return false; |
1166 | } | |
1167 | ||
1168 | return !(s->header.features & QED_F_NEED_CHECK); | |
1169 | } | |
1170 | ||
eabba580 SH |
1171 | /** |
1172 | * Write new data cluster | |
1173 | * | |
1174 | * @acb: Write request | |
1175 | * @len: Length in bytes | |
1176 | * | |
1177 | * This path is taken when writing to previously unallocated clusters. | |
1f01e50b PB |
1178 | * |
1179 | * Called with table_lock held. | |
eabba580 | 1180 | */ |
87f0d882 | 1181 | static int coroutine_fn qed_aio_write_alloc(QEDAIOCB *acb, size_t len) |
eabba580 SH |
1182 | { |
1183 | BDRVQEDState *s = acb_to_s(acb); | |
f13d712b | 1184 | int ret; |
eabba580 | 1185 | |
6f321e93 | 1186 | /* Cancel timer when the first allocating request comes in */ |
0806c3b5 | 1187 | if (s->allocating_acb == NULL) { |
6f321e93 SH |
1188 | qed_cancel_need_check_timer(s); |
1189 | } | |
1190 | ||
eabba580 | 1191 | /* Freeze this request if another allocating write is in progress */ |
0806c3b5 KW |
1192 | if (s->allocating_acb != acb || s->allocating_write_reqs_plugged) { |
1193 | if (s->allocating_acb != NULL) { | |
1f01e50b | 1194 | qemu_co_queue_wait(&s->allocating_write_reqs, &s->table_lock); |
0806c3b5 KW |
1195 | assert(s->allocating_acb == NULL); |
1196 | } | |
1197 | s->allocating_acb = acb; | |
1198 | return -EAGAIN; /* start over with looking up table entries */ | |
eabba580 SH |
1199 | } |
1200 | ||
1201 | acb->cur_nclusters = qed_bytes_to_clusters(s, | |
1202 | qed_offset_into_cluster(s, acb->cur_pos) + len); | |
1b093c48 | 1203 | qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
eabba580 | 1204 | |
0e71be19 SH |
1205 | if (acb->flags & QED_AIOCB_ZERO) { |
1206 | /* Skip ahead if the clusters are already zero */ | |
1207 | if (acb->find_cluster_ret == QED_CLUSTER_ZERO) { | |
d6daddcd | 1208 | return 0; |
0e71be19 | 1209 | } |
e7569c18 | 1210 | acb->cur_cluster = 1; |
0e71be19 | 1211 | } else { |
0e71be19 SH |
1212 | acb->cur_cluster = qed_alloc_clusters(s, acb->cur_nclusters); |
1213 | } | |
1214 | ||
0d09c797 SH |
1215 | if (qed_should_set_need_check(s)) { |
1216 | s->header.features |= QED_F_NEED_CHECK; | |
f13d712b | 1217 | ret = qed_write_header(s); |
a101341a | 1218 | if (ret < 0) { |
d6daddcd | 1219 | return ret; |
a101341a KW |
1220 | } |
1221 | } | |
1222 | ||
e7569c18 | 1223 | if (!(acb->flags & QED_AIOCB_ZERO)) { |
a101341a | 1224 | ret = qed_aio_write_cow(acb); |
e7569c18 PB |
1225 | if (ret < 0) { |
1226 | return ret; | |
1227 | } | |
01979a98 | 1228 | } |
e7569c18 PB |
1229 | |
1230 | return qed_aio_write_l2_update(acb, acb->cur_cluster); | |
eabba580 SH |
1231 | } |
1232 | ||
1233 | /** | |
1234 | * Write data cluster in place | |
1235 | * | |
1236 | * @acb: Write request | |
1237 | * @offset: Cluster offset in bytes | |
1238 | * @len: Length in bytes | |
1239 | * | |
1240 | * This path is taken when writing to already allocated clusters. | |
1f01e50b PB |
1241 | * |
1242 | * Called with table_lock held. | |
