]>
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 | ||
6f321e93 | 15 | #include "qemu-timer.h" |
eabba580 | 16 | #include "trace.h" |
75411d23 | 17 | #include "qed.h" |
10b758e8 | 18 | #include "qerror.h" |
1ed520c6 | 19 | #include "migration.h" |
75411d23 | 20 | |
eabba580 SH |
21 | static void qed_aio_cancel(BlockDriverAIOCB *blockacb) |
22 | { | |
23 | QEDAIOCB *acb = (QEDAIOCB *)blockacb; | |
24 | bool finished = false; | |
25 | ||
26 | /* Wait for the request to finish */ | |
27 | acb->finished = &finished; | |
28 | while (!finished) { | |
29 | qemu_aio_wait(); | |
30 | } | |
31 | } | |
32 | ||
33 | static AIOPool qed_aio_pool = { | |
34 | .aiocb_size = sizeof(QEDAIOCB), | |
35 | .cancel = qed_aio_cancel, | |
36 | }; | |
37 | ||
75411d23 SH |
38 | static int bdrv_qed_probe(const uint8_t *buf, int buf_size, |
39 | const char *filename) | |
40 | { | |
41 | const QEDHeader *header = (const QEDHeader *)buf; | |
42 | ||
43 | if (buf_size < sizeof(*header)) { | |
44 | return 0; | |
45 | } | |
46 | if (le32_to_cpu(header->magic) != QED_MAGIC) { | |
47 | return 0; | |
48 | } | |
49 | return 100; | |
50 | } | |
51 | ||
52 | /** | |
53 | * Check whether an image format is raw | |
54 | * | |
55 | * @fmt: Backing file format, may be NULL | |
56 | */ | |
57 | static bool qed_fmt_is_raw(const char *fmt) | |
58 | { | |
59 | return fmt && strcmp(fmt, "raw") == 0; | |
60 | } | |
61 | ||
62 | static void qed_header_le_to_cpu(const QEDHeader *le, QEDHeader *cpu) | |
63 | { | |
64 | cpu->magic = le32_to_cpu(le->magic); | |
65 | cpu->cluster_size = le32_to_cpu(le->cluster_size); | |
66 | cpu->table_size = le32_to_cpu(le->table_size); | |
67 | cpu->header_size = le32_to_cpu(le->header_size); | |
68 | cpu->features = le64_to_cpu(le->features); | |
69 | cpu->compat_features = le64_to_cpu(le->compat_features); | |
70 | cpu->autoclear_features = le64_to_cpu(le->autoclear_features); | |
71 | cpu->l1_table_offset = le64_to_cpu(le->l1_table_offset); | |
72 | cpu->image_size = le64_to_cpu(le->image_size); | |
73 | cpu->backing_filename_offset = le32_to_cpu(le->backing_filename_offset); | |
74 | cpu->backing_filename_size = le32_to_cpu(le->backing_filename_size); | |
75 | } | |
76 | ||
77 | static void qed_header_cpu_to_le(const QEDHeader *cpu, QEDHeader *le) | |
78 | { | |
79 | le->magic = cpu_to_le32(cpu->magic); | |
80 | le->cluster_size = cpu_to_le32(cpu->cluster_size); | |
81 | le->table_size = cpu_to_le32(cpu->table_size); | |
82 | le->header_size = cpu_to_le32(cpu->header_size); | |
83 | le->features = cpu_to_le64(cpu->features); | |
84 | le->compat_features = cpu_to_le64(cpu->compat_features); | |
85 | le->autoclear_features = cpu_to_le64(cpu->autoclear_features); | |
86 | le->l1_table_offset = cpu_to_le64(cpu->l1_table_offset); | |
87 | le->image_size = cpu_to_le64(cpu->image_size); | |
88 | le->backing_filename_offset = cpu_to_le32(cpu->backing_filename_offset); | |
89 | le->backing_filename_size = cpu_to_le32(cpu->backing_filename_size); | |
90 | } | |
91 | ||
92 | static int qed_write_header_sync(BDRVQEDState *s) | |
93 | { | |
94 | QEDHeader le; | |
95 | int ret; | |
96 | ||
97 | qed_header_cpu_to_le(&s->header, &le); | |
98 | ret = bdrv_pwrite(s->bs->file, 0, &le, sizeof(le)); | |
99 | if (ret != sizeof(le)) { | |
100 | return ret; | |
101 | } | |
102 | return 0; | |
103 | } | |
104 | ||
01979a98 SH |
105 | typedef struct { |
106 | GenericCB gencb; | |
107 | BDRVQEDState *s; | |
108 | struct iovec iov; | |
109 | QEMUIOVector qiov; | |
110 | int nsectors; | |
111 | uint8_t *buf; | |
112 | } QEDWriteHeaderCB; | |
113 | ||
114 | static void qed_write_header_cb(void *opaque, int ret) | |
115 | { | |
116 | QEDWriteHeaderCB *write_header_cb = opaque; | |
117 | ||
118 | qemu_vfree(write_header_cb->buf); | |
119 | gencb_complete(write_header_cb, ret); | |
120 | } | |
121 | ||
122 | static void qed_write_header_read_cb(void *opaque, int ret) | |
123 | { | |
124 | QEDWriteHeaderCB *write_header_cb = opaque; | |
125 | BDRVQEDState *s = write_header_cb->s; | |
01979a98 SH |
126 | |
127 | if (ret) { | |
128 | qed_write_header_cb(write_header_cb, ret); | |
129 | return; | |
130 | } | |
131 | ||
132 | /* Update header */ | |
133 | qed_header_cpu_to_le(&s->header, (QEDHeader *)write_header_cb->buf); | |
134 | ||
ad54ae80 PB |
135 | bdrv_aio_writev(s->bs->file, 0, &write_header_cb->qiov, |
136 | write_header_cb->nsectors, qed_write_header_cb, | |
137 | write_header_cb); | |
01979a98 SH |
138 | } |
139 | ||
140 | /** | |
141 | * Update header in-place (does not rewrite backing filename or other strings) | |
142 | * | |
143 | * This function only updates known header fields in-place and does not affect | |
144 | * extra data after the QED header. | |
145 | */ | |
146 | static void qed_write_header(BDRVQEDState *s, BlockDriverCompletionFunc cb, | |
147 | void *opaque) | |
148 | { | |
149 | /* We must write full sectors for O_DIRECT but cannot necessarily generate | |
150 | * the data following the header if an unrecognized compat feature is | |
151 | * active. Therefore, first read the sectors containing the header, update | |
152 | * them, and write back. | |
153 | */ | |
154 | ||
01979a98 SH |
155 | int nsectors = (sizeof(QEDHeader) + BDRV_SECTOR_SIZE - 1) / |
156 | BDRV_SECTOR_SIZE; | |
157 | size_t len = nsectors * BDRV_SECTOR_SIZE; | |
158 | QEDWriteHeaderCB *write_header_cb = gencb_alloc(sizeof(*write_header_cb), | |
159 | cb, opaque); | |
160 | ||
161 | write_header_cb->s = s; | |
162 | write_header_cb->nsectors = nsectors; | |
163 | write_header_cb->buf = qemu_blockalign(s->bs, len); | |
164 | write_header_cb->iov.iov_base = write_header_cb->buf; | |
165 | write_header_cb->iov.iov_len = len; | |
166 | qemu_iovec_init_external(&write_header_cb->qiov, &write_header_cb->iov, 1); | |
167 | ||
ad54ae80 PB |
168 | bdrv_aio_readv(s->bs->file, 0, &write_header_cb->qiov, nsectors, |
169 | qed_write_header_read_cb, write_header_cb); | |
01979a98 SH |
170 | } |
171 | ||
75411d23 SH |
172 | static uint64_t qed_max_image_size(uint32_t cluster_size, uint32_t table_size) |
173 | { | |
174 | uint64_t table_entries; | |
175 | uint64_t l2_size; | |
176 | ||
177 | table_entries = (table_size * cluster_size) / sizeof(uint64_t); | |
178 | l2_size = table_entries * cluster_size; | |
179 | ||
180 | return l2_size * table_entries; | |
181 | } | |
182 | ||
183 | static bool qed_is_cluster_size_valid(uint32_t cluster_size) | |
184 | { | |
185 | if (cluster_size < QED_MIN_CLUSTER_SIZE || | |
186 | cluster_size > QED_MAX_CLUSTER_SIZE) { | |
187 | return false; | |
188 | } | |
