1
2
3
4
5
6
7
8
9
10
11
12#include "qemu/osdep.h"
13#include "qapi/error.h"
14#include "qemu/sockets.h"
15#include "block/block_int.h"
16#include "qapi/qmp/qdict.h"
17#include "qapi/qmp/qstring.h"
18#include "qemu/cutils.h"
19#include "qemu/option.h"
20
21
22
23#define LOG_FLUSH_FLAG (1 << 0)
24#define LOG_FUA_FLAG (1 << 1)
25#define LOG_DISCARD_FLAG (1 << 2)
26#define LOG_MARK_FLAG (1 << 3)
27#define LOG_FLAG_MASK (LOG_FLUSH_FLAG \
28 | LOG_FUA_FLAG \
29 | LOG_DISCARD_FLAG \
30 | LOG_MARK_FLAG)
31
32#define WRITE_LOG_VERSION 1ULL
33#define WRITE_LOG_MAGIC 0x6a736677736872ULL
34
35
36struct log_write_super {
37 uint64_t magic;
38 uint64_t version;
39 uint64_t nr_entries;
40 uint32_t sectorsize;
41} QEMU_PACKED;
42
43struct log_write_entry {
44 uint64_t sector;
45 uint64_t nr_sectors;
46 uint64_t flags;
47 uint64_t data_len;
48} QEMU_PACKED;
49
50
51
52typedef struct {
53 BdrvChild *log_file;
54 uint32_t sectorsize;
55 uint32_t sectorbits;
56 uint64_t cur_log_sector;
57 uint64_t nr_entries;
58 uint64_t update_interval;
59} BDRVBlkLogWritesState;
60
61static QemuOptsList runtime_opts = {
62 .name = "blklogwrites",
63 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
64 .desc = {
65 {
66 .name = "log-append",
67 .type = QEMU_OPT_BOOL,
68 .help = "Append to an existing log",
69 },
70 {
71 .name = "log-sector-size",
72 .type = QEMU_OPT_SIZE,
73 .help = "Log sector size",
74 },
75 {
76 .name = "log-super-update-interval",
77 .type = QEMU_OPT_NUMBER,
78 .help = "Log superblock update interval (# of write requests)",
79 },
80 { }
81 },
82};
83
84static inline uint32_t blk_log_writes_log2(uint32_t value)
85{
86 assert(value > 0);
87 return 31 - clz32(value);
88}
89
90static inline bool blk_log_writes_sector_size_valid(uint32_t sector_size)
91{
92 return is_power_of_2(sector_size) &&
93 sector_size >= sizeof(struct log_write_super) &&
94 sector_size >= sizeof(struct log_write_entry) &&
95 sector_size < (1ull << 24);
96}
97
98static uint64_t blk_log_writes_find_cur_log_sector(BdrvChild *log,
99 uint32_t sector_size,
100 uint64_t nr_entries,
101 Error **errp)
102{
103 uint64_t cur_sector = 1;
104 uint64_t cur_idx = 0;
105 uint32_t sector_bits = blk_log_writes_log2(sector_size);
106 struct log_write_entry cur_entry;
107
108 while (cur_idx < nr_entries) {
109 int read_ret = bdrv_pread(log, cur_sector << sector_bits, &cur_entry,
110 sizeof(cur_entry));
111 if (read_ret < 0) {
112 error_setg_errno(errp, -read_ret,
113 "Failed to read log entry %"PRIu64, cur_idx);
114 return (uint64_t)-1ull;
115 }
116
117 if (cur_entry.flags & ~cpu_to_le64(LOG_FLAG_MASK)) {
118 error_setg(errp, "Invalid flags 0x%"PRIx64" in log entry %"PRIu64,
119 le64_to_cpu(cur_entry.flags), cur_idx);
120 return (uint64_t)-1ull;
121 }
122
123
124 ++cur_sector;
125
126
127
128
129
130 if (!(cur_entry.flags & cpu_to_le64(LOG_DISCARD_FLAG))) {
131 cur_sector += le64_to_cpu(cur_entry.nr_sectors);
132 }
133
134 ++cur_idx;
135 }
136
137 return cur_sector;
138}
139
140static int blk_log_writes_open(BlockDriverState *bs, QDict *options, int flags,
141 Error **errp)
142{
143 BDRVBlkLogWritesState *s = bs->opaque;
144 QemuOpts *opts;
145 Error *local_err = NULL;
146 int ret;
147 uint64_t log_sector_size;
148 bool log_append;
149
150 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
151 qemu_opts_absorb_qdict(opts, options, &local_err);
152 if (local_err) {
153 ret = -EINVAL;
154 error_propagate(errp, local_err);
155 goto fail;
156 }
157
158
159 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, false,
160 &local_err);