eabba580 | 1243 | */ |
87f0d882 KW |
1244 | static int coroutine_fn qed_aio_write_inplace(QEDAIOCB *acb, uint64_t offset, |
1245 | size_t len) | |
eabba580 | 1246 | { |
1f01e50b PB |
1247 | BDRVQEDState *s = acb_to_s(acb); |
1248 | int r; | |
1249 | ||
1250 | qemu_co_mutex_unlock(&s->table_lock); | |
1251 | ||
0e71be19 SH |
1252 | /* Allocate buffer for zero writes */ |
1253 | if (acb->flags & QED_AIOCB_ZERO) { | |
1254 | struct iovec *iov = acb->qiov->iov; | |
1255 | ||
1256 | if (!iov->iov_base) { | |
48cc565e | 1257 | iov->iov_base = qemu_try_blockalign(acb->bs, iov->iov_len); |
4f4896db | 1258 | if (iov->iov_base == NULL) { |
1f01e50b PB |
1259 | r = -ENOMEM; |
1260 | goto out; | |
4f4896db | 1261 | } |
0e71be19 SH |
1262 | memset(iov->iov_base, 0, iov->iov_len); |
1263 | } | |
1264 | } | |
1265 | ||
eabba580 SH |
1266 | /* Calculate the I/O vector */ |
1267 | acb->cur_cluster = offset; | |
1b093c48 | 1268 | qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
eabba580 | 1269 | |
1f01e50b PB |
1270 | /* Do the actual write. */ |
1271 | r = qed_aio_write_main(acb); | |
1272 | out: | |
1273 | qemu_co_mutex_lock(&s->table_lock); | |
1274 | return r; | |
eabba580 SH |
1275 | } |
1276 | ||
1277 | /** | |
1278 | * Write data cluster | |
1279 | * | |
1280 | * @opaque: Write request | |
0596be7e | 1281 | * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2 or QED_CLUSTER_L1 |
eabba580 SH |
1282 | * @offset: Cluster offset in bytes |
1283 | * @len: Length in bytes | |
1f01e50b PB |
1284 | * |
1285 | * Called with table_lock held. | |
eabba580 | 1286 | */ |
87f0d882 KW |
1287 | static int coroutine_fn qed_aio_write_data(void *opaque, int ret, |
1288 | uint64_t offset, size_t len) | |
eabba580 SH |
1289 | { |
1290 | QEDAIOCB *acb = opaque; | |
1291 | ||
1292 | trace_qed_aio_write_data(acb_to_s(acb), acb, ret, offset, len); | |
1293 | ||
1294 | acb->find_cluster_ret = ret; | |
1295 | ||
1296 | switch (ret) { | |
1297 | case QED_CLUSTER_FOUND: | |
0596be7e | 1298 | return qed_aio_write_inplace(acb, offset, len); |
eabba580 SH |
1299 | |
1300 | case QED_CLUSTER_L2: | |
1301 | case QED_CLUSTER_L1: | |
21df65b6 | 1302 | case QED_CLUSTER_ZERO: |
0596be7e | 1303 | return qed_aio_write_alloc(acb, len); |
eabba580 SH |
1304 | |
1305 | default: | |
0596be7e | 1306 | g_assert_not_reached(); |
d6daddcd | 1307 | } |
eabba580 SH |
1308 | } |
1309 | ||
1310 | /** | |
1311 | * Read data cluster | |
1312 | * | |
1313 | * @opaque: Read request | |
0596be7e | 1314 | * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2 or QED_CLUSTER_L1 |
eabba580 SH |
1315 | * @offset: Cluster offset in bytes |
1316 | * @len: Length in bytes | |
1f01e50b PB |
1317 | * |
1318 | * Called with table_lock held. | |
eabba580 | 1319 | */ |
87f0d882 KW |
1320 | static int coroutine_fn qed_aio_read_data(void *opaque, int ret, |
1321 | uint64_t offset, size_t len) | |
eabba580 SH |
1322 | { |
1323 | QEDAIOCB *acb = opaque; | |
1324 | BDRVQEDState *s = acb_to_s(acb); | |
