]>
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" |
da34e65c | 16 | #include "qapi/error.h" |
1de7afc9 | 17 | #include "qemu/timer.h" |
58369e22 | 18 | #include "qemu/bswap.h" |
eabba580 | 19 | #include "trace.h" |
75411d23 | 20 | #include "qed.h" |
7b1b5d19 | 21 | #include "qapi/qmp/qerror.h" |
8a56fdad | 22 | #include "sysemu/block-backend.h" |
75411d23 | 23 | |
d7331bed | 24 | static const AIOCBInfo qed_aiocb_info = { |
eabba580 | 25 | .aiocb_size = sizeof(QEDAIOCB), |
eabba580 SH |
26 | }; |
27 | ||
75411d23 SH |
28 | static int bdrv_qed_probe(const uint8_t *buf, int buf_size, |
29 | const char *filename) | |
30 | { | |
31 | const QEDHeader *header = (const QEDHeader *)buf; | |
32 | ||
33 | if (buf_size < sizeof(*header)) { | |
34 | return 0; | |
35 | } | |
36 | if (le32_to_cpu(header->magic) != QED_MAGIC) { | |
37 | return 0; | |
38 | } | |
39 | return 100; | |
40 | } | |
41 | ||
42 | /** | |
43 | * Check whether an image format is raw | |
44 | * | |
45 | * @fmt: Backing file format, may be NULL | |
46 | */ | |
47 | static bool qed_fmt_is_raw(const char *fmt) | |
48 | { | |
49 | return fmt && strcmp(fmt, "raw") == 0; | |
50 | } | |
51 | ||
52 | static void qed_header_le_to_cpu(const QEDHeader *le, QEDHeader *cpu) | |
53 | { | |
54 | cpu->magic = le32_to_cpu(le->magic); | |
55 | cpu->cluster_size = le32_to_cpu(le->cluster_size); | |
56 | cpu->table_size = le32_to_cpu(le->table_size); | |
57 | cpu->header_size = le32_to_cpu(le->header_size); | |
58 | cpu->features = le64_to_cpu(le->features); | |
59 | cpu->compat_features = le64_to_cpu(le->compat_features); | |
60 | cpu->autoclear_features = le64_to_cpu(le->autoclear_features); | |
61 | cpu->l1_table_offset = le64_to_cpu(le->l1_table_offset); | |
62 | cpu->image_size = le64_to_cpu(le->image_size); | |
63 | cpu->backing_filename_offset = le32_to_cpu(le->backing_filename_offset); | |
64 | cpu->backing_filename_size = le32_to_cpu(le->backing_filename_size); | |
65 | } | |
66 | ||
67 | static void qed_header_cpu_to_le(const QEDHeader *cpu, QEDHeader *le) | |
68 | { | |
69 | le->magic = cpu_to_le32(cpu->magic); | |
70 | le->cluster_size = cpu_to_le32(cpu->cluster_size); | |
71 | le->table_size = cpu_to_le32(cpu->table_size); | |
72 | le->header_size = cpu_to_le32(cpu->header_size); | |
73 | le->features = cpu_to_le64(cpu->features); | |
74 | le->compat_features = cpu_to_le64(cpu->compat_features); | |
75 | le->autoclear_features = cpu_to_le64(cpu->autoclear_features); | |
76 | le->l1_table_offset = cpu_to_le64(cpu->l1_table_offset); | |
77 | le->image_size = cpu_to_le64(cpu->image_size); | |
78 | le->backing_filename_offset = cpu_to_le32(cpu->backing_filename_offset); | |
79 | le->backing_filename_size = cpu_to_le32(cpu->backing_filename_size); | |
80 | } | |
81 | ||
b10170ac | 82 | int qed_write_header_sync(BDRVQEDState *s) |
75411d23 SH |
83 | { |
84 | QEDHeader le; | |
85 | int ret; | |
86 | ||
87 | qed_header_cpu_to_le(&s->header, &le); | |
d9ca2ea2 | 88 | ret = bdrv_pwrite(s->bs->file, 0, &le, sizeof(le)); |
75411d23 SH |
89 | if (ret != sizeof(le)) { |
90 | return ret; | |
91 | } | |
92 | return 0; | |
93 | } | |
94 | ||
01979a98 SH |
95 | typedef struct { |
96 | GenericCB gencb; | |
97 | BDRVQEDState *s; | |
98 | struct iovec iov; | |
99 | QEMUIOVector qiov; | |
100 | int nsectors; | |
101 | uint8_t *buf; | |
102 | } QEDWriteHeaderCB; | |
103 | ||
104 | static void qed_write_header_cb(void *opaque, int ret) | |
105 | { | |
106 | QEDWriteHeaderCB *write_header_cb = opaque; | |
107 | ||
108 | qemu_vfree(write_header_cb->buf); | |
109 | gencb_complete(write_header_cb, ret); | |
110 | } | |
111 | ||
112 | static void qed_write_header_read_cb(void *opaque, int ret) | |
113 | { | |
114 | QEDWriteHeaderCB *write_header_cb = opaque; | |
115 | BDRVQEDState *s = write_header_cb->s; | |
01979a98 SH |
116 | |
117 | if (ret) { | |
118 | qed_write_header_cb(write_header_cb, ret); | |
119 | return; | |
120 | } | |
121 | ||
122 | /* Update header */ | |
123 | qed_header_cpu_to_le(&s->header, (QEDHeader *)write_header_cb->buf); | |
124 | ||
0d1049c7 | 125 | bdrv_aio_writev(s->bs->file, 0, &write_header_cb->qiov, |
ad54ae80 PB |
126 | write_header_cb->nsectors, qed_write_header_cb, |
127 | write_header_cb); | |
01979a98 SH |
128 | } |
129 | ||
130 | /** | |
131 | * Update header in-place (does not rewrite backing filename or other strings) | |
132 | * | |
133 | * This function only updates known header fields in-place and does not affect | |
134 | * extra data after the QED header. | |
135 | */ | |
097310b5 | 136 | static void qed_write_header(BDRVQEDState *s, BlockCompletionFunc cb, |
01979a98 SH |
137 | void *opaque) |
138 | { | |
139 | /* We must write full sectors for O_DIRECT but cannot necessarily generate | |
140 | * the data following the header if an unrecognized compat feature is | |
141 | * active. Therefore, first read the sectors containing the header, update | |
142 | * them, and write back. | |
143 | */ | |
144 | ||
c41a73ff | 145 | int nsectors = DIV_ROUND_UP(sizeof(QEDHeader), BDRV_SECTOR_SIZE); |
01979a98 SH |
146 | size_t len = nsectors * BDRV_SECTOR_SIZE; |
147 | QEDWriteHeaderCB *write_header_cb = gencb_alloc(sizeof(*write_header_cb), | |
148 | cb, opaque); | |
149 | ||
150 | write_header_cb->s = s; | |
151 | write_header_cb->nsectors = nsectors; | |
152 | write_header_cb->buf = qemu_blockalign(s->bs, len); | |
153 | write_header_cb->iov.iov_base = write_header_cb->buf; | |
154 | write_header_cb->iov.iov_len = len; | |
155 | qemu_iovec_init_external(&write_header_cb->qiov, &write_header_cb->iov, 1); | |
156 | ||
ebb7af21 | 157 | bdrv_aio_readv(s->bs->file, 0, &write_header_cb->qiov, nsectors, |
ad54ae80 | 158 | qed_write_header_read_cb, write_header_cb); |
01979a98 SH |
159 | } |
160 | ||
75411d23 SH |
161 | static uint64_t qed_max_image_size(uint32_t cluster_size, uint32_t table_size) |
162 | { | |
163 | uint64_t table_entries; | |
164 | uint64_t l2_size; | |
165 | ||
166 | table_entries = (table_size * cluster_size) / sizeof(uint64_t); | |
167 | l2_size = table_entries * cluster_size; | |
168 | ||
169 | return l2_size * table_entries; | |
170 | } | |
171 | ||
172 | static bool qed_is_cluster_size_valid(uint32_t cluster_size) | |
173 | { | |
174 | if (cluster_size < QED_MIN_CLUSTER_SIZE || | |
175 | cluster_size > QED_MAX_CLUSTER_SIZE) { | |
176 | return false; | |
177 | } | |
178 | if (cluster_size & (cluster_size - 1)) { | |
179 | return false; /* not power of 2 */ | |
180 | } | |
181 | return true; | |
182 | } | |
183 | ||
184 | static bool qed_is_table_size_valid(uint32_t table_size) | |
185 | { | |
186 | if (table_size < QED_MIN_TABLE_SIZE || | |
187 | table_size > QED_MAX_TABLE_SIZE) { | |
188 | return false; | |
189 | } | |
190 | if (table_size & (table_size - 1)) { | |
191 | return false; /* not power of 2 */ | |
192 | } | |
193 | return true; | |
194 | } | |
195 | ||
196 | static bool qed_is_image_size_valid(uint64_t image_size, uint32_t cluster_size, | |
197 | uint32_t table_size) | |
198 | { | |
199 | if (image_size % BDRV_SECTOR_SIZE != 0) { | |
200 | return false; /* not multiple of sector size */ | |
201 | } | |
202 | if (image_size > qed_max_image_size(cluster_size, table_size)) { | |
203 | return false; /* image is too large */ | |
204 | } | |
205 | return true; | |
206 | } | |
207 | ||
208 | /** | |
209 | * Read a string of known length from the image file | |
210 | * | |
211 | * @file: Image file | |
212 | * @offset: File offset to start of string, in bytes | |
213 | * @n: String length in bytes | |
214 | * @buf: Destination buffer | |
215 | * @buflen: Destination buffer length in bytes | |
216 | * @ret: 0 on success, -errno on failure | |
217 | * | |
218 | * The string is NUL-terminated. | |
219 | */ | |
cf2ab8fc | 220 | static int qed_read_string(BdrvChild *file, uint64_t offset, size_t n, |
75411d23 SH |
221 | char *buf, size_t buflen) |
222 | { | |
223 | int ret; | |
224 | if (n >= buflen) { | |
225 | return -EINVAL; | |
226 | } | |
227 | ret = bdrv_pread(file, offset, buf, n); | |