161 if (local_err) {
162 ret = -EINVAL;
163 error_propagate(errp, local_err);
164 goto fail;
165 }
166
167
168 s->log_file = bdrv_open_child(NULL, options, "log", bs, &child_file, false,
169 &local_err);
170 if (local_err) {
171 ret = -EINVAL;
172 error_propagate(errp, local_err);
173 goto fail;
174 }
175
176 log_append = qemu_opt_get_bool(opts, "log-append", false);
177
178 if (log_append) {
179 struct log_write_super log_sb = { 0, 0, 0, 0 };
180
181 if (qemu_opt_find(opts, "log-sector-size")) {
182 ret = -EINVAL;
183 error_setg(errp, "log-append and log-sector-size are mutually "
184 "exclusive");
185 goto fail_log;
186 }
187
188
189 if (!bdrv_getlength(s->log_file->bs)) {
190 log_sb.magic = cpu_to_le64(WRITE_LOG_MAGIC);
191 log_sb.version = cpu_to_le64(WRITE_LOG_VERSION);
192 log_sb.nr_entries = cpu_to_le64(0);
193 log_sb.sectorsize = cpu_to_le32(BDRV_SECTOR_SIZE);
194 } else {
195 ret = bdrv_pread(s->log_file, 0, &log_sb, sizeof(log_sb));
196 if (ret < 0) {
197 error_setg_errno(errp, -ret, "Could not read log superblock");
198 goto fail_log;
199 }
200 }
201
202 if (log_sb.magic != cpu_to_le64(WRITE_LOG_MAGIC)) {
203 ret = -EINVAL;
204 error_setg(errp, "Invalid log superblock magic");
205 goto fail_log;
206 }
207
208 if (log_sb.version != cpu_to_le64(WRITE_LOG_VERSION)) {
209 ret = -EINVAL;
210 error_setg(errp, "Unsupported log version %"PRIu64,
211 le64_to_cpu(log_sb.version));
212 goto fail_log;
213 }
214
215 log_sector_size = le32_to_cpu(log_sb.sectorsize);
216 s->cur_log_sector = 1;
217 s->nr_entries = 0;
218
219 if (blk_log_writes_sector_size_valid(log_sector_size)) {
220 s->cur_log_sector =
221 blk_log_writes_find_cur_log_sector(s->log_file, log_sector_size,
222 le64_to_cpu(log_sb.nr_entries), &local_err);
223 if (local_err) {
224 ret = -EINVAL;
225 error_propagate(errp, local_err);
226 goto fail_log;
227 }
228
229 s->nr_entries = le64_to_cpu(log_sb.nr_entries);
230 }
231 } else {
232 log_sector_size = qemu_opt_get_size(opts, "log-sector-size",
233 BDRV_SECTOR_SIZE);
234 s->cur_log_sector = 1;
235 s->nr_entries = 0;
236 }
237
238 if (!blk_log_writes_sector_size_valid(log_sector_size)) {
239 ret = -EINVAL;
240 error_setg(errp, "Invalid log sector size %"PRIu64, log_sector_size);
241 goto fail_log;
242 }
243
244 s->sectorsize = log_sector_size;
245 s->sectorbits = blk_log_writes_log2(log_sector_size);
246 s->update_interval = qemu_opt_get_number(opts, "log-super-update-interval",
247 4096);
248 if (!s->update_interval) {
249 ret = -EINVAL;
250 error_setg(errp, "Invalid log superblock update interval %"PRIu64,
251 s->update_interval);
252 goto fail_log;
253 }
254
255 ret = 0;
256fail_log:
257 if (ret < 0) {
258 bdrv_unref_child(bs, s->log_file);
259 s->log_file = NULL;
260 }
261fail:
262 if (ret < 0) {
263 bdrv_unref_child(bs, bs->file);
264 bs->file = NULL;
265 }
266 qemu_opts_del(opts);
267 return ret;
268}
269
270static void blk_log_writes_close(BlockDriverState *bs)
271{
272 BDRVBlkLogWritesState *s = bs->opaque;
273
274 bdrv_unref_child(bs, s->log_file);
275 s->log_file = NULL;
276}
277
278static int64_t blk_log_writes_getlength(BlockDriverState *bs)
279{
280 return bdrv_getlength(bs->file->bs);
281}
282
283static void blk_log_writes_child_perm(BlockDriverState *bs, BdrvChild *c,
284 const BdrvChildRole *role,
285 BlockReopenQueue *ro_q,
286 uint64_t perm, uint64_t shrd,
287 uint64_t *nperm, uint64_t *nshrd)
288{
289 if (!c) {
290 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
291 *nshrd = (shrd & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
292 return;