189 | if (cluster_size & (cluster_size - 1)) { | |
190 | return false; /* not power of 2 */ | |
191 | } | |
192 | return true; | |
193 | } | |
194 | ||
195 | static bool qed_is_table_size_valid(uint32_t table_size) | |
196 | { | |
197 | if (table_size < QED_MIN_TABLE_SIZE || | |
198 | table_size > QED_MAX_TABLE_SIZE) { | |
199 | return false; | |
200 | } | |
201 | if (table_size & (table_size - 1)) { | |
202 | return false; /* not power of 2 */ | |
203 | } | |
204 | return true; | |
205 | } | |
206 | ||
207 | static bool qed_is_image_size_valid(uint64_t image_size, uint32_t cluster_size, | |
208 | uint32_t table_size) | |
209 | { | |
210 | if (image_size % BDRV_SECTOR_SIZE != 0) { | |
211 | return false; /* not multiple of sector size */ | |
212 | } | |
213 | if (image_size > qed_max_image_size(cluster_size, table_size)) { | |
214 | return false; /* image is too large */ | |
215 | } | |
216 | return true; | |
217 | } | |
218 | ||
219 | /** | |
220 | * Read a string of known length from the image file | |
221 | * | |
222 | * @file: Image file | |
223 | * @offset: File offset to start of string, in bytes | |
224 | * @n: String length in bytes | |
225 | * @buf: Destination buffer | |
226 | * @buflen: Destination buffer length in bytes | |
227 | * @ret: 0 on success, -errno on failure | |
228 | * | |
229 | * The string is NUL-terminated. | |
230 | */ | |
231 | static int qed_read_string(BlockDriverState *file, uint64_t offset, size_t n, | |
232 | char *buf, size_t buflen) | |
233 | { | |
234 | int ret; | |
235 | if (n >= buflen) { | |
236 | return -EINVAL; | |
237 | } | |
238 | ret = bdrv_pread(file, offset, buf, n); | |
239 | if (ret < 0) { | |
240 | return ret; | |
241 | } | |
242 | buf[n] = '\0'; | |
243 | return 0; | |
244 | } | |
245 | ||
eabba580 SH |
246 | /** |
247 | * Allocate new clusters | |
248 | * | |
249 | * @s: QED state | |
250 | * @n: Number of contiguous clusters to allocate | |
251 | * @ret: Offset of first allocated cluster | |
252 | * | |
253 | * This function only produces the offset where the new clusters should be | |
254 | * written. It updates BDRVQEDState but does not make any changes to the image | |
255 | * file. | |
256 | */ | |
257 | static uint64_t qed_alloc_clusters(BDRVQEDState *s, unsigned int n) | |
258 | { | |
259 | uint64_t offset = s->file_size; | |
260 | s->file_size += n * s->header.cluster_size; | |
261 | return offset; | |
262 | } | |
263 | ||
298800ca SH |
264 | QEDTable *qed_alloc_table(BDRVQEDState *s) |
265 | { | |
266 | /* Honor O_DIRECT memory alignment requirements */ | |
267 | return qemu_blockalign(s->bs, | |
268 | s->header.cluster_size * s->header.table_size); | |
269 | } | |
270 | ||
eabba580 SH |
271 | /** |
272 | * Allocate a new zeroed L2 table | |
273 | */ | |
274 | static CachedL2Table *qed_new_l2_table(BDRVQEDState *s) | |
275 | { | |
276 | CachedL2Table *l2_table = qed_alloc_l2_cache_entry(&s->l2_cache); | |
277 | ||
278 | l2_table->table = qed_alloc_table(s); | |
279 | l2_table->offset = qed_alloc_clusters(s, s->header.table_size); | |
280 | ||
281 | memset(l2_table->table->offsets, 0, | |
282 | s->header.cluster_size * s->header.table_size); | |
283 | return l2_table; | |
284 | } | |
285 | ||
286 | static void qed_aio_next_io(void *opaque, int ret); | |
287 | ||
6f321e93 SH |
288 | static void qed_plug_allocating_write_reqs(BDRVQEDState *s) |
289 | { | |
290 | assert(!s->allocating_write_reqs_plugged); | |
291 | ||
292 | s->allocating_write_reqs_plugged = true; | |
293 | } | |
294 | ||
295 | static void qed_unplug_allocating_write_reqs(BDRVQEDState *s) | |
296 | { | |
297 | QEDAIOCB *acb; | |
298 | ||
299 | assert(s->allocating_write_reqs_plugged); | |
300 | ||
301 | s->allocating_write_reqs_plugged = false; | |
302 | ||
303 | acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs); | |
304 | if (acb) { | |
305 | qed_aio_next_io(acb, 0); | |
306 | } | |
307 | } | |
308 | ||
309 | static void qed_finish_clear_need_check(void *opaque, int ret) | |
310 | { | |
311 | /* Do nothing */ | |
312 | } | |
313 | ||
314 | static void qed_flush_after_clear_need_check(void *opaque, int ret) | |
315 | { | |
316 | BDRVQEDState *s = opaque; | |
317 | ||
318 | bdrv_aio_flush(s->bs, qed_finish_clear_need_check, s); | |
319 | ||
320 | /* No need to wait until flush completes */ | |
321 | qed_unplug_allocating_write_reqs(s); | |
322 | } | |
323 | ||
324 | static void qed_clear_need_check(void *opaque, int ret) | |
325 | { | |
326 | BDRVQEDState *s = opaque; | |
327 | ||
328 | if (ret) { | |
329 | qed_unplug_allocating_write_reqs(s); | |
330 | return; | |
331 | } | |
332 | ||
333 | s->header.features &= ~QED_F_NEED_CHECK; | |
334 | qed_write_header(s, qed_flush_after_clear_need_check, s); | |
335 | } | |
336 | ||
337 | static void qed_need_check_timer_cb(void *opaque) | |
338 | { | |
339 | BDRVQEDState *s = opaque; | |
340 | ||
341 | /* The timer should only fire when allocating writes have drained */ | |
342 | assert(!QSIMPLEQ_FIRST(&s->allocating_write_reqs)); | |
343 | ||
344 | trace_qed_need_check_timer_cb(s); | |
345 | ||
346 | qed_plug_allocating_write_reqs(s); | |
347 | ||
348 | /* Ensure writes are on disk before clearing flag */ | |
349 | bdrv_aio_flush(s->bs, qed_clear_need_check, s); | |
350 | } | |
351 | ||
352 | static void qed_start_need_check_timer(BDRVQEDState *s) | |
353 | { | |
354 | trace_qed_start_need_check_timer(s); | |
355 | ||
356 | /* Use vm_clock so we don't alter the image file while suspended for | |
357 | * migration. | |
358 | */ | |
359 | qemu_mod_timer(s->need_check_timer, qemu_get_clock_ns(vm_clock) + | |
360 | get_ticks_per_sec() * QED_NEED_CHECK_TIMEOUT); | |
361 | } | |
362 | ||
363 | /* It's okay to call this multiple times or when no timer is started */ | |
364 | static void qed_cancel_need_check_timer(BDRVQEDState *s) | |
365 | { | |
366 | trace_qed_cancel_need_check_timer(s); | |
367 | qemu_del_timer(s->need_check_timer); | |
368 | } | |
369 | ||
75411d23 SH |
370 | static int bdrv_qed_open(BlockDriverState *bs, int flags) |
371 | { | |
372 | BDRVQEDState *s = bs->opaque; | |
373 | QEDHeader le_header; | |
374 | int64_t file_size; | |
375 | int ret; | |
376 | ||
377 | s->bs = bs; | |
eabba580 | 378 | QSIMPLEQ_INIT(&s->allocating_write_reqs); |
75411d23 SH |
379 | |
380 | ret = bdrv_pread(bs->file, 0, &le_header, sizeof(le_header)); | |
381 | if (ret < 0) { | |
382 | return ret; | |
383 | } | |
75411d23 SH |
384 | qed_header_le_to_cpu(&le_header, &s->header); |
385 | ||
386 | if (s->header.magic != QED_MAGIC) { | |
387 | return -EINVAL; | |
388 | } | |