228 | if (ret < 0) { | |
229 | return ret; | |
230 | } | |
231 | buf[n] = '\0'; | |
232 | return 0; | |
233 | } | |
234 | ||
eabba580 SH |
235 | /** |
236 | * Allocate new clusters | |
237 | * | |
238 | * @s: QED state | |
239 | * @n: Number of contiguous clusters to allocate | |
240 | * @ret: Offset of first allocated cluster | |
241 | * | |
242 | * This function only produces the offset where the new clusters should be | |
243 | * written. It updates BDRVQEDState but does not make any changes to the image | |
244 | * file. | |
245 | */ | |
246 | static uint64_t qed_alloc_clusters(BDRVQEDState *s, unsigned int n) | |
247 | { | |
248 | uint64_t offset = s->file_size; | |
249 | s->file_size += n * s->header.cluster_size; | |
250 | return offset; | |
251 | } | |
252 | ||
298800ca SH |
253 | QEDTable *qed_alloc_table(BDRVQEDState *s) |
254 | { | |
255 | /* Honor O_DIRECT memory alignment requirements */ | |
256 | return qemu_blockalign(s->bs, | |
257 | s->header.cluster_size * s->header.table_size); | |
258 | } | |
259 | ||
eabba580 SH |
260 | /** |
261 | * Allocate a new zeroed L2 table | |
262 | */ | |
263 | static CachedL2Table *qed_new_l2_table(BDRVQEDState *s) | |
264 | { | |
265 | CachedL2Table *l2_table = qed_alloc_l2_cache_entry(&s->l2_cache); | |
266 | ||
267 | l2_table->table = qed_alloc_table(s); | |
268 | l2_table->offset = qed_alloc_clusters(s, s->header.table_size); | |
269 | ||
270 | memset(l2_table->table->offsets, 0, | |
271 | s->header.cluster_size * s->header.table_size); | |
272 | return l2_table; | |
273 | } | |
274 | ||
b20123a2 PB |
275 | static void qed_aio_next_io(QEDAIOCB *acb, int ret); |
276 | ||
277 | static void qed_aio_start_io(QEDAIOCB *acb) | |
278 | { | |
279 | qed_aio_next_io(acb, 0); | |
280 | } | |
281 | ||
282 | static void qed_aio_next_io_cb(void *opaque, int ret) | |
283 | { | |
284 | QEDAIOCB *acb = opaque; | |
285 | ||
286 | qed_aio_next_io(acb, ret); | |
287 | } | |
eabba580 | 288 | |
6f321e93 SH |
289 | static void qed_plug_allocating_write_reqs(BDRVQEDState *s) |
290 | { | |
291 | assert(!s->allocating_write_reqs_plugged); | |
292 | ||
293 | s->allocating_write_reqs_plugged = true; | |
294 | } | |
295 | ||
296 | static void qed_unplug_allocating_write_reqs(BDRVQEDState *s) | |
297 | { | |
298 | QEDAIOCB *acb; | |
299 | ||
300 | assert(s->allocating_write_reqs_plugged); | |
301 | ||
302 | s->allocating_write_reqs_plugged = false; | |
303 | ||
304 | acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs); | |
305 | if (acb) { | |
b20123a2 | 306 | qed_aio_start_io(acb); |
6f321e93 SH |
307 | } |
308 | } | |
309 | ||
310 | static void qed_finish_clear_need_check(void *opaque, int ret) | |
311 | { | |
312 | /* Do nothing */ | |
313 | } | |
314 | ||
315 | static void qed_flush_after_clear_need_check(void *opaque, int ret) | |
316 | { | |
317 | BDRVQEDState *s = opaque; | |
318 | ||
319 | bdrv_aio_flush(s->bs, qed_finish_clear_need_check, s); | |
320 | ||
321 | /* No need to wait until flush completes */ | |
322 | qed_unplug_allocating_write_reqs(s); | |
323 | } | |
324 | ||
325 | static void qed_clear_need_check(void *opaque, int ret) | |
326 | { | |
327 | BDRVQEDState *s = opaque; | |
328 | ||
329 | if (ret) { | |
330 | qed_unplug_allocating_write_reqs(s); | |
331 | return; | |
332 | } | |
333 | ||
334 | s->header.features &= ~QED_F_NEED_CHECK; | |
335 | qed_write_header(s, qed_flush_after_clear_need_check, s); | |
336 | } | |
337 | ||
338 | static void qed_need_check_timer_cb(void *opaque) | |
339 | { | |
340 | BDRVQEDState *s = opaque; | |
341 | ||
342 | /* The timer should only fire when allocating writes have drained */ | |
343 | assert(!QSIMPLEQ_FIRST(&s->allocating_write_reqs)); | |
344 | ||
345 | trace_qed_need_check_timer_cb(s); | |
346 | ||
2f47da5f | 347 | qed_acquire(s); |
6f321e93 SH |
348 | qed_plug_allocating_write_reqs(s); |
349 | ||
350 | /* Ensure writes are on disk before clearing flag */ | |
6653a73d | 351 | bdrv_aio_flush(s->bs->file->bs, qed_clear_need_check, s); |
2f47da5f PB |
352 | qed_release(s); |
353 | } | |
354 | ||
355 | void qed_acquire(BDRVQEDState *s) | |
356 | { | |
357 | aio_context_acquire(bdrv_get_aio_context(s->bs)); | |
358 | } | |
359 | ||
360 | void qed_release(BDRVQEDState *s) | |
361 | { | |
362 | aio_context_release(bdrv_get_aio_context(s->bs)); | |
6f321e93 SH |
363 | } |
364 | ||
365 | static void qed_start_need_check_timer(BDRVQEDState *s) | |
366 | { | |
367 | trace_qed_start_need_check_timer(s); | |
368 | ||
bc72ad67 | 369 | /* Use QEMU_CLOCK_VIRTUAL so we don't alter the image file while suspended for |
6f321e93 SH |
370 | * migration. |
371 | */ | |
bc72ad67 | 372 | timer_mod(s->need_check_timer, qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + |
73bcb24d | 373 | NANOSECONDS_PER_SECOND * QED_NEED_CHECK_TIMEOUT); |
6f321e93 SH |
374 | } |
375 | ||
376 | /* It's okay to call this multiple times or when no timer is started */ | |
377 | static void qed_cancel_need_check_timer(BDRVQEDState *s) | |
378 | { | |
379 | trace_qed_cancel_need_check_timer(s); | |
bc72ad67 | 380 | timer_del(s->need_check_timer); |
6f321e93 SH |
381 | } |
382 | ||
a8c868c3 SH |
383 | static void bdrv_qed_detach_aio_context(BlockDriverState *bs) |
384 | { | |
385 | BDRVQEDState *s = bs->opaque; | |
386 | ||
387 | qed_cancel_need_check_timer(s); | |
388 | timer_free(s->need_check_timer); | |
389 | } | |
390 | ||
391 | static void bdrv_qed_attach_aio_context(BlockDriverState *bs, | |
392 | AioContext *new_context) | |
393 | { | |
394 | BDRVQEDState *s = bs->opaque; | |
395 | ||
396 | s->need_check_timer = aio_timer_new(new_context, | |
397 | QEMU_CLOCK_VIRTUAL, SCALE_NS, | |
398 | qed_need_check_timer_cb, s); | |
399 | if (s->header.features & QED_F_NEED_CHECK) { | |
400 | qed_start_need_check_timer(s); | |
401 | } | |
402 | } | |
403 | ||
6653a73d FZ |
404 | static void bdrv_qed_drain(BlockDriverState *bs) |
405 | { | |
406 | BDRVQEDState *s = bs->opaque; | |
407 | ||
408 | /* Fire the timer immediately in order to start doing I/O as soon as the | |
409 | * header is flushed. | |
410 | */ | |
411 | if (s->need_check_timer && timer_pending(s->need_check_timer)) { | |
412 | qed_cancel_need_check_timer(s); | |
413 | qed_need_check_timer_cb(s); | |
414 | } | |
415 | } | |
416 | ||
4e4bf5c4 KW |
417 | static int bdrv_qed_do_open(BlockDriverState *bs, QDict *options, int flags, |
418 | Error **errp) | |
75411d23 SH |
419 | { |
420 | BDRVQEDState *s = bs->opaque; | |
421 | QEDHeader le_header; | |
422 | int64_t file_size; | |
423 | int ret; | |
424 | ||
425 | s->bs = bs; | |
eabba580 | 426 | QSIMPLEQ_INIT(&s->allocating_write_reqs); |
75411d23 | 427 | |
cf2ab8fc | 428 | ret = bdrv_pread(bs->file, 0, &le_header, sizeof(le_header)); |
75411d23 SH |
429 | if (ret < 0) { |
430 | return ret; | |
431 | } | |
75411d23 SH |
432 | qed_header_le_to_cpu(&le_header, &s->header); |
433 | ||
434 | if (s->header.magic != QED_MAGIC) { | |
76abe407 PB |
435 | error_setg(errp, "Image not in QED format"); |
436 | return -EINVAL; | |
75411d23 SH |
437 | } |
438 | if (s->header.features & ~QED_FEATURE_MASK) { | |
10b758e8 | 439 | /* image uses unsupported feature bits */ |
a55448b3 HR |
440 | error_setg(errp, "Unsupported QED features: %" PRIx64, |
441 | s->header.features & ~QED_FEATURE_MASK); | |
10b758e8 | 442 | return -ENOTSUP; |
75411d23 SH |
443 | } |
444 | if (!qed_is_cluster_size_valid(s->header.cluster_size)) { | |
445 | return -EINVAL; | |
446 | } | |
447 | ||
448 | /* Round down file size to the last cluster */ | |
9a4f4c31 | 449 | file_size = bdrv_getlength(bs->file->bs); |
75411d23 SH |
450 | if (file_size < 0) { |
451 | return file_size; | |
452 | } | |
453 | s->file_size = qed_start_of_cluster(s, file_size); | |
454 | ||
455 | if (!qed_is_table_size_valid(s->header.table_size)) { | |
456 | return -EINVAL; | |
457 | } | |
458 | if (!qed_is_image_size_valid(s->header.image_size, | |
459 | s->header.cluster_size, | |
460 | s->header.table_size)) { | |
461 | return -EINVAL; | |
462 | } | |