48cc565e | 1325 | BlockDriverState *bs = acb->bs; |
1f01e50b PB |
1326 | int r; |
1327 | ||
1328 | qemu_co_mutex_unlock(&s->table_lock); | |
eabba580 SH |
1329 | |
1330 | /* Adjust offset into cluster */ | |
1331 | offset += qed_offset_into_cluster(s, acb->cur_pos); | |
1332 | ||
1333 | trace_qed_aio_read_data(s, acb, ret, offset, len); | |
1334 | ||
1b093c48 | 1335 | qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
eabba580 | 1336 | |
1f01e50b PB |
1337 | /* Handle zero cluster and backing file reads, otherwise read |
1338 | * data cluster directly. | |
1339 | */ | |
21df65b6 | 1340 | if (ret == QED_CLUSTER_ZERO) { |
3d9b4925 | 1341 | qemu_iovec_memset(&acb->cur_qiov, 0, 0, acb->cur_qiov.size); |
1f01e50b | 1342 | r = 0; |
21df65b6 | 1343 | } else if (ret != QED_CLUSTER_FOUND) { |
1f01e50b PB |
1344 | r = qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov, |
1345 | &acb->backing_qiov); | |
1346 | } else { | |
1347 | BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); | |
1348 | r = bdrv_co_preadv(bs->file, offset, acb->cur_qiov.size, | |
1349 | &acb->cur_qiov, 0); | |
eabba580 SH |
1350 | } |
1351 | ||
1f01e50b PB |
1352 | qemu_co_mutex_lock(&s->table_lock); |
1353 | return r; | |
eabba580 SH |
1354 | } |
1355 | ||
1356 | /** | |
1357 | * Begin next I/O or complete the request | |
1358 | */ | |
87f0d882 | 1359 | static int coroutine_fn qed_aio_next_io(QEDAIOCB *acb) |
eabba580 | 1360 | { |
eabba580 | 1361 | BDRVQEDState *s = acb_to_s(acb); |
0f21b7a1 KW |
1362 | uint64_t offset; |
1363 | size_t len; | |
dddf8db1 | 1364 | int ret; |
eabba580 | 1365 | |
1f01e50b | 1366 | qemu_co_mutex_lock(&s->table_lock); |
01859874 KW |
1367 | while (1) { |
1368 | trace_qed_aio_next_io(s, acb, 0, acb->cur_pos + acb->cur_qiov.size); | |
eabba580 | 1369 | |
01859874 KW |
1370 | if (acb->backing_qiov) { |
1371 | qemu_iovec_destroy(acb->backing_qiov); | |
1372 | g_free(acb->backing_qiov); | |
1373 | acb->backing_qiov = NULL; | |
1374 | } | |
f06ee3d4 | 1375 | |
01859874 KW |
1376 | acb->qiov_offset += acb->cur_qiov.size; |
1377 | acb->cur_pos += acb->cur_qiov.size; | |
1378 | qemu_iovec_reset(&acb->cur_qiov); | |
eabba580 | 1379 | |
01859874 KW |
1380 | /* Complete request */ |
1381 | if (acb->cur_pos >= acb->end_pos) { | |
48cc565e KW |
1382 | ret = 0; |
1383 | break; | |
01859874 | 1384 | } |
eabba580 | 1385 | |
01859874 KW |
1386 | /* Find next cluster and start I/O */ |
1387 | len = acb->end_pos - acb->cur_pos; | |
1388 | ret = qed_find_cluster(s, &acb->request, acb->cur_pos, &len, &offset); | |
1389 | if (ret < 0) { | |
48cc565e | 1390 | break; |
01859874 | 1391 | } |
0596be7e | 1392 | |
01859874 KW |
1393 | if (acb->flags & QED_AIOCB_WRITE) { |
1394 | ret = qed_aio_write_data(acb, ret, offset, len); | |
1395 | } else { | |
1396 | ret = qed_aio_read_data(acb, ret, offset, len); | |
1397 | } | |
0596be7e | 1398 | |
0806c3b5 | 1399 | if (ret < 0 && ret != -EAGAIN) { |
48cc565e | 1400 | break; |