293 }
294
295 if (!strcmp(c->name, "log")) {
296 bdrv_format_default_perms(bs, c, role, ro_q, perm, shrd, nperm, nshrd);
297 } else {
298 bdrv_filter_default_perms(bs, c, role, ro_q, perm, shrd, nperm, nshrd);
299 }
300}
301
302static void blk_log_writes_refresh_limits(BlockDriverState *bs, Error **errp)
303{
304 BDRVBlkLogWritesState *s = bs->opaque;
305 bs->bl.request_alignment = s->sectorsize;
306}
307
308static int coroutine_fn
309blk_log_writes_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
310 QEMUIOVector *qiov, int flags)
311{
312 return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
313}
314
315typedef struct BlkLogWritesFileReq {
316 BlockDriverState *bs;
317 uint64_t offset;
318 uint64_t bytes;
319 int file_flags;
320 QEMUIOVector *qiov;
321 int (*func)(struct BlkLogWritesFileReq *r);
322 int file_ret;
323} BlkLogWritesFileReq;
324
325typedef struct {
326 BlockDriverState *bs;
327 QEMUIOVector *qiov;
328 struct log_write_entry entry;
329 uint64_t zero_size;
330 int log_ret;
331} BlkLogWritesLogReq;
332
333static void coroutine_fn blk_log_writes_co_do_log(BlkLogWritesLogReq *lr)
334{
335 BDRVBlkLogWritesState *s = lr->bs->opaque;
336 uint64_t cur_log_offset = s->cur_log_sector << s->sectorbits;
337
338 s->nr_entries++;
339 s->cur_log_sector +=
340 ROUND_UP(lr->qiov->size, s->sectorsize) >> s->sectorbits;
341
342 lr->log_ret = bdrv_co_pwritev(s->log_file, cur_log_offset, lr->qiov->size,
343 lr->qiov, 0);
344
345
346 if (lr->log_ret == 0 && lr->zero_size) {
347 cur_log_offset = s->cur_log_sector << s->sectorbits;
348 s->cur_log_sector +=
349 ROUND_UP(lr->zero_size, s->sectorsize) >> s->sectorbits;
350
351 lr->log_ret = bdrv_co_pwrite_zeroes(s->log_file, cur_log_offset,
352 lr->zero_size, 0);
353 }
354
355
356 if (lr->log_ret == 0 && ((lr->entry.flags & LOG_FLUSH_FLAG)
357 || (s->nr_entries % s->update_interval == 0)))
358 {
359 struct log_write_super super = {
360 .magic = cpu_to_le64(WRITE_LOG_MAGIC),
361 .version = cpu_to_le64(WRITE_LOG_VERSION),
362 .nr_entries = cpu_to_le64(s->nr_entries),
363 .sectorsize = cpu_to_le32(s->sectorsize),
364 };
365 void *zeroes = g_malloc0(s->sectorsize - sizeof(super));
366 QEMUIOVector qiov;
367
368 qemu_iovec_init(&qiov, 2);
369 qemu_iovec_add(&qiov, &super, sizeof(super));
370 qemu_iovec_add(&qiov, zeroes, s->sectorsize - sizeof(super));
371
372 lr->log_ret =
373 bdrv_co_pwritev(s->log_file, 0, s->sectorsize, &qiov, 0);
374 if (lr->log_ret == 0) {
375 lr->log_ret = bdrv_co_flush(s->log_file->bs);
376 }
377 qemu_iovec_destroy(&qiov);
378 g_free(zeroes);
379 }
380}
381
382static void coroutine_fn blk_log_writes_co_do_file(BlkLogWritesFileReq *fr)
383{
384 fr->file_ret = fr->func(fr);
385}
386
387static int coroutine_fn
388blk_log_writes_co_log(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
389 QEMUIOVector *qiov, int flags,
390 int (*file_func)(BlkLogWritesFileReq *r),
391 uint64_t entry_flags, bool is_zero_write)
392{
393 QEMUIOVector log_qiov;
394 size_t niov = qiov ? qiov->niov : 0;
395 BDRVBlkLogWritesState *s = bs->opaque;
396 BlkLogWritesFileReq fr = {
397 .bs = bs,
398 .offset = offset,
399 .bytes = bytes,
400 .file_flags = flags,
401 .qiov = qiov,
402 .func = file_func,
403 };
404 BlkLogWritesLogReq lr = {
405 .bs = bs,
406 .qiov = &log_qiov,
407 .entry = {
408 .sector = cpu_to_le64(offset >> s->sectorbits),
409 .nr_sectors = cpu_to_le64(bytes >> s->sectorbits),
410 .flags = cpu_to_le64(entry_flags),
411 .data_len = 0,
412 },
413 .zero_size = is_zero_write ? bytes : 0,
414 };
415 void *zeroes = g_malloc0(s->sectorsize - sizeof(lr.entry));