463 | if (!qed_check_table_offset(s, s->header.l1_table_offset)) { | |
464 | return -EINVAL; | |
465 | } | |
466 | ||
467 | s->table_nelems = (s->header.cluster_size * s->header.table_size) / | |
468 | sizeof(uint64_t); | |
786a4ea8 | 469 | s->l2_shift = ctz32(s->header.cluster_size); |
75411d23 | 470 | s->l2_mask = s->table_nelems - 1; |
786a4ea8 | 471 | s->l1_shift = s->l2_shift + ctz32(s->table_nelems); |
75411d23 | 472 | |
0adfa1ed SH |
473 | /* Header size calculation must not overflow uint32_t */ |
474 | if (s->header.header_size > UINT32_MAX / s->header.cluster_size) { | |
475 | return -EINVAL; | |
476 | } | |
477 | ||
75411d23 SH |
478 | if ((s->header.features & QED_F_BACKING_FILE)) { |
479 | if ((uint64_t)s->header.backing_filename_offset + | |
480 | s->header.backing_filename_size > | |
481 | s->header.cluster_size * s->header.header_size) { | |
482 | return -EINVAL; | |
483 | } | |
484 | ||
cf2ab8fc | 485 | ret = qed_read_string(bs->file, s->header.backing_filename_offset, |
75411d23 SH |
486 | s->header.backing_filename_size, bs->backing_file, |
487 | sizeof(bs->backing_file)); | |
488 | if (ret < 0) { | |
489 | return ret; | |
490 | } | |
491 | ||
492 | if (s->header.features & QED_F_BACKING_FORMAT_NO_PROBE) { | |
493 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), "raw"); | |
494 | } | |
495 | } | |
496 | ||
497 | /* Reset unknown autoclear feature bits. This is a backwards | |
498 | * compatibility mechanism that allows images to be opened by older | |
499 | * programs, which "knock out" unknown feature bits. When an image is | |
500 | * opened by a newer program again it can detect that the autoclear | |
501 | * feature is no longer valid. | |
502 | */ | |
503 | if ((s->header.autoclear_features & ~QED_AUTOCLEAR_FEATURE_MASK) != 0 && | |
04c01a5c | 504 | !bdrv_is_read_only(bs->file->bs) && !(flags & BDRV_O_INACTIVE)) { |
75411d23 SH |
505 | s->header.autoclear_features &= QED_AUTOCLEAR_FEATURE_MASK; |
506 | ||
507 | ret = qed_write_header_sync(s); | |
508 | if (ret) { | |
509 | return ret; | |
510 | } | |
511 | ||
512 | /* From here on only known autoclear feature bits are valid */ | |
9a4f4c31 | 513 | bdrv_flush(bs->file->bs); |
75411d23 SH |
514 | } |
515 | ||
298800ca SH |
516 | s->l1_table = qed_alloc_table(s); |
517 | qed_init_l2_cache(&s->l2_cache); | |
518 | ||
519 | ret = qed_read_l1_table_sync(s); | |
01979a98 SH |
520 | if (ret) { |
521 | goto out; | |
522 | } | |
523 | ||
524 | /* If image was not closed cleanly, check consistency */ | |
058f8f16 | 525 | if (!(flags & BDRV_O_CHECK) && (s->header.features & QED_F_NEED_CHECK)) { |
01979a98 SH |
526 | /* Read-only images cannot be fixed. There is no risk of corruption |
527 | * since write operations are not possible. Therefore, allow | |
528 | * potentially inconsistent images to be opened read-only. This can | |
529 | * aid data recovery from an otherwise inconsistent image. | |
530 | */ | |
9a4f4c31 | 531 | if (!bdrv_is_read_only(bs->file->bs) && |
04c01a5c | 532 | !(flags & BDRV_O_INACTIVE)) { |
01979a98 SH |
533 | BdrvCheckResult result = {0}; |
534 | ||
535 | ret = qed_check(s, &result, true); | |
6f321e93 SH |
536 | if (ret) { |
537 | goto out; | |
538 | } | |
01979a98 SH |
539 | } |
540 | } | |
541 | ||
a8c868c3 | 542 | bdrv_qed_attach_aio_context(bs, bdrv_get_aio_context(bs)); |
6f321e93 | 543 | |
01979a98 | 544 | out: |
298800ca SH |
545 | if (ret) { |
546 | qed_free_l2_cache(&s->l2_cache); | |
547 | qemu_vfree(s->l1_table); | |
548 | } | |
75411d23 SH |
549 | return ret; |
550 | } | |
551 | ||
4e4bf5c4 KW |
552 | static int bdrv_qed_open(BlockDriverState *bs, QDict *options, int flags, |
553 | Error **errp) | |
554 | { | |
555 | bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, | |
556 | false, errp); | |
557 | if (!bs->file) { | |
558 | return -EINVAL; | |
559 | } | |
560 | ||
561 | return bdrv_qed_do_open(bs, options, flags, errp); | |
562 | } | |
563 | ||
3baca891 | 564 | static void bdrv_qed_refresh_limits(BlockDriverState *bs, Error **errp) |
d34682cd KW |
565 | { |
566 | BDRVQEDState *s = bs->opaque; | |
567 | ||
cf081fca | 568 | bs->bl.pwrite_zeroes_alignment = s->header.cluster_size; |
d34682cd KW |
569 | } |
570 | ||
f9cb20f1 JC |
571 | /* We have nothing to do for QED reopen, stubs just return |
572 | * success */ | |
573 | static int bdrv_qed_reopen_prepare(BDRVReopenState *state, | |
574 | BlockReopenQueue *queue, Error **errp) | |
575 | { | |
576 | return 0; | |
577 | } | |
578 | ||
75411d23 SH |
579 | static void bdrv_qed_close(BlockDriverState *bs) |
580 | { | |
298800ca SH |
581 | BDRVQEDState *s = bs->opaque; |
582 | ||
a8c868c3 | 583 | bdrv_qed_detach_aio_context(bs); |
6f321e93 | 584 | |
01979a98 | 585 | /* Ensure writes reach stable storage */ |
9a4f4c31 | 586 | bdrv_flush(bs->file->bs); |
01979a98 SH |
587 | |
588 | /* Clean shutdown, no check required on next open */ | |
589 | if (s->header.features & QED_F_NEED_CHECK) { | |
590 | s->header.features &= ~QED_F_NEED_CHECK; | |
591 | qed_write_header_sync(s); | |
592 | } | |
593 | ||
298800ca SH |
594 | qed_free_l2_cache(&s->l2_cache); |
595 | qemu_vfree(s->l1_table); | |
75411d23 SH |
596 | } |
597 | ||
75411d23 SH |
598 | static int qed_create(const char *filename, uint32_t cluster_size, |
599 | uint64_t image_size, uint32_t table_size, | |
0fea6b79 | 600 | const char *backing_file, const char *backing_fmt, |
4ab15590 | 601 | QemuOpts *opts, Error **errp) |
75411d23 SH |
602 | { |
603 | QEDHeader header = { | |
604 | .magic = QED_MAGIC, | |
605 | .cluster_size = cluster_size, | |
606 | .table_size = table_size, | |
607 | .header_size = 1, | |
608 | .features = 0, | |
609 | .compat_features = 0, | |
610 | .l1_table_offset = cluster_size, | |
611 | .image_size = image_size, | |
612 | }; | |
613 | QEDHeader le_header; | |
614 | uint8_t *l1_table = NULL; | |
615 | size_t l1_size = header.cluster_size * header.table_size; | |
34b5d2c6 | 616 | Error *local_err = NULL; |
75411d23 | 617 | int ret = 0; |
8a56fdad | 618 | BlockBackend *blk; |
75411d23 | 619 | |
4ab15590 | 620 | ret = bdrv_create_file(filename, opts, &local_err); |
75411d23 | 621 | if (ret < 0) { |
0fea6b79 | 622 | error_propagate(errp, local_err); |
75411d23 SH |
623 | return ret; |
624 | } | |
625 | ||
efaa7c4e | 626 | blk = blk_new_open(filename, NULL, NULL, |
55880601 KW |
627 | BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, |
628 | &local_err); | |
8a56fdad | 629 | if (blk == NULL) { |
0fea6b79 | 630 | error_propagate(errp, local_err); |
8a56fdad | 631 | return -EIO; |
75411d23 SH |
632 | } |
633 | ||
8a56fdad KW |
634 | blk_set_allow_write_beyond_eof(blk, true); |
635 | ||
c743849b | 636 | /* File must start empty and grow, check truncate is supported */ |
ed3d2ec9 | 637 | ret = blk_truncate(blk, 0, errp); |
c743849b SH |
638 | if (ret < 0) { |
639 | goto out; | |
640 | } | |
641 | ||
75411d23 SH |
642 | if (backing_file) { |
643 | header.features |= QED_F_BACKING_FILE; | |
644 | header.backing_filename_offset = sizeof(le_header); | |
645 | header.backing_filename_size = strlen(backing_file); | |
646 | ||
647 | if (qed_fmt_is_raw(backing_fmt)) { | |
648 | header.features |= QED_F_BACKING_FORMAT_NO_PROBE; | |
649 | } | |
650 | } | |
651 | ||
652 | qed_header_cpu_to_le(&header, &le_header); | |
8341f00d | 653 | ret = blk_pwrite(blk, 0, &le_header, sizeof(le_header), 0); |
75411d23 SH |
654 | if (ret < 0) { |
655 | goto out; | |
656 | } | |
8a56fdad | 657 | ret = blk_pwrite(blk, sizeof(le_header), backing_file, |
8341f00d | 658 | header.backing_filename_size, 0); |
75411d23 SH |
659 | if (ret < 0) { |
660 | goto out; | |
661 | } | |
662 | ||
7267c094 | 663 | l1_table = g_malloc0(l1_size); |
8341f00d | 664 | ret = blk_pwrite(blk, header.l1_table_offset, l1_table, l1_size, 0); |
75411d23 SH |
665 | if (ret < 0) { |
666 | goto out; | |
667 | } | |
668 | ||
669 | ret = 0; /* success */ | |
670 | out: | |
7267c094 | 671 | g_free(l1_table); |
8a56fdad | 672 | blk_unref(blk); |
75411d23 SH |
673 | return ret; |
