1 /*
2 * Write logging blk driver based on blkverify and blkdebug.
3 *
4 * Copyright (c) 2017 Tuomas Tynkkynen <tuomas@tuxera.com>
5 * Copyright (c) 2018 Aapo Vienamo <aapo@tuxera.com>
6 * Copyright (c) 2018 Ari Sundholm <ari@tuxera.com>
7 *
8 * This work is licensed under the terms of the GNU GPL, version 2 or later.
9 * See the COPYING file in the top-level directory.
10 */
11
12 #include "qemu/osdep.h"
13 #include "qapi/error.h"
14 #include "qemu/sockets.h" /* for EINPROGRESS on Windows */
15 #include "block/block-io.h"
16 #include "block/block_int.h"
17 #include "qapi/qmp/qdict.h"
18 #include "qapi/qmp/qstring.h"
19 #include "qemu/cutils.h"
20 #include "qemu/module.h"
21 #include "qemu/option.h"
22
23 /* Disk format stuff - taken from Linux drivers/md/dm-log-writes.c */
24
25 #define LOG_FLUSH_FLAG (1 << 0)
26 #define LOG_FUA_FLAG (1 << 1)
27 #define LOG_DISCARD_FLAG (1 << 2)
28 #define LOG_MARK_FLAG (1 << 3)
29 #define LOG_FLAG_MASK (LOG_FLUSH_FLAG \
30 | LOG_FUA_FLAG \
31 | LOG_DISCARD_FLAG \
32 | LOG_MARK_FLAG)
33
34 #define WRITE_LOG_VERSION 1ULL
35 #define WRITE_LOG_MAGIC 0x6a736677736872ULL
36
37 /* All fields are little-endian. */
38 struct log_write_super {
39 uint64_t magic;
40 uint64_t version;
41 uint64_t nr_entries;
42 uint32_t sectorsize;
43 } QEMU_PACKED;
44
45 struct log_write_entry {
46 uint64_t sector;
47 uint64_t nr_sectors;
48 uint64_t flags;
49 uint64_t data_len;
50 } QEMU_PACKED;
51
52 /* End of disk format structures. */
53
54 typedef struct {
55 BdrvChild *log_file;
56 uint32_t sectorsize;
57 uint32_t sectorbits;
58 uint64_t cur_log_sector;
59 uint64_t nr_entries;
60 uint64_t update_interval;
61 } BDRVBlkLogWritesState;
62
63 static QemuOptsList runtime_opts = {
64 .name = "blklogwrites",
65 .head = QTAILQ_HEAD_INITIALIZER(runtime_opts.head),
66 .desc = {
67 {
68 .name = "log-append",
69 .type = QEMU_OPT_BOOL,
70 .help = "Append to an existing log",
71 },
72 {
73 .name = "log-sector-size",
74 .type = QEMU_OPT_SIZE,
75 .help = "Log sector size",
76 },
77 {
78 .name = "log-super-update-interval",
79 .type = QEMU_OPT_NUMBER,
80 .help = "Log superblock update interval (# of write requests)",
81 },
82 { /* end of list */ }
83 },
84 };
85
blk_log_writes_log2(uint32_t value)86 static inline uint32_t blk_log_writes_log2(uint32_t value)
87 {
88 assert(value > 0);
89 return 31 - clz32(value);
90 }
91
blk_log_writes_sector_size_valid(uint32_t sector_size)92 static inline bool blk_log_writes_sector_size_valid(uint32_t sector_size)
93 {
94 return is_power_of_2(sector_size) &&
95 sector_size >= sizeof(struct log_write_super) &&
96 sector_size >= sizeof(struct log_write_entry) &&
97 sector_size < (1ull << 24);
98 }
99
blk_log_writes_find_cur_log_sector(BdrvChild * log,uint32_t sector_size,uint64_t nr_entries,Error ** errp)100 static uint64_t blk_log_writes_find_cur_log_sector(BdrvChild *log,
101 uint32_t sector_size,
102 uint64_t nr_entries,
103 Error **errp)
104 {
105 uint64_t cur_sector = 1;