389 | if (s->header.features & ~QED_FEATURE_MASK) { | |
10b758e8 KW |
390 | /* image uses unsupported feature bits */ |
391 | char buf[64]; | |
392 | snprintf(buf, sizeof(buf), "%" PRIx64, | |
393 | s->header.features & ~QED_FEATURE_MASK); | |
394 | qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE, | |
395 | bs->device_name, "QED", buf); | |
396 | return -ENOTSUP; | |
75411d23 SH |
397 | } |
398 | if (!qed_is_cluster_size_valid(s->header.cluster_size)) { | |
399 | return -EINVAL; | |
400 | } | |
401 | ||
402 | /* Round down file size to the last cluster */ | |
403 | file_size = bdrv_getlength(bs->file); | |
404 | if (file_size < 0) { | |
405 | return file_size; | |
406 | } | |
407 | s->file_size = qed_start_of_cluster(s, file_size); | |
408 | ||
409 | if (!qed_is_table_size_valid(s->header.table_size)) { | |
410 | return -EINVAL; | |
411 | } | |
412 | if (!qed_is_image_size_valid(s->header.image_size, | |
413 | s->header.cluster_size, | |
414 | s->header.table_size)) { | |
415 | return -EINVAL; | |
416 | } | |
417 | if (!qed_check_table_offset(s, s->header.l1_table_offset)) { | |
418 | return -EINVAL; | |
419 | } | |
420 | ||
421 | s->table_nelems = (s->header.cluster_size * s->header.table_size) / | |
422 | sizeof(uint64_t); | |
423 | s->l2_shift = ffs(s->header.cluster_size) - 1; | |
424 | s->l2_mask = s->table_nelems - 1; | |
425 | s->l1_shift = s->l2_shift + ffs(s->table_nelems) - 1; | |
426 | ||
427 | if ((s->header.features & QED_F_BACKING_FILE)) { | |
428 | if ((uint64_t)s->header.backing_filename_offset + | |
429 | s->header.backing_filename_size > | |
430 | s->header.cluster_size * s->header.header_size) { | |
431 | return -EINVAL; | |
432 | } | |
433 | ||
434 | ret = qed_read_string(bs->file, s->header.backing_filename_offset, | |
435 | s->header.backing_filename_size, bs->backing_file, | |
436 | sizeof(bs->backing_file)); | |
437 | if (ret < 0) { | |
438 | return ret; | |
439 | } | |
440 | ||
441 | if (s->header.features & QED_F_BACKING_FORMAT_NO_PROBE) { | |
442 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), "raw"); | |
443 | } | |
444 | } | |
445 | ||
446 | /* Reset unknown autoclear feature bits. This is a backwards | |
447 | * compatibility mechanism that allows images to be opened by older | |
448 | * programs, which "knock out" unknown feature bits. When an image is | |
449 | * opened by a newer program again it can detect that the autoclear | |
450 | * feature is no longer valid. | |
451 | */ | |
452 | if ((s->header.autoclear_features & ~QED_AUTOCLEAR_FEATURE_MASK) != 0 && | |
453 | !bdrv_is_read_only(bs->file)) { | |
454 | s->header.autoclear_features &= QED_AUTOCLEAR_FEATURE_MASK; | |
455 | ||
456 | ret = qed_write_header_sync(s); | |
457 | if (ret) { | |
458 | return ret; | |
459 | } | |
460 | ||
461 | /* From here on only known autoclear feature bits are valid */ | |
462 | bdrv_flush(bs->file); | |
463 | } | |
464 | ||
298800ca SH |
465 | s->l1_table = qed_alloc_table(s); |
466 | qed_init_l2_cache(&s->l2_cache); | |
467 | ||
468 | ret = qed_read_l1_table_sync(s); | |
01979a98 SH |
469 | if (ret) { |
470 | goto out; | |
471 | } | |
472 | ||
473 | /* If image was not closed cleanly, check consistency */ | |
474 | if (s->header.features & QED_F_NEED_CHECK) { | |
475 | /* Read-only images cannot be fixed. There is no risk of corruption | |
476 | * since write operations are not possible. Therefore, allow | |
477 | * potentially inconsistent images to be opened read-only. This can | |
478 | * aid data recovery from an otherwise inconsistent image. | |
479 | */ | |
480 | if (!bdrv_is_read_only(bs->file)) { | |
481 | BdrvCheckResult result = {0}; | |
482 | ||
483 | ret = qed_check(s, &result, true); | |
6f321e93 SH |
484 | if (ret) { |
485 | goto out; | |
486 | } | |
487 | if (!result.corruptions && !result.check_errors) { | |
01979a98 SH |
488 | /* Ensure fixes reach storage before clearing check bit */ |
489 | bdrv_flush(s->bs); | |
490 | ||
491 | s->header.features &= ~QED_F_NEED_CHECK; | |
492 | qed_write_header_sync(s); | |
493 | } | |
494 | } | |
495 | } | |
496 | ||
6f321e93 SH |
497 | s->need_check_timer = qemu_new_timer_ns(vm_clock, |
498 | qed_need_check_timer_cb, s); | |
499 | ||
1ed520c6 AL |
500 | error_set(&s->migration_blocker, |
501 | QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, | |
502 | "qed", bs->device_name, "live migration"); | |
503 | migrate_add_blocker(s->migration_blocker); | |
504 | ||
505 | ||
01979a98 | 506 | out: |
298800ca SH |
507 | if (ret) { |
508 | qed_free_l2_cache(&s->l2_cache); | |
509 | qemu_vfree(s->l1_table); | |
510 | } | |
75411d23 SH |
511 | return ret; |
512 | } | |
513 | ||
514 | static void bdrv_qed_close(BlockDriverState *bs) | |
515 | { | |
298800ca SH |
516 | BDRVQEDState *s = bs->opaque; |
517 | ||
1ed520c6 AL |
518 | migrate_del_blocker(s->migration_blocker); |
519 | error_free(s->migration_blocker); | |
520 | ||
6f321e93 SH |
521 | qed_cancel_need_check_timer(s); |
522 | qemu_free_timer(s->need_check_timer); | |
523 | ||
01979a98 SH |
524 | /* Ensure writes reach stable storage */ |
525 | bdrv_flush(bs->file); | |
526 | ||
527 | /* Clean shutdown, no check required on next open */ | |
528 | if (s->header.features & QED_F_NEED_CHECK) { | |
529 | s->header.features &= ~QED_F_NEED_CHECK; | |
530 | qed_write_header_sync(s); | |
531 | } | |
532 | ||
298800ca SH |
533 | qed_free_l2_cache(&s->l2_cache); |
534 | qemu_vfree(s->l1_table); | |
75411d23 SH |
535 | } |
536 | ||
75411d23 SH |
537 | static int qed_create(const char *filename, uint32_t cluster_size, |
538 | uint64_t image_size, uint32_t table_size, | |
539 | const char *backing_file, const char *backing_fmt) | |
540 | { | |
541 | QEDHeader header = { | |
542 | .magic = QED_MAGIC, | |
543 | .cluster_size = cluster_size, | |
544 | .table_size = table_size, | |
545 | .header_size = 1, | |
546 | .features = 0, | |
547 | .compat_features = 0, | |
548 | .l1_table_offset = cluster_size, | |
549 | .image_size = image_size, | |
550 | }; | |
551 | QEDHeader le_header; | |
552 | uint8_t *l1_table = NULL; | |
553 | size_t l1_size = header.cluster_size * header.table_size; | |
554 | int ret = 0; | |
555 | BlockDriverState *bs = NULL; | |
556 | ||
557 | ret = bdrv_create_file(filename, NULL); | |
558 | if (ret < 0) { | |
559 | return ret; | |
560 | } | |
561 | ||
562 | ret = bdrv_file_open(&bs, filename, BDRV_O_RDWR | BDRV_O_CACHE_WB); | |
563 | if (ret < 0) { | |
564 | return ret; | |
565 | } | |
566 | ||
c743849b SH |
567 | /* File must start empty and grow, check truncate is supported */ |
568 | ret = bdrv_truncate(bs, 0); | |
