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/module.h"
20#include "qemu/option.h"
21
22
23
24#define LOG_FLUSH_FLAG (1 << 0)
25#define LOG_FUA_FLAG (1 << 1)
26#define LOG_DISCARD_FLAG (1 << 2)
27#define LOG_MARK_FLAG (1 << 3)
28#define LOG_FLAG_MASK (LOG_FLUSH_FLAG \
29 | LOG_FUA_FLAG \
30 | LOG_DISCARD_FLAG \
31 | LOG_MARK_FLAG)
32
33#define WRITE_LOG_VERSION 1ULL
34#define WRITE_LOG_MAGIC 0x6a736677736872ULL
35
36
37struct log_write_super {
38 uint64_t magic;
39 uint64_t version;
40 uint64_t nr_entries;
41 uint32_t sectorsize;
42} QEMU_PACKED;
43
44struct log_write_entry {
45 uint64_t sector;
46 uint64_t nr_sectors;
47 uint64_t flags;
48 uint64_t data_len;
49} QEMU_PACKED;
50
51
52
53typedef struct {
54 BdrvChild *log_file;
55 uint32_t sectorsize;
56 uint32_t sectorbits;
57 uint64_t cur_log_sector;
58 uint64_t nr_entries;
59 uint64_t update_interval;
60} BDRVBlkLogWritesState;
61
62static QemuOptsList runtime_opts = {
63 .name = "blklogwrites",
64 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
65 .desc = {
66 {
67 .name = "log-append",
68 .type = QEMU_OPT_BOOL,
69 .help = "Append to an existing log",
70 },
71 {
72 .name = "log-sector-size",
73 .type = QEMU_OPT_SIZE,
74 .help = "Log sector size",
75 },
76 {
77 .name = "log-super-update-interval",
78 .type = QEMU_OPT_NUMBER,
79 .help = "Log superblock update interval (# of write requests)",
80 },
81 { }
82 },
83};
84
85static inline uint32_t blk_log_writes_log2(uint32_t value)
86{
87 assert(value > 0);
88 return 31 - clz32(value);
89}
90
91static inline bool blk_log_writes_sector_size_valid(uint32_t sector_size)
92{
93 return is_power_of_2(sector_size) &&
94 sector_size >= sizeof(struct log_write_super) &&
95 sector_size >= sizeof(struct log_write_entry) &&
96 sector_size < (1ull << 24);
97}
98
99static uint64_t blk_log_writes_find_cur_log_sector(BdrvChild *log,
100 uint32_t sector_size,
101 uint64_t nr_entries,
102 Error **errp)
103{
104 uint64_t cur_sector = 1;
105 uint64_t cur_idx = 0;
106 uint32_t sector_bits = blk_log_writes_log2(sector_size);
107 struct log_write_entry cur_entry;
108
109 while (cur_idx < nr_entries) {
110 int read_ret = bdrv_pread(log, cur_sector << sector_bits, &cur_entry,
111 sizeof(cur_entry));
112 if (read_ret < 0) {
113 error_setg_errno(errp, -read_ret,
114 "Failed to read log entry %"PRIu64, cur_idx);
115 return (uint64_t)-1ull;
116 }
117
118 if (cur_entry.flags & ~cpu_to_le64(LOG_FLAG_MASK)) {
119 error_setg(errp, "Invalid flags 0x%"PRIx64" in log entry %"PRIu64,
120 le64_to_cpu(cur_entry.flags), cur_idx);
121 return (uint64_t)-1ull;
122 }
123
124
125 ++cur_sector;
126
127
128
129
130
131 if (!(cur_entry.flags & cpu_to_le64(LOG_DISCARD_FLAG))) {
132 cur_sector += le64_to_cpu(cur_entry.nr_sectors);
133 }
134
135 ++cur_idx;
136 }
137
138 return cur_sector;
139}
140
141static int blk_log_writes_open(BlockDriverState *bs, QDict *options, int flags,
142 Error **errp)
143{
144 BDRVBlkLogWritesState *s = bs->opaque;
145 QemuOpts *opts;
146 Error *local_err = NULL;
147 int ret;
148 uint64_t log_sector_size;
149 bool log_append;
150
151 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
152 qemu_opts_absorb_qdict(opts, options, &local_err);
153 if (local_err) {
154 ret = -EINVAL;
155 error_propagate(errp, local_err);
156 goto fail;
157 }
158
159
160 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file, false,
161 &local_err);
162 if (local_err) {
163 ret = -EINVAL;
164 error_propagate(errp, local_err);
165 goto fail;
166 }
167
168