106 uint64_t cur_idx = 0;
107 uint32_t sector_bits = blk_log_writes_log2(sector_size);
108 struct log_write_entry cur_entry;
109
110 while (cur_idx < nr_entries) {
111 int read_ret = bdrv_pread(log, cur_sector << sector_bits,
112 sizeof(cur_entry), &cur_entry, 0);
113 if (read_ret < 0) {
114 error_setg_errno(errp, -read_ret,
115 "Failed to read log entry %"PRIu64, cur_idx);
116 return (uint64_t)-1ull;
117 }
118
119 if (cur_entry.flags & ~cpu_to_le64(LOG_FLAG_MASK)) {
120 error_setg(errp, "Invalid flags 0x%"PRIx64" in log entry %"PRIu64,
121 le64_to_cpu(cur_entry.flags), cur_idx);
122 return (uint64_t)-1ull;
123 }
124
125 /* Account for the sector of the entry itself */
126 ++cur_sector;
127
128 /*
129 * Account for the data of the write.
130 * For discards, this data is not present.
131 */
132 if (!(cur_entry.flags & cpu_to_le64(LOG_DISCARD_FLAG))) {
133 cur_sector += le64_to_cpu(cur_entry.nr_sectors);
134 }
135
136 ++cur_idx;
137 }
138
139 return cur_sector;
140 }
141
blk_log_writes_open(BlockDriverState * bs,QDict * options,int flags,Error ** errp)142 static int blk_log_writes_open(BlockDriverState *bs, QDict *options, int flags,
143 Error **errp)
144 {
145 BDRVBlkLogWritesState *s = bs->opaque;
146 QemuOpts *opts;
147 Error *local_err = NULL;
148 int ret;
149 uint64_t log_sector_size;
150 bool log_append;
151
152 opts = qemu_opts_create(&runtime_opts, NULL, 0, &error_abort);
153 if (!qemu_opts_absorb_qdict(opts, options, errp)) {
154 ret = -EINVAL;
155 goto fail;
156 }
157
158 /* Open the file */
159 ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
160 if (ret < 0) {
161 goto fail;
162 }
163
164 /* Open the log file */
165 s->log_file = bdrv_open_child(NULL, options, "log", bs, &child_of_bds,
166 BDRV_CHILD_METADATA, false, errp);
167 if (!s->log_file) {
168 ret = -EINVAL;
169 goto fail;
170 }
171
172 log_append = qemu_opt_get_bool(opts, "log-append", false);
173
174 if (log_append) {
175 struct log_write_super log_sb = { 0, 0, 0, 0 };
176
177 if (qemu_opt_find(opts, "log-sector-size")) {
178 ret = -EINVAL;
179 error_setg(errp, "log-append and log-sector-size are mutually "
180 "exclusive");
181 goto fail_log;
182 }
183
184 /* Read log superblock or fake one for an empty log */
185 if (!bdrv_getlength(s->log_file->bs)) {
186 log_sb.magic = cpu_to_le64(WRITE_LOG_MAGIC);
187 log_sb.version = cpu_to_le64(WRITE_LOG_VERSION);
188 log_sb.nr_entries = cpu_to_le64(0);
189 log_sb.sectorsize = cpu_to_le32(BDRV_SECTOR_SIZE);
190 } else {
191 ret = bdrv_pread(s->log_file, 0, sizeof(log_sb), &log_sb, 0);
192 if (ret < 0) {
193 error_setg_errno(errp, -ret, "Could not read log superblock");
194 goto fail_log;
195 }
196 }
197
198 if (log_sb.magic != cpu_to_le64(WRITE_LOG_MAGIC)) {
199 ret = -EINVAL;
200 error_setg(errp, "Invalid log superblock magic");
201 goto fail_log;
202 }
203
204 if (log_sb.version != cpu_to_le64(WRITE_LOG_VERSION)) {
205 ret = -EINVAL;
206 error_setg(errp, "Unsupported log version %"PRIu64,