0596be7e | 1401 | } |
0596be7e | 1402 | } |
eabba580 | 1403 | |
48cc565e KW |
1404 | trace_qed_aio_complete(s, acb, ret); |
1405 | qed_aio_complete(acb); | |
1f01e50b | 1406 | qemu_co_mutex_unlock(&s->table_lock); |
48cc565e | 1407 | return ret; |
89f89709 KW |
1408 | } |
1409 | ||
1410 | static int coroutine_fn qed_co_request(BlockDriverState *bs, int64_t sector_num, | |
1411 | QEMUIOVector *qiov, int nb_sectors, | |
1412 | int flags) | |
1413 | { | |
48cc565e KW |
1414 | QEDAIOCB acb = { |
1415 | .bs = bs, | |
1416 | .cur_pos = (uint64_t) sector_num * BDRV_SECTOR_SIZE, | |
1417 | .end_pos = (sector_num + nb_sectors) * BDRV_SECTOR_SIZE, | |
1418 | .qiov = qiov, | |
1419 | .flags = flags, | |
89f89709 | 1420 | }; |
48cc565e | 1421 | qemu_iovec_init(&acb.cur_qiov, qiov->niov); |
eabba580 | 1422 | |
48cc565e | 1423 | trace_qed_aio_setup(bs->opaque, &acb, sector_num, nb_sectors, NULL, flags); |
eabba580 SH |
1424 | |
1425 | /* Start request */ | |
48cc565e | 1426 | return qed_aio_next_io(&acb); |
75411d23 SH |
1427 | } |
1428 | ||
89f89709 KW |
1429 | static int coroutine_fn bdrv_qed_co_readv(BlockDriverState *bs, |
1430 | int64_t sector_num, int nb_sectors, | |
1431 | QEMUIOVector *qiov) | |
75411d23 | 1432 | { |
89f89709 | 1433 | return qed_co_request(bs, sector_num, qiov, nb_sectors, 0); |
75411d23 SH |
1434 | } |
1435 | ||
89f89709 KW |
1436 | static int coroutine_fn bdrv_qed_co_writev(BlockDriverState *bs, |
1437 | int64_t sector_num, int nb_sectors, | |
e18a58b4 | 1438 | QEMUIOVector *qiov, int flags) |
0e71be19 | 1439 | { |
e18a58b4 | 1440 | assert(!flags); |
89f89709 | 1441 | return qed_co_request(bs, sector_num, qiov, nb_sectors, QED_AIOCB_WRITE); |
0e71be19 SH |
1442 | } |
1443 | ||
49a2e483 EB |
1444 | static int coroutine_fn bdrv_qed_co_pwrite_zeroes(BlockDriverState *bs, |
1445 | int64_t offset, | |
f5a5ca79 | 1446 | int bytes, |
49a2e483 | 1447 | BdrvRequestFlags flags) |
0e71be19 | 1448 | { |
ef72f76e | 1449 | BDRVQEDState *s = bs->opaque; |
0e71be19 SH |
1450 | QEMUIOVector qiov; |
1451 | struct iovec iov; | |
1452 | ||
49a2e483 EB |
1453 | /* Fall back if the request is not aligned */ |
1454 | if (qed_offset_into_cluster(s, offset) || | |
f5a5ca79 | 1455 | qed_offset_into_cluster(s, bytes)) { |
49a2e483 | 1456 | return -ENOTSUP; |
ef72f76e SH |
1457 | } |
1458 | ||
0e71be19 SH |
1459 | /* Zero writes start without an I/O buffer. If a buffer becomes necessary |
1460 | * then it will be allocated during request processing. | |
1461 | */ | |
49a2e483 | 1462 | iov.iov_base = NULL; |
f5a5ca79 | 1463 | iov.iov_len = bytes; |
0e71be19 SH |
1464 | |
1465 | qemu_iovec_init_external(&qiov, &iov, 1); | |
89f89709 | 1466 | return qed_co_request(bs, offset >> BDRV_SECTOR_BITS, &qiov, |
f5a5ca79 | 1467 | bytes >> BDRV_SECTOR_BITS, |
89f89709 | 1468 | QED_AIOCB_WRITE | QED_AIOCB_ZERO); |
0e71be19 SH |
1469 | } |
1470 | ||
061ca8a3 KW |
1471 | static int coroutine_fn bdrv_qed_co_truncate(BlockDriverState *bs, |