674 | } | |
675 | ||
7ab74849 | 676 | static int bdrv_qed_create(const char *filename, QemuOpts *opts, Error **errp) |
75411d23 SH |
677 | { |
678 | uint64_t image_size = 0; | |
679 | uint32_t cluster_size = QED_DEFAULT_CLUSTER_SIZE; | |
680 | uint32_t table_size = QED_DEFAULT_TABLE_SIZE; | |
7ab74849 CL |
681 | char *backing_file = NULL; |
682 | char *backing_fmt = NULL; | |
683 | int ret; | |
684 | ||
c2eb918e HT |
685 | image_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0), |
686 | BDRV_SECTOR_SIZE); | |
7ab74849 CL |
687 | backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE); |
688 | backing_fmt = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FMT); | |
689 | cluster_size = qemu_opt_get_size_del(opts, | |
690 | BLOCK_OPT_CLUSTER_SIZE, | |
691 | QED_DEFAULT_CLUSTER_SIZE); | |
692 | table_size = qemu_opt_get_size_del(opts, BLOCK_OPT_TABLE_SIZE, | |
693 | QED_DEFAULT_TABLE_SIZE); | |
75411d23 SH |
694 | |
695 | if (!qed_is_cluster_size_valid(cluster_size)) { | |
5ff679b4 AG |
696 | error_setg(errp, "QED cluster size must be within range [%u, %u] " |
697 | "and power of 2", | |
698 | QED_MIN_CLUSTER_SIZE, QED_MAX_CLUSTER_SIZE); | |
7ab74849 CL |
699 | ret = -EINVAL; |
700 | goto finish; | |
75411d23 SH |
701 | } |
702 | if (!qed_is_table_size_valid(table_size)) { | |
5ff679b4 AG |
703 | error_setg(errp, "QED table size must be within range [%u, %u] " |
704 | "and power of 2", | |
705 | QED_MIN_TABLE_SIZE, QED_MAX_TABLE_SIZE); | |
7ab74849 CL |
706 | ret = -EINVAL; |
707 | goto finish; | |
75411d23 SH |
708 | } |
709 | if (!qed_is_image_size_valid(image_size, cluster_size, table_size)) { | |
5ff679b4 AG |
710 | error_setg(errp, "QED image size must be a non-zero multiple of " |
711 | "cluster size and less than %" PRIu64 " bytes", | |
712 | qed_max_image_size(cluster_size, table_size)); | |
7ab74849 CL |
713 | ret = -EINVAL; |
714 | goto finish; | |
75411d23 SH |
715 | } |
716 | ||
7ab74849 | 717 | ret = qed_create(filename, cluster_size, image_size, table_size, |
4ab15590 | 718 | backing_file, backing_fmt, opts, errp); |
7ab74849 CL |
719 | |
720 | finish: | |
721 | g_free(backing_file); | |
722 | g_free(backing_fmt); | |
723 | return ret; | |
75411d23 SH |
724 | } |
725 | ||
298800ca | 726 | typedef struct { |
4bc74be9 | 727 | BlockDriverState *bs; |
b7d5a5b8 | 728 | Coroutine *co; |
4bc74be9 PB |
729 | uint64_t pos; |
730 | int64_t status; | |
298800ca | 731 | int *pnum; |
53f1dfd1 | 732 | BlockDriverState **file; |
298800ca SH |
733 | } QEDIsAllocatedCB; |
734 | ||
735 | static void qed_is_allocated_cb(void *opaque, int ret, uint64_t offset, size_t len) | |
736 | { | |
737 | QEDIsAllocatedCB *cb = opaque; | |
4bc74be9 | 738 | BDRVQEDState *s = cb->bs->opaque; |
298800ca | 739 | *cb->pnum = len / BDRV_SECTOR_SIZE; |
4bc74be9 PB |
740 | switch (ret) { |
741 | case QED_CLUSTER_FOUND: | |
742 | offset |= qed_offset_into_cluster(s, cb->pos); | |
743 | cb->status = BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset; | |
53f1dfd1 | 744 | *cb->file = cb->bs->file->bs; |
4bc74be9 PB |
745 | break; |
746 | case QED_CLUSTER_ZERO: | |
747 | cb->status = BDRV_BLOCK_ZERO; | |
748 | break; | |
749 | case QED_CLUSTER_L2: | |
750 | case QED_CLUSTER_L1: | |
751 | cb->status = 0; | |
752 | break; | |
753 | default: | |
754 | assert(ret < 0); | |
755 | cb->status = ret; | |
756 | break; | |
757 | } | |
758 | ||
b7d5a5b8 | 759 | if (cb->co) { |
b9e413dd | 760 | aio_co_wake(cb->co); |
b7d5a5b8 | 761 | } |
298800ca SH |
762 | } |
763 | ||
b6b8a333 | 764 | static int64_t coroutine_fn bdrv_qed_co_get_block_status(BlockDriverState *bs, |
b7d5a5b8 | 765 | int64_t sector_num, |
67a0fd2a FZ |
766 | int nb_sectors, int *pnum, |
767 | BlockDriverState **file) | |
75411d23 | 768 | { |
298800ca | 769 | BDRVQEDState *s = bs->opaque; |
298800ca SH |
770 | size_t len = (size_t)nb_sectors * BDRV_SECTOR_SIZE; |
771 | QEDIsAllocatedCB cb = { | |
4bc74be9 PB |
772 | .bs = bs, |
773 | .pos = (uint64_t)sector_num * BDRV_SECTOR_SIZE, | |
774 | .status = BDRV_BLOCK_OFFSET_MASK, | |
298800ca | 775 | .pnum = pnum, |
53f1dfd1 | 776 | .file = file, |
298800ca SH |
777 | }; |
778 | QEDRequest request = { .l2_table = NULL }; | |
779 | ||
4bc74be9 | 780 | qed_find_cluster(s, &request, cb.pos, len, qed_is_allocated_cb, &cb); |
298800ca | 781 | |
b7d5a5b8 | 782 | /* Now sleep if the callback wasn't invoked immediately */ |
4bc74be9 | 783 | while (cb.status == BDRV_BLOCK_OFFSET_MASK) { |
b7d5a5b8 SH |
784 | cb.co = qemu_coroutine_self(); |
785 | qemu_coroutine_yield(); | |
298800ca SH |
786 | } |
787 | ||
298800ca SH |
788 | qed_unref_l2_cache_entry(request.l2_table); |
789 | ||
4bc74be9 | 790 | return cb.status; |
75411d23 SH |
791 | } |
792 | ||
eabba580 SH |
793 | static BDRVQEDState *acb_to_s(QEDAIOCB *acb) |
794 | { | |
795 | return acb->common.bs->opaque; | |
796 | } | |
797 | ||
798 | /** | |
799 | * Read from the backing file or zero-fill if no backing file | |
800 | * | |
f06ee3d4 KW |
801 | * @s: QED state |
802 | * @pos: Byte position in device | |
803 | * @qiov: Destination I/O vector | |
804 | * @backing_qiov: Possibly shortened copy of qiov, to be allocated here | |
805 | * @cb: Completion function | |
806 | * @opaque: User data for completion function | |
eabba580 SH |
807 | * |
808 | * This function reads qiov->size bytes starting at pos from the backing file. | |
809 | * If there is no backing file then zeroes are read. | |
810 | */ | |
811 | static void qed_read_backing_file(BDRVQEDState *s, uint64_t pos, | |
812 | QEMUIOVector *qiov, | |
f06ee3d4 | 813 | QEMUIOVector **backing_qiov, |
097310b5 | 814 | BlockCompletionFunc *cb, void *opaque) |
eabba580 | 815 | { |
eabba580 SH |
816 | uint64_t backing_length = 0; |
817 | size_t size; | |
818 | ||
819 | /* If there is a backing file, get its length. Treat the absence of a | |
820 | * backing file like a zero length backing file. | |
821 | */ | |
760e0063 KW |
822 | if (s->bs->backing) { |
823 | int64_t l = bdrv_getlength(s->bs->backing->bs); | |
eabba580 SH |
824 | if (l < 0) { |
825 | cb(opaque, l); | |
826 | return; | |
827 | } | |
828 | backing_length = l; | |
829 | } | |
830 | ||
831 | /* Zero all sectors if reading beyond the end of the backing file */ | |
832 | if (pos >= backing_length || | |
833 | pos + qiov->size > backing_length) { | |
3d9b4925 | 834 | qemu_iovec_memset(qiov, 0, 0, qiov->size); |
eabba580 SH |
835 | } |
836 | ||
837 | /* Complete now if there are no backing file sectors to read */ | |
838 | if (pos >= backing_length) { | |
839 | cb(opaque, 0); | |
840 | return; | |
841 | } | |
842 | ||
843 | /* If the read straddles the end of the backing file, shorten it */ | |
844 | size = MIN((uint64_t)backing_length - pos, qiov->size); | |
845 | ||
f06ee3d4 KW |
846 | assert(*backing_qiov == NULL); |
847 | *backing_qiov = g_new(QEMUIOVector, 1); | |
848 | qemu_iovec_init(*backing_qiov, qiov->niov); | |
849 | qemu_iovec_concat(*backing_qiov, qiov, 0, size); | |
850 | ||
820100fd | 851 | BLKDBG_EVENT(s->bs->file, BLKDBG_READ_BACKING_AIO); |
ebb7af21 | 852 | bdrv_aio_readv(s->bs->backing, pos / BDRV_SECTOR_SIZE, |
f06ee3d4 | 853 | *backing_qiov, size / BDRV_SECTOR_SIZE, cb, opaque); |
eabba580 SH |
854 | } |
855 | ||
856 | typedef struct { | |
857 | GenericCB gencb; | |
858 | BDRVQEDState *s; | |
859 | QEMUIOVector qiov; | |
f06ee3d4 | 860 | QEMUIOVector *backing_qiov; |
eabba580 SH |
861 | struct iovec iov; |
862 | uint64_t offset; | |
863 | } CopyFromBackingFileCB; | |
864 | ||
865 | static void qed_copy_from_backing_file_cb(void *opaque, int ret) | |
866 | { | |
867 | CopyFromBackingFileCB *copy_cb = opaque; | |
868 | qemu_vfree(copy_cb->iov.iov_base); | |
869 | gencb_complete(©_cb->gencb, ret); | |
870 | } | |
871 | ||
872 | static void qed_copy_from_backing_file_write(void *opaque, int ret) | |
873 | { | |
874 | CopyFromBackingFileCB *copy_cb = opaque; | |
875 | BDRVQEDState *s = copy_cb->s; | |
eabba580 | 876 | |
f06ee3d4 KW |
877 | if (copy_cb->backing_qiov) { |
878 | qemu_iovec_destroy(copy_cb->backing_qiov); | |