569 | if (ret < 0) { | |
570 | goto out; | |
571 | } | |
572 | ||
75411d23 SH |
573 | if (backing_file) { |
574 | header.features |= QED_F_BACKING_FILE; | |
575 | header.backing_filename_offset = sizeof(le_header); | |
576 | header.backing_filename_size = strlen(backing_file); | |
577 | ||
578 | if (qed_fmt_is_raw(backing_fmt)) { | |
579 | header.features |= QED_F_BACKING_FORMAT_NO_PROBE; | |
580 | } | |
581 | } | |
582 | ||
583 | qed_header_cpu_to_le(&header, &le_header); | |
584 | ret = bdrv_pwrite(bs, 0, &le_header, sizeof(le_header)); | |
585 | if (ret < 0) { | |
586 | goto out; | |
587 | } | |
588 | ret = bdrv_pwrite(bs, sizeof(le_header), backing_file, | |
589 | header.backing_filename_size); | |
590 | if (ret < 0) { | |
591 | goto out; | |
592 | } | |
593 | ||
7267c094 | 594 | l1_table = g_malloc0(l1_size); |
75411d23 SH |
595 | ret = bdrv_pwrite(bs, header.l1_table_offset, l1_table, l1_size); |
596 | if (ret < 0) { | |
597 | goto out; | |
598 | } | |
599 | ||
600 | ret = 0; /* success */ | |
601 | out: | |
7267c094 | 602 | g_free(l1_table); |
75411d23 SH |
603 | bdrv_delete(bs); |
604 | return ret; | |
605 | } | |
606 | ||
607 | static int bdrv_qed_create(const char *filename, QEMUOptionParameter *options) | |
608 | { | |
609 | uint64_t image_size = 0; | |
610 | uint32_t cluster_size = QED_DEFAULT_CLUSTER_SIZE; | |
611 | uint32_t table_size = QED_DEFAULT_TABLE_SIZE; | |
612 | const char *backing_file = NULL; | |
613 | const char *backing_fmt = NULL; | |
614 | ||
615 | while (options && options->name) { | |
616 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
617 | image_size = options->value.n; | |
618 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) { | |
619 | backing_file = options->value.s; | |
620 | } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FMT)) { | |
621 | backing_fmt = options->value.s; | |
622 | } else if (!strcmp(options->name, BLOCK_OPT_CLUSTER_SIZE)) { | |
623 | if (options->value.n) { | |
624 | cluster_size = options->value.n; | |
625 | } | |
626 | } else if (!strcmp(options->name, BLOCK_OPT_TABLE_SIZE)) { | |
627 | if (options->value.n) { | |
628 | table_size = options->value.n; | |
629 | } | |
630 | } | |
631 | options++; | |
632 | } | |
633 | ||
634 | if (!qed_is_cluster_size_valid(cluster_size)) { | |
635 | fprintf(stderr, "QED cluster size must be within range [%u, %u] and power of 2\n", | |
636 | QED_MIN_CLUSTER_SIZE, QED_MAX_CLUSTER_SIZE); | |
637 | return -EINVAL; | |
638 | } | |
639 | if (!qed_is_table_size_valid(table_size)) { | |
640 | fprintf(stderr, "QED table size must be within range [%u, %u] and power of 2\n", | |
641 | QED_MIN_TABLE_SIZE, QED_MAX_TABLE_SIZE); | |
642 | return -EINVAL; | |
643 | } | |
644 | if (!qed_is_image_size_valid(image_size, cluster_size, table_size)) { | |
645 | fprintf(stderr, "QED image size must be a non-zero multiple of " | |
646 | "cluster size and less than %" PRIu64 " bytes\n", | |
647 | qed_max_image_size(cluster_size, table_size)); | |
648 | return -EINVAL; | |
649 | } | |
650 | ||
651 | return qed_create(filename, cluster_size, image_size, table_size, | |
652 | backing_file, backing_fmt); | |
653 | } | |
654 | ||
298800ca | 655 | typedef struct { |
b7d5a5b8 | 656 | Coroutine *co; |
298800ca SH |
657 | int is_allocated; |
658 | int *pnum; | |
659 | } QEDIsAllocatedCB; | |
660 | ||
661 | static void qed_is_allocated_cb(void *opaque, int ret, uint64_t offset, size_t len) | |
662 | { | |
663 | QEDIsAllocatedCB *cb = opaque; | |
664 | *cb->pnum = len / BDRV_SECTOR_SIZE; | |
21df65b6 | 665 | cb->is_allocated = (ret == QED_CLUSTER_FOUND || ret == QED_CLUSTER_ZERO); |
b7d5a5b8 SH |
666 | if (cb->co) { |
667 | qemu_coroutine_enter(cb->co, NULL); | |
668 | } | |
298800ca SH |
669 | } |
670 | ||
b7d5a5b8 SH |
671 | static int coroutine_fn bdrv_qed_co_is_allocated(BlockDriverState *bs, |
672 | int64_t sector_num, | |
673 | int nb_sectors, int *pnum) | |
75411d23 | 674 | { |
298800ca SH |
675 | BDRVQEDState *s = bs->opaque; |
676 | uint64_t pos = (uint64_t)sector_num * BDRV_SECTOR_SIZE; | |
677 | size_t len = (size_t)nb_sectors * BDRV_SECTOR_SIZE; | |
678 | QEDIsAllocatedCB cb = { | |
679 | .is_allocated = -1, | |
680 | .pnum = pnum, | |
681 | }; | |
682 | QEDRequest request = { .l2_table = NULL }; | |
683 | ||
298800ca SH |
684 | qed_find_cluster(s, &request, pos, len, qed_is_allocated_cb, &cb); |
685 | ||
b7d5a5b8 | 686 | /* Now sleep if the callback wasn't invoked immediately */ |
298800ca | 687 | while (cb.is_allocated == -1) { |
b7d5a5b8 SH |
688 | cb.co = qemu_coroutine_self(); |
689 | qemu_coroutine_yield(); | |
298800ca SH |
690 | } |
691 | ||
298800ca SH |
692 | qed_unref_l2_cache_entry(request.l2_table); |
693 | ||
694 | return cb.is_allocated; | |
75411d23 SH |
695 | } |
696 | ||
697 | static int bdrv_qed_make_empty(BlockDriverState *bs) | |
698 | { | |
699 | return -ENOTSUP; | |
700 | } | |
701 | ||
eabba580 SH |
702 | static BDRVQEDState *acb_to_s(QEDAIOCB *acb) |
703 | { | |
704 | return acb->common.bs->opaque; | |
705 | } | |
706 | ||
707 | /** | |
708 | * Read from the backing file or zero-fill if no backing file | |
709 | * | |
710 | * @s: QED state | |
711 | * @pos: Byte position in device | |
712 | * @qiov: Destination I/O vector | |
713 | * @cb: Completion function | |
714 | * @opaque: User data for completion function | |
715 | * | |
716 | * This function reads qiov->size bytes starting at pos from the backing file. | |
717 | * If there is no backing file then zeroes are read. | |
718 | */ | |
719 | static void qed_read_backing_file(BDRVQEDState *s, uint64_t pos, | |
720 | QEMUIOVector *qiov, | |
721 | BlockDriverCompletionFunc *cb, void *opaque) | |
722 | { | |
eabba580 SH |
723 | uint64_t backing_length = 0; |
724 | size_t size; | |
725 | ||
726 | /* If there is a backing file, get its length. Treat the absence of a | |
727 | * backing file like a zero length backing file. | |
728 | */ | |
729 | if (s->bs->backing_hd) { | |
730 | int64_t l = bdrv_getlength(s->bs->backing_hd); | |
731 | if (l < 0) { | |
732 | cb(opaque, l); | |
733 | return; | |
734 | } | |
735 | backing_length = l; | |
736 | } | |
737 | ||
738 | /* Zero all sectors if reading beyond the end of the backing file */ | |
739 | if (pos >= backing_length || | |
740 | pos + qiov->size > backing_length) { | |
741 | qemu_iovec_memset(qiov, 0, qiov->size); | |
742 | } | |
743 | ||
744 | /* Complete now if there are no backing file sectors to read */ | |
745 | if (pos >= backing_length) { | |
746 | cb(opaque, 0); | |
747 | return; | |
748 | } | |
749 | ||