207 le64_to_cpu(log_sb.version));
208 goto fail_log;
209 }
210
211 log_sector_size = le32_to_cpu(log_sb.sectorsize);
212 s->cur_log_sector = 1;
213 s->nr_entries = 0;
214
215 if (blk_log_writes_sector_size_valid(log_sector_size)) {
216 s->cur_log_sector =
217 blk_log_writes_find_cur_log_sector(s->log_file, log_sector_size,
218 le64_to_cpu(log_sb.nr_entries), &local_err);
219 if (local_err) {
220 ret = -EINVAL;
221 error_propagate(errp, local_err);
222 goto fail_log;
223 }
224
225 s->nr_entries = le64_to_cpu(log_sb.nr_entries);
226 }
227 } else {
228 log_sector_size = qemu_opt_get_size(opts, "log-sector-size",
229 BDRV_SECTOR_SIZE);
230 s->cur_log_sector = 1;
231 s->nr_entries = 0;
232 }
233
234 if (!blk_log_writes_sector_size_valid(log_sector_size)) {
235 ret = -EINVAL;
236 error_setg(errp, "Invalid log sector size %"PRIu64, log_sector_size);
237 goto fail_log;
238 }
239
240 s->sectorsize = log_sector_size;
241 s->sectorbits = blk_log_writes_log2(log_sector_size);
242 s->update_interval = qemu_opt_get_number(opts, "log-super-update-interval",
243 4096);
244 if (!s->update_interval) {
245 ret = -EINVAL;
246 error_setg(errp, "Invalid log superblock update interval %"PRIu64,
247 s->update_interval);
248 goto fail_log;
249 }
250
251 ret = 0;
252 fail_log:
253 if (ret < 0) {
254 bdrv_graph_wrlock(NULL);
255 bdrv_unref_child(bs, s->log_file);
256 bdrv_graph_wrunlock(NULL);
257 s->log_file = NULL;
258 }
259 fail:
260 qemu_opts_del(opts);
261 return ret;
262 }
263
blk_log_writes_close(BlockDriverState * bs)264 static void blk_log_writes_close(BlockDriverState *bs)
265 {
266 BDRVBlkLogWritesState *s = bs->opaque;
267
268 bdrv_graph_wrlock(NULL);
269 bdrv_unref_child(bs, s->log_file);
270 s->log_file = NULL;
271 bdrv_graph_wrunlock(NULL);
272 }
273
274 static int64_t coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_getlength(BlockDriverState * bs)275 blk_log_writes_co_getlength(BlockDriverState *bs)
276 {
277 return bdrv_co_getlength(bs->file->bs);
278 }
279
blk_log_writes_child_perm(BlockDriverState * bs,BdrvChild * c,BdrvChildRole role,BlockReopenQueue * ro_q,uint64_t perm,uint64_t shrd,uint64_t * nperm,uint64_t * nshrd)280 static void blk_log_writes_child_perm(BlockDriverState *bs, BdrvChild *c,
281 BdrvChildRole role,
282 BlockReopenQueue *ro_q,
283 uint64_t perm, uint64_t shrd,
284 uint64_t *nperm, uint64_t *nshrd)
285 {
286 if (!c) {
287 *nperm = perm & DEFAULT_PERM_PASSTHROUGH;
288 *nshrd = (shrd & DEFAULT_PERM_PASSTHROUGH) | DEFAULT_PERM_UNCHANGED;
289 return;
290 }
291
292 bdrv_default_perms(bs, c, role, ro_q, perm, shrd,
293 nperm, nshrd);
294 }
295
blk_log_writes_refresh_limits(BlockDriverState * bs,Error ** errp)296 static void blk_log_writes_refresh_limits(BlockDriverState *bs, Error **errp)
297 {
298 BDRVBlkLogWritesState *s = bs->opaque;
299 bs->bl.request_alignment = s->sectorsize;
300 }
301