169 s->log_file = bdrv_open_child(NULL, options, "log", bs, &child_file, false,
170 &local_err);
171 if (local_err) {
172 ret = -EINVAL;
173 error_propagate(errp, local_err);
174 goto fail;
175 }
176
177 log_append = qemu_opt_get_bool(opts, "log-append", false);
178
179 if (log_append) {
180 struct log_write_super log_sb = { 0, 0, 0, 0 };
181
182 if (qemu_opt_find(opts, "log-sector-size")) {
183 ret = -EINVAL;
184 error_setg(errp, "log-append and log-sector-size are mutually "
185 "exclusive");
186 goto fail_log;
187 }
188
189
190 if (!bdrv_getlength(s->log_file->bs)) {
191 log_sb.magic = cpu_to_le64(WRITE_LOG_MAGIC);
192 log_sb.version = cpu_to_le64(WRITE_LOG_VERSION);
193 log_sb.nr_entries = cpu_to_le64(0);
194 log_sb.sectorsize = cpu_to_le32(BDRV_SECTOR_SIZE);
195 } else {
196 ret = bdrv_pread(s->log_file, 0, &log_sb, sizeof(log_sb));
197 if (ret < 0) {
198 error_setg_errno(errp, -ret, "Could not read log superblock");
199 goto fail_log;
200 }
201 }
202
203 if (log_sb.magic != cpu_to_le64(WRITE_LOG_MAGIC)) {
204 ret = -EINVAL;
205 error_setg(errp, "Invalid log superblock magic");
206 goto fail_log;
207 }
208
209 if (log_sb.version != cpu_to_le64(WRITE_LOG_VERSION)) {
210 ret = -EINVAL;
211 error_setg(errp, "Unsupported log version %"PRIu64,
212 le64_to_cpu(log_sb.version));
213 goto fail_log;
214 }
215
216 log_sector_size = le32_to_cpu(log_sb.sectorsize);
217 s->cur_log_sector = 1;
218 s->nr_entries = 0;
219
220 if (blk_log_writes_sector_size_valid(log_sector_size)) {
221 s->cur_log_sector =
222 blk_log_writes_find_cur_log_sector(s->log_file, log_sector_size,
223 le64_to_cpu(log_sb.nr_entries), &local_err);
224 if (local_err) {
225 ret = -EINVAL;
226 error_propagate(errp, local_err);
227 goto fail_log;
228 }
229
230 s->nr_entries = le64_to_cpu(log_sb.nr_entries);
231 }
232 } else {
233 log_sector_size = qemu_opt_get_size(opts, "log-sector-size",
234 BDRV_SECTOR_SIZE);
235 s->cur_log_sector = 1;
236 s->nr_entries = 0;
237 }
238
239 if (!blk_log_writes_sector_size_valid(log_sector_size)) {
240 ret = -EINVAL;
241 error_setg(errp, "Invalid log sector size %"PRIu64, log_sector_size);
242 goto fail_log;
243 }
244
245 s->sectorsize = log_sector_size;
246 s->sectorbits = blk_log_writes_log2(log_sector_size);
247 s->update_interval = qemu_opt_get_number(opts, "log-super-update-interval",
248 4096);
249 if (!s->update_interval) {
250 ret = -EINVAL;
251 error_setg(errp, "Invalid log superblock update interval %"PRIu64,
252 s->update_interval);
253 goto fail_log;
254 }
255
256 ret = 0;
257fail_log:
258 if (ret < 0) {
259 bdrv_unref_child(bs, s->log_file);
260 s->log_file = NULL;
261 }
262fail:
263 if (ret < 0) {
264 bdrv_unref_child(bs, bs->file);
265 bs->file = NULL;
266 }
267 qemu_opts_del(opts);
268 return ret;
269}
270
271static void blk_log_writes_close(BlockDriverState *bs)
272{
273 BDRVBlkLogWritesState *s = bs->opaque;
274
275 bdrv_unref_child(bs, s->log_file);
276 s->log_file = NULL;
277}
278
279static int64_t blk_log_writes_getlength(BlockDriverState *bs)
280{
281 return bdrv_getlength(bs->file->bs);
282}
283
284static void blk_log_writes_child_perm(BlockDriverState *bs, BdrvChild *c,
285 const BdrvChildRole *role,
286 BlockReopenQueue *ro_q,
287 uint64_t perm, uint64_t shrd,
288 uint64_t *nperm, uint64_t *nshrd)
289{
290 if (!c) {
291 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
292 *nshrd = (shrd & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
293 return;
294 }
295
296 if (!strcmp(c->name, "log")) {
297 bdrv_format_default_perms(bs, c, role, ro_q, perm, shrd, nperm, nshrd);