750 | /* If the read straddles the end of the backing file, shorten it */ | |
751 | size = MIN((uint64_t)backing_length - pos, qiov->size); | |
752 | ||
753 | BLKDBG_EVENT(s->bs->file, BLKDBG_READ_BACKING); | |
ad54ae80 PB |
754 | bdrv_aio_readv(s->bs->backing_hd, pos / BDRV_SECTOR_SIZE, |
755 | qiov, size / BDRV_SECTOR_SIZE, cb, opaque); | |
eabba580 SH |
756 | } |
757 | ||
758 | typedef struct { | |
759 | GenericCB gencb; | |
760 | BDRVQEDState *s; | |
761 | QEMUIOVector qiov; | |
762 | struct iovec iov; | |
763 | uint64_t offset; | |
764 | } CopyFromBackingFileCB; | |
765 | ||
766 | static void qed_copy_from_backing_file_cb(void *opaque, int ret) | |
767 | { | |
768 | CopyFromBackingFileCB *copy_cb = opaque; | |
769 | qemu_vfree(copy_cb->iov.iov_base); | |
770 | gencb_complete(©_cb->gencb, ret); | |
771 | } | |
772 | ||
773 | static void qed_copy_from_backing_file_write(void *opaque, int ret) | |
774 | { | |
775 | CopyFromBackingFileCB *copy_cb = opaque; | |
776 | BDRVQEDState *s = copy_cb->s; | |
eabba580 SH |
777 | |
778 | if (ret) { | |
779 | qed_copy_from_backing_file_cb(copy_cb, ret); | |
780 | return; | |
781 | } | |
782 | ||
783 | BLKDBG_EVENT(s->bs->file, BLKDBG_COW_WRITE); | |
ad54ae80 PB |
784 | bdrv_aio_writev(s->bs->file, copy_cb->offset / BDRV_SECTOR_SIZE, |
785 | ©_cb->qiov, copy_cb->qiov.size / BDRV_SECTOR_SIZE, | |
786 | qed_copy_from_backing_file_cb, copy_cb); | |
eabba580 SH |
787 | } |
788 | ||
789 | /** | |
790 | * Copy data from backing file into the image | |
791 | * | |
792 | * @s: QED state | |
793 | * @pos: Byte position in device | |
794 | * @len: Number of bytes | |
795 | * @offset: Byte offset in image file | |
796 | * @cb: Completion function | |
797 | * @opaque: User data for completion function | |
798 | */ | |
799 | static void qed_copy_from_backing_file(BDRVQEDState *s, uint64_t pos, | |
800 | uint64_t len, uint64_t offset, | |
801 | BlockDriverCompletionFunc *cb, | |
802 | void *opaque) | |
803 | { | |
804 | CopyFromBackingFileCB *copy_cb; | |
805 | ||
806 | /* Skip copy entirely if there is no work to do */ | |
807 | if (len == 0) { | |
808 | cb(opaque, 0); | |
809 | return; | |
810 | } | |
811 | ||
812 | copy_cb = gencb_alloc(sizeof(*copy_cb), cb, opaque); | |
813 | copy_cb->s = s; | |
814 | copy_cb->offset = offset; | |
815 | copy_cb->iov.iov_base = qemu_blockalign(s->bs, len); | |
816 | copy_cb->iov.iov_len = len; | |
817 | qemu_iovec_init_external(©_cb->qiov, ©_cb->iov, 1); | |
818 | ||
819 | qed_read_backing_file(s, pos, ©_cb->qiov, | |
820 | qed_copy_from_backing_file_write, copy_cb); | |
821 | } | |
822 | ||
823 | /** | |
824 | * Link one or more contiguous clusters into a table | |
825 | * | |
826 | * @s: QED state | |
827 | * @table: L2 table | |
828 | * @index: First cluster index | |
829 | * @n: Number of contiguous clusters | |
21df65b6 AL |
830 | * @cluster: First cluster offset |
831 | * | |
832 | * The cluster offset may be an allocated byte offset in the image file, the | |
833 | * zero cluster marker, or the unallocated cluster marker. | |
eabba580 SH |
834 | */ |
835 | static void qed_update_l2_table(BDRVQEDState *s, QEDTable *table, int index, | |
836 | unsigned int n, uint64_t cluster) | |
837 | { | |
838 | int i; | |
839 | for (i = index; i < index + n; i++) { | |
840 | table->offsets[i] = cluster; | |
21df65b6 AL |
841 | if (!qed_offset_is_unalloc_cluster(cluster) && |
842 | !qed_offset_is_zero_cluster(cluster)) { | |
843 | cluster += s->header.cluster_size; | |
844 | } | |
eabba580 SH |
845 | } |
846 | } | |
847 | ||
848 | static void qed_aio_complete_bh(void *opaque) | |
849 | { | |
850 | QEDAIOCB *acb = opaque; | |
851 | BlockDriverCompletionFunc *cb = acb->common.cb; | |
852 | void *user_opaque = acb->common.opaque; | |
853 | int ret = acb->bh_ret; | |
854 | bool *finished = acb->finished; | |
855 | ||
856 | qemu_bh_delete(acb->bh); | |
857 | qemu_aio_release(acb); | |
858 | ||
859 | /* Invoke callback */ | |
860 | cb(user_opaque, ret); | |
861 | ||
862 | /* Signal cancel completion */ | |
863 | if (finished) { | |
864 | *finished = true; | |
865 | } | |
866 | } | |
867 | ||
868 | static void qed_aio_complete(QEDAIOCB *acb, int ret) | |
869 | { | |
870 | BDRVQEDState *s = acb_to_s(acb); | |
871 | ||
872 | trace_qed_aio_complete(s, acb, ret); | |
873 | ||
874 | /* Free resources */ | |
875 | qemu_iovec_destroy(&acb->cur_qiov); | |
876 | qed_unref_l2_cache_entry(acb->request.l2_table); | |
877 | ||
878 | /* Arrange for a bh to invoke the completion function */ | |
879 | acb->bh_ret = ret; | |
880 | acb->bh = qemu_bh_new(qed_aio_complete_bh, acb); | |
881 | qemu_bh_schedule(acb->bh); | |
882 | ||
883 | /* Start next allocating write request waiting behind this one. Note that | |
884 | * requests enqueue themselves when they first hit an unallocated cluster | |
885 | * but they wait until the entire request is finished before waking up the | |
886 | * next request in the queue. This ensures that we don't cycle through | |
887 | * requests multiple times but rather finish one at a time completely. | |
888 | */ | |
889 | if (acb == QSIMPLEQ_FIRST(&s->allocating_write_reqs)) { | |
890 | QSIMPLEQ_REMOVE_HEAD(&s->allocating_write_reqs, next); | |
891 | acb = QSIMPLEQ_FIRST(&s->allocating_write_reqs); | |
892 | if (acb) { | |
893 | qed_aio_next_io(acb, 0); | |
6f321e93 SH |
894 | } else if (s->header.features & QED_F_NEED_CHECK) { |
895 | qed_start_need_check_timer(s); | |
eabba580 SH |
896 | } |
897 | } | |
898 | } | |
899 | ||
900 | /** | |
901 | * Commit the current L2 table to the cache | |
902 | */ | |
903 | static void qed_commit_l2_update(void *opaque, int ret) | |
904 | { | |
905 | QEDAIOCB *acb = opaque; | |
906 | BDRVQEDState *s = acb_to_s(acb); | |
907 | CachedL2Table *l2_table = acb->request.l2_table; | |
e4fc8781 | 908 | uint64_t l2_offset = l2_table->offset; |
eabba580 SH |
909 | |
910 | qed_commit_l2_cache_entry(&s->l2_cache, l2_table); | |
911 | ||
912 | /* This is guaranteed to succeed because we just committed the entry to the | |
913 | * cache. | |
914 | */ | |
e4fc8781 | 915 | acb->request.l2_table = qed_find_l2_cache_entry(&s->l2_cache, l2_offset); |
eabba580 SH |
916 | assert(acb->request.l2_table != NULL); |
917 | ||
918 | qed_aio_next_io(opaque, ret); | |
919 | } | |
920 | ||
921 | /** | |
922 | * Update L1 table with new L2 table offset and write it out | |
923 | */ | |
924 | static void qed_aio_write_l1_update(void *opaque, int ret) | |
925 | { | |
926 | QEDAIOCB *acb = opaque; | |
927 | BDRVQEDState *s = acb_to_s(acb); | |
928 | int index; | |
929 | ||
930 | if (ret) { | |
931 | qed_aio_complete(acb, ret); | |