302 static int coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_preadv(BlockDriverState * bs,int64_t offset,int64_t bytes,QEMUIOVector * qiov,BdrvRequestFlags flags)303 blk_log_writes_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
304 QEMUIOVector *qiov, BdrvRequestFlags flags)
305 {
306 return bdrv_co_preadv(bs->file, offset, bytes, qiov, flags);
307 }
308
309 typedef struct BlkLogWritesFileReq {
310 BlockDriverState *bs;
311 uint64_t offset;
312 uint64_t bytes;
313 int file_flags;
314 QEMUIOVector *qiov;
315 int GRAPH_RDLOCK_PTR (*func)(struct BlkLogWritesFileReq *r);
316 int file_ret;
317 } BlkLogWritesFileReq;
318
319 typedef struct {
320 BlockDriverState *bs;
321 QEMUIOVector *qiov;
322 struct log_write_entry entry;
323 uint64_t zero_size;
324 int log_ret;
325 } BlkLogWritesLogReq;
326
327 static void coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_do_log(BlkLogWritesLogReq * lr)328 blk_log_writes_co_do_log(BlkLogWritesLogReq *lr)
329 {
330 BDRVBlkLogWritesState *s = lr->bs->opaque;
331
332 /*
333 * Determine the offsets and sizes of different parts of the entry, and
334 * update the state of the driver.
335 *
336 * This needs to be done in one go, before any actual I/O is done, as the
337 * log entry may have to be written in two parts, and the state of the
338 * driver may be modified by other driver operations while waiting for the
339 * I/O to complete.
340 */
341 const uint64_t entry_start_sector = s->cur_log_sector;
342 const uint64_t entry_offset = entry_start_sector << s->sectorbits;
343 const uint64_t qiov_aligned_size = ROUND_UP(lr->qiov->size, s->sectorsize);
344 const uint64_t entry_aligned_size = qiov_aligned_size +
345 ROUND_UP(lr->zero_size, s->sectorsize);
346 const uint64_t entry_nr_sectors = entry_aligned_size >> s->sectorbits;
347
348 s->nr_entries++;
349 s->cur_log_sector += entry_nr_sectors;
350
351 /*
352 * Write the log entry. Note that if this is a "write zeroes" operation,
353 * only the entry header is written here, with the zeroing being done
354 * separately below.
355 */
356 lr->log_ret = bdrv_co_pwritev(s->log_file, entry_offset, lr->qiov->size,
357 lr->qiov, 0);
358
359 /* Logging for the "write zeroes" operation */
360 if (lr->log_ret == 0 && lr->zero_size) {
361 const uint64_t zeroes_offset = entry_offset + qiov_aligned_size;
362
363 lr->log_ret = bdrv_co_pwrite_zeroes(s->log_file, zeroes_offset,
364 lr->zero_size, 0);
365 }
366
367 /* Update super block on flush or every update interval */
368 if (lr->log_ret == 0 && ((lr->entry.flags & LOG_FLUSH_FLAG)
369 || (s->nr_entries % s->update_interval == 0)))
370 {
371 struct log_write_super super = {
372 .magic = cpu_to_le64(WRITE_LOG_MAGIC),
373 .version = cpu_to_le64(WRITE_LOG_VERSION),
374 .nr_entries = cpu_to_le64(s->nr_entries),
375 .sectorsize = cpu_to_le32(s->sectorsize),
376 };
377 void *zeroes = g_malloc0(s->sectorsize - sizeof(super));
378 QEMUIOVector qiov;
379
380 qemu_iovec_init(&qiov, 2);
381 qemu_iovec_add(&qiov, &super, sizeof(super));
382 qemu_iovec_add(&qiov, zeroes, s->sectorsize - sizeof(super));