416
417 assert((1 << s->sectorbits) == s->sectorsize);
418 assert(bs->bl.request_alignment == s->sectorsize);
419 assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
420 assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
421
422 qemu_iovec_init(&log_qiov, niov + 2);
423 qemu_iovec_add(&log_qiov, &lr.entry, sizeof(lr.entry));
424 qemu_iovec_add(&log_qiov, zeroes, s->sectorsize - sizeof(lr.entry));
425 if (qiov) {
426 qemu_iovec_concat(&log_qiov, qiov, 0, qiov->size);
427 }
428
429 blk_log_writes_co_do_file(&fr);
430 blk_log_writes_co_do_log(&lr);
431
432 qemu_iovec_destroy(&log_qiov);
433 g_free(zeroes);
434
435 if (lr.log_ret < 0) {
436 return lr.log_ret;
437 }
438
439 return fr.file_ret;
440}
441
442static int coroutine_fn
443blk_log_writes_co_do_file_pwritev(BlkLogWritesFileReq *fr)
444{
445 return bdrv_co_pwritev(fr->bs->file, fr->offset, fr->bytes,
446 fr->qiov, fr->file_flags);
447}
448
449static int coroutine_fn
450blk_log_writes_co_do_file_pwrite_zeroes(BlkLogWritesFileReq *fr)
451{
452 return bdrv_co_pwrite_zeroes(fr->bs->file, fr->offset, fr->bytes,
453 fr->file_flags);
454}
455
456static int coroutine_fn blk_log_writes_co_do_file_flush(BlkLogWritesFileReq *fr)
457{
458 return bdrv_co_flush(fr->bs->file->bs);
459}
460
461static int coroutine_fn
462blk_log_writes_co_do_file_pdiscard(BlkLogWritesFileReq *fr)
463{
464 return bdrv_co_pdiscard(fr->bs->file, fr->offset, fr->bytes);
465}
466
467static int coroutine_fn
468blk_log_writes_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
469 QEMUIOVector *qiov, int flags)
470{
471 return blk_log_writes_co_log(bs, offset, bytes, qiov, flags,
472 blk_log_writes_co_do_file_pwritev, 0, false);
473}
474
475static int coroutine_fn
476blk_log_writes_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes,
477 BdrvRequestFlags flags)
478{
479 return blk_log_writes_co_log(bs, offset, bytes, NULL, flags,
480 blk_log_writes_co_do_file_pwrite_zeroes, 0,
481 true);
482}
483
484static int coroutine_fn blk_log_writes_co_flush_to_disk(BlockDriverState *bs)
485{
486 return blk_log_writes_co_log(bs, 0, 0, NULL, 0,
487 blk_log_writes_co_do_file_flush,
488 LOG_FLUSH_FLAG, false);
489}
490
491static int coroutine_fn
492blk_log_writes_co_pdiscard(BlockDriverState *bs, int64_t offset, int count)
493{
494 return blk_log_writes_co_log(bs, offset, count, NULL, 0,
495 blk_log_writes_co_do_file_pdiscard,
496 LOG_DISCARD_FLAG, false);
497}
498
499static const char *const blk_log_writes_strong_runtime_opts[] = {
500 "log-append",
501 "log-sector-size",
502
503 NULL
504};
505
506static BlockDriver bdrv_blk_log_writes = {
507 .format_name = "blklogwrites",
508 .instance_size = sizeof(BDRVBlkLogWritesState),
509
510 .bdrv_open = blk_log_writes_open,
511 .bdrv_close = blk_log_writes_close,
512 .bdrv_getlength = blk_log_writes_getlength,
513 .bdrv_child_perm = blk_log_writes_child_perm,
514 .bdrv_refresh_limits = blk_log_writes_refresh_limits,
515
516 .bdrv_co_preadv = blk_log_writes_co_preadv,
517 .bdrv_co_pwritev = blk_log_writes_co_pwritev,
518 .bdrv_co_pwrite_zeroes = blk_log_writes_co_pwrite_zeroes,
519 .bdrv_co_flush_to_disk = blk_log_writes_co_flush_to_disk,
520 .bdrv_co_pdiscard = blk_log_writes_co_pdiscard,
521 .bdrv_co_block_status = bdrv_co_block_status_from_file,
522
523 .is_filter = true,
524 .strong_runtime_opts = blk_log_writes_strong_runtime_opts,
525};
526
527static void bdrv_blk_log_writes_init(void)
528{
529 bdrv_register(&bdrv_blk_log_writes);
530}
531
532block_init(bdrv_blk_log_writes_init);
533