932 | return; | |
933 | } | |
934 | ||
935 | index = qed_l1_index(s, acb->cur_pos); | |
936 | s->l1_table->offsets[index] = acb->request.l2_table->offset; | |
937 | ||
938 | qed_write_l1_table(s, index, 1, qed_commit_l2_update, acb); | |
939 | } | |
940 | ||
941 | /** | |
942 | * Update L2 table with new cluster offsets and write them out | |
943 | */ | |
944 | static void qed_aio_write_l2_update(void *opaque, int ret) | |
945 | { | |
946 | QEDAIOCB *acb = opaque; | |
947 | BDRVQEDState *s = acb_to_s(acb); | |
948 | bool need_alloc = acb->find_cluster_ret == QED_CLUSTER_L1; | |
949 | int index; | |
950 | ||
951 | if (ret) { | |
952 | goto err; | |
953 | } | |
954 | ||
955 | if (need_alloc) { | |
956 | qed_unref_l2_cache_entry(acb->request.l2_table); | |
957 | acb->request.l2_table = qed_new_l2_table(s); | |
958 | } | |
959 | ||
960 | index = qed_l2_index(s, acb->cur_pos); | |
961 | qed_update_l2_table(s, acb->request.l2_table->table, index, acb->cur_nclusters, | |
962 | acb->cur_cluster); | |
963 | ||
964 | if (need_alloc) { | |
965 | /* Write out the whole new L2 table */ | |
966 | qed_write_l2_table(s, &acb->request, 0, s->table_nelems, true, | |
967 | qed_aio_write_l1_update, acb); | |
968 | } else { | |
969 | /* Write out only the updated part of the L2 table */ | |
970 | qed_write_l2_table(s, &acb->request, index, acb->cur_nclusters, false, | |
971 | qed_aio_next_io, acb); | |
972 | } | |
973 | return; | |
974 | ||
975 | err: | |
976 | qed_aio_complete(acb, ret); | |
977 | } | |
978 | ||
979 | /** | |
980 | * Flush new data clusters before updating the L2 table | |
981 | * | |
982 | * This flush is necessary when a backing file is in use. A crash during an | |
983 | * allocating write could result in empty clusters in the image. If the write | |
984 | * only touched a subregion of the cluster, then backing image sectors have | |
985 | * been lost in the untouched region. The solution is to flush after writing a | |
986 | * new data cluster and before updating the L2 table. | |
987 | */ | |
988 | static void qed_aio_write_flush_before_l2_update(void *opaque, int ret) | |
989 | { | |
990 | QEDAIOCB *acb = opaque; | |
991 | BDRVQEDState *s = acb_to_s(acb); | |
992 | ||
993 | if (!bdrv_aio_flush(s->bs->file, qed_aio_write_l2_update, opaque)) { | |
994 | qed_aio_complete(acb, -EIO); | |
995 | } | |
996 | } | |
997 | ||
998 | /** | |
999 | * Write data to the image file | |
1000 | */ | |
1001 | static void qed_aio_write_main(void *opaque, int ret) | |
1002 | { | |
1003 | QEDAIOCB *acb = opaque; | |
1004 | BDRVQEDState *s = acb_to_s(acb); | |
1005 | uint64_t offset = acb->cur_cluster + | |
1006 | qed_offset_into_cluster(s, acb->cur_pos); | |
1007 | BlockDriverCompletionFunc *next_fn; | |
eabba580 SH |
1008 | |
1009 | trace_qed_aio_write_main(s, acb, ret, offset, acb->cur_qiov.size); | |
1010 | ||
1011 | if (ret) { | |
1012 | qed_aio_complete(acb, ret); | |
1013 | return; | |
1014 | } | |
1015 | ||
1016 | if (acb->find_cluster_ret == QED_CLUSTER_FOUND) { | |
1017 | next_fn = qed_aio_next_io; | |
1018 | } else { | |
1019 | if (s->bs->backing_hd) { | |
1020 | next_fn = qed_aio_write_flush_before_l2_update; | |
1021 | } else { | |
1022 | next_fn = qed_aio_write_l2_update; | |
1023 | } | |
1024 | } | |
1025 | ||
1026 | BLKDBG_EVENT(s->bs->file, BLKDBG_WRITE_AIO); | |
ad54ae80 PB |
1027 | bdrv_aio_writev(s->bs->file, offset / BDRV_SECTOR_SIZE, |
1028 | &acb->cur_qiov, acb->cur_qiov.size / BDRV_SECTOR_SIZE, | |
1029 | next_fn, acb); | |
eabba580 SH |
1030 | } |
1031 | ||
1032 | /** | |
1033 | * Populate back untouched region of new data cluster | |
1034 | */ | |
1035 | static void qed_aio_write_postfill(void *opaque, int ret) | |
1036 | { | |
1037 | QEDAIOCB *acb = opaque; | |
1038 | BDRVQEDState *s = acb_to_s(acb); | |
1039 | uint64_t start = acb->cur_pos + acb->cur_qiov.size; | |
1040 | uint64_t len = | |
1041 | qed_start_of_cluster(s, start + s->header.cluster_size - 1) - start; | |
1042 | uint64_t offset = acb->cur_cluster + | |
1043 | qed_offset_into_cluster(s, acb->cur_pos) + | |
1044 | acb->cur_qiov.size; | |
1045 | ||
1046 | if (ret) { | |
1047 | qed_aio_complete(acb, ret); | |
1048 | return; | |
1049 | } | |
1050 | ||
1051 | trace_qed_aio_write_postfill(s, acb, start, len, offset); | |
1052 | qed_copy_from_backing_file(s, start, len, offset, | |
1053 | qed_aio_write_main, acb); | |
1054 | } | |
1055 | ||
1056 | /** | |
1057 | * Populate front untouched region of new data cluster | |
1058 | */ | |
1059 | static void qed_aio_write_prefill(void *opaque, int ret) | |
1060 | { | |
1061 | QEDAIOCB *acb = opaque; | |
1062 | BDRVQEDState *s = acb_to_s(acb); | |
1063 | uint64_t start = qed_start_of_cluster(s, acb->cur_pos); | |
1064 | uint64_t len = qed_offset_into_cluster(s, acb->cur_pos); | |
1065 | ||
1066 | trace_qed_aio_write_prefill(s, acb, start, len, acb->cur_cluster); | |
1067 | qed_copy_from_backing_file(s, start, len, acb->cur_cluster, | |
1068 | qed_aio_write_postfill, acb); | |
1069 | } | |
1070 | ||
0d09c797 SH |
1071 | /** |
1072 | * Check if the QED_F_NEED_CHECK bit should be set during allocating write | |
1073 | */ | |
1074 | static bool qed_should_set_need_check(BDRVQEDState *s) | |
1075 | { | |
1076 | /* The flush before L2 update path ensures consistency */ | |
1077 | if (s->bs->backing_hd) { | |
1078 | return false; | |
1079 | } | |
1080 | ||
1081 | return !(s->header.features & QED_F_NEED_CHECK); | |
1082 | } | |
1083 | ||
eabba580 SH |
1084 | /** |
1085 | * Write new data cluster | |
1086 | * | |
1087 | * @acb: Write request | |
1088 | * @len: Length in bytes | |
1089 | * | |
1090 | * This path is taken when writing to previously unallocated clusters. | |
1091 | */ | |
1092 | static void qed_aio_write_alloc(QEDAIOCB *acb, size_t len) | |
1093 | { | |
1094 | BDRVQEDState *s = acb_to_s(acb); | |
1095 | ||
6f321e93 SH |
1096 | /* Cancel timer when the first allocating request comes in */ |
1097 | if (QSIMPLEQ_EMPTY(&s->allocating_write_reqs)) { | |
1098 | qed_cancel_need_check_timer(s); | |
1099 | } | |
1100 | ||
eabba580 SH |
1101 | /* Freeze this request if another allocating write is in progress */ |
1102 | if (acb != QSIMPLEQ_FIRST(&s->allocating_write_reqs)) { | |
1103 | QSIMPLEQ_INSERT_TAIL(&s->allocating_write_reqs, acb, next); | |
1104 | } | |
6f321e93 SH |
1105 | if (acb != QSIMPLEQ_FIRST(&s->allocating_write_reqs) || |
1106 | s->allocating_write_reqs_plugged) { | |
eabba580 SH |
1107 | return; /* wait for existing request to finish */ |
1108 | } | |
1109 | ||
1110 | acb->cur_nclusters = qed_bytes_to_clusters(s, | |