1472 | int64_t offset, | |
1473 | PreallocMode prealloc, | |
1474 | Error **errp) | |
75411d23 | 1475 | { |
77a5a000 SH |
1476 | BDRVQEDState *s = bs->opaque; |
1477 | uint64_t old_image_size; | |
1478 | int ret; | |
1479 | ||
8243ccb7 HR |
1480 | if (prealloc != PREALLOC_MODE_OFF) { |
1481 | error_setg(errp, "Unsupported preallocation mode '%s'", | |
977c736f | 1482 | PreallocMode_str(prealloc)); |
8243ccb7 HR |
1483 | return -ENOTSUP; |
1484 | } | |
1485 | ||
77a5a000 SH |
1486 | if (!qed_is_image_size_valid(offset, s->header.cluster_size, |
1487 | s->header.table_size)) { | |
f59adb32 | 1488 | error_setg(errp, "Invalid image size specified"); |
77a5a000 SH |
1489 | return -EINVAL; |
1490 | } | |
1491 | ||
77a5a000 | 1492 | if ((uint64_t)offset < s->header.image_size) { |
f59adb32 | 1493 | error_setg(errp, "Shrinking images is currently not supported"); |
77a5a000 SH |
1494 | return -ENOTSUP; |
1495 | } | |
1496 | ||
1497 | old_image_size = s->header.image_size; | |
1498 | s->header.image_size = offset; | |
1499 | ret = qed_write_header_sync(s); | |
1500 | if (ret < 0) { | |
1501 | s->header.image_size = old_image_size; | |
f59adb32 | 1502 | error_setg_errno(errp, -ret, "Failed to update the image size"); |
77a5a000 SH |
1503 | } |
1504 | return ret; | |
75411d23 SH |
1505 | } |
1506 | ||
1507 | static int64_t bdrv_qed_getlength(BlockDriverState *bs) | |
1508 | { | |
1509 | BDRVQEDState *s = bs->opaque; | |
1510 | return s->header.image_size; | |
1511 | } | |
1512 | ||
1513 | static int bdrv_qed_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
1514 | { | |
1515 | BDRVQEDState *s = bs->opaque; | |
1516 | ||
1517 | memset(bdi, 0, sizeof(*bdi)); | |
1518 | bdi->cluster_size = s->header.cluster_size; | |
d68dbee8 | 1519 | bdi->is_dirty = s->header.features & QED_F_NEED_CHECK; |
95de6d70 | 1520 | bdi->unallocated_blocks_are_zero = true; |
75411d23 SH |
1521 | return 0; |
1522 | } | |
1523 | ||
1524 | static int bdrv_qed_change_backing_file(BlockDriverState *bs, | |
1525 | const char *backing_file, | |
1526 | const char *backing_fmt) | |
1527 | { | |
1528 | BDRVQEDState *s = bs->opaque; | |
1529 | QEDHeader new_header, le_header; | |
1530 | void *buffer; | |
1531 | size_t buffer_len, backing_file_len; | |
1532 | int ret; | |
1533 | ||
1534 | /* Refuse to set backing filename if unknown compat feature bits are | |
1535 | * active. If the image uses an unknown compat feature then we may not | |
1536 | * know the layout of data following the header structure and cannot safely | |
1537 | * add a new string. | |
1538 | */ | |
1539 | if (backing_file && (s->header.compat_features & | |
1540 | ~QED_COMPAT_FEATURE_MASK)) { | |
1541 | return -ENOTSUP; | |
1542 | } | |
1543 | ||
1544 | memcpy(&new_header, &s->header, sizeof(new_header)); | |
1545 | ||
1546 | new_header.features &= ~(QED_F_BACKING_FILE | | |
1547 | QED_F_BACKING_FORMAT_NO_PROBE); | |
1548 | ||
1549 | /* Adjust feature flags */ | |