879 | g_free(copy_cb->backing_qiov); | |
880 | copy_cb->backing_qiov = NULL; | |
881 | } | |
882 | ||
eabba580 SH |
883 | if (ret) { |
884 | qed_copy_from_backing_file_cb(copy_cb, ret); | |
885 | return; | |
886 | } | |
887 | ||
888 | BLKDBG_EVENT(s->bs->file, BLKDBG_COW_WRITE); | |
0d1049c7 | 889 | bdrv_aio_writev(s->bs->file, copy_cb->offset / BDRV_SECTOR_SIZE, |
ad54ae80 PB |
890 | ©_cb->qiov, copy_cb->qiov.size / BDRV_SECTOR_SIZE, |
891 | qed_copy_from_backing_file_cb, copy_cb); | |
eabba580 SH |
892 | } |
893 | ||
894 | /** | |
895 | * Copy data from backing file into the image | |
896 | * | |
897 | * @s: QED state | |
898 | * @pos: Byte position in device | |
899 | * @len: Number of bytes | |
900 | * @offset: Byte offset in image file | |
901 | * @cb: Completion function | |
902 | * @opaque: User data for completion function | |
903 | */ | |
904 | static void qed_copy_from_backing_file(BDRVQEDState *s, uint64_t pos, | |
905 | uint64_t len, uint64_t offset, | |
097310b5 | 906 | BlockCompletionFunc *cb, |
eabba580 SH |
907 | void *opaque) |
908 | { | |
909 | CopyFromBackingFileCB *copy_cb; | |
910 | ||
911 | /* Skip copy entirely if there is no work to do */ | |
912 | if (len == 0) { | |
913 | cb(opaque, 0); | |
914 | return; | |
915 | } | |
916 | ||
917 | copy_cb = gencb_alloc(sizeof(*copy_cb), cb, opaque); | |
918 | copy_cb->s = s; | |
919 | copy_cb->offset = offset; | |
f06ee3d4 | 920 | copy_cb->backing_qiov = NULL; |
eabba580 SH |
921 | copy_cb->iov.iov_base = qemu_blockalign(s->bs, len); |
922 | copy_cb->iov.iov_len = len; | |
923 | qemu_iovec_init_external(©_cb->qiov, ©_cb->iov, 1); | |
924 | ||
f06ee3d4 | 925 | qed_read_backing_file(s, pos, ©_cb->qiov, ©_cb->backing_qiov, |
eabba580 SH |
926 | qed_copy_from_backing_file_write, copy_cb); |
927 | } | |
928 | ||
929 | /** | |
930 | * Link one or more contiguous clusters into a table | |
931 | * | |
932 | * @s: QED state | |
933 | * @table: L2 table | |
934 | * @index: First cluster index | |
935 | * @n: Number of contiguous clusters | |
21df65b6 AL |
936 | * @cluster: First cluster offset |
937 | * | |
938 | * The cluster offset may be an allocated byte offset in the image file, the | |
939 | * zero cluster marker, or the unallocated cluster marker. | |
eabba580 SH |
940 | */ |
941 | static void qed_update_l2_table(BDRVQEDState *s, QEDTable *table, int index, | |
942 | unsigned int n, uint64_t cluster) | |
943 | { | |
944 | int i; | |
945 | for (i = index; i < index + n; i++) { | |
946 | table->offsets[i] = cluster; | |
21df65b6 AL |
947 | if (!qed_offset_is_unalloc_cluster(cluster) && |
948 | !qed_offset_is_zero_cluster(cluster)) { | |
949 | cluster += s->header.cluster_size; | |
950 | } | |
eabba580 SH |
951 | } |
952 | } | |
953 | ||
954 | static void qed_aio_complete_bh(void *opaque) | |
955 | { | |
956 | QEDAIOCB *acb = opaque; | |
1919631e | 957 | BDRVQEDState *s = acb_to_s(acb); |
097310b5 | 958 | BlockCompletionFunc *cb = acb->common.cb; |
eabba580 SH |
959 | void *user_opaque = acb->common.opaque; |
960 | int ret = acb->bh_ret; | |
eabba580 | 961 | |
8007429a | 962 | qemu_aio_unref(acb); |
eabba580 SH |
963 | |
964 | /* Invoke callback */ | |
1919631e | 965 | qed_acquire(s); |
eabba580 | 966 | cb(user_opaque, ret); |
1919631e | 967 | qed_release(s); |
eabba580 SH |
968 | } |
969 | ||
3b7cd9fd KW |
970 | static void qed_resume_alloc_bh(void *opaque) |
971 | { | |
972 | qed_aio_start_io(opaque); | |
973 | } | |
974 | ||
eabba580 SH |
975 | static void qed_aio_complete(QEDAIOCB *acb, int ret) |
976 | { | |
977 | BDRVQEDState *s = acb_to_s(acb); | |
978 | ||
979 | trace_qed_aio_complete(s, acb, ret); | |
980 | ||
981 | /* Free resources */ | |
982 | qemu_iovec_destroy(&acb->cur_qiov); | |
983 | qed_unref_l2_cache_entry(acb->request.l2_table); | |
984 | ||
0e71be19 SH |
985 | /* Free the buffer we may have allocated for zero writes */ |
986 | if (acb->flags & QED_AIOCB_ZERO) { | |
987 | qemu_vfree(acb->qiov->iov[0].iov_base); | |
988 | acb->qiov->iov[0].iov_base = NULL; | |
989 | } | |
990 | ||
eabba580 SH |
991 | /* Arrange for a bh to invoke the completion function */ |
992 | acb->bh_ret = ret; | |
fffb6e12 PB |
993 | aio_bh_schedule_oneshot(bdrv_get_aio_context(acb->common.bs), |
994 | qed_aio_complete_bh, acb); | |
eabba580 SH |
995 | |
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 | */ | |
1002 | if (acb == QSIMPLEQ_FIRST(&s->allocating_write_reqs)) { | |
3b7cd9fd | 1003 | QEDAIOCB *next_acb; |
eabba580 | 1004 | QSIMPLEQ_REMOVE_HEAD(&s->allocating_write_reqs, next); |
3b7cd9fd KW |
1005 | next_acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs); |
1006 | if (next_acb) { | |
1007 | aio_bh_schedule_oneshot(bdrv_get_aio_context(acb->common.bs), | |
1008 | qed_resume_alloc_bh, next_acb); | |
6f321e93 SH |
1009 | } else if (s->header.features & QED_F_NEED_CHECK) { |
1010 | qed_start_need_check_timer(s); | |
eabba580 SH |
1011 | } |
1012 | } | |
1013 | } | |
1014 | ||
1015 | /** | |
1016 | * Commit the current L2 table to the cache | |
1017 | */ | |
1018 | static void qed_commit_l2_update(void *opaque, int ret) | |
1019 | { | |
1020 | QEDAIOCB *acb = opaque; | |
1021 | BDRVQEDState *s = acb_to_s(acb); | |
1022 | CachedL2Table *l2_table = acb->request.l2_table; | |
e4fc8781 | 1023 | uint64_t l2_offset = l2_table->offset; |
eabba580 SH |
1024 | |
1025 | qed_commit_l2_cache_entry(&s->l2_cache, l2_table); | |
1026 | ||
1027 | /* This is guaranteed to succeed because we just committed the entry to the | |
1028 | * cache. | |
1029 | */ | |
e4fc8781 | 1030 | acb->request.l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset); |
eabba580 SH |
1031 | assert(acb->request.l2_table != NULL); |
1032 | ||
b20123a2 | 1033 | qed_aio_next_io(acb, ret); |
eabba580 SH |
1034 | } |
1035 | ||
1036 | /** | |
1037 | * Update L1 table with new L2 table offset and write it out | |
1038 | */ | |
1039 | static void qed_aio_write_l1_update(void *opaque, int ret) | |
1040 | { | |
1041 | QEDAIOCB *acb = opaque; | |
1042 | BDRVQEDState *s = acb_to_s(acb); | |
1043 | int index; | |
1044 | ||
1045 | if (ret) { | |
1046 | qed_aio_complete(acb, ret); | |
1047 | return; | |
1048 | } | |
1049 | ||
1050 | index = qed_l1_index(s, acb->cur_pos); | |
1051 | s->l1_table->offsets[index] = acb->request.l2_table->offset; | |
1052 | ||
1053 | qed_write_l1_table(s, index, 1, qed_commit_l2_update, acb); | |
1054 | } | |
1055 | ||
1056 | /** | |
1057 | * Update L2 table with new cluster offsets and write them out | |
1058 | */ | |
0e71be19 | 1059 | static void qed_aio_write_l2_update(QEDAIOCB *acb, int ret, uint64_t offset) |
eabba580 | 1060 | { |
eabba580 SH |
1061 | BDRVQEDState *s = acb_to_s(acb); |
1062 | bool need_alloc = acb->find_cluster_ret == QED_CLUSTER_L1; | |
1063 | int index; | |
1064 | ||
1065 | if (ret) { | |
1066 | goto err; | |
1067 | } | |
1068 | ||
1069 | if (need_alloc) { | |
1070 | qed_unref_l2_cache_entry(acb->request.l2_table); | |
1071 | acb->request.l2_table = qed_new_l2_table(s); | |
1072 | } | |
1073 | ||
1074 | index = qed_l2_index(s, acb->cur_pos); | |
1075 | qed_update_l2_table(s, acb->request.l2_table->table, index, acb->cur_nclusters, | |
0e71be19 | 1076 | offset); |
eabba580 SH |
1077 | |
1078 | if (need_alloc) { | |
1079 | /* Write out the whole new L2 table */ | |
1080 | qed_write_l2_table(s, &acb->request, 0, s->table_nelems, true, | |
b20123a2 | 1081 | qed_aio_write_l1_update, acb); |
eabba580 SH |
1082 | } else { |
1083 | /* Write out only the updated part of the L2 table */ | |
1084 | qed_write_l2_table(s, &acb->request, index, acb->cur_nclusters, false, | |
b20123a2 | 1085 | qed_aio_next_io_cb, acb); |
eabba580 SH |
1086 | } |
1087 | return; | |
1088 | ||
1089 | err: | |
1090 | qed_aio_complete(acb, ret); | |
1091 | } | |
1092 | ||
0e71be19 SH |
1093 | static void qed_aio_write_l2_update_cb(void *opaque, int ret) |
1094 | { | |
1095 | QEDAIOCB *acb = opaque; | |
1096 | qed_aio_write_l2_update(acb, ret, acb->cur_cluster); | |
1097 | } | |
1098 | ||
eabba580 SH |