383
384 lr->log_ret =
385 bdrv_co_pwritev(s->log_file, 0, s->sectorsize, &qiov, 0);
386 if (lr->log_ret == 0) {
387 lr->log_ret = bdrv_co_flush(s->log_file->bs);
388 }
389 qemu_iovec_destroy(&qiov);
390 g_free(zeroes);
391 }
392 }
393
394 static void coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_do_file(BlkLogWritesFileReq * fr)395 blk_log_writes_co_do_file(BlkLogWritesFileReq *fr)
396 {
397 fr->file_ret = fr->func(fr);
398 }
399
400 static int coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_log(BlockDriverState * bs,uint64_t offset,uint64_t bytes,QEMUIOVector * qiov,int flags,int (* file_func)(BlkLogWritesFileReq * r),uint64_t entry_flags,bool is_zero_write)401 blk_log_writes_co_log(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
402 QEMUIOVector *qiov, int flags,
403 int /*GRAPH_RDLOCK*/ (*file_func)(BlkLogWritesFileReq *r),
404 uint64_t entry_flags, bool is_zero_write)
405 {
406 QEMUIOVector log_qiov;
407 size_t niov = qiov ? qiov->niov : 0;
408 BDRVBlkLogWritesState *s = bs->opaque;
409 BlkLogWritesFileReq fr = {
410 .bs = bs,
411 .offset = offset,
412 .bytes = bytes,
413 .file_flags = flags,
414 .qiov = qiov,
415 .func = file_func,
416 };
417 BlkLogWritesLogReq lr = {
418 .bs = bs,
419 .qiov = &log_qiov,
420 .entry = {
421 .sector = cpu_to_le64(offset >> s->sectorbits),
422 .nr_sectors = cpu_to_le64(bytes >> s->sectorbits),
423 .flags = cpu_to_le64(entry_flags),
424 .data_len = 0,
425 },
426 .zero_size = is_zero_write ? bytes : 0,
427 };
428 void *zeroes = g_malloc0(s->sectorsize - sizeof(lr.entry));
429
430 assert((1 << s->sectorbits) == s->sectorsize);
431 assert(bs->bl.request_alignment == s->sectorsize);
432 assert(QEMU_IS_ALIGNED(offset, bs->bl.request_alignment));
433 assert(QEMU_IS_ALIGNED(bytes, bs->bl.request_alignment));
434
435 qemu_iovec_init(&log_qiov, niov + 2);
436 qemu_iovec_add(&log_qiov, &lr.entry, sizeof(lr.entry));
437 qemu_iovec_add(&log_qiov, zeroes, s->sectorsize - sizeof(lr.entry));
438 if (qiov) {
439 qemu_iovec_concat(&log_qiov, qiov, 0, qiov->size);
440 }
441
442 blk_log_writes_co_do_file(&fr);
443 blk_log_writes_co_do_log(&lr);
444
445 qemu_iovec_destroy(&log_qiov);
446 g_free(zeroes);
447
448 if (lr.log_ret < 0) {
449 return lr.log_ret;
450 }
451
452 return fr.file_ret;
453 }
454
455 static int coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_do_file_pwritev(BlkLogWritesFileReq * fr)456 blk_log_writes_co_do_file_pwritev(BlkLogWritesFileReq *fr)
457 {
458 return bdrv_co_pwritev(fr->bs->file, fr->offset, fr->bytes,
459 fr->qiov, fr->file_flags);
460 }
461
462 static int coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_do_file_pwrite_zeroes(BlkLogWritesFileReq * fr)463 blk_log_writes_co_do_file_pwrite_zeroes(BlkLogWritesFileReq *fr)
464 {
465 return bdrv_co_pwrite_zeroes(fr->bs->file, fr->offset, fr->bytes,
466 fr->file_flags);
467 }
468
469 static int coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_do_file_flush(BlkLogWritesFileReq * fr)470 blk_log_writes_co_do_file_flush(BlkLogWritesFileReq *fr)