1550 | if (backing_file) { | |
1551 | new_header.features |= QED_F_BACKING_FILE; | |
1552 | ||
1553 | if (qed_fmt_is_raw(backing_fmt)) { | |
1554 | new_header.features |= QED_F_BACKING_FORMAT_NO_PROBE; | |
1555 | } | |
1556 | } | |
1557 | ||
1558 | /* Calculate new header size */ | |
1559 | backing_file_len = 0; | |
1560 | ||
1561 | if (backing_file) { | |
1562 | backing_file_len = strlen(backing_file); | |
1563 | } | |
1564 | ||
1565 | buffer_len = sizeof(new_header); | |
1566 | new_header.backing_filename_offset = buffer_len; | |
1567 | new_header.backing_filename_size = backing_file_len; | |
1568 | buffer_len += backing_file_len; | |
1569 | ||
1570 | /* Make sure we can rewrite header without failing */ | |
1571 | if (buffer_len > new_header.header_size * new_header.cluster_size) { | |
1572 | return -ENOSPC; | |
1573 | } | |
1574 | ||
1575 | /* Prepare new header */ | |
7267c094 | 1576 | buffer = g_malloc(buffer_len); |
75411d23 SH |
1577 | |
1578 | qed_header_cpu_to_le(&new_header, &le_header); | |
1579 | memcpy(buffer, &le_header, sizeof(le_header)); | |
1580 | buffer_len = sizeof(le_header); | |
1581 | ||
feba23b1 PB |
1582 | if (backing_file) { |
1583 | memcpy(buffer + buffer_len, backing_file, backing_file_len); | |
1584 | buffer_len += backing_file_len; | |
1585 | } | |
75411d23 SH |
1586 | |
1587 | /* Write new header */ | |
d9ca2ea2 | 1588 | ret = bdrv_pwrite_sync(bs->file, 0, buffer, buffer_len); |
7267c094 | 1589 | g_free(buffer); |
75411d23 SH |
1590 | if (ret == 0) { |
1591 | memcpy(&s->header, &new_header, sizeof(new_header)); | |
1592 | } | |
1593 | return ret; | |
1594 | } | |
1595 | ||
2b148f39 PB |
1596 | static void coroutine_fn bdrv_qed_co_invalidate_cache(BlockDriverState *bs, |
1597 | Error **errp) | |
c82954e5 | 1598 | { |
1f01e50b | 1599 | BDRVQEDState *s = bs->opaque; |
5a8a30db KW |
1600 | Error *local_err = NULL; |
1601 | int ret; | |
c82954e5 BC |
1602 | |
1603 | bdrv_qed_close(bs); | |
3456a8d1 | 1604 | |
61c7887e | 1605 | bdrv_qed_init_state(bs); |
2b148f39 | 1606 | qemu_co_mutex_lock(&s->table_lock); |
4e4bf5c4 | 1607 | ret = bdrv_qed_do_open(bs, NULL, bs->open_flags, &local_err); |
2b148f39 | 1608 | qemu_co_mutex_unlock(&s->table_lock); |
5a8a30db | 1609 | if (local_err) { |
4b576648 MA |
1610 | error_propagate_prepend(errp, local_err, |
1611 | "Could not reopen qed layer: "); | |
5a8a30db KW |
1612 | return; |
1613 | } else if (ret < 0) { | |
1614 | error_setg_errno(errp, -ret, "Could not reopen qed layer"); | |
1615 | return; | |
1616 | } | |
c82954e5 BC |
1617 | } |
1618 | ||
2fd61638 PB |
1619 | static int bdrv_qed_co_check(BlockDriverState *bs, BdrvCheckResult *result, |
1620 | BdrvCheckMode fix) | |
75411d23 | 1621 | { |
01979a98 | 1622 | BDRVQEDState *s = bs->opaque; |
2fd61638 | 1623 | int ret; |
01979a98 | 1624 | |
2fd61638 PB |
1625 | qemu_co_mutex_lock(&s->table_lock); |
1626 | ret = qed_check(s, result, !!fix); | |
1627 | qemu_co_mutex_unlock(&s->table_lock); | |