1099 | /** |
1100 | * Flush new data clusters before updating the L2 table | |
1101 | * | |
1102 | * This flush is necessary when a backing file is in use. A crash during an | |
1103 | * allocating write could result in empty clusters in the image. If the write | |
1104 | * only touched a subregion of the cluster, then backing image sectors have | |
1105 | * been lost in the untouched region. The solution is to flush after writing a | |
1106 | * new data cluster and before updating the L2 table. | |
1107 | */ | |
1108 | static void qed_aio_write_flush_before_l2_update(void *opaque, int ret) | |
1109 | { | |
1110 | QEDAIOCB *acb = opaque; | |
1111 | BDRVQEDState *s = acb_to_s(acb); | |
1112 | ||
9a4f4c31 | 1113 | if (!bdrv_aio_flush(s->bs->file->bs, qed_aio_write_l2_update_cb, opaque)) { |
eabba580 SH |
1114 | qed_aio_complete(acb, -EIO); |
1115 | } | |
1116 | } | |
1117 | ||
1118 | /** | |
1119 | * Write data to the image file | |
1120 | */ | |
1121 | static void qed_aio_write_main(void *opaque, int ret) | |
1122 | { | |
1123 | QEDAIOCB *acb = opaque; | |
1124 | BDRVQEDState *s = acb_to_s(acb); | |
1125 | uint64_t offset = acb->cur_cluster + | |
1126 | qed_offset_into_cluster(s, acb->cur_pos); | |
097310b5 | 1127 | BlockCompletionFunc *next_fn; |
eabba580 SH |
1128 | |
1129 | trace_qed_aio_write_main(s, acb, ret, offset, acb->cur_qiov.size); | |
1130 | ||
1131 | if (ret) { | |
1132 | qed_aio_complete(acb, ret); | |
1133 | return; | |
1134 | } | |
1135 | ||
1136 | if (acb->find_cluster_ret == QED_CLUSTER_FOUND) { | |
b20123a2 | 1137 | next_fn = qed_aio_next_io_cb; |
eabba580 | 1138 | } else { |
760e0063 | 1139 | if (s->bs->backing) { |
eabba580 SH |
1140 | next_fn = qed_aio_write_flush_before_l2_update; |
1141 | } else { | |
0e71be19 | 1142 | next_fn = qed_aio_write_l2_update_cb; |
eabba580 SH |
1143 | } |
1144 | } | |
1145 | ||
1146 | BLKDBG_EVENT(s->bs->file, BLKDBG_WRITE_AIO); | |
0d1049c7 | 1147 | bdrv_aio_writev(s->bs->file, offset / BDRV_SECTOR_SIZE, |
ad54ae80 PB |
1148 | &acb->cur_qiov, acb->cur_qiov.size / BDRV_SECTOR_SIZE, |
1149 | next_fn, acb); | |
eabba580 SH |
1150 | } |
1151 | ||
1152 | /** | |
1153 | * Populate back untouched region of new data cluster | |
1154 | */ | |
1155 | static void qed_aio_write_postfill(void *opaque, int ret) | |
1156 | { | |
1157 | QEDAIOCB *acb = opaque; | |
1158 | BDRVQEDState *s = acb_to_s(acb); | |
1159 | uint64_t start = acb->cur_pos + acb->cur_qiov.size; | |
1160 | uint64_t len = | |
1161 | qed_start_of_cluster(s, start + s->header.cluster_size - 1) - start; | |
1162 | uint64_t offset = acb->cur_cluster + | |
1163 | qed_offset_into_cluster(s, acb->cur_pos) + | |
1164 | acb->cur_qiov.size; | |
1165 | ||
1166 | if (ret) { | |
1167 | qed_aio_complete(acb, ret); | |
1168 | return; | |
1169 | } | |
1170 | ||
1171 | trace_qed_aio_write_postfill(s, acb, start, len, offset); | |
1172 | qed_copy_from_backing_file(s, start, len, offset, | |
1173 | qed_aio_write_main, acb); | |
1174 | } | |
1175 | ||
1176 | /** | |
1177 | * Populate front untouched region of new data cluster | |
1178 | */ | |
1179 | static void qed_aio_write_prefill(void *opaque, int ret) | |
1180 | { | |
1181 | QEDAIOCB *acb = opaque; | |
1182 | BDRVQEDState *s = acb_to_s(acb); | |
1183 | uint64_t start = qed_start_of_cluster(s, acb->cur_pos); | |
1184 | uint64_t len = qed_offset_into_cluster(s, acb->cur_pos); | |
1185 | ||
1186 | trace_qed_aio_write_prefill(s, acb, start, len, acb->cur_cluster); | |
1187 | qed_copy_from_backing_file(s, start, len, acb->cur_cluster, | |
1188 | qed_aio_write_postfill, acb); | |
1189 | } | |
1190 | ||
0d09c797 SH |
1191 | /** |
1192 | * Check if the QED_F_NEED_CHECK bit should be set during allocating write | |
1193 | */ | |
1194 | static bool qed_should_set_need_check(BDRVQEDState *s) | |
1195 | { | |
1196 | /* The flush before L2 update path ensures consistency */ | |
760e0063 | 1197 | if (s->bs->backing) { |
0d09c797 SH |
1198 | return false; |
1199 | } | |
1200 | ||
1201 | return !(s->header.features & QED_F_NEED_CHECK); | |
1202 | } | |
1203 | ||
0e71be19 SH |
1204 | static void qed_aio_write_zero_cluster(void *opaque, int ret) |
1205 | { | |
1206 | QEDAIOCB *acb = opaque; | |
1207 | ||
1208 | if (ret) { | |
1209 | qed_aio_complete(acb, ret); | |
1210 | return; | |
1211 | } | |
1212 | ||
1213 | qed_aio_write_l2_update(acb, 0, 1); | |
1214 | } | |
1215 | ||
eabba580 SH |
1216 | /** |
1217 | * Write new data cluster | |
1218 | * | |
1219 | * @acb: Write request | |
1220 | * @len: Length in bytes | |
1221 | * | |
1222 | * This path is taken when writing to previously unallocated clusters. | |
1223 | */ | |
1224 | static void qed_aio_write_alloc(QEDAIOCB *acb, size_t len) | |
1225 | { | |
1226 | BDRVQEDState *s = acb_to_s(acb); | |
097310b5 | 1227 | BlockCompletionFunc *cb; |
eabba580 | 1228 | |
6f321e93 SH |
1229 | /* Cancel timer when the first allocating request comes in */ |
1230 | if (QSIMPLEQ_EMPTY(&s->allocating_write_reqs)) { | |
1231 | qed_cancel_need_check_timer(s); | |
1232 | } | |
1233 | ||
eabba580 SH |
1234 | /* Freeze this request if another allocating write is in progress */ |
1235 | if (acb != QSIMPLEQ_FIRST(&s->allocating_write_reqs)) { | |
1236 | QSIMPLEQ_INSERT_TAIL(&s->allocating_write_reqs, acb, next); | |
1237 | } | |
6f321e93 SH |
1238 | if (acb != QSIMPLEQ_FIRST(&s->allocating_write_reqs) || |
1239 | s->allocating_write_reqs_plugged) { | |
eabba580 SH |
1240 | return; /* wait for existing request to finish */ |
1241 | } | |
1242 | ||
1243 | acb->cur_nclusters = qed_bytes_to_clusters(s, | |
1244 | qed_offset_into_cluster(s, acb->cur_pos) + len); | |
1b093c48 | 1245 | qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
eabba580 | 1246 | |
0e71be19 SH |
1247 | if (acb->flags & QED_AIOCB_ZERO) { |
1248 | /* Skip ahead if the clusters are already zero */ | |
1249 | if (acb->find_cluster_ret == QED_CLUSTER_ZERO) { | |
b20123a2 | 1250 | qed_aio_start_io(acb); |
0e71be19 SH |
1251 | return; |
1252 | } | |
1253 | ||
1254 | cb = qed_aio_write_zero_cluster; | |
1255 | } else { | |
1256 | cb = qed_aio_write_prefill; | |
1257 | acb->cur_cluster = qed_alloc_clusters(s, acb->cur_nclusters); | |
1258 | } | |
1259 | ||
0d09c797 SH |
1260 | if (qed_should_set_need_check(s)) { |
1261 | s->header.features |= QED_F_NEED_CHECK; | |
0e71be19 | 1262 | qed_write_header(s, cb, acb); |
0d09c797 | 1263 | } else { |
0e71be19 | 1264 | cb(acb, 0); |
01979a98 | 1265 | } |
eabba580 SH |
1266 | } |
1267 | ||
1268 | /** | |
1269 | * Write data cluster in place | |
1270 | * | |
1271 | * @acb: Write request | |
1272 | * @offset: Cluster offset in bytes | |
1273 | * @len: Length in bytes | |
1274 | * | |
1275 | * This path is taken when writing to already allocated clusters. | |
1276 | */ | |
1277 | static void qed_aio_write_inplace(QEDAIOCB *acb, uint64_t offset, size_t len) | |
1278 | { | |
0e71be19 SH |
1279 | /* Allocate buffer for zero writes */ |
1280 | if (acb->flags & QED_AIOCB_ZERO) { | |
1281 | struct iovec *iov = acb->qiov->iov; | |
1282 | ||
1283 | if (!iov->iov_base) { | |
4f4896db KW |
1284 | iov->iov_base = qemu_try_blockalign(acb->common.bs, iov->iov_len); |
1285 | if (iov->iov_base == NULL) { | |
1286 | qed_aio_complete(acb, -ENOMEM); | |
1287 | return; | |
1288 | } | |
0e71be19 SH |
1289 | memset(iov->iov_base, 0, iov->iov_len); |
1290 | } | |
1291 | } | |
1292 | ||
eabba580 SH |
1293 | /* Calculate the I/O vector */ |
1294 | acb->cur_cluster = offset; | |
1b093c48 | 1295 | qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
eabba580 SH |
1296 | |
1297 | /* Do the actual write */ | |
1298 | qed_aio_write_main(acb, 0); | |
1299 | } | |
1300 | ||
1301 | /** | |
1302 | * Write data cluster | |
1303 | * | |
1304 | * @opaque: Write request | |
1305 | * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1, | |
1306 | * or -errno | |
1307 | * @offset: Cluster offset in bytes | |
1308 | * @len: Length in bytes | |
1309 | * | |
1310 | * Callback from qed_find_cluster(). | |
1311 | */ | |