298 } else {
299 bdrv_filter_default_perms(bs, c, role, ro_q, perm, shrd, nperm, nshrd);
300 }
301}
302
303static void blk_log_writes_refresh_limits(BlockDriverState *bs, Error **errp)
304{
305 BDRVBlkLogWritesState *s = bs->opaque;
306 bs->bl.request_alignment = s->sectorsize;
307}
308
309static int coroutine_fn
310blk_log_writes_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
311 QEMUIOVector *qiov, int flags)
312{
313 return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
314}
315
316typedef struct BlkLogWritesFileReq {
317 BlockDriverState *bs;
318 uint64_t offset;
319 uint64_t bytes;
320 int file_flags;
321 QEMUIOVector *qiov;
322 int (*func)(struct BlkLogWritesFileReq *r);
323 int file_ret;
324} BlkLogWritesFileReq;
325
326typedef struct {
327 BlockDriverState *bs;
328 QEMUIOVector *qiov;
329 struct log_write_entry entry;
330 uint64_t zero_size;
331 int log_ret;
332} BlkLogWritesLogReq;
333
334static void coroutine_fn blk_log_writes_co_do_log(BlkLogWritesLogReq *lr)
335{
336 BDRVBlkLogWritesState *s = lr->bs->opaque;
337 uint64_t cur_log_offset = s->cur_log_sector << s->sectorbits;
338
339 s->nr_entries++;
340 s->cur_log_sector +=
341 ROUND_UP(lr->qiov->size, s->sectorsize) >> s->sectorbits;
342
343 lr->log_ret = bdrv_co_pwritev(s->log_file, cur_log_offset, lr->qiov->size,
344 lr->qiov, 0);
345
346
347 if (lr->log_ret == 0 && lr->zero_size) {
348 cur_log_offset = s->cur_log_sector << s->sectorbits;
349 s->cur_log_sector +=
350 ROUND_UP(lr->zero_size, s->sectorsize) >> s->sectorbits;
351
352 lr->log_ret = bdrv_co_pwrite_zeroes(s->log_file, cur_log_offset,
353 lr->zero_size, 0);
354 }
355
356
357 if (lr->log_ret == 0 && ((lr->entry.flags & LOG_FLUSH_FLAG)
358 || (s->nr_entries % s->update_interval == 0)))
359 {
360 struct log_write_super super = {
361 .magic = cpu_to_le64(WRITE_LOG_MAGIC),
362 .version = cpu_to_le64(WRITE_LOG_VERSION),
363 .nr_entries = cpu_to_le64(s->nr_entries),
364 .sectorsize = cpu_to_le32(s->sectorsize),
365 };
366 void *zeroes = g_malloc0(s->sectorsize - sizeof(super));
367 QEMUIOVector qiov;
368
369 qemu_iovec_init(&qiov, 2);
370 qemu_iovec_add(&qiov, &super, sizeof(super));
371 qemu_iovec_add(&qiov, zeroes, s->sectorsize - sizeof(super));
372
373 lr->log_ret =
374 bdrv_co_pwritev(s->log_file, 0, s->sectorsize, &qiov, 0);
375 if (lr->log_ret == 0) {
376 lr->log_ret = bdrv_co_flush(s->log_file->bs);
377 }
378 qemu_iovec_destroy(&qiov);
379 g_free(zeroes);
380 }
381}
382
383static void coroutine_fn blk_log_writes_co_do_file(BlkLogWritesFileReq *fr)
384{
385 fr->file_ret = fr->func(fr);
386}
387
388static int coroutine_fn
389blk_log_writes_co_log(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
390 QEMUIOVector *qiov, int flags,
391 int (*file_func)(BlkLogWritesFileReq *r),
392 uint64_t entry_flags, bool is_zero_write)
393{
394 QEMUIOVector log_qiov;
395 size_t niov = qiov ? qiov->niov : 0;
396 BDRVBlkLogWritesState *s = bs->opaque;
397 BlkLogWritesFileReq fr = {
398 .bs = bs,
399 .offset = offset,
400 .bytes = bytes,
401 .file_flags = flags,
402 .qiov = qiov,
403 .func = file_func,
404 };
405 BlkLogWritesLogReq lr = {
406 .bs = bs,
407 .qiov = &log_qiov,
408 .entry = {
409 .sector = cpu_to_le64(offset >> s->sectorbits),
410 .nr_sectors = cpu_to_le64(bytes >> s->sectorbits),
411 .flags = cpu_to_le64(entry_flags),
412 .data_len = 0,
413 },
414 .zero_size = is_zero_write ? bytes : 0,
415 };
416 void *zeroes = g_malloc0(s->sectorsize - sizeof(lr.entry));
417