1628 | ||
1629 | return ret; | |
75411d23 SH |
1630 | } |
1631 | ||
7ab74849 CL |
1632 | static QemuOptsList qed_create_opts = { |
1633 | .name = "qed-create-opts", | |
1634 | .head = QTAILQ_HEAD_INITIALIZER(qed_create_opts.head), | |
1635 | .desc = { | |
1636 | { | |
1637 | .name = BLOCK_OPT_SIZE, | |
1638 | .type = QEMU_OPT_SIZE, | |
1639 | .help = "Virtual disk size" | |
1640 | }, | |
1641 | { | |
1642 | .name = BLOCK_OPT_BACKING_FILE, | |
1643 | .type = QEMU_OPT_STRING, | |
1644 | .help = "File name of a base image" | |
1645 | }, | |
1646 | { | |
1647 | .name = BLOCK_OPT_BACKING_FMT, | |
1648 | .type = QEMU_OPT_STRING, | |
1649 | .help = "Image format of the base image" | |
1650 | }, | |
1651 | { | |
1652 | .name = BLOCK_OPT_CLUSTER_SIZE, | |
1653 | .type = QEMU_OPT_SIZE, | |
1654 | .help = "Cluster size (in bytes)", | |
1655 | .def_value_str = stringify(QED_DEFAULT_CLUSTER_SIZE) | |
1656 | }, | |
1657 | { | |
1658 | .name = BLOCK_OPT_TABLE_SIZE, | |
1659 | .type = QEMU_OPT_SIZE, | |
1660 | .help = "L1/L2 table size (in clusters)" | |
1661 | }, | |
1662 | { /* end of list */ } | |
1663 | } | |
75411d23 SH |
1664 | }; |
1665 | ||
1666 | static BlockDriver bdrv_qed = { | |
1667 | .format_name = "qed", | |
1668 | .instance_size = sizeof(BDRVQEDState), | |
7ab74849 | 1669 | .create_opts = &qed_create_opts, |
8ee79e70 | 1670 | .supports_backing = true, |
75411d23 SH |
1671 | |
1672 | .bdrv_probe = bdrv_qed_probe, | |
1673 | .bdrv_open = bdrv_qed_open, | |
1674 | .bdrv_close = bdrv_qed_close, | |
f9cb20f1 | 1675 | .bdrv_reopen_prepare = bdrv_qed_reopen_prepare, |
862f215f | 1676 | .bdrv_child_perm = bdrv_format_default_perms, |
959355a4 | 1677 | .bdrv_co_create = bdrv_qed_co_create, |
efc75e2a | 1678 | .bdrv_co_create_opts = bdrv_qed_co_create_opts, |
3ac21627 | 1679 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
b8d739fd | 1680 | .bdrv_co_block_status = bdrv_qed_co_block_status, |
89f89709 KW |
1681 | .bdrv_co_readv = bdrv_qed_co_readv, |
1682 | .bdrv_co_writev = bdrv_qed_co_writev, | |
49a2e483 | 1683 | .bdrv_co_pwrite_zeroes = bdrv_qed_co_pwrite_zeroes, |
061ca8a3 | 1684 | .bdrv_co_truncate = bdrv_qed_co_truncate, |
75411d23 SH |
1685 | .bdrv_getlength = bdrv_qed_getlength, |
1686 | .bdrv_get_info = bdrv_qed_get_info, | |
d34682cd | 1687 | .bdrv_refresh_limits = bdrv_qed_refresh_limits, |
75411d23 | 1688 | .bdrv_change_backing_file = bdrv_qed_change_backing_file, |
2b148f39 | 1689 | .bdrv_co_invalidate_cache = bdrv_qed_co_invalidate_cache, |
2fd61638 | 1690 | .bdrv_co_check = bdrv_qed_co_check, |
a8c868c3 SH |
1691 | .bdrv_detach_aio_context = bdrv_qed_detach_aio_context, |
1692 | .bdrv_attach_aio_context = bdrv_qed_attach_aio_context, | |
f8ea8dac | 1693 | .bdrv_co_drain_begin = bdrv_qed_co_drain_begin, |
75411d23 SH |
1694 | }; |
1695 | ||
1696 | static void bdrv_qed_init(void) | |
1697 | { | |
1698 | bdrv_register(&bdrv_qed); | |
1699 | } | |
1700 | ||
1701 | block_init(bdrv_qed_init); |