1312 | static void qed_aio_write_data(void *opaque, int ret, | |
1313 | uint64_t offset, size_t len) | |
1314 | { | |
1315 | QEDAIOCB *acb = opaque; | |
1316 | ||
1317 | trace_qed_aio_write_data(acb_to_s(acb), acb, ret, offset, len); | |
1318 | ||
1319 | acb->find_cluster_ret = ret; | |
1320 | ||
1321 | switch (ret) { | |
1322 | case QED_CLUSTER_FOUND: | |
1323 | qed_aio_write_inplace(acb, offset, len); | |
1324 | break; | |
1325 | ||
1326 | case QED_CLUSTER_L2: | |
1327 | case QED_CLUSTER_L1: | |
21df65b6 | 1328 | case QED_CLUSTER_ZERO: |
eabba580 SH |
1329 | qed_aio_write_alloc(acb, len); |
1330 | break; | |
1331 | ||
1332 | default: | |
1333 | qed_aio_complete(acb, ret); | |
1334 | break; | |
1335 | } | |
1336 | } | |
1337 | ||
1338 | /** | |
1339 | * Read data cluster | |
1340 | * | |
1341 | * @opaque: Read request | |
1342 | * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1, | |
1343 | * or -errno | |
1344 | * @offset: Cluster offset in bytes | |
1345 | * @len: Length in bytes | |
1346 | * | |
1347 | * Callback from qed_find_cluster(). | |
1348 | */ | |
1349 | static void qed_aio_read_data(void *opaque, int ret, | |
1350 | uint64_t offset, size_t len) | |
1351 | { | |
1352 | QEDAIOCB *acb = opaque; | |
1353 | BDRVQEDState *s = acb_to_s(acb); | |
1354 | BlockDriverState *bs = acb->common.bs; | |
eabba580 SH |
1355 | |
1356 | /* Adjust offset into cluster */ | |
1357 | offset += qed_offset_into_cluster(s, acb->cur_pos); | |
1358 | ||
1359 | trace_qed_aio_read_data(s, acb, ret, offset, len); | |
1360 | ||
1361 | if (ret < 0) { | |
1362 | goto err; | |
1363 | } | |
1364 | ||
1b093c48 | 1365 | qemu_iovec_concat(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); |
eabba580 | 1366 | |
21df65b6 AL |
1367 | /* Handle zero cluster and backing file reads */ |
1368 | if (ret == QED_CLUSTER_ZERO) { | |
3d9b4925 | 1369 | qemu_iovec_memset(&acb->cur_qiov, 0, 0, acb->cur_qiov.size); |
b20123a2 | 1370 | qed_aio_start_io(acb); |
21df65b6 AL |
1371 | return; |
1372 | } else if (ret != QED_CLUSTER_FOUND) { | |
eabba580 | 1373 | qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov, |
b20123a2 | 1374 | &acb->backing_qiov, qed_aio_next_io_cb, acb); |
eabba580 SH |
1375 | return; |
1376 | } | |
1377 | ||
1378 | BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); | |
ebb7af21 | 1379 | bdrv_aio_readv(bs->file, offset / BDRV_SECTOR_SIZE, |
ad54ae80 | 1380 | &acb->cur_qiov, acb->cur_qiov.size / BDRV_SECTOR_SIZE, |
b20123a2 | 1381 | qed_aio_next_io_cb, acb); |
eabba580 SH |
1382 | return; |
1383 | ||
1384 | err: | |
1385 | qed_aio_complete(acb, ret); | |
1386 | } | |
1387 | ||
1388 | /** | |
1389 | * Begin next I/O or complete the request | |
1390 | */ | |
b20123a2 | 1391 | static void qed_aio_next_io(QEDAIOCB *acb, int ret) |
eabba580 | 1392 | { |
eabba580 | 1393 | BDRVQEDState *s = acb_to_s(acb); |
6e4f59bd SH |
1394 | QEDFindClusterFunc *io_fn = (acb->flags & QED_AIOCB_WRITE) ? |
1395 | qed_aio_write_data : qed_aio_read_data; | |
eabba580 SH |
1396 | |
1397 | trace_qed_aio_next_io(s, acb, ret, acb->cur_pos + acb->cur_qiov.size); | |
1398 | ||
f06ee3d4 KW |
1399 | if (acb->backing_qiov) { |
1400 | qemu_iovec_destroy(acb->backing_qiov); | |
1401 | g_free(acb->backing_qiov); | |
1402 | acb->backing_qiov = NULL; | |
1403 | } | |
1404 | ||
eabba580 SH |
1405 | /* Handle I/O error */ |
1406 | if (ret) { | |
1407 | qed_aio_complete(acb, ret); | |
1408 | return; | |
1409 | } | |
1410 | ||
1411 | acb->qiov_offset += acb->cur_qiov.size; | |
1412 | acb->cur_pos += acb->cur_qiov.size; | |
1413 | qemu_iovec_reset(&acb->cur_qiov); | |
1414 | ||
1415 | /* Complete request */ | |
1416 | if (acb->cur_pos >= acb->end_pos) { | |
1417 | qed_aio_complete(acb, 0); | |
1418 | return; | |
1419 | } | |
1420 | ||
1421 | /* Find next cluster and start I/O */ | |
1422 | qed_find_cluster(s, &acb->request, | |
1423 | acb->cur_pos, acb->end_pos - acb->cur_pos, | |
1424 | io_fn, acb); | |
1425 | } | |
1426 | ||
7c84b1b8 MA |
1427 | static BlockAIOCB *qed_aio_setup(BlockDriverState *bs, |
1428 | int64_t sector_num, | |
1429 | QEMUIOVector *qiov, int nb_sectors, | |
097310b5 | 1430 | BlockCompletionFunc *cb, |
7c84b1b8 | 1431 | void *opaque, int flags) |
eabba580 | 1432 | { |
d7331bed | 1433 | QEDAIOCB *acb = qemu_aio_get(&qed_aiocb_info, bs, cb, opaque); |
eabba580 SH |
1434 | |
1435 | trace_qed_aio_setup(bs->opaque, acb, sector_num, nb_sectors, | |
6e4f59bd | 1436 | opaque, flags); |
eabba580 | 1437 | |
6e4f59bd | 1438 | acb->flags = flags; |
eabba580 SH |
1439 | acb->qiov = qiov; |
1440 | acb->qiov_offset = 0; | |
1441 | acb->cur_pos = (uint64_t)sector_num * BDRV_SECTOR_SIZE; | |
1442 | acb->end_pos = acb->cur_pos + nb_sectors * BDRV_SECTOR_SIZE; | |
f06ee3d4 | 1443 | acb->backing_qiov = NULL; |
eabba580 SH |
1444 | acb->request.l2_table = NULL; |
1445 | qemu_iovec_init(&acb->cur_qiov, qiov->niov); | |
1446 | ||
1447 | /* Start request */ | |
b20123a2 | 1448 | qed_aio_start_io(acb); |
eabba580 SH |
1449 | return &acb->common; |
1450 | } | |
1451 | ||
7c84b1b8 MA |
1452 | static BlockAIOCB *bdrv_qed_aio_readv(BlockDriverState *bs, |
1453 | int64_t sector_num, | |
1454 | QEMUIOVector *qiov, int nb_sectors, | |
097310b5 | 1455 | BlockCompletionFunc *cb, |
7c84b1b8 | 1456 | void *opaque) |
75411d23 | 1457 | { |
6e4f59bd | 1458 | return qed_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); |
75411d23 SH |
1459 | } |
1460 | ||
7c84b1b8 MA |
1461 | static BlockAIOCB *bdrv_qed_aio_writev(BlockDriverState *bs, |
1462 | int64_t sector_num, | |
1463 | QEMUIOVector *qiov, int nb_sectors, | |
097310b5 | 1464 | BlockCompletionFunc *cb, |
7c84b1b8 | 1465 | void *opaque) |
75411d23 | 1466 | { |
6e4f59bd SH |
1467 | return qed_aio_setup(bs, sector_num, qiov, nb_sectors, cb, |
1468 | opaque, QED_AIOCB_WRITE); | |
75411d23 SH |
1469 | } |
1470 | ||
0e71be19 SH |
1471 | typedef struct { |
1472 | Coroutine *co; | |
1473 | int ret; | |
1474 | bool done; | |
1475 | } QEDWriteZeroesCB; | |
1476 | ||
49a2e483 | 1477 | static void coroutine_fn qed_co_pwrite_zeroes_cb(void *opaque, int ret) |
0e71be19 SH |
1478 | { |
1479 | QEDWriteZeroesCB *cb = opaque; | |
1480 | ||
1481 | cb->done = true; | |
1482 | cb->ret = ret; | |
1483 | if (cb->co) { | |
b9e413dd | 1484 | aio_co_wake(cb->co); |
0e71be19 SH |
1485 | } |
1486 | } | |
1487 | ||
49a2e483 EB |
1488 | static int coroutine_fn bdrv_qed_co_pwrite_zeroes(BlockDriverState *bs, |
1489 | int64_t offset, | |
1490 | int count, | |
1491 | BdrvRequestFlags flags) | |
0e71be19 | 1492 | { |
7c84b1b8 | 1493 | BlockAIOCB *blockacb; |
ef72f76e | 1494 | BDRVQEDState *s = bs->opaque; |
0e71be19 SH |
1495 | QEDWriteZeroesCB cb = { .done = false }; |
1496 | QEMUIOVector qiov; | |
1497 | struct iovec iov; | |
1498 | ||
49a2e483 EB |
1499 | /* Fall back if the request is not aligned */ |
1500 | if (qed_offset_into_cluster(s, offset) || | |
1501 | qed_offset_into_cluster(s, count)) { | |
1502 | return -ENOTSUP; | |
ef72f76e SH |
1503 | } |
1504 | ||
0e71be19 SH |
1505 | /* Zero writes start without an I/O buffer. If a buffer becomes necessary |
1506 | * then it will be allocated during request processing. | |
1507 | */ | |
49a2e483 EB |
1508 | iov.iov_base = NULL; |
1509 | iov.iov_len = count; | |
0e71be19 SH |
1510 | |
1511 | qemu_iovec_init_external(&qiov, &iov, 1); | |
49a2e483 EB |
1512 | blockacb = qed_aio_setup(bs, offset >> BDRV_SECTOR_BITS, &qiov, |
1513 | count >> BDRV_SECTOR_BITS, | |
1514 | qed_co_pwrite_zeroes_cb, &cb, | |
0e71be19 SH |
1515 | QED_AIOCB_WRITE | QED_AIOCB_ZERO); |
1516 | if (!blockacb) { | |
1517 | return -EIO; | |
1518 | } | |
1519 | if (!cb.done) { | |
1520 | cb.co = qemu_coroutine_self(); | |
1521 | qemu_coroutine_yield(); | |
1522 | } | |
1523 | assert(cb.done); | |
1524 | return cb.ret; | |
1525 | } | |
1526 | ||
4bff28b8 | 1527 | static int bdrv_qed_truncate(BlockDriverState *bs, int64_t offset, Error **errp) |
75411d23 | 1528 | { |
77a5a000 SH |
1529 | BDRVQEDState *s = bs->opaque; |
1530 | uint64_t old_image_size; | |
1531 | int ret; | |
1532 | ||