471 {
472 return bdrv_co_flush(fr->bs->file->bs);
473 }
474
475 static int coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_do_file_pdiscard(BlkLogWritesFileReq * fr)476 blk_log_writes_co_do_file_pdiscard(BlkLogWritesFileReq *fr)
477 {
478 return bdrv_co_pdiscard(fr->bs->file, fr->offset, fr->bytes);
479 }
480
481 static int coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_pwritev(BlockDriverState * bs,int64_t offset,int64_t bytes,QEMUIOVector * qiov,BdrvRequestFlags flags)482 blk_log_writes_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
483 QEMUIOVector *qiov, BdrvRequestFlags flags)
484 {
485 return blk_log_writes_co_log(bs, offset, bytes, qiov, flags,
486 blk_log_writes_co_do_file_pwritev, 0, false);
487 }
488
489 static int coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_pwrite_zeroes(BlockDriverState * bs,int64_t offset,int64_t bytes,BdrvRequestFlags flags)490 blk_log_writes_co_pwrite_zeroes(BlockDriverState *bs, int64_t offset,
491 int64_t bytes, BdrvRequestFlags flags)
492 {
493 return blk_log_writes_co_log(bs, offset, bytes, NULL, flags,
494 blk_log_writes_co_do_file_pwrite_zeroes, 0,
495 true);
496 }
497
498 static int coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_flush_to_disk(BlockDriverState * bs)499 blk_log_writes_co_flush_to_disk(BlockDriverState *bs)
500 {
501 return blk_log_writes_co_log(bs, 0, 0, NULL, 0,
502 blk_log_writes_co_do_file_flush,
503 LOG_FLUSH_FLAG, false);
504 }
505
506 static int coroutine_fn GRAPH_RDLOCK
blk_log_writes_co_pdiscard(BlockDriverState * bs,int64_t offset,int64_t bytes)507 blk_log_writes_co_pdiscard(BlockDriverState *bs, int64_t offset, int64_t bytes)
508 {
509 return blk_log_writes_co_log(bs, offset, bytes, NULL, 0,
510 blk_log_writes_co_do_file_pdiscard,
511 LOG_DISCARD_FLAG, false);
512 }
513
514 static const char *const blk_log_writes_strong_runtime_opts[] = {
515 "log-append",
516 "log-sector-size",
517
518 NULL
519 };
520
521 static BlockDriver bdrv_blk_log_writes = {
522 .format_name = "blklogwrites",
523 .instance_size = sizeof(BDRVBlkLogWritesState),
524
525 .bdrv_open = blk_log_writes_open,
526 .bdrv_close = blk_log_writes_close,
527 .bdrv_co_getlength = blk_log_writes_co_getlength,
528 .bdrv_child_perm = blk_log_writes_child_perm,
529 .bdrv_refresh_limits = blk_log_writes_refresh_limits,
530
531 .bdrv_co_preadv = blk_log_writes_co_preadv,
532 .bdrv_co_pwritev = blk_log_writes_co_pwritev,
533 .bdrv_co_pwrite_zeroes = blk_log_writes_co_pwrite_zeroes,
534 .bdrv_co_flush_to_disk = blk_log_writes_co_flush_to_disk,
535 .bdrv_co_pdiscard = blk_log_writes_co_pdiscard,
536
537 .is_filter = true,
538 .strong_runtime_opts = blk_log_writes_strong_runtime_opts,
539 };
540
bdrv_blk_log_writes_init(void)541 static void bdrv_blk_log_writes_init(void)
542 {
543 bdrv_register(&bdrv_blk_log_writes);
544 }
545
546 block_init(bdrv_blk_log_writes_init);
547