1111 | qed_offset_into_cluster(s, acb->cur_pos) + len); | |
1112 | acb->cur_cluster = qed_alloc_clusters(s, acb->cur_nclusters); | |
1113 | qemu_iovec_copy(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); | |
1114 | ||
0d09c797 SH |
1115 | if (qed_should_set_need_check(s)) { |
1116 | s->header.features |= QED_F_NEED_CHECK; | |
1117 | qed_write_header(s, qed_aio_write_prefill, acb); | |
1118 | } else { | |
01979a98 | 1119 | qed_aio_write_prefill(acb, 0); |
01979a98 | 1120 | } |
eabba580 SH |
1121 | } |
1122 | ||
1123 | /** | |
1124 | * Write data cluster in place | |
1125 | * | |
1126 | * @acb: Write request | |
1127 | * @offset: Cluster offset in bytes | |
1128 | * @len: Length in bytes | |
1129 | * | |
1130 | * This path is taken when writing to already allocated clusters. | |
1131 | */ | |
1132 | static void qed_aio_write_inplace(QEDAIOCB *acb, uint64_t offset, size_t len) | |
1133 | { | |
1134 | /* Calculate the I/O vector */ | |
1135 | acb->cur_cluster = offset; | |
1136 | qemu_iovec_copy(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); | |
1137 | ||
1138 | /* Do the actual write */ | |
1139 | qed_aio_write_main(acb, 0); | |
1140 | } | |
1141 | ||
1142 | /** | |
1143 | * Write data cluster | |
1144 | * | |
1145 | * @opaque: Write request | |
1146 | * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1, | |
1147 | * or -errno | |
1148 | * @offset: Cluster offset in bytes | |
1149 | * @len: Length in bytes | |
1150 | * | |
1151 | * Callback from qed_find_cluster(). | |
1152 | */ | |
1153 | static void qed_aio_write_data(void *opaque, int ret, | |
1154 | uint64_t offset, size_t len) | |
1155 | { | |
1156 | QEDAIOCB *acb = opaque; | |
1157 | ||
1158 | trace_qed_aio_write_data(acb_to_s(acb), acb, ret, offset, len); | |
1159 | ||
1160 | acb->find_cluster_ret = ret; | |
1161 | ||
1162 | switch (ret) { | |
1163 | case QED_CLUSTER_FOUND: | |
1164 | qed_aio_write_inplace(acb, offset, len); | |
1165 | break; | |
1166 | ||
1167 | case QED_CLUSTER_L2: | |
1168 | case QED_CLUSTER_L1: | |
21df65b6 | 1169 | case QED_CLUSTER_ZERO: |
eabba580 SH |
1170 | qed_aio_write_alloc(acb, len); |
1171 | break; | |
1172 | ||
1173 | default: | |
1174 | qed_aio_complete(acb, ret); | |
1175 | break; | |
1176 | } | |
1177 | } | |
1178 | ||
1179 | /** | |
1180 | * Read data cluster | |
1181 | * | |
1182 | * @opaque: Read request | |
1183 | * @ret: QED_CLUSTER_FOUND, QED_CLUSTER_L2, QED_CLUSTER_L1, | |
1184 | * or -errno | |
1185 | * @offset: Cluster offset in bytes | |
1186 | * @len: Length in bytes | |
1187 | * | |
1188 | * Callback from qed_find_cluster(). | |
1189 | */ | |
1190 | static void qed_aio_read_data(void *opaque, int ret, | |
1191 | uint64_t offset, size_t len) | |
1192 | { | |
1193 | QEDAIOCB *acb = opaque; | |
1194 | BDRVQEDState *s = acb_to_s(acb); | |
1195 | BlockDriverState *bs = acb->common.bs; | |
eabba580 SH |
1196 | |
1197 | /* Adjust offset into cluster */ | |
1198 | offset += qed_offset_into_cluster(s, acb->cur_pos); | |
1199 | ||
1200 | trace_qed_aio_read_data(s, acb, ret, offset, len); | |
1201 | ||
1202 | if (ret < 0) { | |
1203 | goto err; | |
1204 | } | |
1205 | ||
1206 | qemu_iovec_copy(&acb->cur_qiov, acb->qiov, acb->qiov_offset, len); | |
1207 | ||
21df65b6 AL |
1208 | /* Handle zero cluster and backing file reads */ |
1209 | if (ret == QED_CLUSTER_ZERO) { | |
1210 | qemu_iovec_memset(&acb->cur_qiov, 0, acb->cur_qiov.size); | |
1211 | qed_aio_next_io(acb, 0); | |
1212 | return; | |
1213 | } else if (ret != QED_CLUSTER_FOUND) { | |
eabba580 SH |
1214 | qed_read_backing_file(s, acb->cur_pos, &acb->cur_qiov, |
1215 | qed_aio_next_io, acb); | |
1216 | return; | |
1217 | } | |
1218 | ||
1219 | BLKDBG_EVENT(bs->file, BLKDBG_READ_AIO); | |
ad54ae80 PB |
1220 | bdrv_aio_readv(bs->file, offset / BDRV_SECTOR_SIZE, |
1221 | &acb->cur_qiov, acb->cur_qiov.size / BDRV_SECTOR_SIZE, | |
1222 | qed_aio_next_io, acb); | |
eabba580 SH |
1223 | return; |
1224 | ||
1225 | err: | |
1226 | qed_aio_complete(acb, ret); | |
1227 | } | |
1228 | ||
1229 | /** | |
1230 | * Begin next I/O or complete the request | |
1231 | */ | |
1232 | static void qed_aio_next_io(void *opaque, int ret) | |
1233 | { | |
1234 | QEDAIOCB *acb = opaque; | |
1235 | BDRVQEDState *s = acb_to_s(acb); | |
1236 | QEDFindClusterFunc *io_fn = | |
1237 | acb->is_write ? qed_aio_write_data : qed_aio_read_data; | |
1238 | ||
1239 | trace_qed_aio_next_io(s, acb, ret, acb->cur_pos + acb->cur_qiov.size); | |
1240 | ||
1241 | /* Handle I/O error */ | |
1242 | if (ret) { | |
1243 | qed_aio_complete(acb, ret); | |
1244 | return; | |
1245 | } | |
1246 | ||
1247 | acb->qiov_offset += acb->cur_qiov.size; | |
1248 | acb->cur_pos += acb->cur_qiov.size; | |
1249 | qemu_iovec_reset(&acb->cur_qiov); | |
1250 | ||
1251 | /* Complete request */ | |
1252 | if (acb->cur_pos >= acb->end_pos) { | |
1253 | qed_aio_complete(acb, 0); | |
1254 | return; | |
1255 | } | |
1256 | ||
1257 | /* Find next cluster and start I/O */ | |
1258 | qed_find_cluster(s, &acb->request, | |
1259 | acb->cur_pos, acb->end_pos - acb->cur_pos, | |
1260 | io_fn, acb); | |
1261 | } | |
1262 | ||
1263 | static BlockDriverAIOCB *qed_aio_setup(BlockDriverState *bs, | |
1264 | int64_t sector_num, | |
1265 | QEMUIOVector *qiov, int nb_sectors, | |
1266 | BlockDriverCompletionFunc *cb, | |
1267 | void *opaque, bool is_write) | |
1268 | { | |
1269 | QEDAIOCB *acb = qemu_aio_get(&qed_aio_pool, bs, cb, opaque); | |
1270 | ||
1271 | trace_qed_aio_setup(bs->opaque, acb, sector_num, nb_sectors, | |
1272 | opaque, is_write); | |
1273 | ||
1274 | acb->is_write = is_write; | |
1275 | acb->finished = NULL; | |
1276 | acb->qiov = qiov; | |
1277 | acb->qiov_offset = 0; | |
1278 | acb->cur_pos = (uint64_t)sector_num * BDRV_SECTOR_SIZE; | |
1279 | acb->end_pos = acb->cur_pos + nb_sectors * BDRV_SECTOR_SIZE; | |
1280 | acb->request.l2_table = NULL; | |
1281 | qemu_iovec_init(&acb->cur_qiov, qiov->niov); | |
1282 | ||
1283 | /* Start request */ | |
1284 | qed_aio_next_io(acb, 0); | |
1285 | return &acb->common; | |
1286 | } | |
1287 | ||
75411d23 SH |
1288 | static BlockDriverAIOCB *bdrv_qed_aio_readv(BlockDriverState *bs, |
1289 | int64_t sector_num, | |
1290 | QEMUIOVector *qiov, int nb_sectors, | |
1291 | BlockDriverCompletionFunc *cb, | |
1292 | void *opaque) | |
1293 | { | |
eabba580 | 1294 | return qed_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, false); |
75411d23 SH |
1295 | } |
1296 | ||
1297 | static BlockDriverAIOCB *bdrv_qed_aio_writev(BlockDriverState *bs, | |
1298 | int64_t sector_num, | |
1299 | QEMUIOVector *qiov, int nb_sectors, | |