418 assert((1 << s->sectorbits) == s->sectorsize);
419 assert(bs->bl.request_alignment == s->sectorsize);
420 assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
421 assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
422
423 qemu_iovec_init(&log_qiov, niov + 2);
424 qemu_iovec_add(&log_qiov, &lr.entry, sizeof(lr.entry));
425 qemu_iovec_add(&log_qiov, zeroes, s->sectorsize - sizeof(lr.entry));
426 if (qiov) {
427 qemu_iovec_concat(&log_qiov, qiov, 0, qiov->size);
428 }
429
430 blk_log_writes_co_do_file(&fr);
431 blk_log_writes_co_do_log(&lr);
432
433 qemu_iovec_destroy(&log_qiov);
434 g_free(zeroes);
435
436 if (lr.log_ret < 0) {
437 return lr.log_ret;
438 }
439
440 return fr.file_ret;
441}
442
443static int coroutine_fn
444blk_log_writes_co_do_file_pwritev(BlkLogWritesFileReq *fr)
445{
446 return bdrv_co_pwritev(fr->bs->file, fr->offset, fr->bytes,
447 fr->qiov, fr->file_flags);
448}
449
450static int coroutine_fn
451blk_log_writes_co_do_file_pwrite_zeroes(BlkLogWritesFileReq *fr)
452{
453 return bdrv_co_pwrite_zeroes(fr->bs->file, fr->offset, fr->bytes,
454 fr->file_flags);
455}
456
457static int coroutine_fn blk_log_writes_co_do_file_flush(BlkLogWritesFileReq *fr)
458{
459 return bdrv_co_flush(fr->bs->file->bs);
460}
461
462static int coroutine_fn
463blk_log_writes_co_do_file_pdiscard(BlkLogWritesFileReq *fr)
464{
465 return bdrv_co_pdiscard(fr->bs->file, fr->offset, fr->bytes);
466}
467
468static int coroutine_fn
469blk_log_writes_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
470 QEMUIOVector *qiov, int flags)
471{
472 return blk_log_writes_co_log(bs, offset, bytes, qiov, flags,
473 blk_log_writes_co_do_file_pwritev, 0, false);
474}
475
476static int coroutine_fn
477blk_log_writes_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset, int bytes,
478 BdrvRequestFlags flags)
479{
480 return blk_log_writes_co_log(bs, offset, bytes, NULL, flags,
481 blk_log_writes_co_do_file_pwrite_zeroes, 0,
482 true);
483}
484
485static int coroutine_fn blk_log_writes_co_flush_to_disk(BlockDriverState *bs)
486{
487 return blk_log_writes_co_log(bs, 0, 0, NULL, 0,
488 blk_log_writes_co_do_file_flush,
489 LOG_FLUSH_FLAG, false);
490}
491
492static int coroutine_fn
493blk_log_writes_co_pdiscard(BlockDriverState *bs, int64_t offset, int count)
494{
495 return blk_log_writes_co_log(bs, offset, count, NULL, 0,
496 blk_log_writes_co_do_file_pdiscard,
497 LOG_DISCARD_FLAG, false);
498}
499
500static const char *const blk_log_writes_strong_runtime_opts[] = {
501 "log-append",
502 "log-sector-size",
503
504 NULL
505};
506
507static BlockDriver bdrv_blk_log_writes = {
508 .format_name = "blklogwrites",
509 .instance_size = sizeof(BDRVBlkLogWritesState),
510
511 .bdrv_open = blk_log_writes_open,
512 .bdrv_close = blk_log_writes_close,
513 .bdrv_getlength = blk_log_writes_getlength,
514 .bdrv_child_perm = blk_log_writes_child_perm,
515 .bdrv_refresh_limits = blk_log_writes_refresh_limits,
516
517 .bdrv_co_preadv = blk_log_writes_co_preadv,
518 .bdrv_co_pwritev = blk_log_writes_co_pwritev,
519 .bdrv_co_pwrite_zeroes = blk_log_writes_co_pwrite_zeroes,
520 .bdrv_co_flush_to_disk = blk_log_writes_co_flush_to_disk,
521 .bdrv_co_pdiscard = blk_log_writes_co_pdiscard,
522 .bdrv_co_block_status = bdrv_co_block_status_from_file,
523
524 .is_filter = true,
525 .strong_runtime_opts = blk_log_writes_strong_runtime_opts,
526};
527
528static void bdrv_blk_log_writes_init(void)
529{
530 bdrv_register(&bdrv_blk_log_writes);
531}
532
533block_init(bdrv_blk_log_writes_init);
534