1533 | if (!qed_is_image_size_valid(offset, s->header.cluster_size, | |
1534 | s->header.table_size)) { | |
f59adb32 | 1535 | error_setg(errp, "Invalid image size specified"); |
77a5a000 SH |
1536 | return -EINVAL; |
1537 | } | |
1538 | ||
77a5a000 | 1539 | if ((uint64_t)offset < s->header.image_size) { |
f59adb32 | 1540 | error_setg(errp, "Shrinking images is currently not supported"); |
77a5a000 SH |
1541 | return -ENOTSUP; |
1542 | } | |
1543 | ||
1544 | old_image_size = s->header.image_size; | |
1545 | s->header.image_size = offset; | |
1546 | ret = qed_write_header_sync(s); | |
1547 | if (ret < 0) { | |
1548 | s->header.image_size = old_image_size; | |
f59adb32 | 1549 | error_setg_errno(errp, -ret, "Failed to update the image size"); |
77a5a000 SH |
1550 | } |
1551 | return ret; | |
75411d23 SH |
1552 | } |
1553 | ||
1554 | static int64_t bdrv_qed_getlength(BlockDriverState *bs) | |
1555 | { | |
1556 | BDRVQEDState *s = bs->opaque; | |
1557 | return s->header.image_size; | |
1558 | } | |
1559 | ||
1560 | static int bdrv_qed_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
1561 | { | |
1562 | BDRVQEDState *s = bs->opaque; | |
1563 | ||
1564 | memset(bdi, 0, sizeof(*bdi)); | |
1565 | bdi->cluster_size = s->header.cluster_size; | |
d68dbee8 | 1566 | bdi->is_dirty = s->header.features & QED_F_NEED_CHECK; |
95de6d70 PB |
1567 | bdi->unallocated_blocks_are_zero = true; |
1568 | bdi->can_write_zeroes_with_unmap = true; | |
75411d23 SH |
1569 | return 0; |
1570 | } | |
1571 | ||
1572 | static int bdrv_qed_change_backing_file(BlockDriverState *bs, | |
1573 | const char *backing_file, | |
1574 | const char *backing_fmt) | |
1575 | { | |
1576 | BDRVQEDState *s = bs->opaque; | |
1577 | QEDHeader new_header, le_header; | |
1578 | void *buffer; | |
1579 | size_t buffer_len, backing_file_len; | |
1580 | int ret; | |
1581 | ||
1582 | /* Refuse to set backing filename if unknown compat feature bits are | |
1583 | * active. If the image uses an unknown compat feature then we may not | |
1584 | * know the layout of data following the header structure and cannot safely | |
1585 | * add a new string. | |
1586 | */ | |
1587 | if (backing_file && (s->header.compat_features & | |
1588 | ~QED_COMPAT_FEATURE_MASK)) { | |
1589 | return -ENOTSUP; | |
1590 | } | |
1591 | ||
1592 | memcpy(&new_header, &s->header, sizeof(new_header)); | |
1593 | ||
1594 | new_header.features &= ~(QED_F_BACKING_FILE | | |
1595 | QED_F_BACKING_FORMAT_NO_PROBE); | |
1596 | ||
1597 | /* Adjust feature flags */ | |
1598 | if (backing_file) { | |
1599 | new_header.features |= QED_F_BACKING_FILE; | |
1600 | ||
1601 | if (qed_fmt_is_raw(backing_fmt)) { | |
1602 | new_header.features |= QED_F_BACKING_FORMAT_NO_PROBE; | |
1603 | } | |
1604 | } | |
1605 | ||
1606 | /* Calculate new header size */ | |
1607 | backing_file_len = 0; | |
1608 | ||
1609 | if (backing_file) { | |
1610 | backing_file_len = strlen(backing_file); | |
1611 | } | |
1612 | ||
1613 | buffer_len = sizeof(new_header); | |
1614 | new_header.backing_filename_offset = buffer_len; | |
1615 | new_header.backing_filename_size = backing_file_len; | |
1616 | buffer_len += backing_file_len; | |
1617 | ||
1618 | /* Make sure we can rewrite header without failing */ | |
1619 | if (buffer_len > new_header.header_size * new_header.cluster_size) { | |
1620 | return -ENOSPC; | |
1621 | } | |
1622 | ||
1623 | /* Prepare new header */ | |
7267c094 | 1624 | buffer = g_malloc(buffer_len); |
75411d23 SH |
1625 | |
1626 | qed_header_cpu_to_le(&new_header, &le_header); | |
1627 | memcpy(buffer, &le_header, sizeof(le_header)); | |
1628 | buffer_len = sizeof(le_header); | |
1629 | ||
feba23b1 PB |
1630 | if (backing_file) { |
1631 | memcpy(buffer + buffer_len, backing_file, backing_file_len); | |
1632 | buffer_len += backing_file_len; | |
1633 | } | |
75411d23 SH |
1634 | |
1635 | /* Write new header */ | |
d9ca2ea2 | 1636 | ret = bdrv_pwrite_sync(bs->file, 0, buffer, buffer_len); |
7267c094 | 1637 | g_free(buffer); |
75411d23 SH |
1638 | if (ret == 0) { |
1639 | memcpy(&s->header, &new_header, sizeof(new_header)); | |
1640 | } | |
1641 | return ret; | |
1642 | } | |
1643 | ||
5a8a30db | 1644 | static void bdrv_qed_invalidate_cache(BlockDriverState *bs, Error **errp) |
c82954e5 BC |
1645 | { |
1646 | BDRVQEDState *s = bs->opaque; | |
5a8a30db KW |
1647 | Error *local_err = NULL; |
1648 | int ret; | |
c82954e5 BC |
1649 | |
1650 | bdrv_qed_close(bs); | |
3456a8d1 | 1651 | |
c82954e5 | 1652 | memset(s, 0, sizeof(BDRVQEDState)); |
4e4bf5c4 | 1653 | ret = bdrv_qed_do_open(bs, NULL, bs->open_flags, &local_err); |
5a8a30db | 1654 | if (local_err) { |
e43bfd9c MA |
1655 | error_propagate(errp, local_err); |
1656 | error_prepend(errp, "Could not reopen qed layer: "); | |
5a8a30db KW |
1657 | return; |
1658 | } else if (ret < 0) { | |
1659 | error_setg_errno(errp, -ret, "Could not reopen qed layer"); | |
1660 | return; | |
1661 | } | |
c82954e5 BC |
1662 | } |
1663 | ||
4534ff54 KW |
1664 | static int bdrv_qed_check(BlockDriverState *bs, BdrvCheckResult *result, |
1665 | BdrvCheckMode fix) | |
75411d23 | 1666 | { |
01979a98 SH |
1667 | BDRVQEDState *s = bs->opaque; |
1668 | ||
4534ff54 | 1669 | return qed_check(s, result, !!fix); |
75411d23 SH |
1670 | } |
1671 | ||
7ab74849 CL |
1672 | static QemuOptsList qed_create_opts = { |
1673 | .name = "qed-create-opts", | |
1674 | .head = QTAILQ_HEAD_INITIALIZER(qed_create_opts.head), | |
1675 | .desc = { | |
1676 | { | |
1677 | .name = BLOCK_OPT_SIZE, | |
1678 | .type = QEMU_OPT_SIZE, | |
1679 | .help = "Virtual disk size" | |
1680 | }, | |
1681 | { | |
1682 | .name = BLOCK_OPT_BACKING_FILE, | |
1683 | .type = QEMU_OPT_STRING, | |
1684 | .help = "File name of a base image" | |
1685 | }, | |
1686 | { | |
1687 | .name = BLOCK_OPT_BACKING_FMT, | |
1688 | .type = QEMU_OPT_STRING, | |
1689 | .help = "Image format of the base image" | |
1690 | }, | |
1691 | { | |
1692 | .name = BLOCK_OPT_CLUSTER_SIZE, | |
1693 | .type = QEMU_OPT_SIZE, | |
1694 | .help = "Cluster size (in bytes)", | |
1695 | .def_value_str = stringify(QED_DEFAULT_CLUSTER_SIZE) | |
1696 | }, | |
1697 | { | |
1698 | .name = BLOCK_OPT_TABLE_SIZE, | |
1699 | .type = QEMU_OPT_SIZE, | |
1700 | .help = "L1/L2 table size (in clusters)" | |
1701 | }, | |
1702 | { /* end of list */ } | |
1703 | } | |
75411d23 SH |
1704 | }; |
1705 | ||
1706 | static BlockDriver bdrv_qed = { | |
1707 | .format_name = "qed", | |
1708 | .instance_size = sizeof(BDRVQEDState), | |
7ab74849 | 1709 | .create_opts = &qed_create_opts, |
8ee79e70 | 1710 | .supports_backing = true, |
75411d23 SH |
1711 | |
1712 | .bdrv_probe = bdrv_qed_probe, | |
1713 | .bdrv_open = bdrv_qed_open, | |
1714 | .bdrv_close = bdrv_qed_close, | |
f9cb20f1 | 1715 | .bdrv_reopen_prepare = bdrv_qed_reopen_prepare, |
862f215f | 1716 | .bdrv_child_perm = bdrv_format_default_perms, |
c282e1fd | 1717 | .bdrv_create = bdrv_qed_create, |
3ac21627 | 1718 | .bdrv_has_zero_init = bdrv_has_zero_init_1, |
b6b8a333 | 1719 | .bdrv_co_get_block_status = bdrv_qed_co_get_block_status, |
75411d23 SH |
1720 | .bdrv_aio_readv = bdrv_qed_aio_readv, |
1721 | .bdrv_aio_writev = bdrv_qed_aio_writev, | |
49a2e483 | 1722 | .bdrv_co_pwrite_zeroes = bdrv_qed_co_pwrite_zeroes, |
75411d23 SH |
1723 | .bdrv_truncate = bdrv_qed_truncate, |
1724 | .bdrv_getlength = bdrv_qed_getlength, | |
1725 | .bdrv_get_info = bdrv_qed_get_info, | |
d34682cd | 1726 | .bdrv_refresh_limits = bdrv_qed_refresh_limits, |
75411d23 | 1727 | .bdrv_change_backing_file = bdrv_qed_change_backing_file, |
c82954e5 | 1728 | .bdrv_invalidate_cache = bdrv_qed_invalidate_cache, |
75411d23 | 1729 | .bdrv_check = bdrv_qed_check, |
a8c868c3 SH |
1730 | .bdrv_detach_aio_context = bdrv_qed_detach_aio_context, |
1731 | .bdrv_attach_aio_context = bdrv_qed_attach_aio_context, | |
6653a73d | 1732 | .bdrv_drain = bdrv_qed_drain, |
75411d23 SH |
1733 | }; |
1734 | ||
1735 | static void bdrv_qed_init(void) | |
1736 | { | |
1737 | bdrv_register(&bdrv_qed); | |
1738 | } | |
1739 | ||
1740 | block_init(bdrv_qed_init); |