1300 | BlockDriverCompletionFunc *cb, | |
1301 | void *opaque) | |
1302 | { | |
eabba580 | 1303 | return qed_aio_setup(bs, sector_num, qiov, nb_sectors, cb, opaque, true); |
75411d23 SH |
1304 | } |
1305 | ||
1306 | static BlockDriverAIOCB *bdrv_qed_aio_flush(BlockDriverState *bs, | |
1307 | BlockDriverCompletionFunc *cb, | |
1308 | void *opaque) | |
1309 | { | |
1310 | return bdrv_aio_flush(bs->file, cb, opaque); | |
1311 | } | |
1312 | ||
1313 | static int bdrv_qed_truncate(BlockDriverState *bs, int64_t offset) | |
1314 | { | |
77a5a000 SH |
1315 | BDRVQEDState *s = bs->opaque; |
1316 | uint64_t old_image_size; | |
1317 | int ret; | |
1318 | ||
1319 | if (!qed_is_image_size_valid(offset, s->header.cluster_size, | |
1320 | s->header.table_size)) { | |
1321 | return -EINVAL; | |
1322 | } | |
1323 | ||
1324 | /* Shrinking is currently not supported */ | |
1325 | if ((uint64_t)offset < s->header.image_size) { | |
1326 | return -ENOTSUP; | |
1327 | } | |
1328 | ||
1329 | old_image_size = s->header.image_size; | |
1330 | s->header.image_size = offset; | |
1331 | ret = qed_write_header_sync(s); | |
1332 | if (ret < 0) { | |
1333 | s->header.image_size = old_image_size; | |
1334 | } | |
1335 | return ret; | |
75411d23 SH |
1336 | } |
1337 | ||
1338 | static int64_t bdrv_qed_getlength(BlockDriverState *bs) | |
1339 | { | |
1340 | BDRVQEDState *s = bs->opaque; | |
1341 | return s->header.image_size; | |
1342 | } | |
1343 | ||
1344 | static int bdrv_qed_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
1345 | { | |
1346 | BDRVQEDState *s = bs->opaque; | |
1347 | ||
1348 | memset(bdi, 0, sizeof(*bdi)); | |
1349 | bdi->cluster_size = s->header.cluster_size; | |
1350 | return 0; | |
1351 | } | |
1352 | ||
1353 | static int bdrv_qed_change_backing_file(BlockDriverState *bs, | |
1354 | const char *backing_file, | |
1355 | const char *backing_fmt) | |
1356 | { | |
1357 | BDRVQEDState *s = bs->opaque; | |
1358 | QEDHeader new_header, le_header; | |
1359 | void *buffer; | |
1360 | size_t buffer_len, backing_file_len; | |
1361 | int ret; | |
1362 | ||
1363 | /* Refuse to set backing filename if unknown compat feature bits are | |
1364 | * active. If the image uses an unknown compat feature then we may not | |
1365 | * know the layout of data following the header structure and cannot safely | |
1366 | * add a new string. | |
1367 | */ | |
1368 | if (backing_file && (s->header.compat_features & | |
1369 | ~QED_COMPAT_FEATURE_MASK)) { | |
1370 | return -ENOTSUP; | |
1371 | } | |
1372 | ||
1373 | memcpy(&new_header, &s->header, sizeof(new_header)); | |
1374 | ||
1375 | new_header.features &= ~(QED_F_BACKING_FILE | | |
1376 | QED_F_BACKING_FORMAT_NO_PROBE); | |
1377 | ||
1378 | /* Adjust feature flags */ | |
1379 | if (backing_file) { | |
1380 | new_header.features |= QED_F_BACKING_FILE; | |
1381 | ||
1382 | if (qed_fmt_is_raw(backing_fmt)) { | |
1383 | new_header.features |= QED_F_BACKING_FORMAT_NO_PROBE; | |
1384 | } | |
1385 | } | |
1386 | ||
1387 | /* Calculate new header size */ | |
1388 | backing_file_len = 0; | |
1389 | ||
1390 | if (backing_file) { | |
1391 | backing_file_len = strlen(backing_file); | |
1392 | } | |
1393 | ||
1394 | buffer_len = sizeof(new_header); | |
1395 | new_header.backing_filename_offset = buffer_len; | |
1396 | new_header.backing_filename_size = backing_file_len; | |
1397 | buffer_len += backing_file_len; | |
1398 | ||
1399 | /* Make sure we can rewrite header without failing */ | |
1400 | if (buffer_len > new_header.header_size * new_header.cluster_size) { | |
1401 | return -ENOSPC; | |
1402 | } | |
1403 | ||
1404 | /* Prepare new header */ | |
7267c094 | 1405 | buffer = g_malloc(buffer_len); |
75411d23 SH |
1406 | |
1407 | qed_header_cpu_to_le(&new_header, &le_header); | |
1408 | memcpy(buffer, &le_header, sizeof(le_header)); | |
1409 | buffer_len = sizeof(le_header); | |
1410 | ||
feba23b1 PB |
1411 | if (backing_file) { |
1412 | memcpy(buffer + buffer_len, backing_file, backing_file_len); | |
1413 | buffer_len += backing_file_len; | |
1414 | } | |
75411d23 SH |
1415 | |
1416 | /* Write new header */ | |
1417 | ret = bdrv_pwrite_sync(bs->file, 0, buffer, buffer_len); | |
7267c094 | 1418 | g_free(buffer); |
75411d23 SH |
1419 | if (ret == 0) { |
1420 | memcpy(&s->header, &new_header, sizeof(new_header)); | |
1421 | } | |
1422 | return ret; | |
1423 | } | |
1424 | ||
1425 | static int bdrv_qed_check(BlockDriverState *bs, BdrvCheckResult *result) | |
1426 | { | |
01979a98 SH |
1427 | BDRVQEDState *s = bs->opaque; |
1428 | ||
1429 | return qed_check(s, result, false); | |
75411d23 SH |
1430 | } |
1431 | ||
1432 | static QEMUOptionParameter qed_create_options[] = { | |
1433 | { | |
1434 | .name = BLOCK_OPT_SIZE, | |
1435 | .type = OPT_SIZE, | |
1436 | .help = "Virtual disk size (in bytes)" | |
1437 | }, { | |
1438 | .name = BLOCK_OPT_BACKING_FILE, | |
1439 | .type = OPT_STRING, | |
1440 | .help = "File name of a base image" | |
1441 | }, { | |
1442 | .name = BLOCK_OPT_BACKING_FMT, | |
1443 | .type = OPT_STRING, | |
1444 | .help = "Image format of the base image" | |
1445 | }, { | |
1446 | .name = BLOCK_OPT_CLUSTER_SIZE, | |
1447 | .type = OPT_SIZE, | |
99cce9fa KW |
1448 | .help = "Cluster size (in bytes)", |
1449 | .value = { .n = QED_DEFAULT_CLUSTER_SIZE }, | |
75411d23 SH |
1450 | }, { |
1451 | .name = BLOCK_OPT_TABLE_SIZE, | |
1452 | .type = OPT_SIZE, | |
1453 | .help = "L1/L2 table size (in clusters)" | |
1454 | }, | |
1455 | { /* end of list */ } | |
1456 | }; | |
1457 | ||
1458 | static BlockDriver bdrv_qed = { | |
1459 | .format_name = "qed", | |
1460 | .instance_size = sizeof(BDRVQEDState), | |
1461 | .create_options = qed_create_options, | |
1462 | ||
1463 | .bdrv_probe = bdrv_qed_probe, | |
1464 | .bdrv_open = bdrv_qed_open, | |
1465 | .bdrv_close = bdrv_qed_close, | |
1466 | .bdrv_create = bdrv_qed_create, | |
b7d5a5b8 | 1467 | .bdrv_co_is_allocated = bdrv_qed_co_is_allocated, |
75411d23 SH |
1468 | .bdrv_make_empty = bdrv_qed_make_empty, |
1469 | .bdrv_aio_readv = bdrv_qed_aio_readv, | |
1470 | .bdrv_aio_writev = bdrv_qed_aio_writev, | |
1471 | .bdrv_aio_flush = bdrv_qed_aio_flush, | |
1472 | .bdrv_truncate = bdrv_qed_truncate, | |
1473 | .bdrv_getlength = bdrv_qed_getlength, | |
1474 | .bdrv_get_info = bdrv_qed_get_info, | |
1475 | .bdrv_change_backing_file = bdrv_qed_change_backing_file, | |
1476 | .bdrv_check = bdrv_qed_check, | |
1477 | }; | |
1478 | ||
1479 | static void bdrv_qed_init(void) | |
1480 | { | |
1481 | bdrv_register(&bdrv_qed); | |
1482 | } | |
1483 | ||
1484 | block_init(bdrv_qed_init); |